An article recently published in the journal Supply Chain Analytics comprehensively explored the use of explainable artificial intelligence (XAI) for predictive maintenance and spare parts optimization of ground support equipment (GSE) in a maintenance, repair, and overhaul (MRO) company. The researchers aimed to improve maintenance strategies by providing clear explanations of machine learning model predictions to enhance decision-making.
Study: An explainable artificial intelligence model for predictive maintenance and spare parts optimization. Image Credit: FabrikaSimf/Shutterstock.com
Background
The rapid advancement of technology has significantly changed maintenance practices. As industries rely more on complex machinery, any malfunction can cause major disruptions and financial losses.
To reduce these risks, companies are adopting innovative maintenance strategies to ensure systems, machines, equipment, and vehicles operate efficiently. Poor maintenance can lead to financial losses, reduced product quality, and increased operational costs, while even minor supply chain disruptions can impact overall productivity.
Developing an effective maintenance process starts with selecting the right strategy, such as reactive, preventive, opportunistic, or aggressive maintenance. The challenge with proactive strategies is predicting faults before they occur.
Technologies like monitoring systems, statistical analysis, data analysis, machine learning, and artificial intelligence help predict potential malfunctions, but their complexity often creates a “black box” effect, making it difficult for users to understand the decision-making process. XAI addresses this issue by providing insights into how these models work, enhancing trust and understanding.
About the Review
In this paper, the authors proposed an XAI methodology for predictive maintenance using the local interpretable model-agnostic explanations (LIME) method. They followed a structured approach, beginning with data collection and preparation, followed by model selection and performance evaluation.
The study used the predictive maintenance dataset (AI4I), a synthetically generated dataset, to train and evaluate the machine learning models. This dataset included various features such as temperature, speed, and torque measurements. It contained 14 features and 10,000 rows of data, which were prepared for analysis through data preprocessing steps like cleaning, normalization, and balancing.
The researchers used PyCaret, an open-source, low-code machine learning library, to automate the workflow. They tested several algorithms, including an extra-tree classifier, a random forest classifier, and an AdaBoost classifier, to identify the best-performing model.
The random forest algorithm was chosen for its high accuracy and F1 score. Furthermore, the LIME method was applied to provide detailed explanations of the model’s predictions, making the results more interpretable for decision-makers.
Research Findings
The outcomes showed that integrating XAI techniques into predictive maintenance significantly improved the interpretability of machine learning models. The random forest classifier emerged as the top performer, achieving a high 98% F1 score on the test data, indicating its reliability in predicting equipment failures.
However, the AdaBoost Classifier, despite performing best on the training data, showed signs of potential overfitting, with its F1 score dropping to around 82% on the test data. Additionally, the LIME method provided clear explanations for the model’s predictions, identifying key features such as tool wear, torque, and rotational speed as influential.
The authors highlighted the importance of balancing precision and recall in predictive maintenance. While high recall ensured that most failures were detected, high precision reduced false positives, minimizing unnecessary maintenance.
The study’s findings suggest that a balanced approach, measured using the F1 score, is most effective for predictive maintenance. LIME’s detailed explanations could help decision-makers understand how different features influenced the model’s predictions, enhancing transparency and supporting better maintenance decisions and spare parts management.
Applications
This research has significant implications for predictive maintenance across various industries. By providing clear explanations of model predictions, XAI enhances the decision-making process, allowing maintenance teams to understand the reasons behind model recommendations for equipment repairs and replacements.
This transparency helps diagnose issues more accurately and prioritize maintenance tasks based on their severity and impact. As a result, maintenance operations become more efficient, with improved scheduling and resource allocation.
Additionally, XAI can help optimize spare parts management by predicting which components are likely to fail and when leading to more strategic inventory management. This approach reduces costs associated with emergency repairs and parts procurement, minimizes downtime, and boosts overall productivity and lifespan of equipment.
Conclusion
In summary, the XAI algorithm proved effective for transforming predictive maintenance strategies by making machine learning models more interpretable. This approach bridged the gap between complex algorithms and human decision-makers.
The researchers emphasized the importance of transparency in AI applications, especially in critical areas like maintenance and manufacturing.
Future work should focus on integrating real-time monitoring systems with XAI techniques to further enhance predictive maintenance and spare parts optimization. Overall, this study contributes to the ongoing evolution of maintenance strategies in the era of Industry 4.0.
Journal Reference
Dereci, U., & Tuzkaya, G. (2024) An explainable artificial intelligence model for predictive maintenance and spare parts optimization. Supply Chain Analytics, 8, 100078. DOI: 10.1016/j.sca.2024.100078, https://www.sciencedirect.com/science/article/pii/S2949863524000219
Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.