پايش فرايندهاي دو‌مرحله‌يي با مشخصه‌ي كيفي پروفايلي در مرحله‌ي دوم

Monitoring Two-Stage Processes with a Profile Quality Characteristic in the Second Stage

در اكثر فرايندها كيفيت محصول حاصل عملكرد مراحل مختلف فرايند روي آن است. اين مراحل معمولاً به يكديگر وابسته‌اند، لذا مستقل فرض‌كردن مراحل سبب بروز خطا در تحليل خروجي مي‌شود. تاكنون تاثير چنين شرايطي بر پايش فرايندهاي چند‌مرحله‌يي تك‌متغيره و چند‌متغيره بررسي شده، در حالي كه شرايطي كه در آن فرايند چند‌مرحله‌يي مشخصه‌ي مورد پايش از جنس پروفايل باشد كم‌تر مورد پژوهش قرار گرفته است. در اين نوشتار علاوه بر معرفي مدلي مبني بر وجود پروفايل در يك فرايند دومرحله‌يي، دو رويكرد مختلف براي پايش اين فرايند پيشنهاد شده است. همچنين عملكرد دو رويكرد پيشنهادي تحت تغييرات مختلف در پارامترهاي پروفايل و مشخصه‌ي كيفي مرحله‌ي اول با استفاده از معيار متوسط طول دنباله در فاز دو مقايسه شده است. نتايج نشان مي‌دهد كه هر دو رويكرد ارايه‌شده عملكرد مناسبي دارند.

In most processes, the quality of a final product depends on the quality of several quality characteristics at previous stages. This is referred to as a cascade property in multi-stage processes. On the other hand, recently, profile monitoring, in which the quality of a process or product is characterized by a relationship between a response variable and one or more explanatory variable(s), has been considered by many researchers. In some applications, the quality of a product in a two-stage process is characterized by a simple linear profile. In this case, monitoring the profile separately leads to misleading results, due to ignoring the cascade property in a two stage process. In this paper, we specifically consider a two stage process where the quality characteristics of the first and second stages are characterized by a univariate normal distribution and a simple linear profile, respectively. Then, we use a model to relate the profile quality characteristic in the second stage to the univariate normal quality characteristic in the first stage. After that, we propose two approaches to account for this problem. In the first approach, the mean and standard deviations of the first stage are monitored by two Shewhart type control charts, namely, and R control charts. In addition, we propose using hotelling, as well as a control chart, as selecting control charts to monitor the estimated parameters of the model relates the first and second stage quality characteristics. This implies that getting a signal from these control charts is equivalent to the shift in the second stage. In the second approach, we use EWMA and R control charts to monitor the mean and standard deviations of the normal quality characteristics in the first stage. Then, we use control charts to monitor the mean and standard deviations of the residuals of the model, which relates the two stages to each other. The control charts are similarly selected control charts. The performance of the two proposed approaches is compared through simulation studies and using an average run length (ARL) criterion. The results show the suitable performance of both proposed approaches.