Research On Reconstruction-PCA Based Transition Process Monitoring Method

In modern industrial production, there are always many operational modes. However, the transition process exists between two adjacent operational modes when the operational modes are switching. Most conventional monitoring methods focus on single mode and ignore the transition process, leading false alarms as the transition process is regarded as the fault of monitoring process. On another hand, the transition process has dynamic character. But most conventional monitoring methods deal with static problems without considering the autocor-relation of monitoring variables, so conventional monitoring methods can not give an effec-tive monitoring for the transition process.Fault direction is assumed for many times by conventional reconstruction-based prin-ciple component analysis algorithm. And the assumed fault directions are submitted into the reconstruction equation. Then PCA method is used to detect the fault of the monitoring process. As fault direction need to be assumed for many times, it needs much work. In addi-tion, RPCA method is also a static method and can be applied directly to the fault detection of transition process.In order to solve these monitoring problems, researches have been done in this paper. The main contributions are listed as follows:(1) Based on existing technology and methods, an improved monitoring method of tran-sition process (IRPCA) based on improved reconstruction-based principle component analysis is proposed to solve the fault detection of transition process. As the common basis matrix contains abundant information of modes, the common basis matrix of starting mode and end-ing mode is selected as the fault detection of the reconstruction algorithm. After the recon-struction PCA method is used to monitor the transition process. Besides, the proposed method and multi-modes PCA method which is a conventional method are applied to monitoring con-tinuous annealing process. The simulation results show the proposed method can not only re-duce false alarms, but also improve the accuracy of fault detection. The proposed method can detect the fault effectively with good detecting performance.(2) As the transition process has dynamic character, which means the current samples of monitoring variables are related with historical samples, the dynamic IRPCA method (DIRPCA) is proposed based on the dynamic latent variable method. The current sample ma-trix is expanded to represent the dynamic character of the transition process. Then the pro-posed IRPCA method is used to monitor the transition process. In addition, this DIRPCA me-thod and IRPCA method are applied to monitoring continuous annealing process and elec-tro-fused magnesia furnace process. The simulation results show that the DIRPCA method can reduce false alarms and improves the fault detection performance further.