55 / 2021-07-04 14:16:05

Fault Diagnosis Based on Data-driven Dynamic Model

The existing quality-related fault detection with dynamic characteristics has the characteristics of low detection rate, complex algorithm, and low applicability. There are few research models for dynamic quality failures. This paper combines the idea of machine reinforcement learning and proposes a dynamic modified partial least squares (DMPLS) model of autoregressive moving average(ARMA). This model introduces the learning factor into the standard MPLS model to flexibly adjust the influence of historical process data on the fault detection of the DMPLS model. While retaining the advantages of the MPLS model, the correlation between input and output data is strengthened. The detection rate of the dynamic system to the fault is improved, and the model obtains more accurate dynamic detection and prediction performance. Establish a more robust dynamic relationship than standard MPLS. On the basis of the DMPLS model, corresponding fault detection and prediction algorithms are proposed. Finally, numerical examples and Tennessee-Eastman(TE) process prove that the DMPLS model has a higher failure detection rate and lower prediction error than the standard MPLS model.