Incipient Fault Diagnosis of Industrial Processes Based on Residual Evaluation Multi-Feature Joint Analysis

In actual industrial engineering processes, external factors such as noise, vibration, and other random interference can affect the normal operation of the system. Therefore, incipient fault detection and real-time alarms are becoming critical for effectively managing these anomalies. To tackle this challenge, we propose a novel method for incipient fault diagnosis that incorporates residual evaluation and multifeature joint analysis. Initially, the approach employs variational mode decomposition to extract residual signals from the data. The next step involves assessing residual information by traversing sliding windows. Following this, an analysis of time-domain properties and data distributions is conducted to extract a range of diverse characteristic indexes. Among these indexes most representative components are then refined into fault features using principal component analysis. For enhanced fault detection, we utilize multiple indexes. The fusion of information for detection is achieved through integrated learning coupled with Bayesian inference, ultimately yielding the final detection outcomes. Numerical simulations have verified the feasibility and validity of the method, and it has been applied to the continuous stirred kettle reactor process and the Tennessee-Eastman process for experimental validation.