Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
Next Article in Journal
Investigation of the Fabrication Parameters’ Influence on the Tensile Strength of 3D-Printed Copper-Filled Metal Composite Using Design of Experiments
Previous Article in Journal
Evaluating Energy Efficiency and Optimal Positioning of Industrial Robots in Sustainable Manufacturing
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
JMMP
Volume 8
Issue 6
10.3390/jmmp8060277
jmmp-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles thumb_up
...
Endorse textsms
...
Comment
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Review

Root Cause Analysis in Industrial Manufacturing: A Scoping Review of Current Research, Challenges and the Promises of AI-Driven Approaches

by
Dominik Pietsch
1,2,*,†,
Marvin Matthes
3,*,†,
Uwe Wieland
1,3,
Steffen Ihlenfeldt
2 and
Torsten Munkelt
3
1
Volkswagen AG, Berliner Ring 2, 38440 Wolfsburg, Germany
2
Institute of Mechatronic Engineering, TUD Dresden University of Technology, Helmholtzstr. 7a, 01069 Dresden, Germany
3
Faculty of Informatics/Mathematics, Hochschule für Technik und Wirtschaft Dresden—University of Applied Sciences, Friedrich-List-Platz 1, 01069 Dresden, Germany
*
Authors to whom correspondence should be addressed.
†
These authors contributed equally to this work.
J. Manuf. Mater. Process. 2024, 8(6), 277; https://doi.org/10.3390/jmmp8060277
Submission received: 4 November 2024 / Revised: 25 November 2024 / Accepted: 28 November 2024 / Published: 2 December 2024
Browse Figures
Versions Notes

Abstract

:
The manufacturing industry must maintain high-quality standards while meeting customer demands for customization, reduced carbon footprint, and competitive pricing. To address these challenges, companies are constantly improving their production processes using quality management tools. A crucial aspect of this improvement is the root cause analysis of manufacturing defects. In recent years, there has been a shift from traditional knowledge-driven approaches to data-driven approaches. However, there is a gap in the literature regarding a systematic overview of both methodological types, their overlaps, and the challenges they pose. To fill this gap, this study conducts a scoping literature review of root cause analysis in manufacturing, focusing on both data-driven and knowledge-driven approaches. For this, articles from IEEE Xplore, Scopus, and Web of Science are examined. This review finds that data-driven approaches have become dominant in recent years, with explainable artificial intelligence emerging as a particularly strong approach. Additionally, hybrid variants of root cause analysis, which combine expert knowledge and data-driven approaches, are also prevalent, leveraging the strengths of both worlds. Major challenges identified include dependence on expert knowledge, data availability, and management issues, as well as methodological difficulties. This article also evaluates the potential of artificial intelligence and hybrid approaches for the future, highlighting their promises in advancing root cause analysis in manufacturing.

1. Introduction

In modern global economy, the manufacturing industry plays a pivotal role in driving economic growth [1]. The manufacturing sector is currently facing a number of significant challenges, including the need to accommodate mass customization, disruptive innovations, and the necessity to meet increasingly stringent quality standards [1]. Furthermore, there is an increasing focus on the ecological sustainability of production and supply chains [2,3]. The necessity for sustainability across all aspects of production is becoming increasingly evident, driven by heightened environmental awareness and evolving customer expectations [4]. Concurrently, new technologies, such as robotics, the Internet of Things, and cyber-physical systems, are revolutionizing the manufacturing industry [5]. These advancements, coupled with existing challenges, such as high complexity, market volatility, and dynamic production conditions [6], underscore the need for continued innovation and adaptation within the sector.
In order to effectively address these challenges, manufacturers invest heavily in research and development, for example, in technologies that are capable of pinpointing the specific causes of machine breakdowns and product defects [7]. Understanding the underlying issues of machine breakdowns and product defects by analyzing the root cause of a problem is crucial, as it allows for the implementation of targeted solutions that enhance equipment reliability and boost overall productivity.
Established in research and the industrial context, root cause analysis (RCA) is a systematic approach used to identify the underlying reasons for problems that arise, facilitating the implementation of corrective actions [8]. It is a methodological tool that helps to uncover the root cause of problems in any production process. RCA is widely used in the manufacturing industry, particularly to determine the causes of repeating machine breakdowns [9,10], product defects [11], and to improve efficiency [12].
Traditionally, experienced workers with in-depth knowledge of the processes have been relied upon to identify root causes of manufacturing issues [13]. They may use quality management tools such as 5-Why analysis or Ishikawa diagrams to trace the problem back to its source [14]. However, the advent of Industry 4.0 has brought about significant implications concerning efficiency, sustainability, production management, etc. [15]. Manufacturing becomes a more complex yet smarter operation due to the integration of technology, including cloud computing as well as data analytics and artificial intelligence (AI) [16]. The availability of extensive data has revolutionized problem detection, advocating for a synergistic approach that melds traditional expertise with modern, data-driven techniques for comprehensive problem solving.
Given the transformative impact of upcoming AI and machine learning (ML) on RCA, it is necessary to conduct a comprehensive analysis of the various RCA approaches employed across the manufacturing sector. The analysis should go beyond traditional and data-driven approaches and explore the wide range of RCA practices that companies and researchers have adopted, modified, and innovated based on practical needs and technological advancements. Therefore, conducting a scoping literature review (ScR) is essential. The purpose of this review is to summarize the various RCA approaches that have emerged and provide an overview of the current trends, challenges, and directions within the field of manufacturing. Focusing exclusively on the manufacturing and production domain is motivated by the sector’s unique demands for precision, consistency, and competitiveness, where even minor defects can lead to significant financial losses, safety hazards, and production delays, making root cause analysis essential for maintaining high standards and competitive advantage. In contrast, this analysis will not be limited to singular RCA approaches, whether knowledge-driven or data-driven, but will instead encompass a broad spectrum of applications and approaches from a holistic perspective. This methodology ensures a thorough understanding of how RCA practices are developing in response to the demands of modern manufacturing and the potential synergies between human insight and technological capability. This ScR’s objectives and review questions (RQs) are outlined in Table 1.
This ScR aims to fill the gap in the literature by conducting a scoping review of RCA in manufacturing. This review includes articles from IEEE Xplore, Scopus, and Web of Science over the recent decade. According to Gusenbauer and Haddaway [17], Web of Science and Scopus are well suited as principal search systems, while IEEE Xplore serves as an effective supplementary system for systematic and structured literature reviews.
The objective is to provide a comprehensive overview of the current research landscape, identify recurring challenges, and outline the potential of RCA for industrial manufacturing. This review differs from previous reviews in that it includes both data-driven and knowledge-driven approaches, without limiting itself to either, thus providing a more comprehensive perspective on RCA approaches. Notable reviews in the manufacturing domain have contributed valuable insights, but they exhibit limitations. For instance, Ito et al. [18] concentrates on resilient production systems, while e Oliveira et al. [19] and Koval et al. [20] examine automatic or data-driven RCA. Papageorgiou et al. [21] explores ML approaches for RCA toward zero-defect manufacturing, while Mokhtarzadeh et al. [22] explores hybrid approaches for failure analysis. However, these reviews either concentrate on particular aspects of RCA or do not utilize an systematic review methodology.
The article is structured as follows. Section 2 describes the review method used, including the identification and selection of articles as well as information extraction. Section 3 deals with the analysis of articles and the answering of review questions. This is followed by a summary of the most important findings of this ScR in the final Section 4.

2. Review Method

The subsequent section describes the methodology used to conduct this ScR, aligning with the PRISMA-ScR (Preferred Reporting Items for Systematic reviews and Meta-Analyses extension for Scoping Reviews) guidelines established by Tricco et al. [23]. The checklist provided by the PRISMA-ScR guidelines was rigorously followed. Additionally, the literature review procedure introduced by Kitchenham and Charters [24] was employed. According to Kitchenham and Charters [24], a review begins with the planning phase, which includes identifying the need for the review, defining review questions (see Section 1), and constructing and evaluating the review protocol. This is followed by the actual implementation of the review, which includes the identification and selection of articles, quality assessment, data extraction, and synthesis. The final stage is reporting, including the writing of the review. Based on these phases, the procedure used here is described in the following sections. Additional guidelines and methodologies utilized in this review are discussed in Jane Webster and Richard T. Watson [25] and vom Brocke et al. [26].

2.1. Identification of Articles

The article selection process in this ScR was guided by specific keywords, utilizing the PCC framework (Population, Concept, Context) (see Pollock et al. [27] for further information) in order to construct the search string. Table 2 presents these categories alongside their associated keywords, including synonyms. For the selection of articles, the Population was determined based on the review questions outlined in Table 1, with a focus on “Manufacturing” and “Production”. In terms of Concept, we included all articles that addressed RCA, which could range from traditional knowledge-driven to modern data-driven and to hybrid approaches. Terms such as fault isolation or fault diagnosis, which are occasionally used interchangeably with RCA, were omitted during an iterative refinement of the search string. The results obtained with these keywords did not align with the intended context and were therefore considered irrelevant. The refined search string is depicted in Table 3. Our Context focuses on the applications and implementations, while extracting the occurred challenges and limitations of the used approaches encountered in the research.
In order to extract articles, IEEE Xplore, Scopus, and Web of Science were employed. The search strings were applied specifically to titles and author-specific keywords, leading to the formulation of the search strings illustrated in Table 4. The decision to restrict the search to author keywords and titles was made subsequent to an initial search, which served as part of the refinement process. By omitting the keyword search in abstract and full text, the alignment with the actual topic was better achieved. The final search was conducted on 4 January 2024 and yielded a total of 245 articles: 166 on Scopus, 22 on IEEE Xplore, and 57 on Web of Science.

2.2. Inclusion and Exclusion of Articles

To ensure a systematic and comprehensible selection of relevant articles, clear inclusion and exclusion criteria were established. Table 5 contains the exclusion criteria that were applied during the review process. To ensure topicality, only articles published from 2014 onward were considered. Duplicates were screened out to avoid redundancy. In addition, only primary studies were considered suitable, with the prerequisite that the authors had access to the full text for evaluation.
Table 6 contains the criteria for including studies in this ScR. Three criteria must be fulfilled by the articles, each independently evaluated and discussed by the first two authors. Firstly, the article’s title must align with the context of RCA in manufacturing. Subsequently, the abstract is assessed, followed by the full text.
Figure 1 provides an outline of the article selection procedure. Of the initial 245 articles under consideration, 56 were excluded due to duplicates (EC1), and 53 based on topicality (EC2). Additionally, 32 articles were discarded due to title mismatch with the scope (IC1), and 27 after abstract assessment (IC2). Two articles were identified as review articles (EC3), four were inaccessible (EC4), and 26 were excluded at the full-text screening stage. Consequently, 35 articles met the defined criteria. This ScR article does not employ a backward search to avoid potential bias in the selection of further articles and to streamline this ScR to the most recent RCA approaches in manufacturing, where authors explicitly indicate this in the title or author keywords.

2.3. Information Extraction Procedure

Information extraction from articles was conducted independently by the first two authors. Eight categories were established to extract relevant information essential for addressing the review questions. These categories include the identification of the industrial sector (semiconductor, automotive, chemical, electronics, others), the subject of the RCA problem (product, machine, process), the employed research method (case study, conceptual work, empirical study), utilized data sources, and both knowledge- and data-driven approaches. Additionally, findings and remarks for every article were extracted. Table 7 illustrates the corresponding categories, utilizing the example article by Pohlmeyer et al. [9]. Furthermore, the table illustrates the categories’ linkage to the individual review questions.
For collaborative work, the authors used an online spreadsheet editor together with an reference management and knowledge organization program. Following the information extraction process by the first two authors, the results were discussed in depth. Any discrepancies were resolved through further evaluation of the relevant literature.

3. Results and Discussion

The results of this ScR are presented in the form of a concept matrix, as illustrated in Table 8. The matrix was designed to facilitate the systematic categorization and analysis of the diverse approaches and application domains in the field of RCA in manufacturing. By organizing the data across multiple dimensions, namely industry relevance and the type of approach, the matrix facilitates the discernment of patterns, trends, and gaps within the current research landscape.
The matrix categorizes articles into several industries—semiconductor, automotive, chemical, and electronics—offering insights into the specific applications and challenges within each sector. Articles with industry-specific applications that do not fall into these predefined categories are listed under “others”. If a study does not specify any industry and the method is validated as industry-independent, it is categorized under general manufacturing.
In addition, the division into knowledge-driven, data-driven, and hybrid approaches facilitates an examination of how traditional approaches, which rely on deep domain expertise, compare against modern, data-intensive computational techniques, as well as the integration of both in hybrid approaches. This distinction is of crucial importance for the evaluation of the algorithmic applications’ evolution and their efficacy in addressing complex problems in RCA.

3.1. RQ1: Current Approaches

As discussed in Section 3, the authors categorize current advancements in RCA within the manufacturing sector into three primary approaches: knowledge-driven, data-driven, and hybrid. Knowledge-driven approaches predominantly depend on domain expertise to identify and solve problems. In contrast, data-driven models harness data to train ML/AI models. Hybrid approaches integrate the strengths of both knowledge-driven and data-driven methodologies, utilizing domain knowledge alongside data-driven techniques to effectively address RCA challenges.
Figure 2 illustrates the distribution of articles over the recent decade. It is evident that the number of articles in RCA in manufacturing has significantly increased over the years, with only 2020 and 2021 deviating from this trend. The authors attribute this deviation to the COVID-19 pandemic. Additionally, Figure 2 depicts the proportions of data-driven, knowledge-driven, and hybrid approaches per year. It can be observed that the relative proportion of data-driven and hybrid techniques, particularly in 2022 and 2023, encompassed all articles. In total, 8 knowledge-driven, 20 data-driven, and 7 hybrid approaches were chosen in the selected articles.
The distribution of approaches used in the articles is presented in Table 9 in the form of a heat map. Particularly, the semiconductor industry seems to heavily rely on data-oriented variants of RCA, including hybrid approaches. However, data-driven and hybrid approaches are also frequently associated with the automotive sector and other industry-independent applications. This may be due to the increasing digitalization in the course of the fourth industrial revolution. With the growing volume of data generated within production processes, utilizing data-driven or hybrid approaches for RCA becomes increasingly plausible. The chemical and electronic categories do not demonstrate a clear trend. This may be due to the limited sample size of the study and the fact that some approaches in these fields may not have been published. All other articles assigned to the ”others” category show a broader spectrum with an almost equal distribution between data-driven and knowledge-driven approaches.

3.1.1. Knowledge-Driven Approaches

The distribution of knowledge-driven approaches employed in the articles is presented in Figure 3. Classical quality management techniques such as Ishikawa diagram, 5-Why analysis, and Pareto analysis are frequently utilized in the identified articles. Particularly prominent is the Ishikawa diagram, serving as a structured representation of cause-and-effect relationships concerning the 6M (Machine, Man, Material, Method, Measure, Milieu) [59]. Additionally, there are more general representations of cause-and-effect relationships, such as graph-based formulations utilized.
In their 2017 study, Braglia et al. [28] focused on RCA by extending the Single-Minute Exchange of Die (SMED) methodology using the 5-Why technique to identify bottlenecks. Cyganiuk et al. [35] conducted research employing Ishikawa diagrams for RCA within the framework of an 8D report. Krishnan et al. [43] examined the root causes of bottlenecks in tire manufacturing, employing Ishikawa, 5-Why analysis, and Pareto analysis. Lee et al. [44] explored RCA in automated soldering of electronic components within the electronics industry, introducing the novel cause–effect chain analysis plus (CECA+) approach as an extension to existing cause–effect chains analysis methodologies. In their 2019 study, Rezaei et al. [51] utilized the Six Sigma methodology, employing failure mode and effects analysis, Ishikawa diagrams, and design of experiments to prevent welding defects in pressure vessel manufacturing. Sooraj et al. [55] investigated bottleneck identification in crane production, utilizing Ishikawa diagrams and 5-Why analysis. Tuninetti et al. [10] examined RCA in veneer manufacturing. They employed Ishikawa diagrams for root cause identification and further investigated the issue through finite element analysis and vibration measurements. Vo et al. [11] conducted a case study in consumer goods manufacturing, specifically focusing on soap dispensers, employing Ishikawa diagrams and the 5-Why technique for analysis.

3.1.2. Data-Driven Approaches

The distribution of data-driven approaches utilized in the articles depicted in Figure 4 indicates a wide spectrum of approaches being employed. It comprises approaches from the explainable artificial intelligence (XAI) domain, as well as neural networks, graph-based models, and others, which are employed for RCA in manufacturing. Also, the combination of individual approaches, such as classification through a tree-based approach followed by the interpretation of feature importance, is applied for RCA [40]. It is important to note that these categories are not mutually exclusive; thus, some of the articles may utilize combinations of these categories. For instance, an article may discuss the interpretation of a random forest (RF), which is a combination of the categories encompassing explainable AI- and tree-based models.
In the XAI domain, Brochado et al. [29] managed to identify person- and shift-dependent influences or causes for performance differences in a production line by interpreting an XGBoost tree via Shapley additive explanations (SHAPs). Soler et al. [54] adopted a similar approach for analyzing production defects in injection molding, using both local and global interpretation approaches from the XAI domain (SEQENS and SHAP TreeExplainer, as well as Individual Condition Expectation for global and local interpretation, respectively) to evaluate an XGBoost model trained beforehand for quality prediction. Features identified as particularly important for prediction through interpretation serve as possible causes for production deviations. In a case study of semiconductor production, Kim et al. [40] demonstrated how the feature importance of a RF can be used to identify critical manufacturing steps that may cause failures. Tan et al. [56] also examined semiconductor industry production, employing binary classifiers (RF and neural networks) for quality assessment based on process data and subsequent multi-feature shuffle operation to extract and interpret the most important features for quality deviations. In 2022, Cho and Kim [33] presented the interpretation of convolutional neural networks (CNNs) for RCA in the steel industry, where process time-series data were processed for quality prediction. Posterior interpretation identified sources for occurring deviations. In the three articles by e Oliveira et al. [36], e Oliveira et al. [37] and e Oliveira et al. [38], the problem of RCA for identifying defect-causing machines in sequential manufacturing with overlap was described. Because multiple machines can perform the same manufacturing step, determining the faulty machine retrospectively is challenging. Using approaches from the field of causal inference and feature ranking, the authors developed a solution to the problem and validated it through a case study in the semiconductor industry. In contrast to the frequently used model interpretation, Kozjek et al. [42], Pohlmeyer et al. [9], Ong et al. [49], and Sariyer et al. [53] employed techniques from the realm of rule mining, where the generated rules are directly interpretable by operators. Kozjek et al. [42] extracted rules from a pre-trained decision tree and performed rule-based root cause analyses through a case study in plastic injection molding. Pohlmeyer et al. [9] utilized association rule mining within a case study in the textile industry for creating rules between failure categories and variables describing the faulty process state. The identified rules directly indicate which variables are potential causes for each failure category. Ong et al. [49] demonstrated that, particularly when working with unbalanced datasets that are common in production environments—where only a few items are not okay (NOK) but many are okay (OK)—weighting rules, such as weighted association rule mining, can be beneficial. Validation was performed through an investigation in the semiconductor industry. Finally, Sariyer et al. [53] also employed association rule mining for identifying hidden rules for root causes. The authors demonstrated an application in the household goods manufacturing sector.
In addition to interpretable or white-box approaches, black-box procedures are also employed for data-driven RCA. For example, Bui [30] used generative adversarial networks for discovering causes of product variation. Validation was conducted through two case studies analyzing the production of cylinder head gaskets in the automotive sector and cylindrical components. Hong et al. [39] utilized CNNs to find relationships between individual manufacturing steps and end-of-line quality assessment in semiconductor production, thus identifying individual process steps as causes for specific quality deviations. A similar goal was pursued by Lee et al. [12] for the RCA of low yield in the semiconductor industry. The approach employed was based on long short-term memory (LSTM) networks with attention, with the objective of identifying features with high attention as causes. Chien and Chuang [31] focused on RCA in sub-batch processing systems in the semiconductor industry, employing a RF for a feature selection process and subsequently stepwise regression.
Furthermore, approaches from the realm of causality and often associated graph-based approaches exist for determining cause-and-effect relationships. Among others, Ma et al. [46,47] developed a procedure for constructing a directed acyclic graph using Granger causality and LSTM or gated recurrent unit (GRU) networks with attention. This approach enabled the determination of causes in a case study in the field of hot rolling (forming technology). A later article by Zhang et al. [58] continued to work on the same problem, utilizing Gonzalo–Granger decomposition and attention-based GRU for causality analysis and subsequent RCA through path propagation analysis.
Beyond the previous thematic blocks, there exist additional, usually very problem-specific approaches. For instance, Otsubo et al. [50] demonstrated how Bayesian inversion can be used for RCA in the assembly of a camera module. Chouichi et al. [34] employed analysis of variance and partial least squares for identifying causes of product quality differences across different machines. Finally, Chien et al. [32] showed how root cause analyses can be performed again in the semiconductor industry using Cramer’s V correlation and logistic regression.

3.1.3. Hybrid Approaches

As shown in Table 8, the hybrid approaches found within the ScR are categorized into three specific groups: initialization and interaction; feature selection and grouping; and post-hoc analysis and validation.
Initialization and interaction approaches often involve the initial setup of models by experts and further refinement through data-driven techniques. For example, a notable approach for RCA was conducted in Kornas et al. [41]. In this study, CERs identified by experts in battery cell manufacturing for automobiles were supplemented by data-driven methods. Using surrogate models like KLIME from the XAI domain, additional production-relevant CERs were determined. Similarly, Lokrantz et al. [45] modeled expert knowledge for multi-stage manufacturing processes in the form of a Bayesian network. Experts defined layers or individual edges within the network, simplifying data-driven learning and optimization. Rippel et al. [52] followed a similar approach in the domain of micro-manufacturing. Additionally, Wehner et al. [57] presented an approach where expert knowledge was modeled via knowledge graphs and causal Bayesian networks, with experts interactively providing feedback on the graphs. Validation took place in the automotive production domain.
Feature selection and grouping approaches involve interactions with machine learning models, such as the selection of important features for RCA by process experts before the actual data-driven model training. Examples of these approaches can be found in Kozjek et al. [42] and Lee et al. [12]. In these studies, expert knowledge was used for feature selection and grouping, enhancing the relevance and accuracy of data-driven models.
Post-hoc analysis and validation approaches involve the analysis and validation of data-driven models by experts after the models have been generated. For instance, Chien et al. [32] employed Cramer’s V correlation and logistic regression for RCA, allowing experts to analyze and validate the generated groups a posteriori.

3.2. RQ2: Recurring Challenges and Limitations

RCA in manufacturing is a central yet complicated process due to the diverse and dynamic nature of modern manufacturing environments. This section addresses the predominant challenges encountered in the reviewed articles, categorized into expert knowledge dependence, data management issues, methodological difficulties, and other overarching challenges.
RCA in manufacturing relies heavily on expert knowledge, which presents several notable challenges. A high level of process knowledge is required to perform RCA within a reasonable time frame, as highlighted by Wehner et al. [57]. Additionally, traditional approaches such as the 5-Why analysis necessitate deep expert understanding but often fall short in accuracy [28,44]. There are also significant difficulties in transferring, storing, and verifying expert knowledge, which can hinder the RCA process [45]. Moreover, expert knowledge is frequently not formalized, complicating communication and validation [9]. The reliance on expertise and experience makes root cause analyses heavily dependent on specific individuals [56]. Consequently, automating RCA for large systems is impractical if reliant solely on expert knowledge, as emphasized by Cho and Kim [33]. In addition, the reliance on expert knowledge makes it difficult to use this information for automated RCA [50].
Effective data management is crucial for RCA, yet it also presents several challenges. As Kim et al. [40] points out, the large volume and complexity of sensor data makes manual investigation difficult. Particularly in the semiconductor industry, production involves hundreds of process steps, requiring significant time and personnel for manual defect detection [39]. In addition, the increasing complexity of manufacturing systems leads to incomplete, inaccurate, heterogeneous, and dynamically changing data, which hinders automated knowledge extraction [42]. Despite advances in data analysis, large datasets are often under-utilized [42]. Furthermore, imbalanced data distribution complicates learning processes, as more production data are available for OK than for NOK states [49]. Additionally, insufficient data quantities increase uncertainty when analyzing, as Tuninetti et al. [10] points out. Combined with the numerous factors in manufacturing, this makes failure prediction even more difficult [53]. In particular, knowledge-driven approaches that heavily rely on expert knowledge do not scale well with the number of factors to be considered [37].
Methodological difficulties significantly hinder the RCA process. There is a challenge in selecting key variables for analysis that provide insights into possible causes [12]. Current approaches for determining the relevance of features, such as ML models in combination with the determination of feature importance, often do not accurately represent the problem or its causes [29]. Furthermore, traditional approaches struggle to handle high-dimensional data, as noted by e Oliveira et al. [36]. Additionally, many approaches, both data-driven and knowledge-driven, rely more on correlation than on causality, which can limit their diagnostic effectiveness [46]. As Lokrantz et al. [45] also noted, distinguishing between symptoms and actual causes remains a persistent difficulty. In addition to these issues, current approaches and models such as cause–effect networks are usually static, which means that they do not easily adapt to changing conditions in production, also making the integration of further data difficult [52].
Additional challenges complicate the RCA process in manufacturing even more. Among others, Brochado et al. [29] identified the lack of robustness in current data-driven approaches as a factor impairing reliability in RCA. Furthermore, the implementation of data-analytical approaches for RCA often fails due to a lack of expertise or personnel in this field [29]. Further complexity arises from rare failures that occur only under specific feature combinations [48]. In addition, non-linearity and non-stationary behavior after failures pose a major diagnostic challenge [58].

3.3. RQ3: Promises of Hybrid- and AI-Driven Approaches

The impact of AI on RCA in manufacturing has recently attracted the attention of researchers. Studies, including those by Otsubo et al. [50], Pohlmeyer et al. [9], and Vo et al. [11], have demonstrated that AI-driven approaches not only address traditional issues but also pave the way for innovative solutions to industrial applications of RCA. The integration of automation, data analytics, and human expertise increases the accuracy and speed of identifying and resolving manufacturing defects.
One of the foremost advantages of AI-driven approaches is their ability to identify the root causes of defects automatically and rapidly. Otsubo et al. [50] mentioned that by automating the detection process of root causes, these approaches significantly reduce process downtime as they eliminate the need for extensive expert analysis. The automation of the root cause and defect identification process not only accelerates the mentioned processes but also increases the probability of detecting anomalies that might otherwise escape the attention of human experts.
Additionally, Otsubo et al. [50] notes that AI-driven approaches are capable of managing more complex failure scenarios than traditional approaches. Complexity in this context refers to productions involving numerous machines, many workers, various materials, and multiple production steps. To tackle the large amounts of data obtained from all these production entities, data-driven approaches provide a more stable and faster identification of defect candidates than is possible through manual expert analysis. Moreover, data-driven approaches are capable of uncovering defect cases that are typically difficult for human experts to detect, as they can detect new patterns faster.
A significant benefit of data-driven approaches is their independence from expert knowledge. By not relying solely on pre-existing expert knowledge, these approaches can uncover previously unknown relationships and insights, adding depth to the analytical process, as emphasized by Pohlmeyer et al. [9]. While AI enables extensive data processing and manifold pattern recognition, the integration of human expertise remains crucial. Some hybrid approaches validate AI-generated results with expert knowledge. Known dependencies can be used to confirm the accuracy of AI results, blending empirical knowledge with automated insights [9,57].
With advancements in sensor and data acquisition technologies, large amounts of process data can now be effectively captured and analyzed. These large amounts of data enable us to utilize the capabilities of data-driven RCA techniques, making them more popular and effective in modern manufacturing contexts [47]. The integration of Industry 4.0 technologies with RCA processes, including the use of advanced analytics and ML, facilitates faster, more accurate, and reliable solutions. These technologies prevent defects from recurring by enabling more precise and data-driven decision-making processes [11].
In summary, AI-driven approaches are redefining the landscape of RCA in manufacturing. The combination of data-driven approaches and human expertise, resulting in hybrid approaches, enhances the speed, accuracy, and depth of analyses, offering a robust solution to the limitations posed by traditional RCA approaches. These advancements streamline manufacturing and promote collaboration between human knowledge and artificial intelligence for better results.

3.4. Future Research Directions

The approaches presented in this ScR, along with the challenges and promises of hybrid and AI-driven methodologies, demonstrate the extensive ongoing research for RCA in manufacturing. The following section addresses several core future research directions, particularly focusing on data-driven and hybrid variants of RCA, which have become predominant in recent years, as illustrated in Figure 2. Additionally, the transition toward predictive and adaptive approaches, as well as the development of standardized guidelines for RCA in manufacturing, is discussed.

3.4.1. Explainable Artificial Intelligence

This ScR and the discussions in Section 3.1.2 indicate that XAI is a fundamental method in data-driven RCA in manufacturing. XAI methods, such as LIME or SHAP, attempt to extract key influencing factors for each model. The identified influencing factors are often the causes of certain decisions by the data-driven model, for instance, in classifications. Consequently, these influencing factors can also be relevant in the actual production environment, indicating specific causes for particular effects. Future methods could be developed or applied to approximate the respective causes for quality-predicting models using XAI methods. Model-agnostic and local methods, in particular, represent a promising research direction.
Beyond the discussed approaches for interpreting black-box models using XAI methods, white-box approaches should also be further investigated. Graph-based methods, such as Bayesian networks or causal variants, as well as rule-based methods like association analysis or decision trees, allow for interpretable mappings between causes and their effects. This can help domain experts identify new or unknown CERs using data-driven methods and eliminate the resulting production errors. The increased transparency provided by white-box models does not contradict their quality, as shown in Rudin [60], making them a promising direction for RCA in manufacturing.

3.4.2. Predictive and Adaptive Approaches

The methods treated within the scope of RCA, especially data-driven methods, are generally reactive, defined following the occurrence of errors. Reactive methods can be divided into descriptive and diagnostic analytics according to the maturity model described in Król and Zdonek [61]. Furthermore, predictive analytics, prescriptive analytics, and cognitive analytics exist [61]. The latter represents the group of predictive methods. Future research directions in data-driven RCA lie in predictive methods. By learning from historical data and continuously adapting approaches to changing production environments, transitioning from reactive to predictive approaches (e.g., RCA for predictive maintenance) can better prevent errors and failures during production, aiding the goal of zero-defect manufacturing.
RCA is also influenced by the advancing digitalization, especially in the context of Industry 4.0. The Industry 4.0 maturity index by Schuh et al. [62] includes six distinct levels: computerization, connectivity, visibility, transparency, predictive capacity, and adaptability. These levels represent additional development opportunities for RCA in manufacturing. Current RCA approaches focus mainly on transparency. This maturity level tries to understand why something happened, similar to the diagnostic maturity level from the previous paragraph. In the future, RCA approaches should increasingly be transformed into higher maturity levels of predictive capacity and adaptability, particularly for the autonomous decision-making capability at the adaptability level, which allows decisions to be made based on identified CERs without the need for manual human intervention. The basis for this is the reliable extraction of CERs, especially through the use of data-driven or hybrid approaches. The digitalization driven by Industry 4.0 results in a vast and diverse array of data, which significantly impacts RCA. These large-scale data provide a rich source of information that can enhance the accuracy and depth of RCA analyses. However, they also necessitate that RCA approaches to be adapted to handle and process these extensive data effectively.

3.4.3. Human-in-the-Loop and Hybrid Approaches

The approaches discussed in Section 3.1.3 regarding the hybridization of RCA approaches represent another future research field. Particularly, the human-in-the-loop approach, which is also emphasized in Industry 5.0 [63], requires further research. For instance, approaches systematized in von Rueden et al. [64] consider numerous possibilities for integrating domain knowledge into ML models and formalizing this knowledge. Moreover, active learning could allow domain experts to interact directly with the model during training, acting in the sense of human in the loop.

3.4.4. Standardized Process Model and Guidelines

The methods identified in this ScR show a wide variety of variations. Despite the specifics of different industries and applications, all methods aim at RCA in manufacturing. However, despite the similarities, there is a lack of appropriate guidelines and a standardized process model for data-driven RCA projects. In contrast, Martinez et al. [65] highlight that these process models significantly contribute to the success of data-driven projects. While process models like KDD [66] and CRISP-DM [67] provide guidelines for data-driven projects, they are not specifically tailored for data-driven RCA. More tailored process models like the ones developed by Schoch [68] or Kornas [69] target quality assurance and RCA in production more specifically. However, these models may be limited by their industry focus and specific problem contexts. Therefore, a generalized approach for data-driven RCA that considers domain knowledge and incorporates human in the loop is still lacking and presents an area for further research.

3.5. Limitations of This Review

Despite rigorously following the PRISMA-ScR guidelines and checklist, the presented review may exhibit bias due to the article selection and evaluation process detailed in Section 2.
A potential source of bias in our study is the restriction of our search system to IEEE Xplore, Scopus, and Web of Science. According to Gusenbauer and Haddaway [17], both Web of Science and Scopus are considered suitable primary sources for article selection. However, they exclude gray literature, which could provide additional insights into the field of RCA in manufacturing. While platforms like Google Scholar can be used to investigate such additional literature, their application in structured literature reviews is limited [17]. Nonetheless, incorporating additional sources, e.g., the ACM Digital Library, could offer a more comprehensive overview of RCA in manufacturing, thereby enhancing the sensitivity of our review.
Another factor that may narrow the scope of this review encompasses the specific inclusion and exclusion criteria. Articles published before 2014 were excluded to focus on recent research in the field from the last decade. However, older research articles could provide additional approaches and key insights into RCA in manufacturing. Additionally, focusing on author keywords and titles for article selection enhanced the specificity but may have reduced sensitivity by potentially overlooking relevant articles that did not explicitly mention RCA in manufacturing in their titles or author keywords.

4. Conclusions

As this work illustrates, the topic of RCA remains a significant concern in manufacturing. Driven by customer orientation and the associated high quality requirements as well as increasing cost pressure, failure-free production is becoming more and more indispensable. To achieve this, occurring failures must be analyzed in detail using RCA approaches.
This ScR of the recent literature on RCA in manufacturing demonstrates the current state of research. In the course of this ScR, a total of 245 articles were extracted from Scopus, IEEE Xplore, and Web of Science. After a fine-grained filtering process, 35 articles were analyzed in detail and compared using a concept matrix.
The analyzed articles indicate an increasing shift toward data-oriented variants of RCA. Advances in digitization within manufacturing, ML, and AI are enabling data-driven and automated RCA. In addition, efforts are being made to integrate expert knowledge into data-driven approaches. This integration allows data-driven approaches to benefit from the existing knowledge of experts. Nevertheless, traditional and expert-based RCA approaches are still in use and are often applied directly by process experts.
Future research directions highlighted in this review include the development and application of XAI methods to improve the interpretability of data-driven RCA models, the usage of white-box models, and the transition from reactive to predictive analytics. Furthermore, the impact of Industry 5.0 and the need for hybrid approaches that integrate domain knowledge is emphasized. Finally, the development of process models for data-driven RCA projects remains an open research area, enabling us to further investigate the application of data-driven RCA, such as a human-in-the-loop approach, in the manufacturing industry.
To conclude, while our review rigorously follows the PRISMA-ScR guidelines, potential biases may arise from the selection of article search systems and the specific inclusion and exclusion criteria used. Future reviews should consider including a wider range of sources and broader criteria to provide an even more complete understanding of RCA in manufacturing. Furthermore, transitioning to a systematic literature review approach could offer another valuable direction for researchers to pursue.

Author Contributions

Conceptualization, D.P. and M.M.; methodology, D.P. and M.M.; investigation, D.P. and M.M.; visualization, D.P. and M.M.; writing—original draft preparation, D.P. and M.M.; writing—review and editing, D.P., M.M., U.W., S.I. and T.M.; supervision, U.W., S.I. and T.M. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by the Federal Ministry for Economic Affairs and Climate Protection on the basis of a resolution of the Deutscher Bundestag with funding code 13IK020F. Funded by the European Union.

Acknowledgments

During the preparation of this manuscript/study, the authors used DeepL Write and GPT-4 chat bots for the purpose of rephrasing and finding synonyms. The authors have reviewed and edited the output and take full responsibility for the content of this publication.

Conflicts of Interest

Authors Dominik Pietsch and Uwe Wieland were employed by the company Volkswagen AG. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The results, opinions and conclusions expressed in this article are not necessarily those of Volkswagen Aktiengesellschaft.

References

  1. Escobar, C.A.; McGovern, M.E.; Morales-Menendez, R. Quality 4.0: A review of big data challenges in manufacturing. J. Intell. Manuf. 2021, 32, 2319–2334. [Google Scholar] [CrossRef]
  2. Fuertes, G.; Zamorano, J.; Alfaro, M.; Vargas, M.; Sabattin, J.; Duran, C.; Ternero, R.; Rivera, R. Opportunities of the Technological Trends Linked to Industry 4.0 for Achieve Sustainable Manufacturing Objectives. Sustainability 2022, 14, 11118. [Google Scholar] [CrossRef]
  3. Camarinha-Matos, L.M.; Rocha, A.D.; Graça, P. Collaborative approaches in sustainable and resilient manufacturing. J. Intell. Manuf. 2022, 35, 499–519. [Google Scholar] [CrossRef] [PubMed]
  4. Prell, B.; Reinhardt, H.; Mauersberger, A.; Rauscher, A.; Reiff-Stephan, J.; Ihlenfeldt, S. A survey on sustainability approaches in manufacturing. Procedia CIRP 2023, 116, 101–106. [Google Scholar] [CrossRef]
  5. Yuan, C.; Li, G.; Kamarthi, S.; Jin, X.; Moghaddam, M. Trends in intelligent manufacturing research: A keyword co-occurrence network based review. J. Intell. Manuf. 2022, 33, 425–439. [Google Scholar] [CrossRef]
  6. Usuga Cadavid, J.P.; Lamouri, S.; Grabot, B.; Pellerin, R.; Fortin, A. Machine learning applied in production planning and control: A state-of-the-art in the era of industry 4.0. J. Intell. Manuf. 2020, 31, 1531–1558. [Google Scholar] [CrossRef]
  7. Ikome, J.M. A dynamic approach to minimise bottlenecks and improve manufacturing competitiveness in a case study manufacturing industry. In Production Management and Process Control; AHFE International: New York, NY, USA, 2023; Volume 104, pp. 127–136. [Google Scholar]
  8. Okes, D. Root Cause Analysis: The Core of Problem Solving and Corrective Action; ASQ Quality Press: Milwaukee, WI, USA, 2009. [Google Scholar]
  9. Pohlmeyer, F.; Kins, R.; Cloppenburg, F.; Gries, T. Interpretable failure risk assessment for continuous production processes based on association rule mining. Adv. Ind. Manuf. Eng. 2022, 5, 100095. [Google Scholar] [CrossRef]
  10. Tuninetti, V.; Alzugaray, R.; González, J.; Valenzuela, M.; Jaramillo, A.; Diez, E. Root cause and vibration analysis to increase veneer manufacturing process efficiency: A case study on an industrial peeling lathe. Eur. J. Wood Wood Prod. 2021, 79, 951–966. [Google Scholar] [CrossRef]
  11. Vo, B.; Kongar, E.; Suarez-Barraza, M.F. Root-Cause Problem Solving in an Industry 4.0 Context. IEEE Eng. Manag. Rev. 2020, 48, 48–56. [Google Scholar] [CrossRef]
  12. Lee, M.Y.; Choi, Y.J.; Lee, G.T.; Choi, J.; Kim, C.O. Attention Mechanism-Based Root Cause Analysis for Semiconductor Yield Enhancement Considering the Order of Manufacturing Processes. IEEE Trans. Semicond. Manuf. 2022, 35, 282–290. [Google Scholar] [CrossRef]
  13. Baier, L.; Frommherz, J.; Nöth, E.; Donhauser, T.; Schuderer, P.; Franke, J. Identifying failure root causes by visualizing parameter interdependencies with spectrograms. J. Manuf. Syst. 2019, 53, 11–17. [Google Scholar] [CrossRef]
  14. Brüggemann, H.; Bremer, P. Grundlagen Qualitätsmanagement; Springer Fachmedien Wiesbaden: Wiesbaden, Germany, 2015. [Google Scholar]
  15. Folgado, F.; Calderón, D.; González, I.; Calderón, A. Review of Industry 4.0 from the Perspective of Automation and Supervision Systems: Definitions, Architectures and Recent Trends. Electronics 2024, 13, 782. [Google Scholar] [CrossRef]
  16. Ustundag, A.; Cevikcan, E. Industry 4.0: Managing The Digital Transformation, 1st ed.; Springer Series in Advanced Manufacturing; Springer International Publishing and Imprint: Cham, Switzerland, 2018. [Google Scholar]
  17. Gusenbauer, M.; Haddaway, N.R. Which academic search systems are suitable for systematic reviews or meta-analyses? Evaluating retrieval qualities of Google Scholar, PubMed, and 26 other resources. Res. Synth. Methods 2020, 11, 181–217. [Google Scholar] [CrossRef] [PubMed]
  18. Ito, A.; Hagström, M.; Bokrantz, J.; Skoogh, A.; Nawcki, M.; Gandhi, K.; Bergsjö, D.; Bärring, M. Improved root cause analysis supporting resilient production systems. J. Manuf. Syst. 2022, 64, 468–478. [Google Scholar] [CrossRef]
  19. e Oliveira, E.; Miguéis, V.L.; Borges, J.L. Automatic root cause analysis in manufacturing: An overview & conceptualization. J. Intell. Manuf. 2023, 34, 2061–2078. [Google Scholar]
  20. Koval, L.; Knollmeyer, S.; Mathias, S.G.; Asif, S.; Uzair Akmal, M.; Grossmann, D.; Bregulla, M. Unlocking the Potential of Information Modeling for Root Cause Analysis in a Production Environment: A Comprehensive State-of-the-Art Review Using the Kitchenham Methodology. IEEE Access 2024, 12, 80266–80282. [Google Scholar] [CrossRef]
  21. Papageorgiou, K.; Theodosiou, T.; Rapti, A.; Papageorgiou, E.I.; Dimitriou, N.; Tzovaras, D.; Margetis, G. A systematic review on machine learning methods for root cause analysis towards zero-defect manufacturing. Front. Manuf. Technol. 2022, 2, 972712. [Google Scholar] [CrossRef]
  22. Mokhtarzadeh, M.; Rodríguez-Echeverría, J.; Semanjski, I.; Gautama, S. Hybrid intelligence failure analysis for industry 4.0: A literature review and future prospective. J. Intell. Manuf. 2024. [Google Scholar] [CrossRef]
  23. Tricco, A.C.; Lillie, E.; Zarin, W.; O’Brien, K.K.; Colquhoun, H.; Levac, D.; Moher, D.; Peters, M.D.J.; Horsley, T.; Weeks, L.; et al. PRISMA Extension for Scoping Reviews (PRISMA-ScR): Checklist and Explanation. Ann. Intern. Med. 2018, 169, 467–473. [Google Scholar] [CrossRef]
  24. Kitchenham, B.A.; Charters, S. Guidelines for Performing Systematic Literature Reviews in Software Engineering; Keele University and Durham University Joint Report; Keele, UK, 2007. [Google Scholar]
  25. Webster, J.; Watson, R.T. Analyzing the Past to Prepare for the Future: Writing a Literature Review. MIS Q. 2002, 26, xiii–xxiii. [Google Scholar]
  26. vom Brocke, J.; Simons, A.; Riemer, K.; Niehaves, B.; Plattfaut, R.; Cleven, A. Standing on the Shoulders of Giants: Challenges and Recommendations of Literature Search in Information Systems Research. Commun. Assoc. Inf. Syst. 2015, 37, 205–224. [Google Scholar] [CrossRef]
  27. Pollock, D.; Peters, M.D.J.; Khalil, H.; McInerney, P.; Alexander, L.; Tricco, A.C.; Evans, C.; de Moraes, É.B.; Godfrey, C.M.; Pieper, D.; et al. Recommendations for the extraction, analysis, and presentation of results in scoping reviews. JBI Evid. Synth. 2023, 21, 520–532. [Google Scholar] [CrossRef] [PubMed]
  28. Braglia, M.; Frosolini, M.; Gallo, M. SMED enhanced with 5-Whys Analysis to improve set-upreduction programs: The SWAN approach. Int. J. Adv. Manuf. Technol. 2017, 90, 1845–1855. [Google Scholar] [CrossRef]
  29. Brochado, A.F.; Rocha, E.M.; Pimentel, C. Understanding and Predicting Process Performance Variations of a Balanced Manufacturing Line at Bosch. In Proceedings of the Advanced Research in Technologies, Information, Innovation and Sustainability, ARTIIS 2022, Pt I; Communications in Computer and Information Science; Guarda, T., Portela, F., Augusto, M.F., Eds.; Springer: Cham, Switzerland, 2022; Volume 1675, pp. 357–371. [Google Scholar]
  30. Bui, A.T. Root cause analysis of manufacturing variation from optical scanning data. Ann. Oper. Res. 2022, 339, 111–130. [Google Scholar] [CrossRef]
  31. Chien, C.F.; Chuang, S.C. A framework for root cause detection of sub-batch processing system for semiconductor manufacturing big data analytics. IEEE Trans. Semicond. Manuf. 2014, 27, 475–488. [Google Scholar] [CrossRef]
  32. Chien, C.F.; Liu, C.W.; Chuang, S.C. Analysing semiconductor manufacturing big data for root cause detection of excursion for yield enhancement. Int. J. Prod. Res. 2017, 55, 5095–5107. [Google Scholar] [CrossRef]
  33. Cho, Y.S.; Kim, S.B. Quality-Discriminative Localization of Multisensor Signals for Root Cause Analysis. IEEE Trans. Syst. Man Cybern. Syst. 2022, 52, 4374–4387. [Google Scholar] [CrossRef]
  34. Chouichi, A.; Blue, J.; Yugma, C.; Pasqualini, F. Heterogranular Multivariate Analytics for Detecting and Controlling the Root Causes of the Mismatching Machines in Semiconductor Manufacturing. In Proceedings of the 2018 29th Annual SEMI Advanced Semiconductor Manufacturing Conference (ASMC), Saratoga Springs, NY, USA, 30 April–3 May 2018; pp. 334–339. [Google Scholar]
  35. Cyganiuk, J.; Kuryło, P.; Idzikowski, A.; Fröhlich, S.; Niemira, P. Solving Selected Quality Management Problems with the Use of the D4 Discipline Algorithm of the G8D Method. Syst. Saf. Hum.-Tech. Facil.-Environ. 2019, 1, 996–1003. [Google Scholar]
  36. e Oliveira, E.; Migueis, V.L.; Borges, J.L. Understanding Overlap in Automatic Root Cause Analysis in Manufacturing Using Causal Inference. IEEE Access 2022, 10, 191–201. [Google Scholar] [CrossRef]
  37. e Oliveira, E.; Miguéis, V.L.; Borges, J.L. On the influence of overlap in automatic root cause analysis in manufacturing. Int. J. Prod. Res. 2022, 60, 6491–6507. [Google Scholar] [CrossRef]
  38. e Oliveira, E.; Miguéis, V.L.; Borges, J.L. Overlap in Automatic Root Cause Analysis in Manufacturing: An Information Theory-Based Approach. Appl. Sci. 2023, 13, 3416. [Google Scholar] [CrossRef]
  39. Hong, W.J.; Shen, C.Y.; Wu, P.Y. Multi-source wafer map retrieval based on contrastive learning for root cause analysis in semiconductor manufacturing. J. Intell. Manuf. 2023. [Google Scholar] [CrossRef]
  40. Kim, E.; An, J.; Cho, H.C.; Cho, S.; Lee, B. A sensor data mining process for identifying root causes associated with low yield in semiconductor manufacturing. Data Technol. Appl. 2023, 57, 397–417. [Google Scholar] [CrossRef]
  41. Kornas, T.; Daub, R.; Karamat, M.Z.; Thiede, S.; Herrmann, C. Data-and expert-driven analysis of cause-effect relationships in the production of lithium-ion batteries. In Proceedings of the 2019 IEEE 15th International Conference on Automation Science and Engineering (CASE), Vancouver, BC, Canada, 22–26 August 2019. [Google Scholar]
  42. Kozjek, D.; Vrabič, R.; Kralj, D.; Butala, P. Interpretative identification of the faulty conditions in a cyclic manufacturing process. J. Manuf. Syst. 2017, 43, 214–224. [Google Scholar] [CrossRef]
  43. Krishnan, S.; Dev, A.S.; Suresh, R.; Sumesh, A.; Rameshkumar, K. Bottleneck identification in a tyre manufacturing plant using simulation analysis and productivity improvement. Mater. Today Proc. 2018, 5, 24720–24730. [Google Scholar] [CrossRef]
  44. Lee, M.G.; Chechurin, L.; Lenyashin, V. Introduction to cause-effect chain analysis plus with an application in solving manufacturing problems. Int. J. Adv. Manuf. Technol. 2018, 99, 2159–2169. [Google Scholar] [CrossRef]
  45. Lokrantz, A.; Gustavsson, E.; Jirstrand, M. Root cause analysis of failures and quality deviations in manufacturing using machine learning. Procedia CIRP 2018, 72, 1057–1062. [Google Scholar] [CrossRef]
  46. Ma, L.; Dong, J.; Peng, K. A Practical Root Cause Diagnosis Framework for Quality-Related Faults in Manufacturing Processes with Irregular Sampling Measurements. IEEE Trans. Instrum. Meas. 2022, 71, 3511509. [Google Scholar] [CrossRef]
  47. Ma, L.; Wang, M.; Peng, K. Nonlinear Dynamic Granger Causality Analysis Framework for Root-Cause Diagnosis of Quality-Related Faults in Manufacturing Processes. IEEE Trans. Autom. Sci. Eng. 2024, 21, 3554–3563. [Google Scholar] [CrossRef]
  48. Mehling, C.W.; Pieper, S.; Ihlenfeldt, S. Concept of a causality-driven fault diagnosis system for cyber-physical production systems. In Proceedings of the 2023 IEEE 21st International Conference on Industrial Informatics (INDIN), Lemgo, Germany, 18–20 July 2023. [Google Scholar]
  49. Ong, P.L.; Choo, Y.H.; Kamilah Muda, A. A manufacturing failure root cause analysis in imbalance data set using PCA weighted association rule mining. J. Teknol. 2015, 77, 103–111. [Google Scholar] [CrossRef]
  50. Otsubo, Y.; Otani, N.; Chikasue, M.; Nishino, M.; Sugiyama, M. Root cause estimation of faults in production processes: A novel approach inspired by approximate Bayesian computation. Int. J. Prod. Res. 2023, 61, 1556–1574. [Google Scholar] [CrossRef]
  51. Rezaei, A.; Ehsanifar, M.; Wood, D.A. Reducing welding repair requirements in refinery pressure vessel manufacturing: A case study applying six sigma principles. Int. J. Interact. Des. Manuf. 2019, 13, 1089–1102. [Google Scholar] [CrossRef]
  52. Rippel, D.; Luetjen, M.; Freitag, M. Enhancing Expert Knowledge Based Cause-Effect Networks Using Continuous Production Data. Procedia Manuf. 2018, 24, 128–134. [Google Scholar] [CrossRef]
  53. Sariyer, G.; Mangla, S.K.; Kazancoglu, Y.; Ocal Tasar, C.; Luthra, S. Data analytics for quality management in Industry 4.0 from a MSME perspective. Ann. Oper. Res. 2021. [Google Scholar] [CrossRef]
  54. Soler, J.P.; Sastre, N.G.; Santacruz, A.L.; Nunez, V.S.; Yuksel, M.C.; Settier, S.G.; Perez-Cortes, J.C. Advancing Towards Zero-Defect Manufacturing in the Plastic Injection Industry with Global and Local Explainability Approaches. In Proceedings of the IECON 2023- 49th Annual Conference of the IEEE Industrial Electronics Society, Singapore, 16–19 October 2023. [Google Scholar]
  55. Sooraj, S.; Vijayakumar, N.; Gaitonde, V.N.; Kotturshettar, B.B.; Zhang, Z. Productivity improvement of a manufacturing industry by process improvement through bottleneck and root cause analysis: Cause-effect relationship between vocal fold physiology and voice production in a three-dimensional phonation model. AIP Conf. Proc. 2019, 2200, 020053. [Google Scholar]
  56. Tan, D.; Xu, X.; Yu, K.; Zhang, S.; Chen, T. Multi-Feature Shuffle Algorithm for Root Cause Detection in Semiconductor Manufacturing. In Proceedings of the 2021 IEEE Intl Conf on Dependable, Autonomic and Secure Computing, Intl Conf on Pervasive Intelligence and Computing, Intl Conf on Cloud and Big Data Computing, Intl Conf on Cyber Science and Technology Congress (DASC/PiCom/CBDCom/CyberSciTech), Virtual, 25–28 October 2021; pp. 130–136. [Google Scholar]
  57. Wehner, C.; Kertel, M.; Wewerka, J. Interactive and Intelligent Root Cause Analysis in Manufacturing with Causal Bayesian Networks and Knowledge Graphs. In Proceedings of the 2023 IEEE 97th Vehicular Technology Conference, VTC2023-Spring, Florence, Italy, 20–23 June 2023. [Google Scholar]
  58. Zhang, H.; Peng, K.; Ma, L. A systematic nonstationary causality analysis framework for root cause diagnosis of faults in manufacturing processes. Control Eng. Pract. 2023, 131, 105404. [Google Scholar] [CrossRef]
  59. Syska, A. Produktionsmanagement: Das A - Z Wichtiger Methoden und Konzepte für die Produktion von Heute; Gabler: Wiesbaden, Germany, 2006. [Google Scholar]
  60. Rudin, C. Stop Explaining Black Box Machine Learning Models for High Stakes Decisions and Use Interpretable Models Instead. Nat. Mach. Intell. 2019, 1, 206–215. [Google Scholar] [CrossRef]
  61. Król, K.; Zdonek, D. Analytics Maturity Models: An Overview. Information 2020, 11, 142. [Google Scholar] [CrossRef]
  62. Schuh, G.; Anderl, R.; Dumitrescu, R.; Krüger, A.; ten Hompel, M. (Eds.) Industrie 4.0 Maturity Index: Die Digitale Transformation von Unternehmen Gestalten, update 2020 ed.; acatech STUDIE, acatech-Deutsche Akademie der Technikwissenschaften: München, Germany; Berlin, Germany; Brüssel, Belgium, 2020. [Google Scholar]
  63. Golovianko, M.; Terziyan, V.; Branytskyi, V.; Malyk, D. Industry 4.0 vs. Industry 5.0: Co-existence, Transition, or a Hybrid. Procedia Comput. Sci. 2023, 217, 102–113. [Google Scholar] [CrossRef]
  64. von Rueden, L.; Mayer, S.; Beckh, K.; Georgiev, B.; Giesselbach, S.; Heese, R.; Kirsch, B.; Walczak, M.; Pfrommer, J.; Pick, A.; et al. Informed Machine Learning—A Taxonomy and Survey of Integrating Prior Knowledge into Learning Systems. IEEE Trans. Knowl. Data Eng. 2021, 35, 614–633. [Google Scholar] [CrossRef]
  65. Martinez, I.; Viles, E.; Olaizola, I.G. A survey study of success factors in data science projects. In Proceedings of the 2021 IEEE International Conference on Big Data (Big Data), Virtual, 15–18 December 2021; pp. 2313–2318. [Google Scholar]
  66. Fayyad, U.; Piatetsky-Shapiro, G.; Smyth, P. Knowledge discovery and data mining: Towards a unifying framework. In Proceedings of the Second International Conference on Knowledge Discovery and Data Mining, Portland, OR, USA, 2–4 August 1996; pp. 82–88. [Google Scholar]
  67. Wirth, R.; Hipp, J. CRISP-DM: Towards a standard process model for data mining. In Proceedings of the 4th International Conference on the Practical Applications of Knowledge Discovery and Data Mining, Manchester, UK, 11–13 April 2000. [Google Scholar]
  68. Schoch, A. Datenzentrierte Künstliche Intelligenz für ein prädiktives Qualitätsmanagement in der Automobilindustrie (DZKI-PQ). Ph.D. Thesis, Universität Kassel, Kassel, Germany, 2023. [Google Scholar]
  69. Kornas, T. Qualitätssicherungssystem für die Anlaufphase und den Serienbetrieb einer Batteriezellenproduktion. Ph.D. Thesis, Technische Universität Braunschweig, Brunswick, Germany, 2021. [Google Scholar]
Jmmp 08 00277 g001
Figure 1. Article selection process.
Figure 1. Article selection process.
Jmmp 08 00277 g001
Jmmp 08 00277 g002
Figure 2. Number of articles by year with cumulative percentage.
Figure 2. Number of articles by year with cumulative percentage.
Jmmp 08 00277 g002
Jmmp 08 00277 g003
Figure 3. Frequency of knowledge-driven approaches.
Figure 3. Frequency of knowledge-driven approaches.
Jmmp 08 00277 g003
Jmmp 08 00277 g004
Figure 4. Frequency of data-driven approaches.
Figure 4. Frequency of data-driven approaches.
Jmmp 08 00277 g004
Table 1. Review questions.
Table 1. Review questions.
Review Question
RQ1What are current approaches for root cause analysis in industrial manufacturing?
RQ2What are the recurring challenges and limitations associated with these current approaches?
RQ3What are the promises for hybrid- and AI-driven approaches to enhance root cause analysis in manufacturing?
Table 2. PCC method for article selection.
Table 2. PCC method for article selection.
CategoryKeywordSynonymsExample
PopulationManufacturingIndustrial, Factory, ProductionAutomotive, Semiconductor, Chemical, Electronic
ConceptRoot Cause AnalysisFault Diagnosis, Fault IsolationRCA, Root Cause, Cause Effect
ContextApplicationsImplementationImplementation, Challenges, Limitations, Promises
Table 3. Construction of search strings.
Table 3. Construction of search strings.
Field of InterestSearch Strings
Manufacturing(“Manufacturing” OR “Production”)
AND
Root Cause Analysis(“Root Cause*” OR “Cause Effect*”)
Table 4. Article source-specific search strings.
Table 4. Article source-specific search strings.
Article SourceSearch String
IEEE Xplore((“Document Title”:manufacturing OR “Document Title”:production) AND (“Document Title”:root cause* OR “Document Title”:cause effect*)) OR ((“Author Keywords”:manufacturing OR “Author Keywords”:production) AND (“Author Keywords”:root cause* OR “Author Keywords”:cause effect*))
Scopus(AUTHKEY((“manufacturing”) OR (“production”) AND (“root cause*”) OR (“cause effect*”)) OR TITLE((“manufacturing”) OR (“production”) AND (“root cause*”) OR (“cause effect*”))
Web of Science(((TI = (manufacturing)) OR TI = (production)) AND (TI = (root cause*) OR TI = (cause effect*))) OR (((AK = (manufacturing)) OR AK = (production)) AND (AK = (root cause*) OR AK = (cause effect*)))
Table 5. Exclusion criteria.
Table 5. Exclusion criteria.
Criteria
EC1Article is a duplicate
EC2Study published before 2014
EC3Article is a review
EC4Full text is not available
Table 6. Inclusion criteria.
Table 6. Inclusion criteria.
Criteria
IC1Title falling in the concept of RCA in manufacturing
IC2Abstract falling in the concept of RCA in manufacturing
IC3Full text falling in the concept of RCA in manufacturing
Table 7. Data extraction categories with example article from Pohlmeyer et al. [9].
Table 7. Data extraction categories with example article from Pohlmeyer et al. [9].
CategoryExampleReview Question
Industry sectorTextile industryRQ1
Subject of problemMachine-
Problem descriptionRCA and failure risk assessment of unplanned machine downtimes-
Research methodConcept and case study-
Used dataMachine data, quality data, environmental dataRQ1
Knowledge-driven approachesNoneRQ1
Data-driven approachesAssociation rule miningRQ1, RQ3
Findings/RemarksExpert knowledge is hard to formalizeRQ2
Table 8. Concept matrix.
Table 8. Concept matrix.
IndustryKnowledge-DrivenData-DrivenHybrid
ArticlesGeneralSemiconductorAutomotiveChemicalElectronicsOthersPareto Analysis5-Why AnalysisIshikawa DiagramCause–Effect (Non-Ishikawa)Graph- and Causality-Based ModelsProbabilistic and Statistical ModelsTree-Based ModelsSupport Vector MachineNeural NetworksExplainable AIInitialization and InteraktionFeature Selection and GroupingPost-Hoc Analysis
Braglia et al. [28]-----•-•-----------
Brochado et al. [29]•-----------•--•---
Bui [30]--•-----------•----
Chien and Chuang [31]-•---------••------
Chien et al. [32]-•---------•------•
Cho and Kim [33]-----•--------••---
Chouichi et al. [34]-•---------•-------
Cyganiuk et al. [35]•-------•----------
e Oliveira et al. [36]-•--------•--------
e Oliveira et al. [37]-•----------•--•---
e Oliveira et al. [38]-•-------------•---
Hong et al. [39]-•------------•----
Kim et al. [40]-•----------•--•---
Kornas et al. [41]--•------•-••••••--
Kozjek et al. [42]---•--------•--•-•-
Krishnan et al. [43]-----••••----------
Lee et al. [44]----•----•---------
Lee et al. [12]-•------------•--•-
Lokrantz et al. [45]•---------•-----•--
Ma et al. [46]-----•----•---•----
Ma et al. [47]-----•----•---•----
Mehling et al. [48]•---------•--------
Ong et al. [49]-•-------------•---
Otsubo et al. [50]•----------•-------
Pohlmeyer et al. [9]-----•---------•---
Rezaei et al. [51]-----•--•----------
Rippel et al. [52]-----•---••-----•--
Sariyer et al. [53]-----•---------•---
Soler et al. [54]-----•------•--•---
Sooraj et al. [55]-----•-••----------
Tan et al. [56]-•----------•-••---
Tuninetti et al. [10]-----•--•----------
Vo et al. [11]-----•-••----------
Wehner et al. [57]--•-------•-----•--
Zhang et al. [58]-----•----•---•----
Table 9. Industry-specific approaches visualized through a heatmap.
Table 9. Industry-specific approaches visualized through a heatmap.
Knowledge-DrivenData-DrivenHybrid
Pareto Analysis5-Why AnalysisIshikawa DiagramCause–Effect (Non-Ishikawa)Graph-Based ModelingGraph and CausalityProb. and Stat. ModelsTree-Based ModelsSupport Vector MachineNeural NetworksExplainable AIInitialization and InteraktionFeature Selection and GroupingPost-Hoc Analysis
General00102211001100
Semiconductor00000134035011
Automotive00011111121200
Chemical00000001001010
Electronics00010000000000
Others14510401044100
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Pietsch, D.; Matthes, M.; Wieland, U.; Ihlenfeldt, S.; Munkelt, T. Root Cause Analysis in Industrial Manufacturing: A Scoping Review of Current Research, Challenges and the Promises of AI-Driven Approaches. J. Manuf. Mater. Process. 2024, 8, 277. https://doi.org/10.3390/jmmp8060277

AMA Style

Pietsch D, Matthes M, Wieland U, Ihlenfeldt S, Munkelt T. Root Cause Analysis in Industrial Manufacturing: A Scoping Review of Current Research, Challenges and the Promises of AI-Driven Approaches. Journal of Manufacturing and Materials Processing. 2024; 8(6):277. https://doi.org/10.3390/jmmp8060277

Chicago/Turabian Style

Pietsch, Dominik, Marvin Matthes, Uwe Wieland, Steffen Ihlenfeldt, and Torsten Munkelt. 2024. "Root Cause Analysis in Industrial Manufacturing: A Scoping Review of Current Research, Challenges and the Promises of AI-Driven Approaches" Journal of Manufacturing and Materials Processing 8, no. 6: 277. https://doi.org/10.3390/jmmp8060277

APA Style

Pietsch, D., Matthes, M., Wieland, U., Ihlenfeldt, S., & Munkelt, T. (2024). Root Cause Analysis in Industrial Manufacturing: A Scoping Review of Current Research, Challenges and the Promises of AI-Driven Approaches. Journal of Manufacturing and Materials Processing, 8(6), 277. https://doi.org/10.3390/jmmp8060277

Article Metrics

Back to TopTop