Process Diagnosis and Monitoring Based on Adaptive Trend Analysis

The paper adopts an adaptive nonlinear algorithm for process monitoring and fault diagnosis, and the purpose of fault diagnosis is obtained through acquiring the sensor data, various basic elements in the identification process, events and trends analysis. The fault class knowledge base is formed by using the automated knowledge base development framework (AKDF), and the algorithm adoptes neural network identification elements. The paper has discussed the minimum and maximum length of adaptive window distinguished by fault signal, the probability of fault is got by referring to the probabilities of all kinds of sensor fault classification in knowledge base. The paper has made comparisons on the diagnosis cases which have 20 sensor in the conditions of simple deviation to complex fault and the percent of noise from 0 to 5%, the result shows that the performance of adaptive trend identification is better than fixed window method, ASTRA can properly diagnose all events included in the historical database, and it can tell the unknown fault class.