Defect Networks and Waste Reduction in Additive Manufacturing

Abstract

This study addresses a key challenge in Additive Manufacturing (AM): while it promises sustainable production, manufacturing defects often lead to significant material and energy waste. The purpose of this research is to apply the Decision-Making Trial and Evaluation Laboratory (DEMATEL) method to identify and map the cause-and-effect relationships among common AM defects. By doing this, the goal is to pinpoint the most influential ‘root’ causes, allowing for more targeted and effective quality improvements. The methodology is based on a qualitative approach using the expert judgment of a panel of six professionals. The DEMATEL analysis successfully sorted the defects into two categories: those that are primary causes and those that are symptoms or effects. The main findings show that contamination is the most significant causal factor, meaning that it strongly influences other defects. In contrast, dimensional inaccuracy is the most affected factor, acting as a symptom of other underlying issues. In conclusion, the study finds that focusing on mitigating root causes like contamination, warping, and porosity is crucial for achieving improvements across the process chain. This framework allows engineers to prioritize quality control efforts on the fundamental problems, rather than on superficial defects, thereby maximizing efficiency and waste reduction. Ultimately, this research provides a clear, actionable framework for improving quality control and promoting more sustainable manufacturing practices.

1. Introduction

Additive manufacturing (AM) is usually presented as an enabler of sustainable production because it supports near-net-shape fabrication, part consolidation, and design freedoms such as topology optimization and light weighting that can reduce material use and logistics burdens across diverse sectors [1,2]. In medical applications, recent studies showed that design decisions and the digital workflow (for example, thermoforming of material-extrusion parts) can reduce printing times while maintaining functional performance, illustrating how “design for use” directly influences the resource profile [3]. These promises, however, are conditional: the sustainability gains only materialize when processes are reliable and repeatable, defects are systematically controlled, and measurement informs design and operation rather than merely documenting deviations after the fact [4,5]. In particular, the mechanical performance penalties associated with porosity, lack of fusion (LOF), and cracking underscore how quality shortfalls translate directly into scrap, rework, and energy-intensive post-processing, thereby eroding the environmental benefits [6].

From a process-science perspective, defect formation in AM is governed by multi-scale thermo-mechanical phenomena and complex process–structure–property linkages that challenge generalization across platforms and materials [7,8]. This has motivated a complementary focus on process planning and geometric fidelity, spanning the digital pipeline from mesh repair and slicing to build strategy, because small decisions upstream can have significant impacts on dimensional accuracy and downstream resource use [9,10]. Early estimation of part accuracy, orientation-aware optimization, and FEM-informed control exemplify how intentional planning and simulation can prevent error accumulation is more effective than attempting to compensate for it later [10,11,12].

In parallel, advancements in sensing and control developments have sought to reduce failure rates and stabilize quality, with optical monitoring architectures and data-driven compensation showing particular promise for in-process detection and corrective action [13,14]. Closed-loop kinematic compensation and sensitivity analyses of geometric errors further illustrate the trajectory towards measurable, correctable processes that reduce variability and waste in practice [15,16]. These quality-oriented technological advances directly support sustainability by reducing material and energy losses associated with build failures and non-conforming parts.

Energy is a critical dimension of AM’s sustainability profile, and empirical modelling work shows that process parameters and strategy can be tuned to reduce consumption without sacrificing quality [17]. Moreover, the relation between geometric complexity and energy demand is non-trivial, suggesting that “design freedom” alone does not guarantee lower energy footprints and reinforcing the need for evidence-based optimization rather than assumptions [18]. Life-cycle and durability-oriented analyses likewise stress that sustainability claims hinge on producing robust, reliable parts that prevent premature failure and rework [19]. At the same time, demonstrations of repeatability in powder bed fusion highlight both the feasibility and the ongoing challenges of consistent quality, which remain pivotal for scaling sustainable production [20].

Unlike subtractive manufacturing methods, which remove material to create a part, AM builds components layer by layer, minimizing waste. Studies in the literature have demonstrated that AM can reduce material consumption by 35–80% in final parts, depending on their complexity [21]. From an energy perspective, AM technology can reduce energy use by approximately 24% and CO2 emissions by 58%, primarily by optimizing part geometries and reducing the volume of material required [22].

Beyond machine and material considerations, sustainability outcomes depend on planning choices such as orientation, packing, and job composition, which belong squarely to the operations research and decision-support domain [23]. Here, multi-criteria decision-making (MCDM) methods, including the Decision-Making Trial and Evaluation Laboratory (DEMATEL) method, offer a robust way to capture expert knowledge about inter-factor influences, separate “cause” from “effect”, and reveal leverage points where interventions deliver significant system-level benefits [24,25]. The broader literature also cautions against over-promising and urges a realistic treatment of AM’s limitations and complementarities with conventional processes, an attitude that is essential when linking defect mitigation to credible sustainability claims [26]. Complementarily, targeted process-planning studies (e.g., path planning in WAAM) show that carefully designed strategies can measurably reduce specific error modes, aligning quality control with resource stewardship [1].

Against this backdrop, the present study applies DEMATEL to a curated set of defect modes that are encountered in polymer and metal AM, with the specific aim of mapping causal relationships that impact sustainability-relevant outcomes (scrap, rework, and energy/time losses) [24,25]. To ground the causal analysis, we supplement expert judgments with a case-based corpus of defect imagery from our laboratory, covering typical phenomena such as LOF, delamination, warping, and surface artefacts, interpreted through established process–structure–property mechanisms and geometric fidelity principles [6,7,27,28,29,30]. By integrating a causal network view (DEMATEL) with visual evidence and process-planning/monitoring insights, we seek to identify “root” defects whose mitigation yields the greatest sustainability return and provide an actionable decision framework that complements in situ monitoring and design-for-quality practices [13,23]. In doing so, we respond to calls for verifiable and transferable methods that couple defect understanding to resource-efficiency outcomes in AM, closing the loop between quality engineering and sustainability performance [20,23]. Computer-vision approaches also show promise for rapid defect detection; for example, YOLO-v8 was trained to identify surface imperfections on FDM/PLA parts produced under varied process parameters [31].

Despite the existing body of work on defect identification and process optimization, a significant research gap remains in understanding the interconnectedness and causal relationships among these defects. While previous studies have focused on mitigating individual defects through targeted process control [15,28,29], they often fail to provide a holistic, systemic view of how defects influence one another and, critically, how addressing a single ‘root cause’ can cascade into broader sustainability benefits. This paper aims to fill this gap by applying the DEMATEL methodology—a tool well-suited for mapping complex causal networks—to a set of common AM defects. We also integrate recent literature from the past three years to ensure that our review is up to date. This is a crucial step that distinguishes our approach from more conventional analyses that treat defects as isolated events. We aim to move beyond simple defect identification towards a deeper, network-based understanding, offering a framework that is both transparent and actionable.

This study contributes a practitioner-informed causal map of eight common AM defect types, with a stable cause–effect partition. By also presenting sustainability-relevant pathways and a tiered intervention order, along with imaging exemplars that ground the map, our work provides an actionable framework for improving quality and resource efficiency. The unique contribution of this paper lies in its application of the DEMATEL method to the specific context of AM defects to provide a clear, quantifiable roadmap for quality improvement, something that is lacking in the current literature. It moves beyond simple defect identification to a deeper understanding of their underlying network, providing a framework that can be applied to various AM processes and materials.

2. Theoretical Framework

The quality and performance of parts fabricated through AM are directly influenced by a series of critical defects that can occur at various stages of the process. These imperfections, from LOF to dimensional inaccuracy, are the result of complex interactions between process parameters, material properties, and part geometry. A deep understanding of the formation mechanisms of these defects is essential for developing effective prevention and control strategies to ensure the reliability and durability of the final products. In this regard, we identified defects specific to the AM process for parts made using a FORMIGA P110 SLS machine with the EOS PA2200 (PA12) material in our rapid prototyping laboratory.

The parts were fabricated under two temperature regimes: R1 (removal chamber at 150 °C/process chamber at 168.5 °C (initial builds)) and R2 (removal chamber at 153 °C/process chamber at 171.5 °C (adjusted builds)). The observed defects were grouped into dimensional deformations, structural defects, surface defects, and process-related failures. For each case, a mechanical explanation and an association with the defects listed above are provided, illustrated by representative images.

In the R1 regime, the macro and micro images showed pronounced layer stratification and local separation between layers, indicating poor adhesion. Probable causes include marginal energy input, local temperature deviations, and high interfacial stresses. This delamination defect is a primary factor for micro-cracks and dimensional inaccuracies, increasing the risk of rejection and requiring additional repairs, such as HIP, with high energy consumption (Figure 1).

In the same regime, porous and friable regions, often located near delaminated areas, indicated incomplete fusion. These extensive LOF zones can lead to micro-cracks and non-conforming densities, requiring scrap or costly energy-intensive treatments (Figure 2).

Also, in R1, a part was stopped after powder replenishment due to contamination, and analysis showed the presence of inclusion-type LOF zones. At the time of the stop, edge curling and delamination were already present, confirming the defect’s propagation to deformation and subsequent dimensional inaccuracy, with powder and machine time loss (Figure 3).

Another case was a large, thin-walled part that exhibited asymmetric curling due to insufficient rigidity relative to shrinkage stresses and the thermal gradient, manifesting as warping that caused a loss of tolerance and consequently required rework or scrapping (Figure 4).

In the R2 regime, images documented embedded impurities, such as a completely encapsulated hair, associated with LOF and delamination—defects that compromise yield and increase the burden of NDT tests and repair operations (Figure 5).

Additionally, a 6 mm sphere, part of a bearing, showed an incomplete zone at the top, resulting from process resolution limitations, which generated LOF and dimensional deviations that necessitate machining or rejection (Figure 6).

A different part showed a positioning error: a cantilevered portion curled, the recoater arm struck and displaced the part, and the build continued with incorrect layer registration, rough surfaces, and global distortions (Figure 7).

In another case, a similarly designed part was analysed in comparison to the CAD model, and the images in Figure 8a,b show layer-by-layer offsets and a loss of concentricity in the ring-like features, confirming the propagation of geometric deviations and surface roughness into dimensional inaccuracies.

Furthermore, a minor edge deformation compromised assembly, generating local adhesion between parts, with geometric losses and time and energy penalties (Figure 9a,b).

The R2 regime also showed nominally flat surfaces with steps and adhering powder particles, i.e., high roughness that promotes the initiation of micro-cracks and leads to dimensional deviations (Figure 10).

On curved surfaces, the stepping effect was also observed, accentuated by the high ratio of layer thickness to curvature radius—a typical manifestation of layer-by-layer fabrication that translates into roughness and increased post-processing effort (Figure 11).

Beyond defects generated directly by the process, a post-processing defect was also documented, originating from air blasting without sufficient cooling of the parts. The images show discrete melting points due to hot residual granules in the form of rounded drops of resolidified material and dirt marks around the impact. These defects do not originate from the physics of the process but from post-processing practices; however, they usually require part rework or scrapping (Figure 12a,b).

Thus, eight defects were identified that will be studied under a network regime to determine the direct and indirect influences between them (

Table 1. Defect set and definitions.

Symbol	Factor	Description
F1	LOF	Regions of unmelted or partially melted powder between scan tracks or successive layers.
F2	Porosity	Internal voids within a part, which are caused by trapped gas or incomplete fusion.
F3	Warping	Macroscopic, out-of-plane distortion of a part’s geometry, such as edge lift.
F4	Delamination	Physical separation or cracking between adjacent printed layers.
F5	Surface roughness	Excessive surface texture irregularity, such as visible stair-stepping or adhered powder.
F6	Contamination	The presence of foreign inclusions (e.g., fibres, oxides, or agglomerates) within the part.
F7	Dimensional inaccuracy	Deviation of the final geometry from the nominal CAD model’s dimensions or tolerances.
F8	Micro-cracks	Sub-millimetre-scale fissures or cracks within the material, often caused by thermal stress.

).

LOF is one of the most critical structural integrity issues in powder-based additive manufacturing [32]. This defect appears as “islands” of unmelted powder or friable regions that fail to bond with the surrounding solidified material. The main causes include insufficient laser or electron beam energy, an excessively high scan speed, or a too-wide spacing between melt tracks. These unmelted zones act as interconnected voids, significantly reducing the material’s density and transforming the affected areas into stress concentration points, which can initiate cracks under loading [32].

Porosity refers to the presence of voids or pores within the part. These can be of two types: LOF pores (irregular and elongated pores caused by incomplete fusion) or “gas” pores (spherical pores formed by gases trapped during solidification) [33]. Regardless of their origin, a high level of porosity reduces the material’s density and mechanical strength, affecting its tensile and fatigue properties [33]. Effective control of process parameters, such as the heat source power and scan speed, is essential for minimizing the formation of these defects [34].

Warping is a macroscopic distortion of the part, which is observable as a curvature or edge lift after the build process is complete [35]. This phenomenon is caused by residual thermal stresses, generated by temperature differences between solidified layers and the new, hot layers deposited on top [35]. As the material cools, uneven contraction leads to warping, which compromises dimensional tolerances and may render the part unusable according to its design, requiring remanufacturing or intensive post-processing [36].

Delamination is the separation of adjacent layers of a part, a defect that severely compromises its structural integrity [37]. It occurs when the interlayer fusion is weak, often due to premature cooling of the lower layer before the next one is deposited. Factors such as inadequate control of ambient or build plate temperature, or excessive delays between layer deposition, can favour the appearance of this defect [37]. A delaminated part has extremely low mechanical strength [38].

Surface roughness is a measure of the part’s surface irregularity, manifested as high Ra/Rz values, a visible “stair-stepping” effect on curves, or residual powder adhesion [39]. This defect is influenced by the part’s orientation on the build plate and the layer thickness [39]. Although it may not directly affect internal mechanical properties, a rough surface can compromise the part’s functionality in applications requiring low friction or a specific aesthetic, often necessitating costly finishing operations [40].

Contamination refers to the presence of foreign inclusions within the part’s material, such as fibres, oxides, or agglomerates of a different powder type [41]. These impurities can come from the working environment or improper material handling. Their presence can reduce the material’s strength, acting as initiation points for micro-cracks or other defects [41]. Rigorous control of the build environment and material purity is essential to prevent this type of defect.

Dimensional inaccuracy represents the deviation of the part from its nominal geometry or specified tolerances [39]. It is a complex defect that can be caused by a multitude of factors, including warping, machine movement errors, or material shrinkage [39]. A high degree of dimensional inaccuracy makes the part non-functional, requiring remanufacturing or subsequent processing, which increases costs and resource consumption [40].

Micro-cracks are microscopic-sized fissures that can be detected visually or with microscopy [42]. They often form as a result of internal thermal stresses and the rapid solidification of the material. Although invisible to the naked eye, they represent structural vulnerabilities that can propagate under loading, leading to the premature failure of the part [42]. Their prevention is crucial for ensuring the long-term durability of the products.

Case studies have shown that defects can lead to the loss of a significant amount of material and energy. For example, in an SLS process using PA2200 powder, powder refreshing is a critical step where unused powder is mixed with new, virgin powder in a recommended ratio of between 50:50 and 70:30 [43]. A single defective build, resulting in scrapped parts, means that this valuable material and the energy invested in heating and fusing it are wasted. While specific cost figures vary, the economic and environmental costs of defects are substantial, with energy consumption for manufacturing potentially reaching up to 100 kWh per kilogram of material [22].

Beyond the individual causes of these defects due to the processes and materials, the current article aims to explore their chain reactions—the network of direct and indirect influences between them. Therefore, understanding these interdependencies is essential for building a robust decision-making framework. A central objective of the present study is to identify the optimal intervention points to reduce material and energy waste.

3. Methodology

This study employed a qualitative research approach grounded in expert judgment to analyse the complex causal relationships among common defects in AM [44]. To navigate the intricate web of interdependencies where one defect can trigger or exacerbate others, a MCDM framework was selected [45]. MCDM methods are analytical tools designed to evaluate complex problems involving multiple, often conflicting, criteria, making them highly suitable for structuring and quantifying expert knowledge in technical domains [24]. They provide a systematic process for breaking down a complex decision into more manageable parts, allowing for a more transparent and robust analysis than purely intuitive assessments [46].

Among the various MCDM techniques, the DEMATEL method was chosen for this research [47]. While other methods like the Analytic Hierarchy Process (AHP) or Analytic Network Process (ANP) are effective for ranking factors or modelling dependencies, the DEMATEL method offers a unique advantage by specifically mapping and quantifying the cause-and-effect relationships within a system [24]. This capability is crucial for the present study, as the primary goal was not to merely rank the severity of AM defects but to understand their underlying causal structure [46]. The DEMATEL method visualizes the system, clearly distinguishing influential “cause” factors, which act as root drivers, from “effect” factors, which are primarily the outcomes of other issues [48]. This distinction provides actionable insights, enabling practitioners to prioritize interventions that address the root causes of quality loss, thereby maximizing the impact on waste reduction and overall process stability [47].

This qualitative study was based on the expertise of professionals with significant, hands-on experience in various AM technologies. Data was collected in June 2025 through a structured questionnaire administered to a panel of six experts. This panel size is consistent with the DEMATEL methodology, which prioritizes the depth and quality of expert insights over statistical sample size [49,50,51]. The selection criteria required the respondents to have a minimum of two years of direct experience in AM, ensuring a high level of practical and theoretical knowledge. The expert panel’s diverse background, spanning multiple application areas, technologies, and materials, provides a comprehensive and well-rounded perspective on defect formation in AM. The detailed profiles of the expert panel are presented in

Table 2. Expert panel.

Expert	Main Application Area(s)	Main AM Technology/Technologies Used	Years of Experience in AM
E1	Industrial/Prototyping, Research	SLM/DMLS, FDM/FFF, SLA/DLP	More than 5 years
E2	Aerospace	DED	2–5 years
E3	Industrial/Prototyping	SLM/DMLS, BJ	More than 5 years
E4	Industrial/Prototyping, Research	FDM/FFF, SLS	2–5 years
E5	Aerospace, Industrial/Prototyping	WAAM	2–5 years
E6	Aerospace, Biomedical, Automotive, Industrial/Prototyping, Educational, Research	WAAM	More than 5 years

.

These evaluations served as input for the DEMATEL analysis, which allows for the identification of key cause–effect relationships for strategic decisions in AM.

The questionnaire was structured as a relationship matrix where the experts were asked to evaluate the direct influence of each defect (factor $i$) on every other defect (factor $j$). The evaluation was performed using a predefined integer scale:

• 0 = no influence;
• 1 = weak influence;
• 2 = moderate influence;
• 3 = strong influence;
• 4 = very strong influence.

The data collected from the expert panel were processed through a series of structured steps to construct the causal model. The next step involved normalizing the matrix of direct relationships $Y$ by adjusting the values in matrix $A$ so that they are suitable for further calculations. The consistency of the expert responses was a critical consideration. The mathematical steps of the DEMATEL method, which involve matrix normalization and calculation of the total influence matrix, ensure that the final results are a stable and consistent representation of the collective expert judgment [52]. To ensure the reproducibility and transparency of the study, the indirect relationships matrix is presented in Annex A. The normalization was carried out using the following formula:

$$Y=A·k$$

(1)

$$\mathrm{where} A=⌊\begin{array}{cccccc}0& a_{12}& \dots & a_{1j}& \dots & a_{1n}\\ a_{21}& 0& \dots & a_{2j}& \dots & a_{2n}\\ \dots & \dots & \dots & \dots & \dots & \dots \\ a_{i1}& a_{i2}& \dots & a_{ij}& \dots & a_{in}\\ \dots & \dots & \dots & \dots & \dots & \dots \\ a_{n1}& a_{n2}& \dots & a_{nj}& \dots & 0\end{array}⌋$$

(2)

$$k={\displaystyle \frac{1}{{max}_{1\le i\le n}\left(\sum _{j=1}^n{a}_{ij}\right)}} (i,j=1, 2, \dots , n)$$

(3)

Here, $n$ represents the number of factors and ${max}_{1\le i\le n}\left(\sum _{j=1}^n{a}_{ij}\right)$ refers to the largest sum among either the row sums or the column sums. For instance, if the largest row sum in matrix $A$ is 15 and the largest column sum is 18, then $k={\displaystyle \frac{1}{18}}$. Each entry in matrix $A$ is then multiplied by this $k$ to produce the normalized matrix $Y$. This process ensures that all values in the normalized matrix are scaled to a range suitable for subsequent operations, usually between 0 and 1.

Next, the total-relation matrix $T$ is computed to capture the overall influences, both direct and indirect, among the factors. This matrix is obtained from the normalized matrix $Y$ and the identity matrix $I$, following the formula

$$T=Y·{\left(I-Y\right)}^{-1}$$

(4)

The $\alpha$ factor threshold is determined using the following formula:

$$\alpha ={\displaystyle \frac{\sum _{i=1}^n\sum _{j=1}^n\left[t_{ij}\right]}{N}}$$

(5)

Here, $N=n^2$ represents the total number of elements, or relationships, contained in the influence matrix.

The $\alpha$ threshold is commonly applied to exclude minor influences in the total-relation matrix, highlighting only the more meaningful relationships for building the causal map. Any relationship with a value in the total-relation matrix exceeding $\alpha$ is deemed significant. In this study, an $\alpha$ threshold of 0.255 was determined using Formula (5) based on the total-relation matrix. The selection of the $\alpha$ threshold is a crucial step in the DEMATEL methodology as it allows for the distinction between significant and negligible causal relationships. In this study, the α-threshold value was determined by calculating the arithmetic mean of all the elements in the total influence matrix ($T$). This approach is a widely accepted standard in the literature as it serves as an objective filter, ensuring that only influences that exceed the average total effect are retained. Consequently, any relationship with a value in the $T$ matrix that is greater than the $\alpha$ threshold is considered a significant causal or effect factor. By excluding weaker relationships that could introduce ‘noise’ into the analysis, this method reinforces the validity and clarity of the causal map, allowing us to focus on the most essential interdependencies within the system.

This positive threshold ensures that only influences above the average total effect are included, enhancing clarity and emphasizing the most important causal links within the map.

After obtaining the total influence matrix, two important measures can be determined for each criterion: the impact score $D_i$ and the causal score $R_j$.

The impact score $D_i$ for a particular factor indicates the overall influence it has on all the other factors and is computed using the following formula:

$$D_i= {\left[\sum _{i=1}^n{t}_{ij}\right]}_{nx1}={\left[t_i\right]}_{{nx1}^{\prime }}$$

(6)

$$\mathrm{for} \mathrm{each} i\in \{\mathrm{1,2},\dots ,n\}.$$

On the other hand, the causal score $R_j$ represents the total influence that a specific factor receives from all other factors and is calculated as follows:

$$R_j= {\left[\sum _{j=1}^n{t}_{ij}\right]}_{1xn}={\left[t_j\right]}_{{nx1}^{\prime }}$$

(7)

$$\mathrm{for} \mathrm{each} j\in \{\mathrm{1,2},\dots ,n\}.$$

4. Results

The first step of the DEMATEL analysis generated the normalized direct relation matrix, which reflects the relative strength of direct influences among the eight identified defects (F₁–F₈). Normalizing the values ensures that all influences are scaled comparably, eliminating distortions caused by differences in measurement scale and providing a coherent foundation for systemic analysis. This matrix ensures that all values are scaled between 0 and 1, providing a consistent basis for subsequent calculations (

Table 3. Normalized direct relationship matrix.

Factor	F1	F2	F3	F4	F5	F6	F7	F8
F1	0.000	0.115	0.069	0.126	0.092	0.057	0.138	0.126
F2	0.092	0.000	0.023	0.057	0.138	0.092	0.069	0.138
F3	0.103	0.034	0.000	0.184	0.057	0.069	0.184	0.126
F4	0.138	0.046	0.126	0.000	0.080	0.069	0.218	0.138
F5	0.092	0.092	0.057	0.103	0.000	0.069	0.138	0.103
F6	0.138	0.126	0.080	0.103	0.092	0.000	0.161	0.172
F7	0.103	0.034	0.103	0.092	0.057	0.034	0.000	0.057
F8	0.057	0.103	0.046	0.103	0.080	0.069	0.092	0.000

).

Based on the normalized matrix $Y$, the total-relation matrix $T$ was computed using Equation (4). This matrix captures both direct and indirect influences among defects, thus offering a comprehensive representation of interdependencies. By quantifying the cumulative effects, this matrix allows for the identification of defect interactions that might not be evident in pairwise analyses, supporting a more strategic approach to defect mitigation. By applying the $\alpha$ threshold (0.255), only the most significant relationships were retained for the causal analysis.

Table 4. Total-relation matrix highlighting factors with significant influence (*).

Factor	F1	F2	F3	F4	F5	F6	F7	F8
F1	0.197	0.254	0.213	0.320 (*)	0.249	0.183	0.387 (*)	0.337 (*)
F2	0.251	0.137	0.148	0.232	0.267 (*)	0.195	0.288 (*)	0.317 (*)
F3	0.305 (*)	0.190	0.165	0.384 (*)	0.224	0.197	0.448 (*)	0.348 (*)
F4	0.342 (*)	0.209	0.283 (*)	0.237	0.252	0.203	0.485 (*)	0.367 (*)
F5	0.265 (*)	0.222	0.191	0.284 (*)	0.150	0.181	0.364 (*)	0.299 (*)
F6	0.355 (*)	0.295 (*)	0.249	0.341 (*)	0.281 (*)	0.152	0.452 (*)	0.418 (*)
F7	0.235	0.139	0.202	0.237	0.168	0.124	0.194	0.214
F8	0.211	0.213	0.161	0.256 (*)	0.205	0.166	0.293 (*)	0.178

highlights the factors that exerted notable influence across the system with an asterisk (*).

Using the calculated the $D$ and $R$ vectors, the role of each defect within the system was determined. The $D+R$ value indicates the total importance or prominence of factor $i$. A higher value signifies greater involvement in the system. The $D-R$ value determines the factor’s net role (

Table 5. Assessment of the influence and causality of the factors.

Factor	D	Factor	D + R	D − R	Dominant Characteristic
F1	2.140	F1	4.303	−0.022	Effect
F2	1.836	F2	3.496	0.175	Cause
F3	2.261	F3	3.875	0.647	Cause
F4	2.378	F4	4.669	0.087	Cause
F5	1.956	F5	3.754	0.158	Cause
F6	2.544	F6	3.945	1.142	Cause
F7	1.517	F7	4.428	−1.394	Effect
F8	1.686	F8	4.166	−0.792	Effect

). If $D-R$ is positive, the factor is a net “cause” exerting primary influence on other defects, whereas a negative $D-R$ denotes a net “effect,” indicating that the factor is predominantly influenced by the system. This distinction provides actionable insight for prioritizing interventions in additive manufacturing processes, ensuring that resources are used to target the most impactful defects first.

The causal relationships identified in our analysis can be effectively visualized in a causal diagram, which is provided in Figure 13. This diagram plots the net influence ($D-R)$ on the X-axis and the net relation ($D+R$) on the Y-axis, providing a clear visual representation of each defect’s role in the network. Defects with a positive $D-R$ value are classified as ‘causes’, while those with a negative value are ‘effects’. The diagram shows the position of each defect and the interdependencies between them, offering a comprehensive and intuitive overview of the findings. Arrows indicate the causal relationships and the direction of influence between the defects.

Defects such as contamination (F₆), warping (F₃), delamination (F₄), surface roughness (F₅), and porosity (F₂) showed positive $D-R$ values, indicating that they act predominantly as causes. Among these, contamination (F₆) stood out with the highest $D-R$ (1.142), suggesting that it is the most influential driver of the other defects. Similarly, warping (F₃) and delamination (F₄) play significant causal roles, highlighting the importance of thermal stress management and interlayer bonding quality in the AM process.

In contrast, dimensional inaccuracy (F₇), micro-cracks (F₈), and LOF (F₁) presented negative $D-R$ values, classifying them as effect factors. Dimensional inaccuracy (F₇) had the lowest $D-R$ (–1.394), making it the most sensitive to the other defects, especially those related to process instabilities such as warping or delamination. Micro-cracks (F₈) also appeared as downstream outcomes of stress concentration and contamination, while LOF (F₁), although critical for mechanical integrity, was more often the result of other systemic issues rather than a primary driver.

When considering the $D+R$ values, which reflect the total involvement of each factor in the network, delamination (F₄), dimensional inaccuracy (F₇), and LOF (F₁) emerged as the most prominent defects, with scores above 4.3. This underlines their central position within the defect system: even if some act as effects, their strong interconnections make them crucial points of attention.

The DEMATEL analysis suggests that addressing root causes such as contamination (F₆), warping (F₃), and porosity (F₂) could lead to significant reductions in downstream defects like dimensional inaccuracy (F₇) and micro-cracks (F₈). Therefore, strategies aimed at stricter powder handling, optimized thermal management, and improved process parameter control are expected to yield the greatest improvements in AM part quality while also reducing rework and waste.

In summary, the DEMATEL-based analysis has enabled the identification of both influential root-cause defects and the more reactive, downstream defects in the additive manufacturing process. This structured evaluation highlights the interconnected nature of defects and underlines the necessity of addressing upstream issues to achieve system-wide quality improvements. By prioritizing interventions targeting dominant causal factors such as contamination, warping, and delamination, it is possible to reduce the incidence of critical effect-type defects like dimensional inaccuracies and micro-cracks. These insights form the foundation for a more targeted and efficient quality control strategy in AM, paving the way for improved process stability, reduced material waste, and enhanced part reliability.

5. Discussion

Recent work has consolidated a mechanism-level understanding of laser powder bed fusion (LPBF) defects, yet these studies rarely quantified inter-defect causality. Authoritative reviews mapped defect types, origins, and property implications, but they tend to treat these phenomena in isolation rather than as a coupled system. This includes Mostafaei et al. [53], who catalogued LPBF anomalies and their formation mechanisms, and Haribaskar et al. [54], who synthesized defect classes in metal AM and linked them to performance, both without ranking defects by network influence.

Complementary strands of research emphasize process-defect regimes and real-time detection. Gordon et al. [55] introduced defect–structure–process maps that delineate parameter windows for common defects like LOF, keyholes, and balling. In addition, in situ and machine-learning studies, such as those by Feng et al. [56] and Zhao et al. [57], achieved layer-wise defect prediction and multi-defect detection. Recent reviews of monitoring technologies also underlined the shift toward closed-loop control.

Against this backdrop, the present DEMATEL-driven analysis contributes a network perspective that captures both direct and indirect interactions among defects. By doing so, it yielded $D$, $R$, $D+R$, and $D-R$ scores that allow for the prioritization of root-cause nodes. This approach complements mechanism-centric and detection-centric studies by providing a quantified basis for cross-study comparisons and targeted quality-control interventions with clear sustainability relevance.

Previous case studies quantified how raising first-pass yield and reducing scrap translate into measurable sustainability gains in AM. At the factory scale, Cozzolino et al. [58] performed a gate-to-gate LCA of AlSi10Mg LPBF jobs and reported a per-job printing energy of 233–651 MJ plus 173–482 MJ for the chiller, together with 500–800 kg of argon and 6–9 kg of solid waste; by definition, each avoided failed build preserves these burdens and proportionally lowers the global warming potential (GWP) and cumulative energy demand per delivered part. Complementarily, Rotter and Fagerberg [59] showed, using primary industrial data, that argon can dominate LPBF’s GWP and that process optimization can reduce the footprint by more than 75%, underscoring the higher yield and fewer reprints. At the component level, Puukko et al. [60] quantified a 40% reduction in use-phase carbon footprint for an LPBF nozzle; holding the manufacturing route constant, achieving first-time-right printing reduced the manufacturing share of the footprint in direct proportion to avoided scrap. Process-design studies likewise indicate that stabilizing parameter windows to prevent porosity/LOF reduces time and energy per good part while sustaining properties: Mercurio et al. [61] increased AlSi10Mg throughput by approximately 65% at 90 μm layers with CT-verified density while emphasizing powder recycling and lower waste generation. Taken together, these case-based results indicate that interventions that demonstrably decrease reject rates (e.g., parameter optimization, in-process monitoring, and feedstock-handling protocols) have quantifiable sustainability benefits, lower cumulative energy demands, greenhouse-gas emissions, argon use, and material discard, aligning with the mitigation priorities identified in our DEMATEL-guided causal network.

The DEMATEL-driven causal mapping of defects in AM offers a novel systemic perspective that complements and extends prior research by moving beyond isolated defect analysis. While most studies have traditionally focused on individual defect mechanisms, our approach situates these defects within an interconnected network, capturing both direct and indirect interactions and allowing for more nuanced comparisons across studies. Porosity, for instance, emerged as a primary causal factor in our model, corroborating the findings of Al-Maharma et al. [62], Laleh et al. [63], and Frazier [64], who emphasized the influence of pore morphology and distribution on fracture toughness, stiffness, and fatigue strength. Similarly, Li et al. [65] highlighted the disproportionate impact of LOF pores on vertically built AM components due to stress concentration effects, accelerating crack initiation. Unlike these prior studies, which primarily quantify the consequences of porosity on mechanical properties, our analysis demonstrated that porosity functions as a systemic defect capable of propagating secondary failures such as micro-cracks and dimensional inaccuracies, reinforcing its role as a root cause rather than a mere byproduct of processing errors.

Contamination similarly emerged as a significant causal defect in our network, echoing the findings of Slotwinski et al. [66], Yi et al. [67], and Pleass et al. [68], who showed that micro-scale powder contamination can lead to inclusions that compromise both homogeneity and process repeatability. Other studies, such as those by Singh et al. [69] and Mostafaei et al. [53], have emphasized how cross-contamination or inconsistent feedstock quality can undermine reliability, particularly in titanium alloys and nickel-based superalloys, where even minor impurities drastically affect mechanical performance. While previous works have treated contamination largely as a material characterization issue, our DEMATEL-based framework situates it within a system of interdependent defects, revealing that contamination not only introduces immediate imperfections but also catalyses cascading effects that amplify dimensional deviations, porosity formation, and microstructural inconsistencies. This systemic insight underscores the critical importance of feedstock handling protocols and aligns with empirical observations from multiple powder-based AM studies [53,66,67,68,69] while simultaneously providing a quantifiable measure of causal leverage through $D-R$ scores.

Warping and thermal distortions, another set of influential causal defects identified in our analysis, are well-documented in the literature. Finite element studies of fused deposition modelling and laser powder bed fusion by Cao et al. [70], Mercelis and Kruth [71], and Mostafaei et al. [72] consistently showed that poor infill orientation, insufficient support structures, or uneven thermal gradients exacerbate warping, leading to compromised dimensional accuracy and surface defects. Our networked analysis confirmed these findings but extends them by illustrating how warping acts as a hub defect that propagates through the system, indirectly contributing to porosity exacerbation and micro-crack formation. This integrated perspective reveals feedback loops that traditional experimental or simulation-based studies rarely capture, highlighting the interplay between thermal, mechanical, and geometric factors in defect proliferation.

The novelty of our work lies not merely in reaffirming the significance of porosity, contamination, and warping, but in embedding them within a mathematically quantified causal network. By applying a significance threshold ($\alpha = 0.255$) and analysing $D+R$ and $D-R$ scores, we filter out negligible interactions and prioritize defects based on systemic influence rather than occurrence frequency alone. This contrasts with conventional studies, which typically describe defect severity in isolation or focus on single-factor interventions. For example, while Al-Maharma et al. [62], Laleh et al. [63], and Frazier [64] provided detailed assessments of pore effects, they did not contextualize porosity within a network that includes contamination and warping, nor did they quantify the relative leverage of each defect in propagating failures. Our framework addresses this gap, offering both theoretical and practical guidance for prioritizing mitigation strategies.

Comparing our findings with the broader AM literature further highlights the robustness of our approach. Studies on powder recycling by Yi et al. [67], Singh et al. [69], and Mostafaei et al. [53] demonstrated that repeated use can increase contamination and pore formation, consistent with our identification of these defects as primary drivers. Similarly, research on build orientation and thermal management [53,67,69] corroborates our observation that warping has cascading effects on downstream defects. Moreover, by integrating both direct and indirect influences, our causal network revealed interactions that might remain hidden in experimental observations or single-defect simulations, such as the interplay between thermal distortions and pore coalescence, or between contamination-induced inclusions and micro-crack initiation. This systemic lens, grounded in quantitative analysis, not only aligns with diverse experimental and computational studies but also transcends them by providing a unified framework to assess defect propagation and prioritize interventions.

Although our DEMATEL-based methodology relies on expert judgment, the resulting defect causality network is strongly supported by both qualitative and quantitative evidence from the academic literature. This section provides a comprehensive validation, aligning our findings with established empirical data and statistical analyses.

Our model identified porosity as a significant causal factor with a direct influence on the final part’s state. This finding is validated by research that has quantified the impact of porosity on structural integrity. A detailed study [73] focusing on Ti-6Al-4V alloys produced via additive manufacturing showed a direct negative correlation between porosity and a material’s dynamic properties. Specifically, it was found that the Hugoniot elastic limit and spall strength decrease monotonically as porosity increases. This is a critical observation as it demonstrates that even a low porosity level, such as 3%, can substantially reduce the material’s mechanical strength and compromise a component’s performance under dynamic loading conditions [73]. This quantitative data from external research confirms our qualitative assessment of porosity as a root cause.

Furthermore, our analysis established a clear causal relationship between warping and delamination, a prevalent issue in Fused Deposition Modelling (FDM) processes. The literature fully supports this mechanism, explaining that delamination is often a direct consequence of the thermal residual stresses that cause part warping. A study on this topic explicitly states that warping is caused by the “residual thermal strain accumulated during the printing,” while delamination occurs when “the adhesion between two layers is weak” [74]. These interlayer imperfections are a major challenge in additive manufacturing as they can lead to print job failures, resulting in significant waste of both material and energy, which is a central focus of our work.

The issue of defect accumulation across layers also validates the importance of a root-cause approach. A Six-Sigma quality management study in additive manufacturing provided an essential statistical perspective: for a part with 100 layers, even with a low probability of defects per layer (0.0114), the overall probability of the final build containing at least one defect was found to be a high 68.23% [75]. This powerful statistic demonstrates how minor, seemingly independent defects can accumulate from one layer to the next, leading to a non-conforming final product. This reinforces why a methodology like the DEMATEL method, which identifies and prioritizes root causes over symptoms, is vital for preventing defect propagation and substantially improving production yield. This quantitative evidence validates our core hypothesis that a preventive, cause-focused approach is key to enhancing the sustainability of additive manufacturing processes.

The economic and environmental impacts of AM defects are significant. In an ideal scenario, AM can save over 90% of materials compared to subtractive methods [21]. However, as mentioned earlier, each defect can result in a scrapped part, leading to considerable waste. Studies have shown that a single defective build can result in the loss of several kilograms of valuable powder. For example, using a standard build volume on an EOS P 396 system, a failed build can lead to the loss of up to 4 kg of PA2200 powder. Considering the cost of raw powder and the energy consumed during the printing process, which can be up to 100 kWh per kilogram of material [22], the economic and environmental costs of defects are substantial. By reducing these defects, companies can significantly reduce material consumption, energy use, and operational costs, thereby improving the overall sustainability and profitability of their AM operations.

A major strength of our methodology is its generalizability beyond the specific powder-based systems studied. While our expert panel and case study focused on Powder Bed Fusion (PBF), the DEMATEL framework is technology-agnostic. Defects such as porosity and dimensional inaccuracy are universal challenges across AM technologies, from Stereolithography (SLA) to Wire Arc Additive Manufacturing (WAAM) and Electron Beam Melting (EBM). For example, porosity is a critical defect in EBM, similar to PBF, and warping is a central issue in both SLA and Material Extrusion. Therefore, even though the specific causes may vary, the fundamental causal relationships and effects identified in our study provide a valuable perspective that can be extended and adapted to a wide range of AM processes. Future research could apply this DEMATEL model to other technologies to determine if the identified root causes remain the most influential, thereby providing a more comprehensive framework for sustainable quality management in the broader AM industry.

Our study connects the fundamental mechanisms of individual defects to their broader, system-level effects. While previous research has detailed the formation of flaws like porosity and warping in isolation, our work situates them within a causal network that quantifies their interdependencies and relative influence. This network approach provides a strategic, evidence-based framework for AM quality control, allowing resources to be focused on mitigating the root defects that have the greatest systemic impact—ultimately improving reliability, reducing waste, and enhancing part performance. Our results are not only consistent with established findings but also demonstrate the unique value of this quantified methodology, marking a critical shift toward a more holistic and impactful approach to quality management in additive manufacturing.

6. Conclusions

By using the DEMATEL methodology, this study successfully mapped the complex network of cause-and-effect relationships among eight critical defects encountered in additive manufacturing. The analysis clearly distinguishes between fundamental, “cause”-type defects and “effect”-type defects, offering a systemic perspective on how quality issues propagate during the manufacturing process. The results unequivocally identified contamination (F₆) as the most influential root cause, followed by warping (F₃), porosity (F₂), delamination (F₄), and surface roughness (F₅). In contrast, dimensional inaccuracy (F₇), micro-cracks (F₈), and LOF (F₁) were classified as “effect”-type defects, which are primarily the result of upstream problem propagation.

The originality of this study lies in its systemic approach, which goes beyond the traditional, isolated analysis of defects. By applying a multi-criteria decision-making method based on expert judgment, the paper provides a quantifiable causal map of the interdependencies between defects. This network perspective is innovative as it directly links defect formation mechanisms with tangible sustainability outcomes, such as reductions in waste, rework, and energy consumption. Thus, the study does not merely describe the problems but also hierarchizes their root causes, offering a strategic and novel perspective on quality control in additive manufacturing.

From a practical standpoint, the results offer a clear and effective framework for action for engineers and operators in the additive manufacturing field. The causal map demonstrates that quality control efforts must be primarily focused on preventing root-cause defects. Implementing rigorous material handling protocols (to eliminate contamination), optimizing build strategies (to manage thermal stress and minimize warping), and fine-tuning process parameters (to reduce porosity) will have the greatest impact in preventing a chain of secondary defects. This proactive, cause-focused approach is fundamentally more efficient than reactive efforts to correct final defects, such as dimensional inaccuracy. Therefore, the study provides a directly applicable decision-support tool for increasing process reliability, reducing costs, and promoting more sustainable production.

While the present research offers a valuable perspective on the causality network of defects in additive manufacturing through the application of the DEMATEL method, it is essential to acknowledge its limitations. Firstly, the validity of our conclusions relies exclusively on the judgments of a panel of six experts. Although this methodology is standard for the DEMATEL method, the absence of quantitative data from real production environments or independent experimental validation represents a critical limitation. Consequently, the resulting cause-and-effect map reflects a qualitative understanding based on specialized knowledge, but it has not been validated with direct empirical evidence. Another limitation is that our methodology was applied to a predefined set of eight common defects. Variations in material types, AM technologies (e.g., EBM, SLA), and process conditions could alter the structure of the causality network.

Based on our findings and the identified limitations, we propose several directions for future research. Firstly, an experimental validation of our conclusions is imperative. This could involve collecting data from mass production and recording the frequency and severity of each defect to verify whether the causal factors identified by the DEMATEL method (such as contamination and warping) do indeed have a proportional impact on effect-type defects (such as dimensional inaccuracy). Furthermore, it would be relevant to expand the study by applying the DEMATEL method to a wider variety of materials and additive manufacturing technologies. An interdisciplinary approach could combine DEMATEL modelling with quantitative risk analysis or artificial intelligence methods, such as neural networks, to predict the occurrence of defects based on process parameters. This study’s findings suggest several promising directions for future research. A key focus should be developing real-time predictive models using machine learning or AI to anticipate defect formation. The framework could then be extended to a broader range of AM processes and materials, like metal powder bed fusion, to compare defect causality networks across different technologies. To validate these models, longitudinal studies are needed to test the identified causal relationships with empirical data from real-world production runs. Additionally, targeted case studies in critical sectors, such as the biomedical and aerospace industries, would provide crucial insights into performance in safety-critical applications. Future work should also investigate how to systematically integrate this production feedback into decision-making models, allowing for dynamic, adaptive quality control strategies that minimize waste and optimize process efficiency.

In conclusion, by identifying the critical levers within the defect network, this research contributes to the maturation of additive manufacturing, transforming knowledge about quality into a quantifiable strategy for resource efficiency and industrial performance.