Intrinsic and post-hoc XAI approaches for fingerprint identification and response prediction in smart manufacturing processes

• DOI: 10.1007/s10845-023-02266-2
• Publication Date: 2024-02-28T07:02:32Z
• Authors: Abhilash Puthanve...; Xichun Luo; Qi Liu; Charles Walker; Rajeshkumar Madarkar; Yukui Cai; Zhanqiang Liu; Wenlong Chang; Yi Qin
• Keywords: Identification; Categorical variable

Abstract In quest of improving the productivity and efficiency manufacturing processes, Artificial Intelligence (AI) is being used extensively for response prediction, model dimensionality reduction, process optimization, monitoring. Though having superior accuracy, AI predictions are unintelligible to end users stakeholders due their opaqueness. Thus, building interpretable inclusive machine learning (ML) models a vital part smart paradigm establish traceability repeatability. The study addresses this fundamental limitation AI-driven processes by introducing novel Explainable (XAI) approach develop product fingerprints. Here explainability implemented in two stages: developing representations fingerprints, posthoc explanations. Also, first time, concept fingerprints extended an probabilistic bottleneck events during processes. demonstrated using datasets: nanosecond pulsed laser ablation produce superhydrophobic surfaces wire EDM real-time monitoring dataset machining Inconel 718. fingerprint identification performed global Lipschitz functions optimization tool (MaxLIPO) stacked ensemble prediction. proposed robust change can responsively handle both continuous categorical responses alike. Implementation XAI not only provided useful insights into physics but also revealed decision-making logic local predictions.