Abstract: To address the issue of low fault detection accuracy in intermittent processes due to complex dynamic characteristics and multi-stage features, a fault detection method for multi-stage intermittent processes based on Bottom-Up Derivative Dynamic Time Warping (BU-DDTW) and Semi-Positive Definite Sparse Principal Component Analysis (SPCA-PSD) is proposed. Initially, an encoder-decoder model is utilized to capture the dynamic characteristics of the time series after batching. Subsequently, by integrating the BU-DDTW merging strategy, the dynamic structural similarity between different subsequences is measured, achieving precise phase division. Then, the SPCA-PSD method introduced sparsity and semi-positive definite constraints to accurately identify key variables representing the characteristics of each stage, constructing a multi-stage fault detection model. Experiments are conducted on data from the penicillin supplementation batch fermentation process, and the results demonstrate that the proposed method achieves an average fault detection rate of 95.2% for six different types of faults, significantly outperforming other methods. The results not only prove the effectiveness of the proposed method in phase division of intermittent processes, but also enhance the accuracy and interpretability of the multi-stage fault detection model.

Key words: batch process, fault monitoring, phase division, nonlinearity, dynamic nature