Research On Fault Detection Algorithm Of Process Industry Based On Data-driven

Fault detection system is one of the key elements of monitor in process industry monitoring. In order to improve product quality and economic benefit, the process conditions should be closely monitored and faults should be timely detected. Neverthelessthe technique of fault detection based on data–driven have became one of the focus in the process control field, which also is the key content of this thesis.This thesis mainly applied principal component analysis (PCA), and introduced the algorithm of wavelet transform as well as Time-Frequency Analysis, To different industrial processes objects, some improvements to traditional PCA have been made at different degree, and some new integrated algorithms of fault detection are also proposed,All the strategies are based on data-driven technique and independent of mathematical model.Dynamic Principal Component Analysis (DPCA) is an extend method of static PCA, it is used for monitoring dynamic multivariate process. However, the large computing load of DPCA limits its application to industrial process. The improved DPCA menthoned in the thesis can make certain the time-lag variable and time-lag length according to the autocorrelation remarkable criterion, comparing with traditional DPCA, it not only decreased the overhead for model detection, but also reduced tcomputing work. The paper demonstrated the effectiveness of DPCA for monitoring dynamic multivariate process.Traditional PCA is a method only observes datas on all scale from single scale aspect, so the resolving power for the event happened in some scale is not high. In the paper, a new multiscale principal component analysis (MSPCA) approach is presented to resolve the problem.It is a kind of fault detections method that combines the ability of wavelet analysis to extract deterministic features in different scales and the ability of PCA decorrelate to the variables.The new method can set up the corresponding PCA model for the wavelet coefficient in the each scale, then monitor the process matter in the defferent scale. Comparing with the traditional single-scale PCA, it has the higher resolving power.Recently, there is no a kind of algorithm can resolve the all problems in the fault detection system of industry process, so it is important to integrate the exist algorithms, By studying the inner relationship and characteristic. An integrated framework for process fault detection is...