Process quality anomaly detection method based on deep learning

Abstract

This invention relates to the field of process quality anomaly detection technology, specifically a deep learning-based process quality anomaly detection method. It acquires multi-source sensor and operational data to generate a time-series dataset, identifies peak values, extracts amplitude directions to generate a phase feature sequence, analyzes feature time intervals and performs translation interpolation to generate an alignment correction feature set, compares deviations with a qualified benchmark to identify and accumulates propagation paths to generate a deviation sequence, extracts window features and inputs them into a deep network for probability discrimination, and finally generates process quality anomaly detection results. This invention constructs a unified time benchmark to form a continuous process sequence, extracts cross-channel peak phase consistency features, achieves multi-source response synchronization alignment through time translation interpolation, constructs a deviation propagation accumulation representation based on benchmark comparison, integrates multi-dimensional features of the time window to output anomaly results, structurally characterizes the multi-source response relationship, enhances correlation expression, and improves the precision and tracking capability of anomaly identification and localization.

Images