Research on Automated Fault Diagnosis of Chemical Process Time-series Data under Fault Conditions

Abstract

The safe and stable operation of chemical processes is of significant importance to industrial production. However, due to the complexity, multi-variable coupling, and nonlinear dynamic characteristics of chemical processes, traditional fault diagnosis methods. This paper proposes an automated fault diagnosis method based on deep learning for fault time-series data in chemical processes, integrating an improved time-series feature extraction algorithm with dynamic pattern recognition technology. Specifically, this paper introduces an attention mechanism-based improved temporal convolutional network (ATCN), which can effectively extract key fault features from time-series data and enhance the ability to capture long-term dependencies. Additionally, by combining transfer learning strategies, the model can quickly adapt to different operating conditions, significantly improving the model’s generalization and robustness. Experiments were conducted on a chemical process dataset, showing that compared to traditional methods, this method significantly improves fault detection and diagnosis accuracy, especially demonstrating higher reliability under complex operating conditions. This research provides a new approach for real-time fault diagnosis in chemical processes, not only effectively reducing the production risks caused by faults but also laying an important foundation for the construction of intelligent industrial systems.