clustering-based detection rule

Detection of change time of the process parameters is a crucial problem in statistical process control (SPC), because more detailed information on the time and the pattern of a change can provide process managers with more eective clues for root-cause analysis and corresponding corrective actions. Parameter changes may take dierent forms including monotonic, trend, step shift, and so on. The issue frequently considered in the relevant studies involves only a single shift, whereas an out-of-control condition may be caused by multiple changes occurring in dierent points. On the other hand, recently, the issue of prole monitoring in which the quality of a process or product is represented by a functional relationship between a dependent and a number of explanatory variables has attracted a great deal of attention as witnessed by the growing number of publications in this area. Our investigation showed that the studies dealing with change point estimation in prole monitoring had neglected the case of multiple change points. This gap is noticed as the primary subject of this research and a clustering-based algorithm is proposed for estimating the number, as well as the location of the change points, while monitoring a simple linear prole. This clustering-based method, which is implemented in an iterative manner, is an extension of a similar method in monitoring univariate individual quality measures using Shewhart control charts. A decision rule determined via simulation using a pre-specied signicance level enables the algorithm to detect multiple change points of the parameters in addition to identifying out-of-control conditions. The proposed method is applied in the phase I of process monitoring, where a historical dataset is available and the ultimate goal is to nd reliable estimates of the process parameters, including the intercept and the slope of a linear prole model. Extensive simulation scenarios were devised to declare the performance of the aforementioned method.