Enhanced Anomaly Detection in Compressor Components Using Deep Learning and an Attribute Updating Model

With increased focus on the safety of chemical plants, detecting anomalies in core equipment like compressors has become crucial for stable operations, preventive maintenance, and optimizing production efficiency. However, accurately setting anomaly thresholds for multidimensional data, pinpointing abnormal components, and fully considering the interdependence among various components remain challenging. Hence, this paper proposes an anomaly detection method integrating deep learning and an attribute updating model. It comprises an attribute update model, a dimensionality reduction and structural reorganization model, and an SL-RegNet detection model enhanced by SE and LKA. A set of detection methods for complex anomalies (collective anomalies) is developed in the end. Experimental results demonstrate an accuracy of 95.68%, effectively identifying abnormal states of compressor components. Simultaneously, we conduct validity experiments on the attribute updating model, ablation experiments, and comparison experiments to demonstrate the superiority of our proposed method.