Pattern-based feature extraction for fault detection in quality relevant process control

Statistical quality control (SQC) applies multivariate statistics to monitor production processes over time and detect changes in their performance in terms of meeting specification limits on key product quality metrics. These limits are imposed by customers and typically assumed to be a single target value, however, for some products, it is more reasonable to target a range of values. Under this assumption we propose a multi-stage approach for mapping operating conditions to product quality classes. We use principal component analysis (PCA) and a pattern mining algorithm to reduce dimensionality and identify predictive patterns in time series of operating conditions in order to improve the performance of the classifier. We apply this approach to an industrial machining process and obtain significant improvements over models trained using features based on the last value of each process variable.