Yüzdeliğe Dayalı Kontrol Kartları

Abstract

In this thesis study, the process monitoring performance of percentile-based control charts, which aim to overcome the limitations of classical quality control methods, has been evaluated on non-normally distributed data. The analyses were conducted using energy consumption measurements obtained under real production conditions, with the goal of reducing false alarm risks caused particularly by skewed distribution structures and improving the balance of signal generation. Two scenarios were constructed using an energy consumption dataset. In the first scenario, a combined X ̄–S control scheme was designed using the logic of percentile based control charts (PBCC) and enhanced with optimization techniques. In this scenario, both the process mean and variance are monitored simultaneously, and the goal is for the control chart to reliably generate signals under a specific percentile condition. As an alternative to the traditional Average Run Length approach, control limit designs were created at a specific confidence level using the percentile-based control chart methodology. In this study, for the optimal design of the PBCC X ̄–S control chart, the control limit coefficient was optimized using a deterministic search method, while the lower and upper limits were optimized using a genetic algorithm. In the second scenario, the Shewhart median control scheme was examined and evaluated using a percentile-based approach under different sample sizes and design scenarios. In this context, run length distributions under both in-control and out-of control conditions were analyzed; in addition to traditional metrics such as average run length and median run length, run length percentiles were also considered for comparison. As a result of the application, it was observed that while the percentile based approach enabled more sensitive signal generation for small sample sizes, delays in signal generation occurred for larger sample sizes due to the increasing stability of the median. When the results of both scenarios are evaluated together, it is seen that percentile based approaches offer more flexible systems with a lower likelihood of false alarms compared to classical control schemes. These findings support the preference for percentile-based control charts in modern production environments, especially for datasets with unknown distribution structures or those that do not meet standard assumptions. The study contributes to academic literature and also provides practical recommendations for developing more reliable process monitoring techniques in industrial applications.