Abstract:The control chart is very useful and efficient to detect abnormal changes of a process, however it is difficult to identify out-of-control variables that are responsible for the change. Accurate fault diagnosis has become increasingly important, which helps engineer quickly eliminate the root causes of abnormal changes. However, most of existing diagnostic methods were developed under the assumption of normality, of which the performance is affected by the nonnormality. To overcome these limitations, this paper attempts to combine Bootstrap technique to propose a diagnosis method that is robust to underlying distribution, but also applicable to the case of small out-of-control sample size. Bootstrap technique is used to improve the diagnostic ability of the diagnostic method, which is our first attempt. Compared with the existing procedures, numerical simulations favour the proposed procedure. Finally, an example from a bolt production process is provided to illustrate the implementation of the proposed procedure.

A robust diagnostic approach in mean shifts for multivariate statistical process control