Small Fault Detection and Classification Method for Complex Production Process

In order to monitor the condition of a complex production process, based on the multivariate statistical process control and support vector machine theory, the cumulative control chart principle was extended to multivariate form for data preprocessing, and the principal component analysis (PCA) was utilized to extract significant information from the complex production process. The effective data of small fault was obtained. Then, the statistical threshold values of normal data, statistic T2, and squared prediction error (SPE) of corresponding fault data were constructed to realize the small fault detection of the complex production process, and the support vector machine (SVM) multi-classification method was used to classify the detected fault modes. The simulation results of a hot mix asphalt production process show that in cases of small fluctuation, cyclical rise and fall respectively, the small fault detection recognition rate is about 95%, with an average improvement of 75% over the PCA method; the classification accuracy rate reaches 92.5%, improved 19.3% compared with the BP neural networks method.