Progress in Biomass Combustion Systems for Ultra-Low Emissions

Abstract

Biomass combustion, as a key technology for achieving a low-carbon transformation of the energy system, faces multiple challenges in its efficient and clean utilization, including the high heterogeneity of fuels, the complex multi-scale coupling of the combustion process, and the attainment of ultra-low emissions. Traditional research methods have significant disconnections between microscopic mechanism understanding, macroscopic performance prediction of reactors, and end-of-pipe pollution control, which restricts the improvement of system performance. This review presents recent advances in advanced numerical simulation, pollutant control strategies, and bioenergy with carbon capture and storage (BECCS) pathways targeting ultra-low emissions in biomass combustion. This work synthesizes progress across three interconnected domains. First, methodologies are examined for integrating detailed chemical kinetics, particle-scale models, and reactor-scale simulations to develop high-fidelity predictive tools. Second, low-nitrogen combustion and synergistic pollutant control strategies for primary furnace types (e.g., grate, fluidized bed) are evaluated, alongside process optimization from fuel pretreatment to flue gas purification. Third, the potential for integrated design of biomass energy systems with carbon capture is assessed, emphasizing that system efficiency hinges on holistic “fuel-combustion-capture” chain optimization rather than isolated unit improvements. Future research directions are highlighted, including the development of physics-informed AI modeling paradigms, deeper co-design of multiple processes, and the establishment of robust life-cycle assessment frameworks. This review aims to provide a structured reference to inform both fundamental research and the practical development of next-generation clean biomass combustion technologies.

1. Introduction

In the process of the global energy system’s transition to a low-carbon model, biomass energy has emerged as a focus due to its renewability and carbon neutrality potential [1,2]. Efficient and clean utilization of biomass combustion not only converts its potential into an actual energy supply but also combines with carbon capture and storage technologies to achieve negative emission goals [3]. However, the high heterogeneity of biomass fuels and the complexity of their combustion processes pose challenges to traditional combustion technologies’ pursuit of high efficiency and ultra-low emissions. Achieving ultra-low emissions in biomass combustion faces several core bottlenecks. These include the generation and control of nitrogen oxides. Alkali metals also cause issues of slagging and corrosion. Furthermore, a contradiction exists between maintaining combustion stability and ensuring broad fuel adaptability [4]. To overcome these problems, research methods have shifted from relying on experience to relying on mechanistic understanding and numerical simulation-driven refined analysis [5]. At the fundamental research level, a multi-scale simulation tool system helps us understand the biomass combustion process. This system integrates methods from chemical reaction kinetics. It also incorporates principles of single-particle heat and mass transfer. Furthermore, it applies computational fluid dynamics at the reactor level. Together, these approaches cover the process across different scales [6,7]. At the application level, low-nitrogen combustion optimization, pollutant co-deletion, and ash and slag management strategies for mainstream furnace types such as grate furnaces and fluidized beds are constantly being improved and enhanced [8,9]. Meanwhile, the integration of biomass combustion with carbon capture technology, known as bioenergy and carbon capture and storage (BECCS), is regarded as an important direction for achieving negative carbon emissions [10].

Current research has advanced in several subfields. However, the field remains fragmented. This fragmentation is evident in a key challenge: insights from microscale studies—such as the formation of pollutant precursors [11] and alkali metal migration mechanisms [12]—cannot effectively guide the design of macro-scale equipment or the optimization of its operation. Separately, high-fidelity fundamental models exist, but their high computational cost prevents their direct application in engineering design [13]. Additionally, research on combustion process optimization [14], end-of-pipe pollutant control [15], ash disposal [16], and carbon capture units [17] often occurs in isolation, resulting in a lack of systematic understanding of their cross-influence and collaborative potential. Biomass fuel combustion systems exhibit strong nonlinearity and uncertainty. Traditional physical models encounter a challenge: balancing computational efficiency with prediction accuracy. Meanwhile, purely data-driven methods suffer from inherent limitations. Their explanatory power and generalizability beyond the training data are constrained [18].

Although significant progress has been made in existing research, there remains a clear gap in current reviews regarding the integration of the “mechanism–process–system” chain. For instance, the review by Ansari et al. [19] provides limited discussion on quantifying microscopic oxygen carrier reaction mechanisms into macroscopic predictive models, while the work of Yang et al. [6] overly focuses on particle-scale deformation, failing to effectively extend to pollutant control and BECCS system optimization. To address these gaps, this review aims to establish a systematic cognitive framework to integrate the currently fragmented knowledge base. Specifically, this paper will (1) strive to build a cross-scale simulation bridge connecting microscopic reaction kinetics with macroscopic reactor performance; (2) systematically explore synergistic optimization pathways for combustion process regulation, pollutant suppression, and ash management to overcome the limitations of isolated studies on individual aspects; (3) comprehensively evaluate BECCS technological pathways and clarify that their efficacy fundamentally depends on the holistic co-design of the “fuel–combustion–capture” chain. Through critical integration of existing research, this paper aims to provide a clear roadmap and decision-making reference for planning the next generation of efficient, clean, and carbon-negative biomass combustion technologies.

2. Multi-Scale Characteristics of Biomass Fuels and the Basis of Combustion Chemical Transformation

To establish a theoretical framework that describes the entire process of biomass combustion, multiple sub-models focusing on different physical and chemical processes need to be integrated. The complex combustion behavior is shown in Figure 1. These models are not separate “parts”, but rather tools that depict the same complex system from different perspectives, jointly serving to provide a complete description of the combustion phenomenon. Typically, a comprehensive theoretical modeling system will cover the following key modules: (1) pyrolysis model [20], which describes the process of biomass fuel decomposition and the generation of combustible gases under heating conditions; (2) gas-phase combustion model [7], depicting the chemical reaction kinetics and flame structure of the reaction between pyrolysis products and oxygen in the gas phase; (3) heat and mass transfer model [21], used to calculate the transfer of heat and mass between solid, liquid, and gas phases during the combustion process; (4) flow and radiation model [22], used to describe the influence of fluid motion in the combustion environment on the reaction rate and mixing process.

2.1. Chemical Components, Structures and Reactivity at the Molecular Level

At the molecular scale, biomass is mainly composed of three natural high-molecular polymers: cellulose, hemicellulose, and lignin. Their chemical structures, bonding methods, and relative contents directly determine the initial pyrolysis behavior and the distribution of volatile products. Numerous studies have emphasized the decisive role of the chemical composition of the raw material. Optical diagnostic studies reveal that biomass undergoes rapid pyrolysis. This release stage is primarily linked to cellulose decomposition. In contrast, the subsequent combustion stage of char is influenced more strongly by the residues from lignin decomposition. Furthermore, the relative proportions of cellulose, hemicellulose, and lignin are critical. These proportions directly impact the ignition characteristics and combustion stability of biomass [23]. This correlation implies that biomass from different sources (such as woody, herbaceous, or agricultural waste) will exhibit distinct combustion kinetics due to differences in their molecular composition.

Furthermore, at the molecular level, interactions are also crucial in specific advanced conversion technologies. For example, in chemical looping combustion (CLC) or chemical looping gasification (CLG), the reductive gases derived from biomass (such as CO and H₂) need to undergo redox reactions with the surface of oxygen carriers (such as Fe₂O₃ and CuO), as shown in Figure 2. Density functional theory (DFT) studies have indicated that an external electric field can significantly alter the oxidation energy barrier of CO on the CuO surface, suggesting the controllability of molecule–surface interactions [19]. The catalytic gasification mechanism of ethanol on the Fe₂O₃ surface is also reliant on surface defect states. This dependence indirectly reflects the complexity of interfacial chemistry. This chemistry occurs between the oxygenated compounds from biomass pyrolysis and the oxygen carriers [19]. Therefore, understanding and regulating the reaction pathways at the molecular scale is key to optimizing product selectivity and suppressing side reactions.

2.2. Morphology, Pores and Heat and Mass Transfer Characteristics at the Particle Scale

When the scale is magnified to a single particle, its geometric shape, size, internal pore network, and thermophysical properties become the core factors controlling the rates of internal heat transfer, mass transfer, and chemical reactions. Real biomass particles typically have irregular geometric shapes and complex multi-level pore structures, which pose challenges to traditional simplified models of spherical or cylindrical shapes.

To address this issue, researchers have developed various high-fidelity particle models. Thomas Sheehan et al. [25] proposed an automated, geometrically accurate pore network model. This model can directly extract pore topology from X-ray micro-CT data. It is used to precisely simulate the diffusion of reactive gases, like oxygen, within the pores. This enables a more accurate prediction of the reaction rate and internal temperature distribution of the particle. Yang et al. [6] further verified this point. They compared a structural model based on a biomass particle model (BPM) with micro-CT measurements. They also simulated the evolution of particle temperature and carbon content in a fluidized bed gasifier. Their work demonstrates the importance of considering real particle structure. It is crucial for capturing local non-uniform reactions.

The influence of particle size is also significant. Chen et al. [26] found that larger particles can cause a temperature rise lag in the central region. This lag occurs due to reduced efficiency in internal heat conduction. Consequently, the volatiles release time is prolonged. This prolonged release may increase the formation of incomplete combustion products, such as CO. Conversely, smaller particles can reach thermal equilibrium more quickly, promoting combustion intensity and reducing CO formation. Additionally, the bulk density and porosity of the particles are also key parameters. In a CFD model of a packed bed for particle combustion, researchers adjusted porosity parameters within the compaction algorithm. The adjustment set a maximum porosity of 0.6 for the carbon layer and 0.79 for the ash layer. This was done to match experimental observations. Their work highlights a critical point: the local physical structure of the particle bed plays a regulatory role. It governs the overall flow and reactions within the system [27].

2.3. Porous Medium Structure of the Fuel Bed Scale and Its Macroscopic Transport Behavior

At the fuel bed scale, the random stacking of biomass particles constitutes a complex porous medium. Its macroscopic properties include volume fraction distribution, permeability, effective thermal conductivity, and bed height. These properties jointly dominate the gas phase flow and heat transfer. They also govern the propagation behavior of the combustion or gasification front [27]. For fixed-bed or moving-bed systems, a typical approach is to treat them as a porous continuous medium and simulate them using the filling-bed CFD method. This method treats parameters like porosity and permeability as spatial variables. This approach enables the efficient simulation of three-dimensional combustion for entire bed spruce particles. The prediction results from this simulation show good agreement with experimental data. The agreement is specifically in terms of gas temperature, CO, and NO_x concentrations [27].

In fluidized bed systems, the simulation of gas–solid two-phase flow is more complex. Dharmendra et al. [28] conducted research on bubbling fluidized bed gasification systems. In this study, they used Computational Fluid Dynamics (CFD) to analyze the system. Their analysis focused on the distribution and movement patterns of sand particles. It also examined various biomass particles, such as peanut shells, rice husks, wood chips, and discarded masks. This analysis was performed under different equivalence ratios (ER). The study shows that fuel particles of different densities and sizes exhibit unique volume fraction distributions and residence times at specific fluidization velocities, which directly determine the burnout rate and system thermal efficiency [28]. In fluidized bed gasifier simulations, multi-scale coupling strategies are critically important. One common approach combines mesoscale methods. These methods include DEM (Discrete Element Method) or MP-PIC (Multiphase Particle-In-Cell). They are used to describe particle dynamics. This is coupled with macroscale models. The macroscale models are Euler–Euler or Lagrangian CFD frameworks. These solve for gas-phase flow and reactions. This integrated strategy enables the simultaneous analysis of interactions. It connects microscopic behaviors, like particle collisions and agglomeration, with the macroscopic fluidization characteristics of the bed [29]. Based on such models, design space methods can be applied. Latin hypercube sampling is an example. These methods allow for the systematic optimization of key operational parameters. These parameters include gasification agent flow rate and equivalence ratio. Such optimization aims to improve energy efficiency and carbon conversion rate [30].

Current research trends aim to break the barriers between different scales from molecules to particles to bed layers. Deep learning methods, such as convolutional neural networks, have been applied to flame image classification. This application has achieved high accuracy. It has also significantly accelerated CFD calculations. These advances provide new tools for online monitoring and process optimization [31]. Future work will focus more on integrating high-resolution in situ characterization with physics-based multi-scale simulations. These simulations include methods like direct numerical simulation and DEM-CFD [29,32]. The integration will also incorporate data-driven artificial intelligence methods. The aim is to build a full-chain digital model spanning from the microscopic to the macroscopic level. The ultimate goal is to achieve more efficient, clean, and intelligent design and operation of biomass energy systems.

Researchers have employed molecular-scale DFT calculations. They have also used particle-scale micro-CT scans. Bed-scale CFD simulations are another tool they have utilized. However, seamlessly and bidirectionally coupling these tools poses a significant challenge. This challenge arises because the tools have different resolutions. They also employ different physical descriptions. For instance, converting the surface reaction energy barriers calculated by DFT into effective reaction rate constants in the particle model often relies on numerous empirical assumptions, which are one of the main sources of uncertainty in model predictions.

2.4. Thermal Decomposition and Combustion Kinetics from Empirical Models to Mechanistic Cognition

Biomass pyrolysis, as the initial and crucial step of the combustion process, has reaction pathways and product distributions that directly determine the characteristics of subsequent volatile matter combustion, coke oxidation, and pollutant generation. Therefore, accurately modeling the pyrolysis process is the foundation for in-depth understanding and optimization of the entire combustion system [33,34]. The construction and selection of a model should follow a specific cognitive logic. This logic progresses from simplification to complexity. It also moves from a macroscopic to a mechanistic perspective. The typical coverage spans several levels. It begins with computationally efficient empirical models. It then advances to multi-step competitive models, which distinguish component reactions. The progression extends up to detailed chemical mechanism models. These detailed models are based on molecular structure [34]. Evaluating a pyrolysis kinetics model requires a comprehensive, multi-dimensional assessment. First, the model must accurately reproduce the pyrolysis weight loss process. This accuracy is typically measured by indicators like the coefficient of determination and root mean square error (RMSE). Second, the model must reliably predict the yields of coke, bio-oil, and gas products. The relative error for these predictions usually needs to be within 10–15%. Third, the model should accurately describe the release kinetics of key gas components, such as H₂, CO, and CH₄. Finally, the computational efficiency of the model is critical, especially when it is coupled with large-scale simulations like computational fluid dynamics [33,34,35,36]. The comparison of various models is shown in

Table 1. Method evolution of biomass pyrolysis kinetic models.

Model Method	Core Modeling Concept	Representative Methods	Key Progress and Academic Value
Empirical/Single step model [37]	Simplify the macroscopic total package reaction into a single or a few epigenetic reactions, and perform phenomenological fitting using apparent kinetic parameters. (A, E, n).	N-level reaction model	A mathematical description method for the pyrolysis process was established, providing a fundamental tool for rapid engineering estimation.
Multi-step competitive response model [38]	Construct a parallel competitive reaction network based on the intrinsic reaction characteristics of the three components of biomass (cellulose, hemicellulose, lignin).	Broido–Shafizadeh framework [39], Ranzi model [40]	Transitioning from “phenomenon fitting” to “component mapping” significantly enhances the ability to predict and interpret complex weightlessness curves, becoming a bridge connecting engineering and mechanism.
Distributed Activation Energy Model (DAEM) [41]	Characterize the heterogeneity of the chemical structure of reactants and the infinite parallelism of reactions using the continuous distribution function of activation energy (E).	Various distributed functions (such as Gaussian distribution [42])	It provides a powerful mathematical framework for handling complex solid reactions, and its distribution function has physical and chemical connotations, making it a key tool for high-precision analysis of experimental data.
Detailed chemical structure model	Based on the initial chemical structure of fuel macromolecules (functional groups, bridging bonds), simulate elementary process networks such as bond breaking and free radical reactions.	FG-DVC [43], CPD model [44]	The leap from “phenomenon simulation” to “molecular scale mechanism prediction” has been achieved, which is the computational basis for a deeper understanding of the nature of pyrolysis and the directional regulation of products.

.

The selection of the model must strictly serve the specific research goals and application scenarios. When it comes to engineering process simulation (such as using Aspen Plus V12 for system simulation), the parameterized multi-step competitive reaction model has become the preferred choice due to its good balance between computational efficiency and prediction accuracy [36]. When the research focus is on the pore evolution or reaction activity of coke, a detailed model that can track the chemical structure of solid residues needs to be adopted [34]. When conducting reactor-scale combustion or gasification simulations, the core challenge lies in achieving multi-scale coupling of particle-scale pyrolysis and reactor-scale flow and heat transfer. At this time, embedding an appropriately simplified pyrolysis sub-model as a source term into the macroscopic CFD solver is an effective strategy [35,36]. The comparison of various pyrolysis kinetic models is shown in

Table 2. Engineering role, core characteristics, and applicable boundaries of pyrolysis kinetics models.

Model Method	Core Logic and Characteristics	Computational-Mechanism Trade-Off	Engineering Roles and Applicable Boundaries	Ref.
Experience/Single-step model	Mathematical fitting of macro package response.	The calculation cost is extremely low, and the clarity of the mechanism is lacking.	Role: Rapid Estimation Tool. Boundary: Suitable for low-precision requirements in conceptual design and initial screening.	[45,46]
Multi-step competitive response model	A parallel competitive reaction network based on three components (cellulose, hemicellulose, lignin).	The calculation cost is moderate, the mechanism is clear, and the best engineering compromise is achieved.	Role: The main tool for engineering design optimization. Boundary: A wide range of options for coupling CFD for reactor-level combustion/gasification simulations.	[34,35]
Distributed Activation Energy Model (DAEM) [41]	Characterization of reaction heterogeneity using a continuous distribution of activation energy.	The mathematical framework is powerful and can accurately describe experiments without providing specific reaction pathways.	Role: Advanced experimental data analysis and parameter calibration tool. Boundary: Extract intrinsic kinetic parameters from thermogravimetric (TG) and other data to provide input for engineering models.	[47]
Detailed chemical structure model	Simulation of an elementary reaction network based on the initial structure of fuel macromolecules.	The mechanism has the highest clarity, can predict molecular products, and has extremely high computational costs.	Role: Mechanism analysis and knowledge discovery tool. Boundary: used in basic research and the path to explore, its value is to guide engineering model after the order reduction.	[48]

.

Although the existing model system is already relatively rich, it still faces several key challenges. The multi-step competitive reaction model can predict total product yield well. However, its description of volatile matter’s secondary reactions is often simplified. This simplification may lead to deviations in predicting tar cracking behavior. It can also cause inaccuracies in forecasting the final light gas yield. These issues are particularly relevant under high-temperature and long residence time conditions [35]. Future development should focus on integrating more detailed gaseous secondary reaction mechanisms and using theoretical calculations and model inversion methods to obtain more reliable kinetic parameters. Moreover, many detailed mechanism models are currently mainly applied at the single-particle scale and are disconnected from actual reactor engineering design [34]. Therefore, bridging the gap between fundamental mechanism understanding and engineering practice requires a key approach. This approach is the construction of a multi-scale coupling method and simulation tool. The tool must seamlessly connect “particle-scale detailed pyrolysis” with “reactor-scale macroscopic transport.” This effort will also provide a core simulation tool. The tool is intended for the design and optimization of next-generation biomass energy systems. These systems must be both highly efficient and achieve ultra-low emissions [34,36].

Current model selection research mostly remains at the stage of feature listing, lacking quantitative decision criteria based on error propagation and computational cost. The core contradiction lies in the disconnection between the complexity of mechanism research and the simplicity requirement of engineering applications. In the future, an urgent task is to establish a mapping framework. This framework should connect key order parameters from micro-dynamics to macro-performance. Concurrently, a quantitative selection system must be built. This system needs to balance physical fidelity with computational efficiency. Together, these efforts will promote a transformation. The goal is to shift model selection from qualitative description to scientific decision-making.

2.5. Summary and Discussion

The combustion of biomass fuel is a typical complex process involving multiple scales and multiple physical fields. Mechanistic models at the molecular scale can reveal reaction pathways. However, a significant gap exists between their idealized conditions and the actual deactivation mechanisms in macroscopic reactors. High-fidelity models at the particle scale improve prediction accuracy. Yet, their extremely high computational costs make them difficult to use directly for engineering-scale CFD simulations. This presents a fundamental contradiction between fidelity and computational efficiency. Regarding model selection, empirical models and detailed chemical models represent two extremes. They differ greatly in computational cost and mechanism depth. The multi-step competitive models or distributed activation energy models are widely used in engineering. These models are essentially compromises between accuracy and efficiency. The compromise is made based on specific research goals, such as rapid screening, system optimization, or mechanistic analysis. The current core challenge is not the lack of models, but the lack of a coupling framework that can effectively connect these scales and achieve the forward feedback of mechanism knowledge to engineering parameters. The future breakthrough direction lies in integrating high-resolution characterization, physical simulation, and artificial intelligence to build a digital twin system that spans the scale gap.

3. Multi-Scale Modeling and Numerical Simulation Methods for Biomass Combustion

3.1. Particle Scale Model from Single Particle Reaction to Group Behavior

In the multi-scale modeling system for biomass combustion, the particle-scale model acts as a crucial bridge. It connects the intrinsic dynamics to the macroscopic reactor performance [49,50]. This model is also a core element. It determines the overall simulation accuracy and engineering applicability [51]. Although various modeling approaches have been established in current research, there are still significant differences in model selection, the rationality of assumptions, computational efficiency, and the matching degree with actual operating conditions.

3.1.1. Mass Transfer and Structural Evolution Model Within a Single Particle

The single-particle model aims to describe the internal heat and mass transfer, chemical reactions, and dynamic evolution of the microstructure during the thermal conversion process of biomass particles. In the early models, to simplify the calculation, the isothermal assumption and uniform contraction mechanism were often adopted [52], treating the particles as homogeneous spheres and fitting the experimental data with empirical rate constants [53]. Although these models have low computational costs, they are only suitable for preliminary estimations involving small-sized particles (<1 mm). Furthermore, they ignore real temperature gradients and pore evolution. Their physical foundation is weak. Consequently, they are difficult to extrapolate to actual, complex conditions.

With a deeper understanding of the heat-reaction coupling mechanism, non-isothermal multi-dimensional models have gradually become the mainstream [54,55]. When simulating the steam vaporization of large carbon particles (>5 mm), one-dimensional and two-dimensional non-isothermal models are used. These models clearly reveal significant temperature gradients inside the particles. They also show the distribution of reaction fronts. The results confirm a severe distortion of the isothermal assumption for particles larger than a few millimeters [56]. Figure 3, for example, provides a schematic diagram of the internal steam vaporization reaction for a single particle. This illustration shows gases like water vapor (H₂O) diffusing into the particle interior. There, they undergo vaporization reactions with carbon. This reaction produces gases such as H₂ and CO. The physical process depicted in this schematic is based on the conceptual framework constructed by such non-isothermal models. The setting of the length of micro elements in the radial direction is precisely used to analyze such temperature and concentration gradients. Figure 4 presents a single particle shape diagram. This diagram distinguishes the distinct areas of carbon, volatile matter, moisture, and ash content. It also indicates the directions of key processes, including evaporation, volatilization, carbon oxidation, and gasification. Consequently, the diagram depicts a more complex conversion path involving the coexistence and transformation of multiple components. It intuitively presents the dynamic distribution of the internal structure and reactions of biomass particles during thermal conversion. To further realistically reflect the complex porous structure of biomass char and its influence on the diffusion–reaction process, a geometrically faithful pore network model based on high-resolution X-ray micro-computed tomography (micro-CT) images has been developed [25]. These models can simulate based on the real three-dimensional structure and have significant advantages in the mechanism research level, but they rely on high-precision imaging and are computationally expensive, limiting their direct application in engineering design.

Although high-fidelity models demonstrate excellent physical fidelity, their application still faces significant challenges. The models are highly sensitive to input parameters such as initial porosity, effective thermal conductivity, and reaction kinetics parameters, and the significant batch differences inherent in biomass raw materials lead to inherent uncertainties in these parameters. Moreover, most models still simplify particles into regular geometric shapes (such as spheres or cylinders), ignoring the influence of the irregular morphology of natural biomass on local heat transfer and reaction pathways. Studies have shown that considering the true shape of particles can significantly reduce the error in predicting combustion behavior, highlighting the importance of not ignoring the systematic deviations introduced by geometric simplification [59].

The development of single-particle models shows an evolutionary trend from “empirical fitting” to “mechanism-driven”. Future breakthroughs will depend on balancing model fidelity with practicality. For example, model reduction methods can be developed to control computational costs [60]. Furthermore, systematic uncertainty quantification research is needed [61]. This work would assess and constrain the impact of parameter uncertainties on predictions. Such advances are essential for enhancing model reliability and applicability in engineering design and optimization [25,53,56,59]. The different modeling methods and their comparisons are shown in

Table 3. Trade-off framework for the computational cost, prediction accuracy and applicability of single-particle thermal conversion models.

Model Type	Key Assumption	Calculation Cost	Prediction Accuracy	Applicable Particle Size/Conditions	Typical Application Scenarios
Isothermal + uniform shrinkage model	The internal temperature of the particles is uniform. The reaction interface is clear. The pore structure remains unchanged.	Extremely low (analytical solution or simple ODE (Ordinary Differential Equation))	Low (only applicable to extremely small particles or strong external control)	<1 mm; high heating rate; Thin bed	Rapid parameter fitting, teaching examples, and preliminary process screening [53]
Non-isothermal + multidimensional shrinkage model	Consider radial/axial temperature gradients; Continuous distribution of reaction zone; Empirical variation in porosity with conversion rate.	Medium-high (requires numerical solution of PDE (Partial Differential Equation))	Medium high (capable of capturing internal gradients and reaction frontiers)	1–10 mm; conventional pyrolysis/gasification conditions	Design of a large particle gasifier and estimation of reactor residence time [56]
Non-isothermal + pore network model	Based on real or random pore structures; Dynamically update the connectivity of holes and throats; Multi-component diffusion-reaction coupling.	Extremely high (requiring graph theory algorithms and large-scale iterations)	High (especially during the high conversion rate stage)	<5 mm (limited by imaging resolution); Slow gasification/combustion	Mechanism research, ash content impact analysis, catalyst-loaded particle simulation [25]

.

The failure of the particle size assumption mainly stems from the highly heterogeneous physical and chemical properties of biomass particles. Specifically, a large Peclet number causes significant internal thermal resistance in particles. This occurs, for instance, with millimeter-sized particles or in extremely high heat flow environments. Under these conditions, the isothermal assumption fails completely. A non-isothermal model becomes necessary to capture the substantial internal–external temperature gradients. For real particles with highly irregular and anisotropic morphologies, such as straw and tree bark, the assumption of regular geometric shapes is problematic. It significantly distorts predictions of the heat transfer surface area and mass diffusion paths. This leads to systematic deviations. Furthermore, the assumption of a homogeneous pore structure is inadequate for fuels with complex multi-level pore structures. Examples include wood vessel lenticels or fuels whose pores change significantly after pre-treatment. This assumption cannot accurately describe local diffusion limitations. Consequently, it affects the prediction accuracy of the coke formation rate and gas product composition. In engineering, the “non-isothermal multi-dimensional contraction model” has become mainstream. This status stems from not only its relatively high accuracy, but also its “moderate” computational cost. The model achieves the optimal balance between two key factors. The first is the critical engineering requirement of predicting the internal gradient of particles. The second is the realistic constraint of controlling the computing time. Although the “porous network model” has “high” accuracy, its extremely high computational cost and imaging dependence determine that it can only be a tool for “mechanism research” at present and cannot be used for engineering design. This reveals the essence of engineering models, which is to seek a compromise between acceptable simplification and necessary fidelity.

3.1.2. DEM-CFD and Statistical Methods for Multi-Particle Simulation

Multi-particle modeling aims to overcome the limitations of single-particle idealization. It captures the collective behavior of particle groups, such as flow, mixing, and collisions. This approach also accounts for agglomeration and particle size distribution evolution. By doing so, it enables a more realistic simulation of multiphase flow and reaction processes within reactors [62]. Current mainstream methods can be classified into discrete methods based on individual tracking and continuous methods based on statistical averaging.

Individual-tracking-based methods, like DEM-CFD, can precisely reproduce non-uniform phenomena. These phenomena include particle segregation, local concentration fluctuations, and residence time distribution. This capability is achieved by directly solving the Newtonian motion equations for each particle. The methods also solve the contact mechanics for each particle [62,63]. For example, in mechanical fluidized reactors, this method reveals how double helix belt agitation affects particle heating rate uniformity. This guides the optimization of rotation speed and biochar residue management [62]. In fluidized beds, the method can also evaluate the impact of air distribution plate design. It specifically assesses the effects on initial fluidization and mixing efficiency [63]. However, its computational cost increases superlinearly with the number of particles, and even with high-performance computing, it is difficult to directly simulate particle systems of billions of particles at industrial scales [64]. Moreover, the model’s results exhibit high sensitivity. This sensitivity applies to both the chosen contact force models and key physical property parameters. Such parameters include the elastic modulus and friction coefficient of the biomass particles. These parameters are often difficult to accurately obtain. Consequently, this difficulty directly affects the reproducibility of the simulation results [64].

In contrast, the Euler–Euler two-fluid model (TFM) treats the particle phase as a continuous medium, akin to the fluid. It describes the average motion and reaction behavior of the particle phase. This is achieved by solving the macroscopic conservation equations for the phase’s mass, momentum, and energy. This model possesses high computational efficiency. It is suitable for simulating both transient and steady-state conditions in large-scale cyclic fluidized bed reactors. These reactors include engineering-scale applications like boilers and gasifiers. The model currently serves as the mainstream numerical tool. It is widely used in engineering applications and industrial design [65,66]. To compensate for the neglect of individual differences in TFM, a population balance model is often coupled to track the evolution of particle size distribution (PSD) [29,67]. However, TFM itself relies on empirical models. These models describe coalescence, fragmentation, and other functions. Furthermore, TFM faces difficulties in finely coupling with the transient characteristics of local flow fields. This results in a limited predictive ability for non-uniform transient phenomena. Examples of such phenomena include bubble dynamics and dead zone formation [29,67].

From an application perspective, DEM-CFD and TFM are not mutually exclusive but complementary tools serving different research goals and scales. DEM-CFD is more suitable for basic mechanism exploration, pilot-scale optimization, and providing constitutive relationships for macroscopic models [62,64], while TFM dominates the process design and overall performance analysis of industrial reactors [65,66]. The core challenge is the lack of an effective cross-scale bridging framework. It is difficult to systematically elevate detailed DEM-CFD simulation results to macroscopic TFM constitutive relationships, such as particle phase viscosity and solid stress [64]. Conversely, TFM’s macroscopic results cannot effectively guide modifications to local assumptions in discrete models, like contact models. Developing theoretical strategies is therefore an important future direction. These strategies, such as “filtering—reconstruction” methods or data-driven machine learning agent models, aim to achieve high-fidelity information transfer and integration from microscopic details to macroscopic behavior in multi-particle models [29,64,67].

The evolution process of the modeling approach at the particle scale clearly indicates that the selection of the model is essentially a precise balance between computational cost, prediction accuracy, and the degree of physical mechanism restoration. The widely used “non-isothermal multi-dimensional contraction model” in engineering is the optimal solution under this balance, as it can capture the key internal gradients that affect engineering performance at an acceptable computational cost. The “porous network model” represents the boundary of mechanism exploration, and its extremely high cost limits its direct application in engineering. At the multi-particle simulation level, two complementary approaches form a key tool set. The first is the DEM-CFD method, which is based on individual particle tracking. It provides the possibility for understanding microscopic mechanisms, such as mixing and segregation. It also serves to verify constitutive relationships. The second is the two-fluid model, which relies on statistical averaging. This model represents a feasible means for simulating industrial-scale reactor systems. However, there is a lack of effective scale bridging theory between the two, making it difficult to convert the insights from fine-scale simulation into improved macroscopic models that have been fully verified.

3.2. Turbulence, Reaction and Radiation Coupling Model at Reactor Scale

3.2.1. Turbulence Chemical Reaction Interaction Model

The interaction between turbulence and chemical reactions is a core physical and chemical process that determines the combustion rate, pollutant generation, and flame stability. Accurate simulation of this process is crucial for predicting the performance of biomass combustion devices. Current mainstream simulation methods include the eddy dissipation concept (EDC) model, the flame surface model, and the finite rate/eddy dissipation model [68].

The EDC model is based on the Kolmogorov micro-scale assumption and confines chemical reactions within the smallest turbulent structures, suitable for diffusion-dominated non-premixed combustion [68]. The main advantage of this model is that it does not require detailed chemical kinetic mechanisms and can operate based on the total reaction, resulting in relatively lower computational costs [69]. However, this cost advantage relies on a simplified chemical description. When the EDC model is combined with the detailed heptane mechanism (EDC heptane mech), the calculation time increases sharply to over 300 h. Once complex chemical reactions are introduced, its computational efficiency will sharply decline. Moreover, its ability to capture the chemical kinetic behavior of local flameout, re-ignition, or partially premixed regions is limited. In highly turbulent or low-oxygen conditions, the error may significantly increase [26]. The improvement in computational efficiency often comes at the expense of the detailed chemical mechanism. When using the simple mechanism, EDC has extremely high efficiency, but when using the detailed mechanism, the time consumption increases dramatically. SFM significantly enhances efficiency by pre-calculating the database while maintaining certain chemical details. This indicates that the trade-off between “efficiency” and “depth of mechanism” is the core when choosing a model. SFM is applicable in scenarios where the assumptions hold (complete flame surface structure), but once the assumptions (such as strong swirl flow) fail, its computational efficiency advantage will significantly decrease due to the failure of the prediction.

The flame surface model introduces mixed fractions and scalar dissipation rates. It uses these to construct a pre-calculated laminar flame database. This database characterizes the turbulent flame state. The model is applicable to non-premixed or weakly premixed flames. This application is valid under moderate to high turbulence intensities [65]. The core premise is that turbulence only stretches but does not destroy the flame surface structure. However, under strong swirl, highly premixed, or MILD combustion conditions, this premise may no longer hold. Here, the flame surface can be strongly stretched or even locally extinguished. This results in deviations in model predictions [70]. Nevertheless, this model achieves a good balance between computational efficiency and accuracy, and is widely used for temperature field prediction in large boilers and gasifiers [71].

The finite-rate/eddy-dissipation model (FR/EDM) combines Arrhenius finite rate chemical kinetics with the turbulent dissipation time scale. It determines the net reaction rate by taking the slower of these two controlling processes. This approach thereby balances the effects of chemical kinetics and turbulent mixing [68]. This model has a moderate requirement for the chemical mechanism and can handle complex processes such as multi-step volatile release and coke oxidation in coal–biomass co-combustion [69]. Its limitation lies in its strong empirical dependence on the turbulent time scale, and it may overestimate the reaction rate in low Reynolds numbers or strong recirculation zones [72]. Its computational cost is usually between the above two models [26].

The selection of the model requires a trade-off between chemical kinetics fidelity, adaptability to turbulent conditions, and computational resources. The EDC model is suitable for rapid engineering estimations, as it offers high computational efficiency [26,68,69]. The flame surface model can provide higher accuracy, provided its structural integrity assumption is satisfied. In contrast, the FR/EDM model demonstrates better robustness. This is particularly evident when handling complex scenarios, such as multi-fuel co-combustion [26,68,69]. A comparison of mainstream turbulent combustion models is shown in

Table 4. Core concepts, evolution history and characteristics of mainstream turbulent combustion models.

Model	Core Idea	Requirements for Chemical Reaction Mechanisms	Typical Computational Cost	The Most Suitable Scenario	Main Limitations
EDC	Reaction occurs within the smallest turbulence scale	Overall package response is sufficient	Low	Non-premixed biomass combustion, fixed bed [68]	Unable to handle local stalling and premixing effects [26]
FSM	Based on a precomputed laminar flame database	Need to build a flame surface library	Mid	Moderate turbulent non-premixed flame [65]	Failure under strong swirl, MILD or high premix [73]
FR/EDM	Choose the slower one in the mixture of chemical kinetics and turbulence	Simplify the mechanism	Midhigh	Coal–biomass co-combustion, multiphase combustion [69]	Experience parameters are sensitive, not accurate at low Re [72]

.

3.2.2. Radiative Heat Transfer Modeling and Its Efficiency Trade-Offs

Radiation heat transfer is a dominant heat transfer method in high-temperature biomass combustion and gasification reactors. This is especially true in the volatile combustion zone and the coke bed. In these regions, its heat flux contribution can exceed 60% of the total heat flux. Related CFD models and experimental studies have shown that the penetration behavior of radiation in the bed and the radiative heat transfer between gas particles have a critical impact on the temperature distribution and reaction rate of the combustion process [27]. Accurate modeling of radiative heat transfer is crucial for predicting temperature distribution, combustion efficiency, and pollutant generation inside the furnace. The core involves the solution method of radiative transfer equations and the selection of medium radiative property models.

The discrete ordinates (DO) method and the P1 model are two main methods for solving the radiation transport equation [27]. The DO model can accurately handle complex geometric structures and anisotropic radiation fields by solving equations in discrete solid angle directions and is suitable for simulating laboratory burners or industrial furnaces with complex internal structures [74]. The temperature and carbon monoxide distribution simulation on the central axis of the biomass combustion furnace in Figure 5 was completed using the DO method. The comparison of this simulation result with experimental data verified the capability of the DO model. The model successfully captures the peak temperature distribution trend at the furnace center. It also captures the attenuation law of CO concentration. Although there are differences in absolute values, the model successfully reproduced the bimodal distribution characteristics of the temperature curve along the axial direction. However, its computational cost is high, reaching several times that of the flow field solution, and it is sensitive to the quality of the computational grid [27]. The P1 model employs first-order spherical harmonic functions for approximation. This simplifies the radiation transport equation into a diffusion-type equation, which is easier to couple and solve. The model offers higher computational efficiency. It is therefore well-suited for large-scale systems that have relatively uniform radiation fields [68]. However, its simplifying assumptions may lead to significant errors when dealing with strong shading or local high-temperature hotspots [68].

No matter which solution method is adopted, accurately describing the non-gray gas radiation characteristics of the flue gas (mainly H₂O and CO₂) is another key point. The weighted sum of gray gases model (WSGG) approximates the real gas absorption coefficient. It does this by combining multiple gray gases through weighting. This model is commonly used in engineering. It handles the non-gray gas characteristics of gases effectively [27]. However, its model parameters depend on the composition of the flue gas and temperature, and its applicability may be limited for components such as alkali metal vapors that may be released during biomass combustion [72]. Currently, most simulations use a one-step empirical model based on acetylene concentration, but the prediction accuracy of such models heavily relies on empirical constants. The diversity of biomass fuel types leads to significant differences in volatile composition, making the prediction of carbon black yield difficult and becoming one of the important sources of uncertainty in radiation heat transfer simulation [76,77].

Radiative heat transfer modeling requires a trade-off between accuracy and efficiency, as well as between the depth of physical process representation and practical feasibility, shown in

Table 5. Selection criteria of radiation heat transfer model for biomass combustion: based on computational objectives and system complexity.

Model	Adaptability to the Complexity of Geometric/Radiation Fields	Calculation Cost	Ability to Handle Non-Gray Media (Gases/Particles)	Recommended Usage Scenarios
DO	High (complex geometry, strong directionality)	High	Gas: requires cooperation with WSGG; Particles: Coupled Mie Theory [27]	Laboratory burner, small-scale verification [74]
P1	Low (suitable for uniform and unobstructed fields)	Low	Gas: Gray gas assumption is required; Particles: approximate treatment [68]	Preliminary Design of Large Boiler [78]
WSGG (as a physical property model)	Not applicable (needs to be coupled with DO/P1)	Low additional cost	Gas: Excellent (CO2/H2O/CH4); Particles: no treatment [27]	All biomass systems containing gas-phase radiation [72]

. The DO model and the P1 model are designed for different application scenarios. They place different emphases on accuracy versus computational efficiency. Furthermore, the selection of the gas radiative property model and the carbon black model is critical. This selection directly affects the reliability of simulation results, particularly under specific fuels and operating conditions [23,27,68]. Future improvement directions include developing more universal non-gray gas models and introducing more refined carbon black kinetics models or data-driven methods to reduce the uncertainty of key sub-models [23,72,77].

At the reactor scale, accurately simulating the interaction between turbulence and chemical reactions as well as radiation heat transfer is a core challenge. The choice of the model is highly dependent on the extent to which it satisfies the basic assumptions for specific combustion conditions (such as premixing degree and turbulence intensity). Any misuse of the model will lead to systematic deviations in the predictions. The simulation of radiation heat transfer requires a choice. This choice lies between the high accuracy of the DO method and the computational efficiency of the P1 model. Furthermore, the simulation’s fidelity is strongly dependent on the description of the physical sub-models. These sub-models describe the non-gray properties of the gas and the generation of soot. These sub-models, especially the soot generation model for the diverse volatile components of biomass, are important sources of uncertainty in current radiation simulations.

3.3. Multi-Scale Coupling Strategy and Model Validation

In the reactor-scale modeling of biomass energy conversion systems, the multi-physics field coupling of turbulence, chemical reactions and radiative heat transfer is the core challenge for achieving high-fidelity simulations. In recent years, CFD technology has advanced considerably. This development has enabled significant progress in key research areas. Researchers have improved model selection, coupling strategies, and industrial adaptability. These advancements provide crucial theoretical support. Specifically, they support the design of combustion and gasification systems that are both highly efficient and achieve ultra-low emissions.

3.3.1. Numerical Implementation Path of Cross-Scale Coupling

Cross-scale coupling is implemented numerically. Its aim is to connect multi-scale correlations. These correlations span from molecular reaction mechanisms to reactor system performance. This numerical implementation forms the core methodology. It is essential for accurately simulating the biomass combustion process. There is a strong bidirectional coupling between microstructure evolution and macroscopic flow and heat transfer [79]. For instance, the porosity of coke particles changes dynamically. This significantly affects oxygen diffusion and local exothermic reactions. These effects subsequently disturb the temperature and component fields throughout the furnace [80]. Therefore, a multi-scale coupling strategy must be adopted to achieve the scientific depiction of the entire chain behavior.

Multiscale modeling typically covers molecular/single-particle scale, particle group scale, and reactor system scale [81]. Information is transmitted between these scales via a bidirectional mechanism. Effective parameters from the fine-scale model are calculated. These include the apparent reaction rate and volatile fraction release characteristics. These parameters are “up-scaled” and used as input for the coarse-scale model. Concurrently, local environmental conditions are obtained from the system-scale simulation. These conditions, such as temperature and oxygen concentration, are fed back to the fine-scale model. They serve as boundary conditions. This process achieves a dynamic self-consistent solution [79,82]. This coupling is manifested in multiple key stages of biomass conversion. In gasification applications, the single-particle pyrolysis model serves a key function. It provides source terms for reactor-level CFD simulation, such as coke yield and volatile fraction composition [83]. Conversely, reactor-level CFD simulations also inform the particle model. They provide the local conditions around the particle [84]. These conditions determine the surface reaction degree of the single particles. For NO_x emission prediction, the microscopic revelation of NH₃ and HCN conversion paths based on detailed chemical mechanisms (such as the Glarborg mechanism), its simplified form can be embedded in the furnace-scale simulation to predict overall emissions [85]. In mixer optimization, the prediction of ash melting deposition [80] and the design of chemical looping combustion systems [81] require similar cross-scale information transmission. This information pertains to particle motion, mineral phase changes, and the surface reaction kinetics of oxygen carriers. Similarly, solar-driven pyrolysis [82] requires the transmission of cross-scale information. This involves the microscopic structure of porous media. This bidirectional information transmission can be understood through the cloud diagrams of reactor-scale DEM-CFD coupled simulation. In complex reactor design, achieving precise control and coordinated optimization of the thermal decomposition and gasification reduction zones relies on a core foundation [79,86]. This foundation involves the iterative and coupled solution of particle-scale and reactor-scale models. It requires closely linking detailed particle reaction models, such as for combustion and pyrolysis, with the macroscopic transport phenomena of the entire reactor.

The realization of the aforementioned coupling relies on an integrated numerical simulation platform. The core of this platform is a robust solver, such as ANSYS Fluent 21R2, based on the finite volume method, which has been widely used to handle complex multiphase flow, chemical reactions, and radiative heat transfer coupling problems in biomass combustion [87,88]. A dedicated chemical kinetics solver (such as CHEMKIN) is used for detailed mechanism verification and parameter analysis [87]. Reliable fuel physical property and reaction kinetics databases form the foundation of the platform. For instance, in ammonia-biomass co-combustion studies, different simplified mechanisms predict varying self-ignition delay times. This discrepancy highlights the critical importance of accurate kinetic parameters [89]. Industrial analyses of the fuel and ash data directly affect the setting of the pyrolysis model and the calculation of radiation characteristics [90,91]. The modular architecture enhances the platform’s flexibility. It allows the configuration of different sub-models based on research goals. For instance, one can choose a mixed rate model for kinetics/diffusion control in coke combustion [90]. Alternatively, one can independently configure the constants for multi-path heterogeneous reactions involving carbon and O₂, CO₂, and H₂O [92]. The introduction of transport equations for surface area density parameters can expand the modeling from single particles to local statistical average scales [93]. The platform’s complete pre- and post-processing capabilities are built upon several key components. It employs a well-developed grid generation strategy, which often utilizes three-dimensional non-uniform grids, and conducts grid independence verification. The solver applies second-order upwind discretization and appropriate pressure–velocity coupling algorithms, such as COUPLED. The post-processing and verification procedures are comprehensive, covering temperature, components, and emissions, such as the total fixed nitrogen conversion rate [85,87]. For transient processes involving phase changes, specialized solution settings are required [94].

Currently, there are mainly two strategies for coupling particle-scale models with macroscopic reactor simulations: offline parameterization and online coupling. The offline method calculates “effective kinetics” parameters in advance. These parameters are then input into the macroscopic model as constitutive relations or lookup tables. This approach offers the advantage of high computational efficiency. Consequently, it is a common solution in current industrial CFD simulations. However, this method typically fails to reflect the dynamic influence of the actual flow and temperature fields on the particle reaction process [53]. Online coupling, also known as bidirectional tight coupling, has higher physical fidelity. It requires real-time data exchange and a synchronous solution between the particle and fluid phases at each time step. However, it faces severe computational challenges. This is due to the large difference in time and spatial scales. The pyrolysis process inside particles occurs on a millisecond scale, while the flow simulation in the reactor operates on a second scale. This significant difference in time scales creates a conflict in time step matching for the numerical solution. Additionally, solving the control equations for a large number of particles within each computing unit in parallel would lead to astonishing computational resource consumption [29]. Existing platforms, such as MFiX, utilize strategies like “particle packages” or “representative particles” to alleviate computational pressure. However, these strategies may introduce statistical biases when dealing with wide particle size distributions or multi-component raw materials [29]. Additionally, most coupling frameworks still assume independent particle behavior, ignoring collective effects such as radiation shielding between particles and interaction of volatile components in high-fill-bed layers [56]. Therefore, the core issue of current multi-scale coupling is clear. High-fidelity bidirectional coupling offers theoretical completeness. However, it is constrained by prohibitive computational costs. Conversely, offline parametric methods are feasible for engineering applications. Yet, these methods struggle to capture the essential nonlinear feedback. The future breakthrough path should not be confined to the iterative optimization of complex coupling algorithms. Instead, it should shift towards constructing intelligent reduced-order models or AI agent models. The aim is to precisely reconstruct the input–output mapping of the high-fidelity model. This reconstruction must be achieved with extremely low computational overhead. Consequently, this will effectively bridge the gap between theoretical accuracy and engineering efficiency.

The current multi-scale coupling paradigm still mostly remains at the “nested solution” level. Future development must evolve towards “cooperative modeling.” The ideal framework should possess adaptive capabilities. It enables the use of high-fidelity models in key areas, such as the reaction front. In secondary areas, it can switch to simplified models. The framework also needs to establish standardized model interfaces. This is essential to achieve true interoperability. This will promote the particle-scale model from an “academic display” tool for mechanism research to a powerful means that can substantially empower engineering design and optimization.

3.3.2. Model Validation, Uncertainty Quantification and Industrial Application Guidelines

Model validation, uncertainty quantification, and engineering application are the key steps to ensure that the numerical simulation research on biomass combustion is both scientifically rigorous and practically applicable [95]. Biomass fuel exhibits high heterogeneity and variability. Furthermore, the combustion process involves strong, multi-scale, and multi-physical field nonlinear coupling. Given these complexities, the traditional, isolated “point-to-point” data comparison method is no longer sufficient. It cannot comprehensively evaluate the predictive reliability of the model [96,97]. Therefore, a systematic approach must be established to enhance the predictive credibility of the model when dealing with fuel characteristic fluctuations and complex mechanisms.

Uncertainty quantification is the foundation of this process, and its core lies in systematically identifying and quantifying the impact of various uncertainty sources on the model’s prediction results. A complete quantification process requires comprehensively identifying input parameter variability. This must especially include uncertainties arising from biomass fuel’s inherent properties. Key properties are moisture, ash content, and alkali metal content. These properties significantly affect pyrolysis, ignition, and pollutant generation pathways [98]. Through global sensitivity analysis, the most significant dominant parameters that have the most significant impact on key outputs can be selected [26]. Subsequently, methods like Monte Carlo simulation or polynomial chaos expansion can be used. They propagate the joint probability distribution of the input parameters to the model output. This process yields the statistical characteristics of the predicted values. It also provides their confidence intervals [99,100]. Its ultimate value lies in providing probabilistic risk assessment for engineering practice. This assessment can quantify, for instance, the probability of pollutant emissions exceeding standards. It can also quantify the probability of severe coking occurrence. These probabilities are evaluated within a given range of fuel characteristic fluctuations. Consequently, this approach supports robust design and operation optimization [100,101].

These methods have been applied in the research of biomass energy systems. Researchers develop oxygen-rich coal combustion technology. They conduct numerical simulations to study combustion characteristics. Experimental data, like particle burnout degree, are used to constrain model uncertainty. A three-dimensional numerical model of a pressurized oxygen-rich burner is established. It uses computational fluid dynamics methods. The model’s reliability is verified primarily through coal char burnout data. This enables a systematic analysis of biomass blending under pressurized conditions. The analysis focuses on specific impacts regarding combustion temperature, gas composition, and pollutant generation pathways, such as NO [102]. In the study of the co-combustion of nitrogen oxide generation mechanisms, hierarchical combustion experimental data are used to distinguish and calibrate different conversion paths of volatile nitrogen and coke nitrogen [103]. For complex gas–solid two-phase flow systems, advanced visualization techniques can be used to invert and verify key model parameters such as particle flow [101]. Comprehensive verification from local flames to overall emissions of small-scale combustion devices demonstrates the necessity of aligning verification depth with engineering goals [26]. These practices indicate that embedding structured verification and uncertainty quantification processes in research is the fundamental approach to improving the predictive credibility of the model [104].

When applied to industrial applications, the core lies in achieving a balance between prediction accuracy and computational cost for different goals. For rapid design iterations, a combination of computationally robust and efficient methods is typically adopted. A common approach uses Reynolds-averaged Navier–Stokes (RANS) simulation combined with turbulent dissipation concept combustion models. This combination is suitable for the rapid screening of furnace structure and air distribution schemes. However, it provides a relatively coarse depiction of details, such as pollutant generation [105]. When the research focus is on pollutant generation mechanisms, a high-fidelity model combination is required. For example, large eddy simulation can be combined with detailed chemical reaction mechanisms. This approach can precisely capture key formation processes. However, it demands extremely high computational resources [76]. For assessing combustion stability issues, a compromise strategy is to use a validated flame surface model within the RANS framework [73].

Numerical simulation of industrial-scale biomass boilers encounters inherent multi-physics coupling challenges. These include significant time-scale differences between chemical reactions, turbulence, and radiation heat transfer. Another challenge is the strong nonlinear feedback among processes like carbon black formation, radiation, and flow fields [53]. For such complex problems, it is generally recommended to start the analysis with the widely validated RANS + EDC + DO model combination [26,69]. Any model application must be supplemented by strict grid independence verification [74] and necessary experimental data validation [106]. If there is a concern about slag formation or corrosion related to alkali metals, a specialized ash chemical sub-model must be coupled [72]. The selection of the model and the combination strategy should always serve the clear engineering requirements, and an optimal solution should be sought between predictive capability, computational efficiency, and engineering practicability.

Offline parametrization dominates engineering practice due to its efficiency, while the more physically complete online bidirectional coupling is limited to basic research due to the huge computational challenges. This contradiction highlights the core dilemma of multi-scale modeling: the most rigorous physical descriptions are often computationally infeasible, while the feasible engineering methods have to introduce significant simplifications and assumptions. The future breakthrough direction lies in developing intelligent reduction methods and data-driven surrogate models to achieve efficient and reliable transmission of high-fidelity physical information to engineering computable models.

3.4. Summary and Discussion

Based on the systematic elaboration of the multi-scale modeling system for biomass combustion in this chapter, the fidelity of the numerical simulation shows significant imbalance across different physical processes and scales. CFD simulations have shown reliable predictive capabilities for the overall flow field and average temperature distribution in biomass combustion devices. However, the “Achilles’ heel” of their prediction failure is typically not apparent in these macroscopic averages. Instead, it is deeply exposed in specific physical and chemical details. These details are highly dependent on poorly understood microscopic mechanisms. Alternatively, they are extremely sensitive to initial and boundary conditions. The failure of CFD prediction is typically not seen in the overall flow field or average temperature. Instead, it is concentrated in certain critical details. These details heavily rely on microscopic mechanisms that are not yet fully understood. Alternatively, they are extremely sensitive to the imposed initial and boundary conditions. Specifically, CFD often shows significant deviations from measured values in predicting alkali metal release, migration, transformation, and final distribution. This inaccuracy stems from the complex multiphase, multi-component thermodynamic and kinetic processes involved, combined with an incomplete sub-model database. Furthermore, its predictions for low-volatile fuel ignition delay, flame residence position, and extinction limit are often inaccurate. This limitation is due to an insufficient understanding of solid surface reactions and the turbulence–chemical reaction coupling mechanism. Moreover, conventional Reynolds averaging or large eddy simulation approaches smooth out the strong random coupling details between turbulence pulsation and detailed chemical reactions. Consequently, it becomes difficult to accurately capture the instantaneous peak concentration and spatial distribution of pollutants. This processing also renders CFD unable to quantitatively predict the initial formation rate and long-term growth morphology of the dynamic evolution processes for slagging and fouling.

4. Characteristics, Optimization, and Emission Control of Mainstream Biomass Combustion Furnace Types

4.1. Furnace Grate Combustion Technology

The grate furnace combustion technology holds an important position in biomass energy utilization due to its wide adaptability to fuels and relatively low pre-treatment requirements. This technology can handle various heterogeneous fuels including municipal solid waste, crop straw, and various shaped fuels, and has excellent adaptability to high moisture, high ash content, and low calorific value fuels [107]. Even when faced with mixed fuels such as municipal sludge and solid waste with fluctuating moisture and calorific values, stable combustion can be achieved by optimizing the feed and air distribution systems [108]. This is primarily due to two advantages. First, the technology exhibits good adaptability to fluctuations in fuel particle size. Second, its grate structure facilitates the rapid ignition of high-volatile biomass. This structure also promotes the formation of a stable combustion front [105].

The combustion process of the grate furnace exhibits distinct spatial sequence characteristics, with the fuel passing through stages such as drying, pyrolysis, combustion, and complete combustion on the grate [108,109]. However, this process shows significant non-uniformity along the width of the grate, with the area near the water-cooled wall having poorer heat transfer and air dynamic conditions, and its combustion process often lags behind the central area [107,110]. The structural form of the grate (such as vibration [107], reciprocation [111,112], or new rotating grate [113]) and operating parameters are key to regulating the combustion process and improving fuel mixing and residence time. Numerical simulations demonstrate a key advantage. Models that finely consider changes in the fuel bed structure can more accurately predict the internal state of the furnace [114]. Furthermore, thickening the fuel bed enhances the reducing atmosphere of the coke layer. This action has a positive effect, as it improves combustion efficiency and helps control pollutant generation [9].

In terms of pollutant emissions, nitrogen oxides, carbon monoxide, and particulate matter are the main control targets. NO_x generation is closely related to local oxygen concentration [107]. Its emissions can be effectively reduced through several methods. These include optimizing fuel distribution [107], adjusting air distribution [9], or adopting new combustion organization strategies. Examples of such strategies are “generation-reduction-combustion” or decoupled combustion technology [109,115]. The concentration of CO can be controlled by optimizing the segmented air supply mode [31]. The generation of particulate matter is closely related to the volatilization and condensation process of alkali metals in the fuel, and its deposition can also affect heat transfer and potentially form secondary particulate matter [116,117].

The improvement of combustion efficiency is the core of technological development. By optimizing fuel and primary air distribution [31,107], innovating grate structure design [109,113], and fuel pre-treatment [118], the degree of complete combustion can be significantly improved. From an economic perspective, the initial investment for a grate furnace is relatively low [119]. Its operational economy benefits significantly from two key factors. First, it can directly utilize cheap, low-quality fuels, such as high-moisture straw bales [113]. Second, it allows for feasible co-combustion with other waste materials, like municipal sludge [108,109]. Through optimization of operation (such as waste heat recovery [120]) and reduction in slagging and ash accumulation [116,117], energy consumption and maintenance costs can be further controlled, demonstrating good overall feasibility [113].

4.2. Fluidized Bed Combustion Technology

Fluidized bed combustion technology holds a crucial position in biomass energy utilization. This technology includes both bubbling fluidized bed and circulating fluidized bed. Its importance is due to outstanding fuel flexibility, a uniform bed temperature, and high combustion efficiency [121]. This technology demonstrates significant adaptability to biomass fuels with high moisture content, high ash content, and high alkali metal content, such as wet straw, rice husk, and sugarcane bagasse [122,123]. Its strong gas–solid mixing and heat and mass transfer effectively promote the drying and volatile release of high moisture fuels [122]. Additionally, by selecting bed materials such as silica sand, ilmenite, or rich iron coal ash, their dilution, adsorption, or catalytic effects can alleviate the risk of bed material agglomeration and heat surface clogging caused by alkali metals [123,124,125].

The core advantages of the fluidized bed lie in its intense gas–solid mixing and efficient internal heat transfer, which results in a uniform bed temperature distribution, typically stabilizing at the ideal range of 800–900 °C. This is beneficial for the combustion reaction and can inhibit the formation of nitrogen oxides and the melting of ash [121,123,126]. The combustion process has multi-stage characteristics, involving rapid pyrolysis, gas-phase combustion, and heterogeneous oxidation of coke [126,127]. Parameters such as fluidization speed, bed material properties, and air stratification jointly regulate the residence time, mixing efficiency, and reaction process of the fuel [121,125,128]. For example, additives such as iron-based oxygen carriers can promote oxygen transfer to enhance the carbon conversion rate [127].

In terms of pollutant control, the lower bed temperature naturally inhibits the formation of thermal NO_x, and combined with air stratification and flue gas recirculation technologies, it can further reduce emissions [121]. Optimizing excess air and achieving uniform bed temperature are effective measures. These measures can suppress the volatilization and migration of trace elements, such as K, Pb, and Cd. This suppression works by minimizing local high-temperature zones and excessive oxygen concentrations. Both local high temperatures and high oxygen levels are key factors that drive the release of trace elements into the flue gas [129]. The oxygen carrier-assisted combustion technology can change the conversion path of fuel nitrogen under specific conditions [128]. For SO₂, it can be desulfurized in the furnace by adding limestone or utilizing calcium and iron components in the fuel ash [130,131,132]. The emission characteristics of particulate matter differ from those of grate furnaces, and their generation and alkali metal volatilization behavior are closely related. They need to be controlled through optimized operation and efficient dust removal.

The fluidized bed typically achieves high combustion efficiency, which is attributed to its good gas–solid contact, long fuel residence time, and flexible operation control [121,123,124,126]. Efficiency is influenced by the combined effects of fuel characteristics, bed material properties, and operating parameters [125,130,133]. From an economic perspective, the initial investment of the system is usually higher than that of grate furnaces [119], but the high combustion efficiency and lower initial pollutant emissions help to reduce operating costs [124,134]. This technology has loose requirements for fuel pre-treatment and can directly utilize various inexpensive waste materials. The application of co-combustion technology can further optimize fuel economy [119,134,135]. System energy consumption mainly concentrates on the air supply and material circulation units. System integration optimization can improve energy efficiency [134]. Maintenance costs are related to bed material agglomeration and heat surface ash deposition issues, which need to be managed through optimizing bed material selection and process control [123,124].

4.3. Multi-Dimensional Comparative Analysis of Grate Furnace and Fluidized Bed

The grate furnace and the fluidized bed are the two main types of biomass combustion furnaces. They have significant differences in terms of applicable fuel characteristics, combustion performance, pollutant emissions, and energy consumption costs.

Table 6. Comparison table of biomass combustion furnace type selection.

Comparison Item	Circulating Fluidized Bed Boiler	Grate Furnace	Ref.
Physical properties of applicable fuels	Fuel with small, uniform particles and good fluidity	Large particle fuels with uneven size, irregular shape, and poor flowability	[109,119]
Applicable fuel chemical characteristics	High alkali metals, high moisture content, mixed fuel	High-volatility fuel	[109,115,131,132]
Combustion efficiency	Usually very high. Due to the long residence time and good mixing of particles, the carbon content of fly ash is low. The carbon conversion rate can exceed 96.85%.	After optimization, the carbon content in the slag can be reduced to about 10%. The combustion efficiency is closely related to the operational level, and after optimization, it can approach the CFB level.	[109]
NOx emissions	Intrinsic low NOx characteristics. Low-temperature combustion reduces thermal NOx. The air stratification inside the furnace is conducive to creating a reduction zone to suppress fuel-type NOx. Pressurized co-combustion can significantly reduce the conversion rate of NO.	Adopting decoupling combustion, GRB and other technologies can achieve ultra-low emissions (50–60 mg/m3)	[109,115,131]
CO emissions	Usually low. Due to the intense mixing effect, the combustion is complete.	Relying on an optimized design to ensure the combustion of volatile matter. Advanced technology can control CO at a level of 35 mg/m3.	[115]
SO2 emissions	Desulfurization can be directly carried out in the bed by adding inexpensive desulfurizers (such as limestone), which is highly efficient. When biomass is co-burned with coal, there is a synergistic effect of self-sulfur fixation.	Desulfurization capacity inside the furnace is limited and mainly relies on the purification of tail flue gas	[131]
Equipment wear and tear	High-speed particle flow may cause wear on the heating surface, so it is necessary to optimize the design (such as the secondary air angle) and select wear-resistant materials.	Structure is relatively simple, and the wear issue is not prominent	[136]
Investment cost	System is complex and requires a high initial investment, especially for large CFB units	Structure is relatively simple, and the initial investment is usually lower than that of CFB boilers with the same capacity	[119]
Operating cost	Wide adaptability of fuel, can use low-priced and low-quality fuel, good environmental performance, may reduce the cost of end-of-pipe treatment	Easy maintenance and low operating costs when using suitable fuels	[119]
Biomass fuel consumption per unit of electricity generation	Due to its high heat transfer efficiency and advantages in large-scale production, the power generation efficiency may be slightly higher when burning medium-quality biomass.	Related to fuel calorific value and net efficiency of power plants, the gap between high-efficiency grate boilers and CFB is narrowing	[134,137]

provides a comprehensive comparison to offer a reference basis for the selection and optimization of furnace types in practical applications.

From the comparison results, it can be seen that the core advantages of the grate furnace lie in its wide fuel adaptability, low initial investment, and no need for complex pre-treatment. It is suitable for processing large particles, high moisture content, and complex component biomass fuels, especially for small- and medium-sized biomass combustion projects and scenarios with unstable fuel supply [108,109,113]. However, its combustion efficiency and pollutant control effect are relatively poor. During operation, measures such as optimizing fuel distribution, air supply mode, and grate structure need to be taken to improve performance [31,107,115].

The fluidized bed, on the other hand, has the main advantages of efficient combustion, good pollutant control effect, and flexible co-combustion capability. It is suitable for large-scale biomass power generation and heating projects, especially for scenarios with stable fuel supply and strict environmental protection requirements [121,124,134]. Although its initial investment is higher, the economic and environmental benefits during long-term operation are more significant [119]. Moreover, the fluidized bed has greater potential in the integration of advanced technologies such as oxygen-fuel combustion and carbon capture, and is an important direction for the low-carbon development of biomass energy in the future [134,138].

The wide adaptability of grate furnace technology is essentially reflected in its adaptability to the physical form differences in fuels. This characteristic leads to the combustion efficiency and pollution control being highly dependent on operational parameters such as fuel distribution and air supply, and it is difficult to achieve a uniform and efficient mass and heat transfer process. In contrast, the adaptability of fluidized bed technology stems from the utilization and transformation of the chemical properties of fuels. The strong mixing mechanism and uniform temperature field inside it form the intrinsic basis for efficient and clean combustion. However, the realization of this advantage strictly depends on pretreatment processes such as crushing and screening to ensure that the physical form of the fuel meets the requirements of fluidization. Therefore, the core difference between the two technical routes lies in a specific choice. One option is grate furnace technology. This technology can accept a rough physical form of raw materials. However, it suffers from poor performance stability and has limited optimization potential. The alternative is fluidized bed technology. This technology imposes strict entry standards for the physical form of the fuel. Yet, if its requirements are met, it can deliver better performance. It also provides higher operational stability and stronger predictability. This choice essentially reflects a deep trade-off between project scale, stability of the fuel supply chain, environmental protection standards and economic benefits. Grate furnaces can be economically efficient in certain applications. This efficiency is applicable to small-scale, intermittent waste disposal projects. In such projects, fuel is often free or even has a negative cost. In contrast, fluidized bed systems hold a more convincing economic advantage. This is true for large-scale, continuous operation projects. These projects typically require fuel procurement and involve power generation or heating. Furthermore, they are often subject to strict environmental regulations.

4.4. Exploration of Cutting-Edge Furnace Types Based on New Combustion Principles

In addition to the structural improvements of the mainstream furnace types, the cutting-edge combustion technology based on new reaction principles provides an innovative approach to solving the efficient and clean conversion of biomass and waste. Among them, CLC technology [139,140,141] overcomes the high energy consumption of gas separation processes in traditional combustion or post-combustion capture by fundamentally changing the reaction pathway. This technology is based on the cyclic oxidation–reduction principle of chemical reaction engineering. Its core utilizes lattice oxygen from metal oxide oxygen carriers (OC) as the oxygen source. For example, Huber et al. [137] studied a system featuring a fuel reactor (FR). In this reactor, the fuel first undergoes gasification. The resulting gasification products then contact oxygen carrier particles and are oxidized. This process achieves a hierarchical release of the fuel’s chemical energy. It also accomplishes the internal separation of CO₂. Another important innovation is the integrated technology of gasification and combustion, which is designed based on the optimization of thermodynamic equilibrium and reaction kinetics, aiming to enhance the efficiency of the entire energy conversion chain. Liu et al. [142] developed a three-dimensional numerical model of an industrial-scale downward-flow fixed-bed gasifier for poultry manure fuel, systematically studying the feasibility of its oxygenation. The research shows that in atmospheres of Oxy-21 (21% O₂/79% CO₂) and Oxy-30 (30% O₂/70% CO₂), stable operation of industrial-scale gasifiers can be achieved, while a pure CO₂ atmosphere makes it difficult to maintain a stable self-heating gasification process [142]. These technologies fundamentally alter the conversion path of the fuel. They also change the distribution and transfer efficiency of energy and matter within the reaction space. This encompasses volatile matter, coke, and pollutant generation and transfer. The changes are achieved by precisely regulating the reactivity of the oxygen carrier. They also optimize the gasifier agent composition, such as the O₂/CO₂ ratio. Furthermore, they control the reactor’s temperature field and pressure. This approach opens new possibilities for deep decarbonization. It also paves the way for improving the efficiency of biomass energy utilization.

4.5. Mechanism of Combustion Pollutant Generation and Whole Process Control Strategy

4.5.1. Furnace Control and Chemical Kinetics of Low-Nitrogen Combustion

The internal control strategy for low-nitrogen combustion relies on a core chemical principle. This principle involves creating a “fuel-rich reduction zone.” Within this zone, a strong reducing atmosphere is established. Specifically, the equivalence ratio in this zone is maintained at less than 1. Under these conditions, the fuel-bound nitrogen is directed toward conversion into N₂. The result is that less nitrogen is converted into NO_x [121]. The nitrogen in the fuel is mainly released in the pyrolysis stage in the form of HCN and NH₃ [143]. These precursors undergo homogeneous reduction reactions with the carbon–hydrogen free radicals generated during pyrolysis in the reduction zone [144,145], as shown in reaction Equations (1) and (2).

HCN + CH₃ → NCN → NCO → N₂

(1)

NH₃ + OH → NH₂ → NNH → N₂

(2)

The kinetic characteristics of these reactions, such as the relatively high activation energy of the NCN → NCO step, determine that the reduction zone needs to maintain a temperature of 1100–1300 K to ensure efficient conversion [121,146]. To achieve accurate predictions of denitrification performance, computational fluid dynamics simulations must be coupled with detailed chemical mechanisms. Alternatively, highly reduced simplified mechanisms can also be used. These mechanisms must be capable of accurately resolving the aforementioned complex reaction network [147,148]. Simulation analysis shows that the denitrification efficiency of the reduction zone is highly sensitive to the local equivalence ratio and gas residence time. For example, consider conditions where the local equivalence ratio is maintained at 0.75–0.85. When the gas residence time exceeds 0.35 s, the conversion selectivity of fuel nitrogen to N₂ can reach 90%. Conversely, if the gas residence time is less than 0.2 s, the subsequent burnout zone will oxidize unreacted active intermediates. This oxidation leads to significant deviations in NO_x emission predictions [148,149].

This fine-scale chemical kinetics regulation imposes high demands for multi-scale and multi-physical-field coupling in numerical simulations. Firstly, a non-isothermal model needs to be adopted at the particle scale to provide real-time release rates of HCN/NH₃ during pyrolysis, serving as the key boundary condition for the gas-phase reaction network [56,150]. Meanwhile, the coupling simulation of CFD and DEM reveals that the heat conduction within large-sized particles leads to a lag effect in HCN release, which, if not corrected in the macroscopic model, can result in NO emission prediction errors as high as 40% [151,152]. Additionally, radiation heat transfer (such as simulation using a DO model) affects the reduction reaction rate by influencing the particle surface temperature [153,154]. Therefore, a CFD model must reliably predict the denitration effect of staged combustion. This requires the integration of detailed nitrogen chemical mechanisms and advanced turbulence–chemical reaction interaction models. Furthermore, the model must comprehensively consider the coupling effects of particle-scale dynamics, multiphase flow, and radiation heat transfer [147,148,149].

Transforming these chemical concepts into engineering reality has given rise to an integrated design paradigm for furnace structure and pollutant control. The core logic is to “solidify” the chemical processes such as staged combustion and re-combustion through innovative geometric structures and operating parameters within the spatial and temporal dimensions of the combustion device [121,145]. In grate furnaces, the primary air supply can be reduced. This creates fuel-rich volatile matter. “Z-shaped” or “J-shaped” furnace arches are designed in the upper part of the furnace. These arches force the volatiles towards the high-temperature recirculation zone. This design significantly prolongs the volatile matter’s residence time. The prolonged residence occurs within the reducing atmosphere. This furnace arch is essentially a “chemical reactor” that regulates temperature, oxygen concentration, residence time, and free radical pools in multiple dimensions. In circulating fluidized bed (CFB) boilers, a coordinated strategy is employed. This strategy involves the downward tilt of secondary air nozzles and the assistance of oxygen carriers, such as Fe₂O₃. It aims to form a rich fuel wall flow on the dense phase zone wall. The lattice oxygen released by the oxygen carrier is then used. This oxygen selectively oxidizes CO. Consequently, this process enhances the carbon conversion rate. Simultaneously, it inhibits the formation of NO_x [127,136]. For high-sodium fuels like chicken manure, compact castable refractory combustion chambers are employed. These chambers feature an integrated design. They utilize the heat storage–radiation coupling of the refractory materials. This system allows for precise control of the combustion temperature below 850 °C. Such control can inhibit the catalytic effect of sodium chloride on NO_x formation. It also helps to avoid slagging.

These cases collectively demonstrate that the development of modern low-nitrogen combustion technologies has transcended simple air-fuel optimization and entered a deep integration stage of “structure as control”. The design essence translates the chemical requirements of low-nitrogen combustion into specific geometric, material, and flow field languages. This applies to the grate furnace’s special furnace arch, the CFB’s downward tilt nozzles coupled with oxygen carriers, and special refractory combustion chambers. Consequently, the furnace structure itself becomes an active “reactor.” It intervenes in the spatial and temporal conversion path of fuel nitrogen. The ultimate goal is to synergistically optimize efficient combustion and achieve ultra-low emissions at the source [110,121,127,145].

4.5.2. Collaborative Control from Furnace Optimization to Backend Purification

The selection of biomass flue gas treatment technologies should primarily consider the diversity of fuel chemical composition to achieve coordinated control from the furnace interior to the end-stage purification. In terms of nitrogen oxide control, the choice of technical path is highly dependent on the nitrogen and chlorine contents of the fuel (unless otherwise specified, elemental content refers to the received basis mass fraction). For lignocellulosic fuels with high nitrogen (N > 1.5%) and low chlorine (Cl < 0.1%), the “air stratification and selective non-catalytic reduction (SNCR)” combined process is an economically efficient option. This process typically achieves a 35–55% NOx removal efficiency. The efficiency is attained by injecting the reducing agent within a temperature range of 850–1000 °C [155]. However, when dealing with fuels with high nitrogen and high chlorine (Cl > 0.3%, such as straw), the SNCR process is susceptible to alkali metal influence, resulting in increased ammonia escape. In this case, the “fuel stratification and high-temperature selective catalytic reduction (SCR)” technology should be adopted. Anti-poisoning catalysts are placed in the tail-end flue gas temperature range of 300–400 °C. This configuration achieves a nitrogen removal efficiency of over 90% [146,156]. However, the catalyst lifespan is shortened by the effect of alkali metals. Consequently, the investment cost increases accordingly [146,156].

In terms of particulate matter removal, the choice of technology is closely related to the ash content characteristics and alkali metal content of the fuel. For wood fuels with low ash content (<5%) and ash composition mainly composed of silicon and aluminum, low-temperature electrostatic precipitators (ESP) can achieve a good dust removal effect and low operating power consumption [157,158,159]. For fuels such as rice husks and wheat straw with high ash content (>8%) and high alkali metals (based on fly ash mass fraction, >15%), the high specific resistance of their fly ash can easily cause ESP back corona. In this case, the “countercurrent bag filter (BF) combined with activated carbon injection” process is more suitable, which can not only achieve lower particulate matter emissions, but also synergistically remove dioxins [157,158,159]. For fuels with a high chlorine content (Cl > 0.5%) and a significant sulfur dioxide concentration, a wet electrostatic precipitator (WESP) can be highly effective. It is typically connected in series after a bag filter. This configuration effectively removes condensable particulate matter, hydrogen chloride, and enables associated oxidation and mercury removal [157,160]. However, the wastewater generated by this process requires proper treatment.

For the control of sulfur dioxide, due to the low sulfur content of biomass itself (S < 0.1%), there is generally no need for specialized desulfurization. However, when the sulfur content increases (S > 0.3%) due to the co-firing of waste, the semi-dry desulfurization process can be used to meet the standard for sulfur dioxide emissions by injecting calcium-based or sodium-based absorbents [161].

Based on the above technical characteristics, a systematic selection decision logic can be summarized. The core lies in first accurately measuring the N, S, Cl, and ash content of the fuel, and then matching them according to the target emission limit, equipment scale, and economic considerations. For small and economically sensitive projects, a certain combination demonstrates good cost-effectiveness. This combination is “air stratification + SNCR + countercurrent BF” [146,155,162]. For large units requiring ultra-low emissions, a different combination provides more reliable protection. This configuration is “fuel grading + high-temperature SCR + BF + WESP” [146,155,162]. This collaborative control strategy based on fuel characteristics is the key to achieving efficient and low-cost purification of biomass combustion flue gas.

4.6. Summary and Discussion

This chapter conducts a system analysis of mainstream biomass combustion furnace types. The analysis reveals a core difference in the choice of technical paths. This difference lies in the strategies adopted to deal with fuel complexity. One strategy is to adopt technical solutions that can adapt to the inherent heterogeneity of raw materials. The alternative strategy is to change fuel properties through pre-treatment. This adaptation aims to meet the requirements of a high-performance combustion system. The grate furnace improves its adaptability to the original biomass through precise design, but its non-uniform combustion mode limits the potential for source control of pollutants and the integration with future BECCS. The fluidized bed and chemical loop combustion establish the physical foundation for achieving efficient, clean, and carbon capture-friendly combustion through intense mixing and uniform reaction fields. However, this comes at the cost of moving the technical challenges to the fuel pre-treatment stage, achieving the standardization and predictability of combustion control.

5. Frontier Trends and System Integration of Biomass Combustion Technology

5.1. Intelligent Combustion Based on Digital Twin and AI Physical Hybrid Model

The modeling and optimization of biomass combustion systems face a practical contradiction. High-fidelity mechanistic models can deeply reveal the essence of the process. However, their high computational costs make it difficult to meet the requirements for real-time control and rapid optimization [13]. Furthermore, the experimental data available for verifying these models are often limited [163]. Therefore, this section explores an integrated method within the digital twin framework [164]. This method combines physical mechanisms with data-driven approaches. The aim is to construct a computationally efficient surrogate model. This model must also be physically reliable. Its purpose is to support the online application of the system. To achieve the transition from high-precision offline simulation to high-real-time online application, developing AI-physical hybrid models based on digital twin is a key approach to achieving high-fidelity real-time prediction. This model combines mechanism depth and data efficiency, and can provide the core computing engine for subsequent dynamic optimization and control. On this basis, a closed-loop optimization and predictive intelligent control system can be further constructed to effectively address the challenges of fuel characteristic fluctuations and multi-objective optimization requirements in actual operation.

5.1.1. High Fidelity Real-Time Prediction and Model Generalization

Achieving high-precision real-time prediction for biomass combustion systems requires strong model generalization. This necessitates addressing dual challenges. The first challenge is the high computational cost of high-precision mechanistic models. The second challenge is the insufficient credibility of pure data-driven models [18,72,165]. The “hybrid modeling” paradigm developed in recent years provides an effective path for this [166,167,168]. The core of this method is an intelligent agent model driven by both “mechanism” and “data”. First, it uses high-fidelity computational fluid dynamics (CFD) simulation to systematically scan key parameter spaces. These parameters include fuel type, moisture content, particle size, air distribution, and load. This process generates high-dimensional datasets covering a wide range of operating conditions. The datasets include information such as temperature fields, component concentration fields, and conversion rates [24,165,168,169,170]. Subsequently, this dataset is used for offline training of deep neural networks. Physical conservation constraints, such as mass, energy, and momentum residuals, are embedded in the loss function. Through this process, the complex physical and chemical mechanisms revealed by CFD are implicitly encoded into the network weights [53,165,171,172]. These mechanisms include alkali metal migration, coke morphology evolution, and the coupling of turbulence and reaction. The trained agent model can complete intelligent calculations within milliseconds to seconds. Its prediction accuracy is comparable to that of high-fidelity CFD models. Furthermore, it demonstrates superior generalization ability to unseen fuels or operating conditions. This performance surpasses that of traditional empirical correlation methods [165,171,172].

This hybrid model can be directly embedded into the online system. It takes real-time feeding characteristics and operating parameters as inputs. The model outputs the three-dimensional state field of the furnace, such as temperature and NO_x distribution, in seconds. This output thereby drives real-time optimization control of the air valve, feeding, and re-ignition systems. The process achieves a significant reduction in pollutant emissions while improving combustion efficiency [72,165,173]. Its value is twofold. It not only compresses computational time from hours to seconds, but also integrates the physical depth of mechanistic models with the computational efficiency of data models. This integration is achieved through a specific technical link: “CFD generates data, neural network distillation, and online continuous fine-tuning” [174]. This method can also leverage transfer learning and other technologies to enable stable generalization of a single agent model across different capacity boilers and multiple fuels [18,165,168,175]. Its core particularly incorporates the physical mechanism of the high-fidelity model. This incorporation enables reasonable trend extrapolation, even under extreme working conditions with scarce experimental data. It can predict slagging tendencies or pollutant generation paths. The reliability of this approach far exceeds that of pure black-box models [18,72,176]. Therefore, the hybrid modeling approach successfully bridges a critical gap. It connects high-fidelity offline simulation with high-real-time online applications. This approach provides an expandable and interpretable intelligent core. This core is essential for building a biomass combustion digital twin system. The system features real-time closed-loop optimization capabilities [18,72,165,169,172].

5.1.2. Closed-Loop Optimization and Predictive Intelligent Control

Modern biomass combustion systems face dual challenges: the “dual carbon” target and the high volatility of fuels. In response, these systems are evolving. They are moving away from the traditional “feedback-regulation” mode. The evolution is towards an intelligent stage characterized by “closed-loop optimization-predictive intelligent control” [7,177]. The core is to build an intelligent framework. This framework is centered on a digital twin and uses millisecond-level agent models as its computing engine [178]. It achieves instantaneous trade-offs among conflicting objectives, such as combustion efficiency versus pollutant emissions. Furthermore, through online data assimilation, the framework enables the model’s lifelong learning and adaptability [165,179]. The realization of this intelligent control closed-loop begins with the construction of high-fidelity agent models. Advanced network architectures are trained on multi-holistic datasets. These datasets can include coupled two-dimensional and high-precision three-dimensional simulation fields [177]. The trained model can rapidly reconstruct the three-dimensional combustion field. This reconstruction achieves extremely low errors [180]. The model also transfers its learned shared features to other prediction tasks. These tasks include forecasting key scalars, such as NOx concentration. Consequently, the model can generate high-precision response surfaces. These surfaces cover all operating conditions. The entire generation process is completed within milliseconds [165,181]. Subsequently, this response surface is coupled with multi-objective optimization algorithms, such as NSGA-III [182]. This coupling enables the search for Pareto-optimal solutions satisfying constraints like complete combustion rate and ultra-low NOx emissions. The search is performed within an extremely short time. The system then extracts the corresponding optimal operating parameters, which include ratios for primary air, secondary air, and fuel quantity. These parameters are issued to the distributed control system for execution via standard industrial protocols, such as OPC UA. This process forms an optimization closed-loop [165,183].

To ensure the robustness of this closed-loop system in long-term operation in response to the slow drift of fuel characteristics (such as changes in moisture and alkali metal content), a real-time data assimilation mechanism needs to be introduced [184]. Distributed high-frequency sensing arrays, such as distributed optical fiber temperature measurement and tunable diode laser absorption spectroscopy, can be deployed. This deployment enables the continuous acquisition of real data, including furnace cross-sectional temperature and key gas component concentrations. The key parameters of the agent model can then be online fine-tuned. This tuning utilizes ensemble Kalman filter—variational hybrid assimilation frameworks [173,179,185]. This dynamic calibration enables the digital twin to adaptively extrapolate [186,187], maintaining the long-term accuracy of the prediction and optimization chain. In practical applications, this mechanism has demonstrated significant value, effectively reducing slagging rates and simultaneously improving combustion efficiency and reducing the average NO_x emissions over a period of several months [72,165]. Physical information neural networks are deeply integrated with computational fluid dynamics frameworks [174,188]. In this integration, key mechanism equations, such as those for alkali metal migration–deposition, are embedded as soft constraints during the agent model’s training process. This approach can significantly improve the model’s physical consistency. It also enhances the model’s extrapolation credibility under unseen conditions. Consequently, it promotes the intelligent upgrade of the core combustion model. The upgrade shifts the model paradigm from “offline high precision” to “online high credibility” [165,189,190]. This complete technical system provides a replicable digital core. The system encompasses rapid prediction, multi-objective real-time optimization, and online adaptive learning. It is specifically designed to build the next generation of intelligent biomass combustion systems. These systems are characterized by high efficiency, operational flexibility, and ultra-low emissions.

The advantages of AI hybrid models are not unconditional superiority over empirical models. Instead, they can integrate complex physics more efficiently and demonstrate potential in highly data-concentrated scenarios only when there is sufficient, broad and reliable high-fidelity data. For new fields with scarce data, their values may actually be lower than those of a theoretically simplified model constructed based on a small number of key experiments.

5.2. Collaborative Design of Fuel Adaptability and Negative Carbon System for BECCS

Driven by the “dual carbon” goal, the value orientation of biomass combustion systems has evolved from “high efficiency and cleanliness” to a three-dimensional synergy of “high efficiency-cleanliness-negative carbon”. BECCS are the key system levers to achieve this goal [191,192,193,194]. The economic feasibility of these systems depends on multiple factors. It relies not only on the performance of the carbon capture unit, but also on the main combustion process. This process must continuously provide flue gas characterized by “high concentration, low impurities, and stable flow.” Simultaneously, it must cope with the high volatility of the fuel. Achieving these conditions reduces the capture energy consumption. The reduction must bring it to an economically acceptable range [191,192,193,195]. The traditional “boiler + end capture” mode is a simple superimposition. In biomass combustion, the flue gas typically has a low CO₂ concentration (12–16 vol%) and high moisture content (>20 vol%). This often leads to huge regeneration energy consumption in the amine absorption capture system, ranging from 3.6 to 4.2 GJ/t CO₂. Consequently, it significantly increases the levelized electricity cost and weakens the negative carbon benefits [192,196,197]. Therefore, the new generation of systems must adopt the pre-design logic of “fuel-process-capture” collaboration [198].

The fuel–process integration design for BECCS requires deep collaboration between pre-treatment and combustion reactors. Taking baking pre-treatment as an example, straw is treated at 280 °C under anoxic conditions. This treatment decreases its O/C ratio and reduces the volatile matter/carbon ratio. Consequently, the peak combustion temperature shifts backward. Furthermore, the concentration of CO₂ at the furnace outlet increases to approximately 20 vol%. At the same time, it reduces the water vapor content and decreases the total flue gas volume [192,199,200]. The increase in flue gas CO₂ concentration can significantly reduce the regeneration heat load of the subsequent chemical absorption method, thereby greatly reducing the system efficiency penalty and negative carbon marginal cost [191,193,201]. Moreover, baking can reduce the alkali metal content in the fuel. This alleviates the negative effects, such as oxygen carrier deactivation, caused by alkali metals in the subsequent “combustion capture” technology route. Importantly, baking maintains the flexibility of technical choice [202,203,204].

For “combustion capture” technologies such as CLC or pressurized oxygen-enriched combustion, their negative carbon potential is more significant [205]. CLC technology transfers oxygen through an oxygen carrier, and the inherent flue gas at the fuel reactor outlet is a mixture of high-concentration CO₂ and water vapor. After condensation, it can be directly compressed and sealed, avoiding the high-energy-consuming adsorbent regeneration process, with a system efficiency penalty much lower than that of post-combustion capture [202,206,207]. However, high-alkali metal biomass ash is prone to form low-melting-point eutectics with the oxygen carrier, leading to bed material agglomeration and deactivation. Studies have shown that by adding additives such as kaolin to fix the alkali metals in situ, it can effectively maintain high capture efficiency and oxygen carrier activity [202,208]. In a pressurized oxygen-enriched combustion circulating fluidized bed system, the high-moisture biomass combustion flue gas contains latent heat. This latent heat is utilized through efficient heat exchange and recovery. The recovered heat is used to preheat the reaction gas. This process improves the flame temperature and combustion efficiency. It also reduces the compression power consumption of the air separation unit. Consequently, the system achieves a higher net efficiency. Furthermore, it leads to more competitive negative carbon costs [209,210,211].

It must be emphasized that the negative carbon effectiveness of any technical route must be judged through strict life-cycle assessment (LCA) [212]. The upstream stages such as raw material collection, transportation, and pre-treatment may generate significant “carbon leakage”, which is sufficient to offset the capture benefits at the end [206,213,214]. For example, an excessively large raw material collection radius or an excessively high straw removal rate may cause the carbon emissions throughout the life cycle to shift from negative to positive [213,214,215]. Therefore, the fuel adaptability design and optimization of the combustion–capture process must be developed through iteration. This iteration must be done in conjunction with models of geography, logistics, and soil carbon dynamics. Furthermore, life-cycle assessment (LCA) must serve as a rigid constraint for this entire process. This integrated approach is essential to ensure that BECCS genuinely contributes to the carbon neutrality goal. Its contribution must be validated on a full life-cycle scale [192,213,216,217]. Comparison of different technology routes for biomass combustion coupled with BECCS is shown in

Table 7. Multi-dimensional comparison table of biomass combustion coupled with BECCS technology route.

Comparative Dimension	Post Combustion Capture (MEA Method)	CLC	Capture During Combustion—Pressurized Oxygen Enriched Combustion (POxy CFB)	Distributed Modularization CLC-BECCS
Technical concept and maturity	Decoupling combustion and capture. The most mature technology is suitable for power station renovation.	Intrinsic separation. Replacing air with oxygen carriers produces high concentrations of CO2 during combustion. In the pilot demonstration stage.	Process reinforcement. Increase the concentration of CO2 in flue gas by pressurized oxygen enrichment. In the pilot stage.	Full chain integration. Modular coupling of pre-processing, combustion, and capture. It belongs to the concept of forward-looking systems.
Key performance trade-offs	Capture energy consumption is high (~3–4 GJ/t CO2), resulting in significant efficiency penalties (7–12%).	Capture energy consumption is the lowest and the efficiency penalty is small (2.5–4.0 percentage points).	Net efficiency of the system can lead, but it is limited by the high energy consumption and investment of the air separation unit.	Lowest carbon intensity throughout the entire lifecycle, but the economic feasibility of distributed scale remains to be verified.
Fuel adaptability core constraint	Relying on fuel baking pretreatment to improve economy, sensitive to fluctuations in raw materials.	Highly sensitive to alkali metals (K) in fuel, requiring additives (such as kaolin) to modify and prevent oxygen carrier deactivation.	Especially suitable for high moisture fuels (such as sludge), it can utilize latent heat.	Strictly control the fuel collection radius to maintain soil carbon balance and low-carbon transportation.
Core strengths	Compatible with existing facilities, with clear renovation paths. Low technical risk.	Capture the lowest energy consumption and cost potential. Avoid using chemical absorbents.	System has high net efficiency. Can handle difficult-to-dispose-of fuels.	Risk of carbon leakage throughout the entire chain is minimized. Modular deployment is flexible.
Main challenges	Solvent degradation and equipment corrosion. Preprocessing increases complexity and cost.	Long-term activity and anti-poisoning ability of oxygen carriers in real flue gas. Reactor amplification and wear.	High investment and energy consumption of air separation units. Materials and safety challenges under pressurized oxygen-enriched conditions.	Lack of economies of scale. Relying on precise, sustainable supply chains and regional ecological management.
Strategic positioning and applicable scenarios	Rapid BECCS retrofit of existing power plants; Large centralized projects.	Optimal route for building a dedicated negative carbon power station; Suitable for low alkali metal woody biomass.	Processing high moisture and difficult-to-pretreat waste biomass; A project that demands the utmost efficiency.	Negative carbon solutions for dispersed areas of biomass resources; Ecological projects that pursue full chain carbon credibility.

.

5.3. Closed-Loop Management of the Entire Chain from Fuel to Ash

5.3.1. Prediction of Ash Characteristics and Combustion Feedforward Optimization

To achieve the true sustainability of the biomass energy system, the final solid product—ash—must be transformed from a “terminal burden” to a “resource entry” [203,216,218,219]. The production and characteristics of ash are determined by multiple factors. These characteristics include chemical composition, mineral phase, and fusibility. The fuel type is a primary factor, such as wood, herbaceous, or waste [216,220]. The combustion technology is another determinant, for example, grate furnace, fluidized bed, or oxygen carrier-assisted combustion. Operating conditions also play a crucial role. These conditions encompass temperature, excess air coefficient, and the use of additives [221,222,223]. It is rich in potential nutrients such as potassium and phosphorus, but it may also concentrate toxic heavy metals such as cadmium and lead [218,224,225,226]. The properties of ash must be accurately predicted and directionally regulated. Otherwise, high alkali metal ash will cause serious slagging and corrosion in the furnace. Concurrently, ash enriched in heavy metals may be classified as hazardous waste. This classification leads to a significant increase in disposal costs. These increased costs can offset the negative carbon benefits of the system [193,224,227,228,229].

Therefore, the future direction is to establish a precise resource utilization path based on the chemical–mineralological “fingerprint” of ash, rather than seeking universal recycling technologies [203,218,230,231]. This requires the combustion system to have the ability to predict and actively regulate the characteristics of ash. A feasible approach is to build a digital twin framework of “fuel–operating conditions–ash–product” multi-dimensional data coupling. The underlying layer utilizes high-fidelity computational fluid dynamics coupled with thermodynamic equilibrium calculations to generate a “virtual ash” composition and phase database offline covering a wide range of fuels and operating conditions [216,218,230,232]. The middle layer utilizes residual neural networks and other machine learning methods. It distills lightweight proxy models from massive data. This achieves real-time, second-by-second prediction of key ash quality indicators, such as alkali metal content, sintering tendency, and heavy metal concentration [203,233,234]. The top layer then inputs these prediction results into multi-objective optimization algorithms. It adjusts operating variables, including primary air ratio, bed temperature, or additive injection volume. This performs feedforward control of the combustion process. The goal is to make the ash composition directly fall into a preset “target window.” This window is conducive to subsequent high-value utilization, like preparing potassium-rich fertilizers or silicon-aluminum-based materials, or safe disposal [203,233,235].

To ensure the long-term reliability of this framework, online laser-induced breakdown spectroscopy and other real-time ash content monitoring data can be introduced. The framework’s proxy model is then continuously fine-tuned through data assimilation algorithms. This adaptive process is designed to cope with the slow drift of fuel properties [178]. In this way, digital twins can be upgraded from “post explanation” tools to “feedforward closed-loop” intelligent agents. This evolution truly builds a bridge. This bridge connects the understanding of micro combustion mechanisms to macro-level decision-making regarding ash resource utilization. They can achieve directional control of ash characteristics at the source, providing core support for building a complete biomass energy closed-loop from “cradle” to “cradle” [193,236,237].

5.3.2. Integration of Ash Resource Utilization and BECCS Negative Carbon Value Chain

The resource utilization of ash is a key link to achieve the “carbon-nutrient” dual circulation closure of the biomass energy system and further enhance the overall efficiency of its negative carbon value chain [193,238]. Based on its distinct chemical “fingerprint,” ash can be directed to different high-value products. Rich silicon-aluminum ash can be transformed into geopolymer building materials. This transformation occurs through alkali fusion and hydrothermal synthesis. These materials can replace ordinary silicate cement, thereby reducing implicit carbon emissions [203,230,239]. Separately, high calcium ash can be used to improve acidic soils. It aids in fixing toxic metals while also enhancing soil productivity [225,233,240]. The full life-cycle assessment shows that when the on-site resource substitution rate of ash exceeds 35%, the overall carbon intensity of the system can be further reduced, and the negative carbon benefits can be significantly improved [216,221,241]. As shown in

Table 8. Optimization path and collaborative value of the “fuel combustion ash” closed-loop in the BECCS-oriented biomass combustion system.

Key Closed-Loop Processes	Core Technologies and Innovative Concepts	Core Results and Indicators	Synergistic Value of BECCS/Negative Carbon Systems
(1) Fuel end pre-processing closed-loop	Concept: Control the physical and chemical properties of fuel from the source to achieve “design fuel”. Technology: Baking pretreatment (~280 °C), regulating O/C ratio and alkali metal content.	The O/C ratio decreased from 0.7 to 0.4, and the energy yield was >90%. Alkali metal (K) decreases by 30–40%, reducing the risk of slagging.	1. Increase the concentration of CO2 in flue gas to ~20 vol% and reduce subsequent capture energy consumption by ~24%. 2. Preserve fuel flexibility for advanced technologies such as CLC.
(2) Intelligent closed-loop combustion process	Concept: “Structure is Control” and “Digital Twin + AI” dynamic optimization. Technology: Real-time prediction and optimization of air distribution using hybrid AI/physical models, precise control of ash chemical fingerprints.	Combustion efficiency > 92%, NOx emissions < 120 mg/Nm3. Prediction error of key components (K2O, etc.) in ash is less than 3%.	1. Producing “high concentration, low impurity, stable flow” flue gas is a prerequisite for reducing capture costs. 2. Reduce carbon leakage during the combustion process and reduce greenhouse gas emissions by 12–15% throughout the entire lifecycle.
(3) Precise directional sorting and closed-loop resource utilization of ash	Concept: Transforming ash from an “end of pipe treatment object” to a “starting point for resource-based design”. Technology: LIBS intelligent sorting; Classify and modify according to mineral composition (rich in potassium, rich in silicon and aluminum, high in calcium).	Utilization rate of ash resources is greater than 85%, and the landfill rate of hazardous waste is significantly reduced. It can selectively obtain potassium-rich ash or highly active calcium-based materials with K2O > 20%.	1. Avoid carbon sequestration failure caused by landfilling and ensure the authenticity of net negative emissions. 2. Modified ash can be used as an auxiliary material for flue gas purification, reducing the impurity load of the capture unit.
(4) System-level “soil biomass” carbon cycle closed-loop	Concept: Using agricultural soil as the ultimate carbon sink and feedback node to achieve full chain sustainability. Technology: Ash is used for soil improvement/carbon sequestration; Optimize straw removal rate (<35%) and crop rotation system.	The carbon sequestration of ash can reach 0.3 t CO2/t ash. Maintain or even enhance soil organic carbon and eliminate the negative impacts of indirect land use change (iLUC).	1. Provide additional geological/soil carbon sinks to enhance the negative carbon efficiency of BECCS. 2. Ensure net negative emissions of the system throughout its entire lifecycle (with a carbon intensity as low as −1.2 t CO2/MWh) and improve the overall economic efficiency of the entire chain.

, to achieve this goal, it is necessary to build a complete closed-loop optimization path covering “fuel pretreatment–intelligent control of combustion—ash resource utilization–soil carbon cycle”. This path systematically enhances the overall efficiency and economy of BECCS through the collaborative design of each link.

The systematic integration of ash resource utilization and BECCS technology can generate significant synergy effects [217]. The system can further utilize the captured low-temperature waste heat to drive the resource utilization transformation process of ash, achieving the “dual circulation” of energy and materials [218,230,242]. At the policy level, both the European Union and China have issued relevant regulations and guidelines. Their aim is to promote the market entry of biomass ash-based products that meet established standards. These products are intended for the fertilizer or building materials markets. This policy framework provides crucial institutional guarantees. It supports the high-value utilization of biomass ash [222,228,243,244].

Based on big data of ash characteristics, the combustion process can be optimized in real time using digital twin technology. This real-time optimization regulates ash quality. The optimized ash can then be precisely guided to multiple negative carbon pathways. These pathways include soil carbon sequestration and building material substitution, coupled with BECCS. This approach realizes the ultimate vision of a complete closed-loop operation for the biomass energy system. The system operates from “sustainable fuel” to “ash cycle management” [193,216,218,221,230].

The current mainstream method for controlling ash and slag is feedback control. The prerequisite for feedforward optimization is to conduct real-time, precise and online detection of the fuel composition entering the furnace. However, the existing online detection technologies are expensive, complex to maintain, and difficult to perform global representative measurements of the unevenly composed biomass on the feeding belt. Therefore, the biggest challenge for “feedforward optimization” is not the algorithm, but reliable, economical and real-time hardware for fuel property perception. Without reliable input, even the most advanced digital twins can only perform speculative feedforward. In contrast, although passive, using additive blending to broaden the fuel adaptability window is a more robust and more easily implementable ash and slag management strategy in engineering.

5.4. Summary and Discussion

Based on the in-depth discussion of BECCS in this chapter, the real bottleneck hindering its industrial-scale deployment is a complex network that integrates technology, economy, resources, and environmental integrity. Any single technological breakthrough is insufficient to drive its large-scale deployment. The core challenge lies in the lack of system integration and multi-dimensional coordination. At the technical level, the challenge has shifted from principle verification to long-term operational reliability at the industrial scale. The core lies in the durability of materials and the stability of the entire system. At the economic level, its core negative carbon value is difficult to monetize in the current market, lacking strong carbon pricing and subsidy mechanisms. The high additional costs significantly undermine the commercial feasibility of the project. More profoundly, its environmental benefits depend on a sustainable biomass supply chain. If the acquisition of raw materials triggers indirect land use changes leading to carbon leakage, it may offset its negative carbon gains. Therefore, a strict, full life-cycle carbon management framework must be established. Additionally, large-scale deployment is also constrained by the severe lack of carbon capture and storage infrastructure, which constitutes a key obstruction from capture to final storage. Thus, the development path for BECCS is not solely about pursuing breakthrough unit technologies. It must shift toward constructing a full-chain collaborative design encompassing “fuel–process–capture–storage.” This shift requires simultaneously advancing the systematic construction of supportive policies, market mechanisms, and essential infrastructure.

6. Conclusions

This article systematically reviews the research progress in numerical simulation, pollutant control, and system integration of biomass combustion technologies aimed at achieving ultra-low emissions targets. Currently, a multi-scale cognitive framework has been established in this field. It spans from molecular reaction mechanisms to reactor macroscopic performance. Furthermore, strategies for low-nitrogen combustion and pollutant co-control have been developed. These strategies target mainstream furnace types. The research shows that achieving efficient, clean, and even negative carbon emission utilization of biomass energy requires breaking through multiple technical bottlenecks from basic mechanism understanding to engineering system integration. The improvement of multi-scale numerical simulation methods provides an important tool for understanding the combustion process, but there are still challenges in predicting performance from microscopic mechanisms to macroscopic scales. Pollutant control strategies need to shift from end-of-pipe treatment to source regulation, and the feasibility of technologies such as BECCS fundamentally depends on the collaborative optimization of the combustion side and the capture side.

At the simulation methodology level, the improvement of the multi-scale modeling system provides a systematic framework for the study of the combustion process. At the single-particle scale, models have evolved significantly. Early models were simple, isothermal, contracting-core types. Modern models are high-fidelity and non-isothermal. These advanced models can analyze the temperature gradient within particles. They also simulate component diffusion. Furthermore, they incorporate the evolution of the real pore structure. At the particle group scale, the discrete element method effectively tracks individual particle movement and their interactions. This approach is well-suited for fundamental mechanism exploration. In contrast, continuum methods are more appropriate for large-scale, engineering-level simulations. These two methodologies are not mutually exclusive; rather, they complement each other. At the reactor and system scale, model development has focused on key areas. More refined models for turbulent–chemical reaction interactions have been created. Improved radiation heat transfer models have also been developed. Enhanced sub-models for pollutant generation are now available. This progress has collectively led to a significant improvement in predictive accuracy. Specifically, the prediction of the combustion flow field is now more precise. The forecasting of temperature distribution has become significantly better. The simulation of the pollutant generation process has also been greatly enhanced. In addition, coupling strategies that connect different scales are being increasingly emphasized. Model validation methods for systems are also receiving more focus. Similarly, uncertainty quantification frameworks are gaining greater attention. These efforts concentrate on solving problems of correlation and prediction, specifically from micro mechanisms to macro performance. This continuous work aims to improve the reliability of multi-scale models. It also seeks to enhance their engineering guidance value. These methodological advancements form an important bridge from fundamental cognition to technological optimization.

In the future, the development of biomass combustion technology will mainly focus on three core directions: intelligence, negative carbonization, and resource closed-loop. In terms of intelligence, the AI physical hybrid model deeply integrates digital twins, artificial intelligence, and real-time perception data. This model will become the key to achieving accurate prediction, dynamic regulation, and adaptive optimization of combustion processes. It promotes the evolution of the system towards an intelligent operation mode, characterized by flexibility, high efficiency, and ultra-low emissions. In terms of negative carbonization, the integration of BECCS technology routes is particularly important. Its efficiency improvement depends on a deep collaborative design and system optimization. This process spans various links, from fuel pretreatment and combustion control to carbon capture. It is not merely a simple assembly of technologies. The ultimate goal is to balance economic feasibility while maximizing negative carbon benefits. In terms of the resource closed-loop, the research boundary is expanding from a single energy conversion to the full material flow management of “fuel combustion ash resource”. A predictive model for ash characteristics can be established. Its directional conversion technology can also be developed. These advancements allow ash to be converted into high-value-added materials. These materials can then be coupled with other approaches, such as soil carbon sequestration and building material substitution. This coupling helps to construct a complete biomass carbon cycling system. The ultimate goal is to achieve a system upgrade from mere energy output to comprehensive material cycling.

Looking forward to the future, the development of biomass combustion technologies will rely on the deep integration of interdisciplinary methods. Materials science must develop anti-scaling coatings or high-stability oxygen carriers. Advanced sensing technologies are needed to provide real-time, multi-parameter measurement data. These data are essential to drive digital twin systems. Policy formulation must rely on full life-cycle carbon accounting. Furthermore, it must establish carbon market mechanisms. These mechanisms should incentivize slag resource utilization and BECCS deployment. Therefore, the advancement of biomass combustion technology requires coordinated progress in several key areas. First, multi-scale cognitive and simulation methods must be deepened. Second, intelligent control should be achieved through AI-physical integration. Third, a collaborative design for the entire fuel–burning–capture chain is essential. Fourth, policies and market mechanisms need improvement, based on full life-cycle assessment. These efforts will work together. Their goal is to promote biomass combustion technology to become a stable, clean, and economically competitive carbon-neutral solution within the future energy system.