Out of Distribution Detection for Efficient Continual Learning in Quality Prediction for Arc Welding

TL;DR

The paper tackles the challenge of distribution shifts in welding quality prediction by extending the VQ-VAE-Tr framework to use autoregressive loss as an out-of-distribution indicator. It introduces a unified OOD score and ROC-based thresholding to robustly separate ID and OOD data, and couples this with selective continual learning to trigger model adaptation only when necessary, reducing labeling costs. Empirical results on real GMAW data show that autoregressive OOD signals outperform reconstruction- and quantization-based cues and that OOD-triggered continual learning achieves performance comparable to full continual learning while dramatically cutting labeling requirements. This work offers a practical, explainable approach for robust, adaptive quality monitoring in dynamic manufacturing environments, with potential applicability to other time-series industrial processes where temporal patterns encode critical dynamics.

Abstract

Modern manufacturing relies heavily on fusion welding processes, including gas metal arc welding (GMAW). Despite significant advances in machine learning-based quality prediction, current models exhibit critical limitations when confronted with the inherent distribution shifts that occur in dynamic manufacturing environments. In this work, we extend the VQ-VAE Transformer architecture - previously demonstrating state-of-the-art performance in weld quality prediction - by leveraging its autoregressive loss as a reliable out-of-distribution (OOD) detection mechanism. Our approach exhibits superior performance compared to conventional reconstruction methods, embedding error-based techniques, and other established baselines. By integrating OOD detection with continual learning strategies, we optimize model adaptation, triggering updates only when necessary and thereby minimizing costly labeling requirements. We introduce a novel quantitative metric that simultaneously evaluates OOD detection capability while interpreting in-distribution performance. Experimental validation in real-world welding scenarios demonstrates that our framework effectively maintains robust quality prediction capabilities across significant distribution shifts, addressing critical challenges in dynamic manufacturing environments where process parameters frequently change. This research makes a substantial contribution to applied artificial intelligence by providing an explainable and at the same time adaptive solution for quality assurance in dynamic manufacturing processes - a crucial step towards robust, practical AI systems in the industrial environment.

Figures (4)

• Figure 1: Sample images of the weld with different weld types.
• Figure 2: Representative current and voltage patterns from single welding cycles for overlap (left) and T-joint (right) configurations.
• Figure 3: Selective adaptation framework using OOD detection (green pathway) as a trigger for continual learning, optimizing model updates only when process parameters change significantly, with the blue pathway showing the framework without OOD.
• Figure 4: Continual learning performance over 53 sequential experiences comparing three deployment strategies: static baseline (No CL), continuous adaptation (Replay), and OOD-triggered selective adaptation (OOD Replay). Green regions indicate OOD detection triggers and the I standing for the initial performance.