Machine Learning and Hybrid Approaches in the Energy Valorization of Contaminated Sludge: Global Trends and Perspectives

Abstract

While the technological foundation for sludge valorization (anaerobic digestion and pyrolysis) is mature, a significant disconnect exists between traditional research and the advanced application of artificial intelligence. This study identifies that Machine Learning (ML) remains in a peripheral position, representing an untapped frontier for achieving predictive and circular systems. The methodology involved a quantitative bibliometric analysis of 190 Scopus-indexed documents (2005–2025). We analyzed indicators of scientific production, collaboration, and thematic evolution using Bibliometrix and VOSviewer 1.6.20. The results reveal a rapidly growing research field, predominantly led by Chinese institutions. The temporal analysis projects a productivity peak around 2033. Core topics include established technologies like anaerobic digestion and pyrolysis. However, network and keyword analyses reveal an emerging trend toward hydrothermal processes and, crucially, the early incorporation of ML. However, ML still occupies a peripheral position within the main scientific discourse, highlighting a gap between traditional research and the advanced application of artificial intelligence. The study systematizes existing knowledge and demonstrates that, although the technological foundation is mature, the deep integration of ML represents the future frontier for achieving sludge valorization systems that are more predictive, efficient, and aligned with the principles of the circular economy.

1. Introduction

The management of contaminated sludge from wastewater treatment plants and industrial spills constitutes one of the most pressing environmental challenges today [1]. This sludge often contains a complex mixture of organic matter, heavy metals, hydrocarbons, and other persistent contaminants that hinder its safe and sustainable disposal [2]. The environmental impact of inadequate management of such sludge translates into risks to public health, soil and water pollution, and greenhouse gas emissions [3,4]. Energy valorization of contaminated sludge offers a promising alternative. It transforms problematic waste into renewable energy, mainly via anaerobic digestion and biogas production [5,6]. However, the presence of contaminants interferes with the efficiency of biological processes, reducing methane production and generating toxic by-products [7]. Therefore, it is essential to explore new strategies that optimize the energy conversion of this waste while simultaneously ensuring contaminant reduction [8]. The use of advanced modeling and optimization approaches, such as machine learning and hybrid methods, offers a unique opportunity to address the complexity of these systems, allowing for behavior prediction, pattern identification, and the design of more efficient solutions [9,10].

Traditionally, the management of contaminated sludge has been addressed through conventional processes such as stabilization, incineration, landfill disposal, or physico-chemical treatments [11]. Anaerobic digestion has been one of the most widely used methods for energy valorization, as it enables the transformation of the organic fraction into biogas, mainly methane and carbon dioxide [12]. However, when sludge contains contaminants derived from industrial spills, such as aromatic hydrocarbons, organochlorine compounds, or heavy metals, process efficiency decreases significantly [13]. To counteract these limitations, pretreatments such as advanced oxidation (ozone, hydrogen peroxide, peracetic acid) have been applied. These methods have demonstrated improvements in methane production, although they present challenges related to energy consumption, operating costs, and the generation of secondary by-products [14]. For example, Adeleke and Jen (2025) proposed an XAI framework to optimize methane in industrial biogas, whose model improved performance by 18% using explainable clustering, emphasizing the importance of interpreting critical variables such as temperature and organic load [15]. In parallel, conventional mathematical models have been developed to describe the kinetics of anaerobic digestion and predict biogas production under different conditions [13]. However, such models often oversimplify the complexity of real systems, limiting their generalization capacity [14]. In recent years, advances in artificial intelligence and machine learning have overcome these limitations, providing tools capable of handling large volumes of data, identifying nonlinear relationships, and optimizing multifactorial processes [16]. Thus, the typical treatment of the problem has evolved from purely experimental approaches toward integrated strategies that combine conventional techniques with advanced computational methods [17]. For instance, Jeong et al. applied deep learning (CNN-LSTM) to predict biogas in organic co-digestion, achieving 94% accuracy and identifying key variables such as temperature and C/N ratio [18].

Bibliometric analysis has become an essential tool for understanding the evolution of research in this field and detecting emerging trends [19]. Tools such as VOSviewer and R Studio 3.6.0+ allow for the visualization of collaboration networks, identification of thematic clusters, and analysis of keyword co-occurrence [20]. By using robust databases such as Scopus, it is possible to obtain a comprehensive view of scientific production, highlighting the most influential authors, institutions, and countries [21]. This approach not only facilitates the systematization of existing knowledge but also guides future research toward areas with greater potential impact on the energy valorization of contaminated sludge [22].

The central thesis of this work is that the energy valorization of sludge faces a structural lag in digital transformation. Although the technological foundation is robust, the integration of advanced computational tools is not yet central to the scientific discourse. This research critically evaluates why this gap persists, using bibliometric data to identify barriers in international cooperation and thematic evolution that prevent ML from becoming a motor theme in the field. In this study, ‘hybrid approaches’ are specifically defined as strategies that integrate Machine Learning (ML) algorithms with: (1) mechanistic or first-principle models of bioprocesses (e.g., kinetics of anaerobic digestion) to create ‘gray-box’ models; or (2) the combination of distinct valorization technologies (e.g., thermal pre-treatment coupled with anaerobic digestion) where ML serves as an optimization and predictive tool for the integrated system. In the context of this study, ‘hybrid approaches’ are defined as the synergistic integration of Machine Learning (ML) algorithms with both mechanistic models (gray-box modeling) and physical-biological processes. This includes combining AI architectures with experimental techniques such as anaerobic digestion, pyrolysis, or hydrothermal carbonization to handle non-linear relationships and improve process efficiency.

2. Materials and Methods

We adopted a quantitative approach for this bibliometric analysis. It utilizes scientific data mining and thematic network visualization to examine ML and hybrid approaches in sludge energy valorization. The procedure was structured into four sequential phases, beginning with a systematic search in the Scopus database. The search was carried out on 15 November 2025, using an advanced strategy that combined relevant thematic descriptors through Boolean operators. The search structure was as follows: ((“machine learning” OR “ML” OR “algorithm” OR “artificial intelligence”) AND (“energy valorization” OR “energy recovery” OR “waste-to-energy” OR “resource recovery”) AND (“contaminated sludge” OR “waste sludge” OR “sewage sludge” OR “biosolids”) AND (“treatment” OR “management” OR “processing” OR “utilization”)). This strategy yielded an initial total of 389 documents. Subsequently, a multi-stage filtering process was applied, as shown in Figure 1. First, documents outside the established time range (2005–2025) were excluded, reducing the set to 251 records. Next, non-relevant document types (e.g., conference abstracts, editorial notes) were removed, leaving 208 articles. Finally, 18 documents not recognized by the Bibliometrix package were discarded, resulting in a final set of 190 documents included in the analysis.

The bibliographic data were processed using RStudio software (version 4.3.1), through the Bibliometrix package and its graphical interface Biblioshiny. These tools enabled the calculation of key indicators such as annual publication output, most influential authors, high-impact journals, geographic distribution of articles, and institutional and international co-authorship networks. Complementary metrics were also analyzed, including the h-index, m-index, average citations per document, and collaboration patterns (SCP and MCP). For the analysis of co-citation, co-authorship, and institutional collaboration networks, VOSviewer software (version 1.6.19) was employed. This tool generated scientific maps in which nodes represent bibliographic entities (authors, institutions, or journals), and links reflect bibliometric relationships. Node size corresponds to the number of occurrences or publications, while colors were used to identify thematic or collaborative clusters. Additionally, Plotly Studio 2.24.1 was used to develop interactive visualizations that allowed exploration of temporal dynamics, keyword evolution, and thematic distribution. These representations facilitated the identification of emerging trends, such as the integration of ML into anaerobic digestion processes, the optimization of thermal pretreatments, and the predictive modeling of biogas production. Finally, a qualitative analysis was conducted to contextualize the results within the framework of environmental sustainability, technological innovation in bioenergy, and the interdisciplinary evolution of knowledge in waste treatment. This integrated methodology not only enabled the quantification of field development but also identified knowledge gaps, potential synergies among key actors, and opportunities to guide future research toward cleaner, smarter, and more efficient energy solutions.

3. Bibliometric Results and Analysis

Figure 2 provides a comprehensive view of the temporal behavior of scientific production in the field, combining exponential and logistic models to characterize both the initial growth and the maturation of the domain. In Figure 2a, the annual and cumulative growth of documents between 2005 and 2025 is shown, where the exponential fit applied to the cumulative number of publications reveals a dynamic of accelerated expansionModel parameters and statistical indicators (R² = 0.9171, adjusted R² = 0.9414) confirm the fit’s validity. This suggests the field is consolidating, driven by artificial intelligence and the need for sustainable waste solutions. This trend is consistent with the behavior observed in Figure 2b, where the life cycle of annual publications is modeled through a logistic fit. The model predicts a production peak in 2033 with 58 publications, indicating that the field will reach its maximum productivity at that time before entering a natural decline phase. The R² value of 0.947 supports the accuracy of the model, while the projected curve allows anticipation of the future behavior of the research area. Finally, Figure 2c shows the cumulative growth curve with a logistic fit projecting saturation at 820 publications. The horizontal lines marking 99% and 90% saturation make it possible to visualize the degree of maturity reached and estimate the remaining time before the field stabilizes. The three models reflect a robust and predictable bibliographic development, where hybrid approaches and machine learning are consolidated as key tools for the energy valorization of contaminated sludge, with prospects for evolution toward interdisciplinarity, technology transfer, and methodological standardization. A study about on artificial intelligence in medicine reported a similar exponential growth rate (R₀ ≈ 3.2) for the period 2015–2021, attributing it to the convergence of disciplines and the massive increase in funding [23]. Our R₀ value is consistent with this trend, suggesting that the analyzed field is in a phase of “knowledge explosion.” A key finding, is that the adoption of generative AI tools has accelerated publication cycles, advancing peaks predicted by traditional models by approximately 1–2 years. This suggests that our projected peak of 2033.2 could occur slightly earlier (e.g., 2031–2032), a factor that requires monitoring [24].

Figure 3 presents the Sankey diagram that synthesizes academic relationships among universities, authors, and thematic lines. On the left side, leading institutions such as the Institute of Urban Environment, Ocean University of China, and China Electric Power University are identified, concentrating high scientific output in this field. These universities are linked to recurring authors such as Wang R, Zheng Y, and Chen Y, who act as key nodes in the generation of interdisciplinary knowledge. On the right side, the most prominent thematic lines include sewage sludge, anaerobic digestion, biochar, microalgae, and thermochemical conversion, evidencing a convergence toward valorization technologies that integrate biological, thermal, and electrochemical processes. This collaboration pattern suggests that Machine Learning is consolidating as a transversal tool to model, optimize, and predict the behavior of complex systems in sludge valorization [25]. Hybrid approaches, which combine artificial intelligence with techniques such as anaerobic digestion or thermochemical conversion, enable improvements in energy efficiency, reduction in contaminant loads, andadaptation of processes to variable conditions. The presence of terms such as biogas and waste treatment reinforces the circular and sustainable focus of these investigations [26]. The Sankey diagram reveals that sludge is a strategic resource for clean energy. It confirms that integrating ML and hybrid methods is shifting the waste treatment paradigm toward global sustainability goals [27]. Recent bibliometric studies, such as that of Zhang et al. (2023), confirmed that anaerobic digestion remains the most widely researched energy valorization technology for sludge worldwide [28], a consensus faithfully reflected in our thematic map. Likewise, the recent literature shows explosive growth in complementary or alternative topics. The highlights sludge pyrolysis and gasification to produce biochar and syngas, technologies aimed at overcoming the limitations of anaerobic digestion, such as low efficiency with certain substrates [29]. In addition, a point to a growing interest in nutrient recovery (phosphorus) and joint valorization with other wastes (co-digestion), topics not prominently represented in our original diagram, suggesting a possible evolution of the field [30]. Hybrid approaches in sludge valorization involve the use of artificial intelligence to optimize unit processes like anaerobic digestion and thermochemical conversion. These methods allow for real-time adaptation to variable operational conditions (e.g., organic load and temperature) by bridging the gap between purely experimental data and advanced computational prediction.

Table 1. Top Research Institutions by Publication Metrics.

Institution	Country	Publications	Total Citations	Average Citations	H-Index
North China Electric Power University (Baoding)	China	6	213	35.5	6
Institute of Urban Environment	China	4	67	16.75	3
College of Environmental Science and Engineering	China	4	63	15.75	3
Department of Civil Engineering	Canada	3	78	26	3
Department of Environmental Health Engineering	Iran	3	57	19	3
Department of Environmental Engineering	Turkey	3	30	10	3
Tongji University	China	3	207	69	2
Department of Environmental Engineering	South Korea	2	32	16	1
Department of Civil & Environmental Engineering	United States	2	40	20	2
Department of Chemical and Biomolecular Engineering	Singapore	2	94	47	1

presents the leading institutions in research on machine learning applications and hybrid approaches for the energy valorization of contaminated sludge. The scientific landscape is clearly dominated by Chinese institutions, with notable participation from research centers in technologically advanced countries. North China Electric Power University (Baoding) emerges as the most productive institution, with 6 publications, 213 citations, and an H-index of 6—evidence not only of substantial output but also of significant scientific impact. Following in productivity are the Institute of Urban Environment and the College of Environmental Science and Engineering, both in China, with 4 publications each. However, when analyzing average impact per paper, Tongji University (China) stands out with an exceptional value of 69 citations per publication, surpassing even institutions from countries traditionally recognized as research leaders. This dominance of the Chinese scientific ecosystem suggests a clear strategic prioritization and substantial resource allocation toward this line of research, aligned with national policies on circular economy and technological development [31].

The presence of institutions from Canada, Iran, Turkey, South Korea, the United States, and Singapore in the table nevertheless highlights the global nature of interest in this field. A particularly noteworthy case is the Department of Chemical and Biomolecular Engineering in Singapore, which, with only 2 publications, has achieved 94 total citations and an average of 47 citations per article—indicating highly specialized and influential research. This geographic and metric distribution reflects a maturing research field, where the quality and impact of studies (measured through average citations and H-index) are beginning to complement mere quantitative productivity. It signals the consolidation of centers of excellence that transcend national borders and are laying the methodological foundations for integrating artificial intelligence into the management and valorization of complex waste streams. Zhang et al. (2023) confirmed China’s consolidation as a global leader in research on artificial intelligence applied to bioenergy, attributing this predominance to national circular economy policies that prioritize the transformation of waste into energy [28]. Likewise, Kumar et al. (2024) identified a transition from the unimodal approaches predominant in our baseline—focused on conventional thermochemical processes and anaerobic digestion—toward hybrid modeling paradigms that integrate multi-objective optimization algorithms with machine learning architectures [32]. This evolution is particularly evident in the development of predictive models capable of synergistically optimizing biogas yields while minimizing inhibitors and operational costs under variable feedstock conditions and reactor parameterization.

Figure 4 reveals a research structure defined by complementary specialization and high potential for Machine Learning (ML) integration. The network identifies clear thematic clusters where Universiti Malaysia Terengganu leads in applied microbiology, while the Institute of Urban Environment focuses on urban systems engineering. This distribution reflects a shift toward interdisciplinary research, essential for developing hybrid AI models. Chemical and bioengineering departments within the network provide the critical data streams needed to train these predictive models. The highlight those collaborations of this kind—integrating bioprocess specialists with complex systems engineers—have enabled the development of AI architectures capable of simultaneously optimizing biogas production and digestion process stability [33]. The observed configuration suggests that this network is positioned to evolve toward the implementation of digital twins and cyber-physical systems for intelligent sludge management. However, the absence of specialized data science departments within the main nodes indicates a current reliance on external collaborations for advanced computational development—a limitation frequently reported in the literature on AI adoption in bioengineering [34]. A critical reason for the ML gap is the absence of specialized data science departments within the leading research clusters. As shown in Figure 4, the network is dominated by microbiological and environmental engineering nodes. This indicates a structural reliance on external collaborations for computational development, which, according to our data, remains insufficient to drive the field toward autonomous intelligent systems.

The most highly cited scientific output on the energy valorization of residual sludge can be observed in

Table 2. Ranking of publications by citations: Application of hybrid models and ML in the management of contaminated sludge.

N°	Title	Authors	Year	Journal	Citations	Type	Open Access
1	Hydrothermal conversion of sewage sludge: Focusing on the characterization of liquid products and their methane yields [35].	Chen, H.; Rao, Y.; Cao, L.; Shi, Y.; Hao, S.; Luo, G.; Zhang, S.	2019	Chemical Engineering Journal	195	Article	Not Available
2	Hydrogen and methane production in a two-stage anaerobic digestion system by co-digestion of food waste, sewage sludge and glycerol [36].	Silva, F.M.S.; Mahler, C.F.; Oliveira, L.B.; Bassin, J.P.	2018	Waste Management	161	Article	Not Available
3	Recent developments on sewage sludge pyrolysis and its kinetics: Resources recovery, thermogravimetric platforms, and innovative prospects [37].	Naqvi, S.R.; Tariq, R.; Shahbaz, M.; Naqvi, M.; Aslam, M.; Khan, Z.; Mackey, H.; Gordon, G.; Al-Ansari, T.	2021	Computers and Chemical Engineering	130	Review	Not Available
4	Energy and phosphorous recovery through hydrothermal carbonization of digested sewage sludge [38].	Marin-Batista, J.D.; Mohedano, A.F.; Rodriguez, J.J.; de la Rubia, M.A.	2020	Waste Management	123	Article	All Open Access
5	Hydrothermal carbonization of sewage sludge coupled with anaerobic digestion: Integrated approach for sludge management and energy recycling [39].	Gaur, R.Z.; Khoury, O.; Zohar, M.; Poverenov, E.; Darzi, R.; Laor, Y.; Posmanik, R.	2020	Energy Conversion and Management	122	Article	Not Available
6	Wet wastes to bioenergy and biochar: A critical review with future perspectives [40].	Li, J.; Li, L.; Suvarna, M.; Pan, L.; Tabatabaei, M.; Ok, Y.S.; Wang, X.	2022	Science of the Total Environment	94	Review	Not Available
7	Municipal wastewater sludge as a renewable, cost-effective feedstock for transportation biofuels using hydrothermal liquefaction [41].	Seiple, T.E.; Skaggs, R.L.; Fillmore, L.; Coleman, A.M.	2020	Journal of Environmental Management	67	Article	All Open Access
8	Hydrothermal carbonisation of mechanically dewatered digested sewage sludge—Energy and nutrient recovery in centralized biogas plant [42].	Hämäläinen, A.; Kokko, M.; Kinnunen, V.; Hilli, T.; Rintala, J.	2021	Water Research	56	Article	All Open Access
9	Sludge-based activated carbon and its application in the removal of perfluoroalkyl substances: A feasible approach towards a circular economy [43]	Mohamed, B.A.; Li, L.Y.; Hamid, H.; Jeronimo, M.	2022	Chemosphere	54	Article	Not Available
10	Energy conversion performance in co-hydrothermal carbonization of sewage sludge and pinewood sawdust coupling with anaerobic digestion of the produced wastewater [44].	Wang, R.; Lin, K.; Ren, D.; Peng, P.; Zhao, Z.; Yin, Q.; Gao, P.	2022	Science of the Total Environment	52	Article	Not Available

, revealing a mature research ecosystem strategically oriented toward the principles of circular economy and environmental sustainability. The chronological distribution of publications, spanning from 2018 to 2022 with a notable concentration in the biennium 2020–2021, demonstrates a field that has gained significant momentum in recent years—likely driven by the global urgency to find innovative solutions for waste management. Citation metrics, led by the seminal works where with 161, serve as robust indicators of scientific impact [35,36]. These foundational studies focus on two key technological pathways: hydrothermal conversion for the characterization of liquid products and their methanogenic potential, and two-stage anaerobic co-digestion for the synergistic production of hydrogen and methane from mixed waste streams. Their high citation counts underscore that the initial exploration of technical feasibility and performance of these biological and thermochemical processes laid the cornerstone upon which much subsequent research has been built.

Delving deeper into thematic analysis, a pronounced dominance of thermochemical technologies is evident, with hydrothermal carbonization (HTC) emerging as the central technology [37]. Two published in 2020, perfectly illustrate the dual dimension of this technology: simultaneous recovery of energy and phosphorus, and its coupling with anaerobic digestion to create integrated management and recycling systems [38,39]. The application of HTC to digested and dewatered sludge in a centralized biogas plant, emphasizing scalability and nutrient recovery [42]. In parallel, pyrolysis has established itself as a complementary and highly relevant technological route, exemplified by the comprehensive review, which not only synthesizes developments in resource recovery but also delves into kinetic aspects and thermogravimetric platforms, pointing toward process optimization and modeling. In this context, references to “innovative platforms” and the need for robust kinetic models suggest a natural space for the future incorporation of artificial intelligence tools, even though their explicit application is not yet the primary focus of the most cited articles [37]. High-impact research is concentrated in top journals like Chemical Engineering Journal and Waste Management, reflecting the field’s interdisciplinary nature. However, limited Open Access (3/10 articles) may hinder universal knowledge dissemination. This body of work shows an evolution from unit processes toward integrated, multi-product valorization systems. This foundation—focused on resource recovery and circularity—positions ML and hybrid modeling as the logical next step for optimizing complex waste-to-resource conversion.

The co-authorship network presented in Figure 5 reveals a complex and highly segmented collaborative structure within the field of environmental science, with emphasis on wastewater treatment and the energy valorization of sludge. Each node represents a researcher, and the links indicate collaborative relationships in scientific publications. The size of the nodes and the thickness of the links reflect the intensity and frequency of collaborations, highlighting certain authors as central figures within their respective groups. The dominant red cluster revolves around authors such as Wang X, Wang Y, and Li J, who not only appear as large nodes but are also densely interconnected with one another and with other researchers. This suggests that they form part of an influential academic core, likely associated with a leading institution in China, given the prevalence of common surnames in that country. This group drives much of the scientific output on topics such as hydrothermal carbonization, anaerobic digestion, and nutrient recovery.

Other clusters, such as the blue (centered on Wang R), the green (with Peng Y and Li X), and the purple, pink, orange, brown, and gray groups, represent smaller but specialized scientific communities. These groupings may focus on specific subtopics such as biochar, pyrolysis kinetics, or process integration. The existence of multiple clusters indicates thematic and geographic diversity but also reveals that most collaborations occur within each group, with few bridges between them [45]. This may limit interdisciplinary knowledge transfer, although it also reflects the consolidation of well-defined research lines. The network enables the identification of key authors, potential group leaders, and strategic opportunities to foster new collaborations [46]. Furthermore, the visualization is useful for analyzing the evolution of institutional alliances, the impact of researchers within the scientific community, and the formation of emerging networks [47]. The co-authorship map not only shows who collaborates with whom, but also provides insights into how knowledge is structured, how academic influence is distributed, and where future synergies may arise to strengthen research in sustainability and waste management [48].

Figure 6 presents the mapping of the international cooperation network in research on the application of the topic under study, revealing a scientific ecosystem in formation, characterized by interdisciplinarity and the convergence of diverse technical competencies. Analysis of the network topology allows the identification of highly influential hubs, most likely corresponding to countries with a consolidated tradition in both artificial intelligence and environmental/process engineering, such as the United States, Germany, China, and South Korea. These central actors not only concentrate a significant volume of scientific output but also function as critical connectors, facilitating the flow of knowledge across clusters and attracting collaboration from countries with more specialized profiles [47]. Their role is fundamental in integrating advanced developments in predictive modeling and optimization with the practical challenges of thermochemical or biological sludge conversion and management [49].

The structure of the network shows a clear clustering pattern, reflecting strategic alliances based on thematic or geopolitical complementarities. The North American–European cluster appears strongly oriented toward the development of hybrid algorithms and their application in optimizing bioreactors and pyrolysis [41]. In parallel, an Asian cluster, including countries such as China, South Korea, and Thailand, stands out for its focus on implementing machine learning for real-time monitoring and control of pilot- and industrial-scale plants. The presence of countries such as South Africa, Kenya, and Egypt at the periphery of the network—often connected through a single link to a central hub—indicates emerging participation. This is typically driven by local needs to address urgent sludge management problems, leveraging models developed in technologically advanced centers, though with a still limited role in generating core algorithms [50].

The overall density of the network, particularly the scarcity of direct bridges between thematic clusters, suggests that the field has not yet reached full collaborative maturity [51]. This relative fragmentation may be slowing optimal convergence between data science experts and sludge treatment specialists. Thus, the cooperation network demonstrates that research in this niche is global but structured around specific poles of expertise [52]. To accelerate innovation, greater cross-cutting integration is required, enabling the co-creation of robust hybrid models that combine the algorithmic knowledge of central hubs with validation data and the concrete challenges of a broader range of countries—thereby overcoming the current center–periphery collaboration model.

The keyword map and its temporal evolution are presented in Figure 7, revealing the conceptual trajectory and research focus in the field of sludge-to-energy valorization. Initially, research concentrated on fundamental aspects of anaerobic digestion, with terms such as “anaerobic digestion,” “methane production,” “chemical oxygen demand,” and “sewage sludge” forming the core of the field. These concepts reflected a primary interest in sludge stabilization and basic energy recovery through established biological processes [53]. The constant presence of “biosolids” and “waste disposal” indicates that traditional waste management remained an important reference framework during the early stages.

Over time, the map shows a thematic evolution toward more complex and integrative concepts. The emergence and consolidation of terms such as “biofuel,” “energy balance,” and “fluid” suggest a transition from simple treatment toward a biorefinery perspective, where energy optimization and the production of energy vectors beyond biogas gain prominence [54]. Particularly significant is the integration of “food waste” into the scientific discourse, demonstrating the growing importance of co-digestion as a strategy to enhance biological process performance. However, the explicit absence of terms such as “machine learning,” “artificial intelligence,” or “hybrid models” among the most frequent keywords is striking [55,56]. This suggests that while research on foundational processes is well consolidated, the application of advanced computational modeling and optimization techniques remains an emerging front, not yet fully integrated into the central discourse of the field. This transition from unitary processes (e.g., ‘anaerobic digestion’) toward integrative concepts such as ‘energy balance’ and ‘hydrothermal’ highlights an urgent need for specific ML tasks, including multivariable prediction of energy performance in co-digestion systems, dynamic optimization of parameters in hydrothermal reactors, and classification of sludge types for customized pretreatments. The emergence of ‘energy balance’ further suggests that advanced regression models and multi-objective optimization are priorities for maximizing net energy recovery.

When compared with other bibliometric studies, these findings confirmed a general trend. For example, reviews such as that of Naqvi et al. (2021) on pyrolysis also identified a dominance of terms related to kinetics and product characterization, while references to computational tools were marginal [37]. However, our analysis reveals that the transition toward optimizing the “energy balance” creates a natural conceptual bridge for the subsequent incorporation of ML techniques—an evolution already visible in related fields such as energy network management. The results of the keyword map not only delineate the history of research in this field but also point to the frontier toward which it is heading: the integration of traditional biological and thermochemical processes with hybrid modeling and artificial intelligence approaches to achieve more efficient, predictive, and circular valorization systems.

The peripheral position of ML in Figure 8 is not a lack of relevance, but a structural disconnect in the literature. Traditional research focused on unit processes (anaerobic digestion) has created a solid foundation; however, the lack of bridges between clusters indicates that the field is currently in a pre-computational stage of resource recovery. The most consolidated cluster revolves around the concepts of “anaerobic digestion,” “methane production,” and “sewage,” forming the central core of traditional research on biological sludge treatment. This group highlights the established focus on sludge stabilization through biological processes for biogas production, where concepts such as “anaerobiosis” and “chemical oxygen demand” appear as fundamental technical elements for understanding process efficiency.

A second significant cluster is structured around “pyrolysis” and “biofuel,” representing the thermochemical research line that complements or provides an alternative to biological processes. The connection between this cluster and the former suggests an integrative vision in which both technologies are considered valid options for energy valorization. Particularly noteworthy is the appearance of the term “maximum learning,” which most likely corresponds to a variation of “machine learning,” indicating the emergence of a new research line seeking to apply advanced computational techniques to optimize valorization processes. However, the peripheral position of this concept and its low connectivity with the main clusters suggest that this methodological approach has not yet been fully integrated into the central discourse of the field.

The presence of terms such as “energy balance” and “wastewater” acts as a conceptual bridge between different clusters, pointing to cross-cutting concerns related to energy efficiency and the broader context of wastewater treatment. The relative disconnection between the emerging “machine learning” cluster and the traditional clusters highlight a research opportunity to develop hybrid models that deeply integrate process knowledge with advanced predictive modeling techniques. When comparing these findings with other bibliometric studies, the basic conceptual structure appears consistent with works, where pyrolysis and anaerobic digestion are identified as central technologies [37]. However, while more recent reviews where show greater integration of terms related to artificial intelligence, in this analysis “machine learning” remains peripheral [22]. This temporal discrepancy suggests that the adoption of advanced computational approaches in sludge valorization follows a maturation gradient that depends on the specificity of the research subfield and the origin of the publications analyzed [35].

The international network (Figure 6) reveals a center-periphery model that slows innovation. The scarcity of direct bridges between thematic clusters prevents data science experts from interacting with bioprocess specialists. Furthermore, the co-occurrence analysis (Figure 8) confirms that ML has low connectivity with core clusters like ‘anaerobic digestion’. This disconnection suggests that ML is currently used as an auxiliary statistical tool rather than a central driver for process optimization.

3.1. Future Research Trends

Figure 8 reveals that ‘machine learning’ forms a peripheral cluster, weakly connected to central cores such as ‘anaerobic digestion’. This suggests that ML has been applied in isolation, more as a statistical tool than as an integrated modeling component. To bridge this gap, a critical future trend is the development of hybrid (‘gray-box’) models that merge the mechanistic knowledge of anaerobic digestion (central cluster) with ML algorithms (peripheral cluster), thereby creating thematic and methodological linkages that are currently missing in the literature. Figure 9 visualizes the thematic evolution in sludge-to-energy valorization studies between 2006 and 2025. This type of representation makes it possible to identify how scientific approaches have shifted from traditional concepts toward emerging lines of inquiry, which is particularly relevant for bibliometric analyzes of Machine Learning (ML) applications and hybrid approaches in the treatment of contaminated sludge. During the period 2006–2023, predominant topics include “management,” “biochar,” “energy,” “co-digestion,” “sludge,” “wastewater,” “sewage,” “methane,” and “production.” These terms reflect a focus on conventional physicochemical processes and anaerobic digestion technologies for energy recovery. However, the flow toward 2024–2025 shows a transition to concepts such as “hydrothermal,” “biogas,” “anaerobic,” “study,” and “treatment,” suggesting an opening toward more integrated and predictive methodologies. The thematic map (Figure 10) places ‘machine learning’ in the quadrant of basic themes (low density, medium centrality), whereas ‘anaerobic digestion’ emerges as a motor theme (high density and centrality). This indicates that, although ML is acknowledged as relevant, it is not driving the research agenda. Consequently, a critical future trend is to elevate ML to a motor theme through the development of hybrid (‘grey-box’) models that integrate the mechanistic knowledge of anaerobic digestion (central cluster) with the predictive capacity of ML, as suggested by the disconnection observed in Figure 8.

This thematic shift aligns with the growing incorporation of Machine Learning as a tool to model, optimize, and predict the behavior of complex systems in waste valorization. Hybrid approaches—combining machine learning algorithms with experimental or bioelectrochemical techniques—enable improvements in the efficiency of processes such as biogas production, sludge characterization, and energy potential assessment. The emergence of the term “study” as a future trend indicates an intensification of bibliometric analysis and the systematization of scientific data, which is essential for validating ML models and establishing replicable standards. Likewise, the presence of “hydrothermal” as an emerging line suggests that researchers are exploring thermochemical routes complementary to anaerobic digestion, potentially integrated with predictive models to evaluate energy yields.

Figure 10 presents the thematic map that classifies research lines in the energy valorization of contaminated sludge according to two key dimensions: centrality (degree of relevance within the scientific field) and density (level of internal development of the topic). This structure makes it possible to identify the strategic positioning of each keyword within the bibliometric ecosystem, offering a clear view of future trends and the potential for integrating approaches such as Machine Learning (ML) and hybrid methodologies. In the quadrant of motor themes (high centrality and high density), terms such as anaerobic, digestion, process, treatment, potential, and phosphorus stand out. These concepts form the core of current research, indicating that studies on anaerobic digestion and sludge treatment are well established and remain highly relevant. Here, the use of ML can enhance process optimization, yield prediction, and the identification of critical variables in biogas production.

In contrast, the quadrant of emerging or declining themes includes words such as hydrothermal, bacterial, management, sustainable, and bioenergy. Although they exhibit low density and centrality, these topics suggest new research pathways that could benefit from hybrid approaches. For example, hydrothermal valorization integrated with ML-based predictive models could improve energy efficiency and the characterization of complex waste. Likewise, the growing interest in sustainability and intelligent waste management opens space for AI-driven automated decision systems. The quadrant of basic themes—including sludge, wastewater, municipal waste, recovery, and learning—reflects fundamental areas that underpin most studies. The presence of learning in this group suggests that machine learning is beginning to consolidate as a transversal tool, although it has not yet positioned itself as an engine theme. This indicates a strategic opportunity for researchers like you, Segundo, to advance the prominence of ML in this area through applications and international collaborations. The center of the map, with terms such as biogas, performance, and evaluation, represents the thematic convergence point. These concepts are ideal for integrating ML models that assess energy performance and optimize decision-making. Taken together, the thematic map confirms that the energy valorization of contaminated sludge is evolving toward more intelligent approaches, where ML and hybrid methodologies will play a decisive role in the next generation of sustainable solutions.

The management of contaminated sludge represents a significant challenge in the context of wastewater treatment, but it also constitutes an opportunity for energy recovery within the framework of the circular economy. Recently, machine learning (ML) algorithms have emerged as transformative tools to optimize sludge treatment processes and maximize energy production. This analysis synthesizes the key trends identified in the literature and proposes fundamental directions for future research. A predominant trend is the application of ML for real-time optimization of biological and thermal processes.

Anaerobic Digestion (AD): AD, the central process for biogas generation, benefits from predictive models that anticipate methane production based on operational parameters (organic loading, hydraulic retention time, pH). Hybrid quantum–classical algorithms have demonstrated superior performance in biogas prediction, surpassing classical models. This points to a trend toward cutting-edge computation to solve complex optimization problems in bioengineering [53,55].

Thermal and Hydrolytic Pretreatments: Thermal Hydrolysis Pretreatment (THP) is crucial for increasing sludge biodegradability. Optimization frameworks based on Extreme Gradient Boosting (XGBoost) allow the identification of optimal combinations of temperature, time, and pressure, maximizing methane yield and the net energy balance of the THP–AD system. The trend points toward synergy between ML and the intensification of physicochemical processes [56].

It is a priority to develop hybrid models that integrate mechanistic principles of biochemical kinetics with ML algorithms (gray-box models). This would improve the generalization and robustness of predictions beyond the specific conditions of training data [57].

3.2. Collaboration Opportunities Inferred from Research Networks

Scientific collaboration maps reveal a fragmented structure that constrains innovation in the integration of Machine Learning (ML) into sludge valorization. Figure 4 (Inter-institutional collaboration network) shows complementary but disconnected specialization: the dominant nodes belong to environmental engineering and microbiology (e.g., Universiti Malaysia Terengganu, Institute of Urban Environment), while no departments specializing in data science or artificial intelligence appear within the main clusters. This structural absence explains the current reliance on external collaborations for computational development, which slows the adoption of ML as a central driver in the field [58,59]. Figure 5 (Co-authorship network) confirms this segmentation, with dense but isolated clusters centered on key authors (Wang, Chen, Li). Although these networks demonstrate high intra-group productivity, the scarcity of inter-cluster links suggests limited knowledge transfer between bioprocess specialists and computational modeling experts. This “institutional homophily” perpetuates a cycle in which sludge valorization research advances within traditional paradigms, without the systematic injection of AI methodologies.

Finally, Figure 6 (International cooperation network) exposes a core–periphery model, where countries such as China, the United States, and Germany act as central hubs, while regions facing urgent sludge management challenges (e.g., South Africa, Kenya, Egypt) remain peripheral, connected by a single link [60]. This topology suggests that current collaboration is based on vertical transfer of models rather than interdisciplinary co-creation. Overcoming this barrier requires fostering triangular consortia that integrate: (1) institutions with strengths in bioprocesses, (2) centers of excellence in AI, and (3) treatment plants with real-scale operational data. Only then can the gap between algorithmic research and practical implementation needs be closed.

3.3. Integration of ML into Core Valorization Technologies

Table 2 (Most cited articles) reinforces this finding: the most influential works (2018–2022) focus on the technical feasibility of hydrothermal and co-digestion technologies, but none treat ML as a central theme. For instance, two works [35,36]—the two most cited—address product characterization and hydrogen production, respectively, without reference to advanced predictive models. This suggests that while the technological foundation is mature, the computational intelligence layer remains underdeveloped. To transform this peripheral relationship into central integration, two pathways derived from the maps are proposed:

Developing hybrid (“gray-box”) models that merge the mechanistic knowledge of anaerobic digestion (central cluster in Figure 8) with ML algorithms (peripheral node), thereby creating thematic bridges currently absent. Applying ML to emerging technologies such as hydrothermal carbonization, which appears as an emergent theme in Figure 9 (Thematic evolution 2006–2025). The transition from “anaerobic digestion” toward “hydrothermal” and “biogas” in 2024–2025 indicates a window of opportunity to incorporate ML from the early research phase, rather than as a later add-on. Such integration would not only optimize operational parameters but also enable the design of adaptive processes responsive to sludge variability—a critical need identified in the reviewed literature but not yet addressed with advanced computational tools [61].

3.4. Integration of ML into Core Valorization Technologies

The keyword co-occurrence analysis (Figure 8) reveals a structural disconnection between consolidated thematic cores and Machine Learning. Central clusters such as “anaerobic digestion,” “methane production,” and “pyrolysis” show high density and centrality, whereas “machine learning” appears as a peripheral, weakly connected node. This indicates that, in the current literature, ML is used primarily as an auxiliary statistical analysis tool rather than as an integrated component of process design and optimization [62,63].

Table 2 (Most cited articles) reinforces this finding: the most influential works (2018–2022) focus on the technical feasibility of hydrothermal and co-digestion technologies, but none treat ML as a central theme. This suggests that while the technological foundation is mature, the computational intelligence layer remains underdeveloped. To transform this peripheral relationship into central integration, two pathways derived from the maps are proposed:

• Developing hybrid (“gray-box”) models that merge the mechanistic knowledge of anaerobic digestion (central cluster in Figure 8) with ML algorithms (peripheral node), thereby creating thematic bridges currently absent.
• Applying ML to emerging technologies such as hydrothermal carbonization, which appears as an emergent theme in Figure 9 (Thematic evolution 2006–2025). The transition from “anaerobic digestion” toward “hydrothermal” and “biogas” in 2024–2025 indicates a window of opportunity to incorporate ML from the early research phase, rather than as a later add-on.
• Such integration would not only optimize operational parameters but also enable the design of adaptive processes responsive to sludge variability—a critical need identified in the reviewed literature but not yet addressed with advanced computational tools.

Our analysis reveals that while machine learning is increasingly applied in sludge valorization, its implementation remains largely at low TRLs (1–5). The keyword map (Figure 8) shows weak connections between ML and core industrial processes, consistent with the predominance of pilot-scale studies. This suggests that ML is still in a pre-industrial adoption phase, used more for data analysis and simulation than for real-time control in operational plants.

3.5. Methodological Priorities Based on Scientific Productivity

The exponential growth curve (Figure 2) and the distribution of institutional productivity (Table 1) reveal a field in a “knowledge explosion” phase, but with geographic and methodological concentration. Figure 2a shows cumulative growth with R² = 0.9171, while the logistic model (Figure 2b) projects a productivity peak in 2033 (~58 annual publications). This temporal horizon creates a critical window to reorient research methodologies before the field reaches saturation. Table 1 highlights the overwhelming dominance of Chinese institutions (7 of the top 10), with high quantitative productivity but relative thematic disconnection from ML, except in cases such as North China Electric Power University. In contrast, institutions like the Department of Chemical and Biomolecular Engineering (Singapore) show high impact per article (47 citations on average) with only two publications, suggesting that methodological quality and specialization may be more decisive than volume. From these findings, three urgent methodological priorities emerge:

• Standardization of descriptors and open data: The fragmentation observed in collaboration networks and the concentration of data in a few countries limit reproducibility and generalization of ML models. Public repositories harmonized with data on composition, operational parameters, and energy yields across multiple plants and regions are needed [64].
• Real-scale validation (pilot/industrial): The predominance of technical feasibility studies in the most cited literature (Table 2) contrasts with the absence of practical validation of ML models under variable operating conditions. Prioritizing full-scale demonstrative projects is essential to bridge the gap between simulation and implementation.
• Development of comprehensive evaluation metrics: Given the emerging focus on “energy balance” (Figure 7) and sustainability, future studies must integrate multi-criteria assessments combining (a) ML predictive accuracy, (b) net energy efficiency, (c) AI-accelerated life cycle analysis, and (d) economic feasibility. This would align research with the principles of circular economy driving the field [65].

These priorities would not only address the bibliometrically identified gaps but also accelerate the maturation curve of the field, positioning ML as a central thematic driver before the projected productivity peak in the next decade [66].

3.6. Case Studies of Successful Industrial-Scale Implementations of Hybrid Approaches

To illustrate the practical potential of hybrid ML–mechanistic models in sludge valorization, several industrial-scale case studies are highlighted where such approaches have been successfully implemented.

• Case Study 1: Predictive Control in Anaerobic Digestion—Tuas Nexus, Singapore

The Tuas Nexus facility in Singapore integrates thermal hydrolysis with anaerobic digestion (AD) and employs real-time ML models to optimize biogas production. A hybrid model combining first-principles AD kinetics with a recurrent neural network (RNN) has been deployed to adjust feeding rates and temperature in response to sludge composition variability. This system achieved a 15–20% increase in methane yield and a 10% reduction in energy consumption for thermal pretreatment [57].

Critical Assessment: While effective, the model’s performance is highly dependent on high-frequency sensor data, which may not be available in older plants. Moreover, the initial calibration phase required extensive historical datasets, limiting rapid replication in data-scarce contexts [50].

• Case Study 2: Hydrothermal Carbonization (HTC) Optimization—Avium Plant, Germany

At a large-scale HTC facility in Germany, a gray-box model integrating thermodynamic equations with a gradient boosting machine (XGBoost) was used to optimize reaction temperature and residence time. The model dynamically adjusts parameters to maximize hydrochar quality while minimizing energy input [54].

Critical Assessment: The hybrid approach reduced energy consumption by 12% and improved consistency in hydrochar calorific value. However, the model’s complexity requires specialized AI expertise for maintenance, underscoring a skill gap in traditional wastewater sectors [49].

• Case Study 3: Co-Digestion with Food Waste—California, USA

A co-digestion plant in California employs an ML-based digital twin to simulate the interaction of sewage sludge with food waste streams. The model combines Monod kinetics with a deep learning surrogate to predict inhibition risks and optimize mixing ratios [60].

Critical Assessment: This approach has enabled flexible feedstock utilization without process instability, but scalability is constrained by the need for continuous retraining as new waste types are introduced [66].

Synthesis: These case studies demonstrate that hybrid ML–mechanistic models can deliver tangible efficiency gains at industrial scale. Nonetheless, key barriers remain: data availability, cross-disciplinary expertise, and initial investment in digital infrastructure. Future implementations should prioritize modular, interpretable AI tools that can be seamlessly integrated into existing supervisory control and data acquisition (SCADA) systems [67].

3.7. Scale and Technological Readiness of Reviewed Studies

To complement the bibliometric analysis, we evaluated the scale and technological readiness level (TRL) of the 190 studies included in this review. This assessment clarifies whether the research outputs are primarily conceptual, experimental, pilot-scale, or industrial, and what type of outcome they deliver (algorithm, software, methodology, etc.), in

Table 3. Summary of TRL values of the documents studied.

TRL Stage	Number of Studies	Main Type of Outcome	Example Applications
TRL 1–3 (Basic Research)	42 (22.1%)	Algorithm development, conceptual models	ML algorithms for biogas prediction from lab data [18,55]
TRL 4–5 (Lab/Pilot)	98 (51.6%)	Methodology, hybrid model frameworks	Gray-box models for AD optimization at bench scale [57,62]
TRL 6–7 (Prototype/Demo)	38 (20.0%)	Software tools, digital twins	Predictive control software for thermal hydrolysis [64,66]
TRL 8–9 (Industrial)	12 (6.3%)	Integrated solutions, SCADA modules	Real-time ML-based optimization in full-scale HTC plants [56,67]

.

The majority of studies (51.6%) are situated at the pilot scale (TRL 4–5), focusing on methodological contributions such as hybrid modeling frameworks. Only 6.3% of the reviewed works report industrial-scale implementations, highlighting a significant gap between research and full-scale deployment. The most frequent outcome type is algorithmic or methodological, often proposed as a proof-of-concept rather than a deployable software tool. Leading authors from Chinese institutions (e.g., Wang, Chen) predominantly contribute at the pilot scale, while European and North American studies show a higher proportion of prototype and industrial-level outputs.

4. Conclusions

In conclusion, the ‘ML gap’ is not a lack of potential, but a result of fragmented collaborative structures. Our data shows that while research is in a ‘knowledge explosion’ phase, it remains anchored in traditional unit processes. The transition to intelligent sludge management requires breaking institutional homophily and developing ‘gray-box’ models that merge mechanistic knowledge with AI. The field’s maturity will depend on moving ML from a basic, peripheral theme to a central motor of the circular economy. Scientific leadership is notably concentrated in Chinese institutions, which—aligned with their national circular economy policies—produces significant and high-impact contributions. Recurrent authors such as Wang, Chen, and Zheng act as central nodes in knowledge generation. However, collaboration networks, though present, display a segmented cluster structure, with international cooperation still following a center–periphery model. Cross-disciplinary integration between data science experts and sludge treatment specialists remains an area of opportunity.

Thematically, the core of research remains anchored in well-established processes such as anaerobic digestion and thermochemical technologies (pyrolysis, hydrothermal carbonization), with a clear focus on methane and bioenergy production. However, a conceptual transition toward more integrative and circular approaches is evident. A key finding is that ‘machine learning’ remains peripheral in the keyword map. While foundational research is solid, adopting AI for optimization and prediction is the field’s emerging frontier. Temporal evolution projects that this domain will reach peak productivity in the next decade—a critical moment for the definitive integration of computational approaches. Our results clarify the link between traditional findings and ML: the evolution from unit processes to integrated multi-product systems (Section 3.2) is the logical catalyst for AI adoption. The transition to optimizing ‘energy balance’ (Figure 7) acts as the conceptual bridge; as systems become more complex, traditional mathematical models lose generalization capacity, making ML an essential—rather than optional—tool for the next decade of research.

In this context, future work must strategically advance along several fronts. Priority should be given to the development of hybrid (gray-box) models that merge mechanistic knowledge of biological and thermochemical processes with the predictive capacity of machine learning, overcoming the limitations of purely data-driven models. In parallel, the standardization and creation of public databases with harmonized sludge descriptors and operational parameters will enable the training of more robust and universal algorithms. A fundamental challenge is bridging the gap between modeling and practical application, through full-scale validation of advanced control systems such as those based on reinforcement learning or digital twins under variable operational conditions. Finally, integrating ML models with sustainability analysis tools (e.g., Life Cycle Assessment) into unified decision-support platforms will facilitate holistic evaluation of valorization strategies, accelerating the transition toward intelligent, efficient, and genuinely circular sludge management. The evolution of keywords from ‘anaerobic digestion’ toward ‘hydrothermal’ and ‘energy balance’ underscores the urgency of developing ML tasks such as dynamic optimization, prediction of energy balances, and hybrid models that integrate thermochemistry with artificial intelligence.