Diagnosing Out-of-Control Signals of Multivariate Control Chart based on Variable Length PSO-SVM

Multivariate statistical process control is an essential procedure employed to deliver quality products in modern manufacturing and service industries. Multivariate control charts are an extensively used tool to determine whether a process is performing as intended. Once the control chart detects an abnormal process variable, one difficulty encountered is to interpret the source(s) of the out-of-control signal. Therefore, a novel approach for diagnosing the out-of-control signals in the multivariate process is developed in this paper. The proposed methodology uses the optimized support vector machines (SVM) based on variable-length particle swarm optimization to recognize subclasses of multivariate abnormal process patterns and to identify the responsible variable(s) for their occurrence. Based on a simulation experiment, the proposed approach verifies its capability in accurate classification of the source(s) of out-of-control signal and outperforms the conventional multivariate control scheme based on SVM.