Journal of Quality Engineering and Management

Abstract In most cases, the alert received from a control chart does not indicate the actual time of the process change due to the delay between the actual change time and the time of receiving the alert from the control chart. As a result, it is necessary to examine the real time of change, which is referred to as the "point of change". By reviewing the literature on identifying real-time process changes, it can be concluded that most research in this field focuses on univariate processes and little research is devoted to multivariate processes. In addition, most research in the field of estimating change time in multivariate processes has focused on changes in the mean process vector, and only one research has been done on the covariance matrix. In this paper, a model based on artificial neural network is proposed to estimate the point of change in the covariance matrix of multivariate normal processes. The method presented in phase 2 is control diagrams and the type of change that occurred in the variance of qualitative characteristics is assumed to be the type of step changes. The performance of the proposed method in estimating the change point is evaluated based on two criteria of experimental distribution of estimates as well as the mean and standard deviation of the change point estimator for different step shifts in the variance of process variables in a simulation study. Finally, in order to further explain the proposed method, a numerical example is provided. The results show the proper performance of the proposed method in estimating the change point in the covariance matrix of multivariate normal processes.

Abstract Sometimes, a quality characteristic is expressed as a specific functional relationship, which is referred to as a profile in statistical quality control. In this paper, linear profile analysis in Phase I control charts is investigated from a new perspective. In statistical quality control, it is usually assumed that the process has a constant mean. However, there may be processes where the mean of the parameters is not constant over time, and this variability is an inherent characteristic of the process. For controlling such processes, most existing control chart methods are ineffective. To statistically monitor such processes in Phase I, it is necessary to segment the process into sections where it is stable and consistent before performing statistical process control. For process segmentation, few studies have proposed the use of time-based clustering techniques, and the existing studies have mainly focused on univariate data. In this paper, we develop a process segmentation method for profile monitoring. Different segmentation algorithms for functional data are compared, and the results are reported. Studies show that the functional data clustering algorithm based on the Fuzzy c-means method performs well for profile segmentation.

Abstract Control charts are among the most important tools in statistical process control, used to monitor the extent of variation in a process. When a assignable cause deviation is observed in a control chart, identifying the root causes of the change and determining the time at which the deviation began—referred to as the change point—is crucial and impactful. In some statistical process control problems, the quality of a product or the performance of a process is described by the relationship between a response variable and one or more independent variables, known as a profile. In many applications, such as calibration, this relationship is described by a linear profile, while in other situations more complex models, such as Poisson regression profiles, are required. In this paper, the maximum likelihood estimation method is employed to detect change points in Phase II monitoring of Poisson regression profiles, and its performance is evaluated through simulation.