The LASSO-based Diagnostic Problems For Distribution-free Multivariate Control Chart

Statistical Process Control (SPC) techniques have been developed to wide range ofindustrial applications, and a large number of classical univariate control charts have been widelyaccepted in industry. However, with the fierce market competition in the recent years, in order tomeet the rising demand about product quality and service quality of the consumer, Manufacturersand service industries are required to monitor several quality characteristics simultaneously inthe processes of production and service. Therefore, the multivariate statistical process control(MSPC), which is based on the research of the traditional statistical process control techniquedevelopment and multivariate statistical analysis methods, has been formed gradually.While MSPC has been researched in many literatures and obtained abundant researchachievements, it is always a great challenge about the design of distribution-free control schemes.A distribution-free multivariate control chart regarding to the location parameters for themonitoring process is proposed in the recent study. This control chart used dynamic controllimits which are determined online and depend on the current and past sample observations, anddemands not too more about the distribution the observations follows. The studies show that thecontrol chart above can obtain very satisfactory in-control run-length performance whenhistorical data is insufficient. Therefore, it plays vital role in the practical application.However, apart from quick detection of abnormal changes of system performance and keyparameters, accurate fault diagnosis of main parameters has become increasingly important in avariety of applications that involve rich process data in monitoring complex system. But theproblem about the fault diagnosis after the control chart alarms didn’t get further study in theresearch of distribution-free multivariate control chart proposed. That is to say, the majorobjective focus on the monitoring stage and does not identify the definite reason causing thecontrol chart alarms. Therefore, the main purpose of this paper is to utilize the LASSO-baseddiagnostic framework, which is on account of the research of distribution-free multivariatestatistics process control, to diagnose after the control chart alarms. It can accurately determinewhen a detected shift has occurred and identify the major components of that shift or factorsleading to the change. This makes a better practical application of the distribution-freemultivariate statistical process control methods. It can not only monitor a production processwell, but also help business managers and engineers to identify and eliminate the root of faultquickly and accurately, which results in the improvement of products’quality and production.