Research On PCA Monitoring And Fault Diagnosis For Batch Processes

Molding is a widely used method in plastic processing. Injection molding machine is the main processing equipment. Injection molding process is a typical batch process which is nonlinear, dynamic and multi-stage. Different from general industrial processes, mechanism of the Injection molding process is rather complex, and the complicit of its operation is outweighed the continuous process. In order to improve the security of the injection molding process and the control system, it’s necessary to establish the monitoring system model to monitor and diagnose the production process.Currently the multivariate statistical modeling method for industrial processes, using principal component analysis (PCA) as the core technology, based on the process data, gradually becomes the important tool of online monitoring and fault diagnosis. And the Multi-way principal component analysis (MPCA) is a kind of effective method applied in batch production process monitoring and fault diagnosis. The method takes all the data of an intermittent operation as a sample. Although this method can effectively monitor the whole condition of the operation process, in batch process it’s hard to further analyze the character of the data for each sub-stage. Besides, the online process monitoring algorithm largely depends on the accuracy of the prediction of the future measured values.In view of the multi-stage characteristics of the batch process and the advantages and disadvantages of the ordinary MPCA method, a new stages separation method of the production process is provided in the paper. Based on the changes in loading matrixes and principal component matrixes, which reveal the evolvement of the underlying process behavior, a two-step sub-stage separation is realized. To reflect more objectively the similarity between the loading matrixes and the similarity between the principal component matrixes, two similarity measurement methods, based on weighted cosine of the angle between loading vectors and weighted absolute value of the singular value change respectively, are proposed to judge more accurately the changes of process characteristic. Process sub-stage separation is realized according to the same operation sub-stage has the greater similarity of loading matrixes and singular value matrixes. The PCA modeling, based on the improved stages separation method, is applied in online monitoring and fault detection of the Injection molding process. The result of the experiment has verified the effectiveness of the method and has certain directive significance in the actual production.Finally in this thesis, the process monitoring and fault diagnosis of injection molding has been implemented with C#. At the same time a sub-system is realized based on PCA theory.