The Research On The Modified Multivariate Statistics Based Fault Diagnosis Technology

With the improvement of humans life,they have higher demands on the quality of the products,which leads to the introduce of the complex control system in the industrial system.It makes the industrial system become more intelligent and intricate,which brings the system suffering from higher security risks.Therefore,it is important to carry out fault diagnosis schemes and fault-tolerant control to sustain the system working in a reliable environment.The traditional fault diagnosis scheme is based on the analytic model,which makes the most of the deepest recognition of the system,and has preferable monitoring performance.Nevertheless,the precise model of the process is hard to be known as a prior because of the complexity of the system.Therefore,the model-based approach has its obstacle.On the other hand,with the development of the control science and electronic technology,there are varieties of sensors equipped with the system,which sample lots of process data.The abnormal state of the system can be reflected by these data indirectly,and the work needed to be done is to mining the information from the data,that is,data-based fault diagnosis.First,the Principal Component Analysis(PCA)based projection is studied.Based on it,the statistics-based scheme is discussed to monitor the process,and the statistics variables are improved to simplify the results.Furthermore,the Empirical Mode Decomposition(EMD)is utilized to carry out data-filtering,which improves the fault diagnosis performance.Then,the Partial Least Squares(PLS)based process monitoring method is studied to deal with the quality-based fault detection problem.Furthermore,the Total Projection to Latent Structures(TPLS)is studied to completely illustrate the relation between the process variables and quality variables.Focused on the deficiency of TPLS,this thesis proposes a modified method to improve the monitoring efficiency of the system.Finally,the Tennessee Eastman(TE)process and a numerical simulation are introduced,and two methods mentioned above are adopted within the process to validate the effectiveness of the both.