MOXAI - Manufacturing Optimization through Model-Agnostic Explainable AI and Data-Driven Process Tuning

Modern manufacturing equipment offers numerous configurable parameters for optimization, yet operators often underutilize them. Recent advancements in machine learning (ML) have introduced data-driven models in industrial settings, integrating key equipment characteristics. This paper evaluates the performance of ML models in classification tasks, revealing nuanced observations. Understanding model decisionmaking processes in failure detection is crucial, and a guided approach aids in comprehending model failures, although human verification is essential. We introduce MOXAI, a datadriven approach leveraging existing pre-trained ML models to optimize manufacturing machine parameters. MOXAI underscores the significance of explainable artificial intelligence (XAI) in enhancing data-driven process tuning for production optimization and predictive maintenance. MOXAI assists operators in adjusting process settings to mitigate machine failures and production quality degradation, relying on techniques like DiCE for automatic counterfactual generation and LIME to enhance the interpretability of the ML model's decision-making process. Leveraging these two techniques, our research highlights the significance of explaining the model and proposing the recommended parameter setting for improving the process.