Recognition of quality abnormal patterns for a dynamic
process has seen increasing demands nowadays in the real-time
process fault detection and diagnosis. As the dynamic data from a
quality abnormal process is linearly inseparable, the recognition
efficiency of a support vector machine (SVM) model mainly depends
on the selection of the kernel functions and the optimizing of
their parameters. Based on the analysis of the quality abnormal
patterns in a dynamic process, this paper presents a recognition
framework of quality abnormal patterns by using a multi-SVM
(MSVM). For the different quality abnormal patterns, the
simulation results indicate that the recognition accuracies of the
MSVM classifiers with the selected kernel functions are quite
different. A MSVM recognition model for quality abnormal patterns
in a dynamic process is proposed by the kernel functions being
of high accuracies. It is shown that this MSVM model with suitable kernel functions can increase the recognition accuracy.