Method Of Abnormal Information Extraction During Continuous Casting Mould Based On PCA And NN

Keeping mould in a stable and efficient state is the key factor to continuous casting production. Research on process monitoring of mould has been done for a long time. However, with the trend of high-speed in continuous casting, the defects on surface quality, breakout and other abnormalities have never been resolved satisfactorily. Further study should be taken on methods of mould monitoring and extracting abnormalities. In this paper, based on principal component analysis (PCA) and Neural Network (NN), the measured data of the Mould Friction (MDF) has been analyzed, and appropriate method of abnormal information extraction of mould process has been explored.PCA and improved PCA methods are used to extract abnormal information from mould, and the principle and modeling process of PCA method are elaborated in this paper. Based on the measured data of MDF in slab continuous casting, a PCA model has been built, and the corresponding control charts, T~2 chart and Q chart, are generated from the calculation of the test of T~2 and Q. Appling PCA model to analyze a large number of abnormal data and the fault detection is performed by means of the T~2 and Q charts, the MDF and its root-mean-square charts are used to analyze the condition of process. Simulation results show that most process faults can be diagnosed by PCA model, and it is feasible to monitor mould process in continuous casting with the application of multivariate statistics method. Principal component related variable residuals (PVR) statistic of improved PCA can be used to make further diagnosis when T~2 plot diagnoses fault but Q plot. Analysis of the abnormal data shows that Changes of T~2 statistic, which is caused by processing conditions and process fault, can be distinguished by the improved PCA. Compared with the conventional PCA, the improved PCA is more effective and sensible in fault diagnosis and process monitoring.The technique, integrating PCA and NN, is applied to monitor abnormal mould process in continuous casting. PCA is used to pretreated measured data and get PCs which are as inputs of BP network. NN is employed to remove the nonlinear and dynamic characteristics of PCs and build NN model, which can expresses characters of normal production. The measured data has been compared with the outputs of NN model driven in the same observation, and the multivariable residuals derived from the differences between measured data and an output are evaluated and simple monitoring charts are generated from the application of PCA. Based on the data analysis of the measured mould friction data and...