Performance of Cluster-Based Logistic Profile Monitoring Under Existence of Different Linkage Functions

During industrial process monitoring, a common practice involves analyzing the relationship between a measured outcome (response variable) and other relevant factors (descriptive variables), which is called a profile. However, the perceptible challenge in this issue is the reliable estimation of profile parameters that can deviate significantly under the influence of outliers. Saremian et al. (2021) addressed the challenge of parameter estimation within generalized linear profiles during Phase I of a research investigation. They proposed a robust methodology for this purpose. Their results showed that incorporating a clustering approach, particularly with a complete linkage function, yields superior control charts parameter for monitoring binary logistic profiles compared to the traditional, non-clustering method. The performance of cluster-based control charts in monitoring logistic profiles is evaluated under varying linkage function conditions in this paper. The aim is to improve the performance of cluster-based method by evaluating the effect of using different linkage functions, including complete, average, single, weighted, centroid, median, and ward linkage. The simulation runs demonstrated a significant improvement in the Hotelling 𝑻𝟐 control chart s ability to detect process deviations when combined with a clustering approach. Furthermore, employing various linkage methods, such as average, centroid, and ward s linkage, demonstrably yields more accurate control chart parameter estimates compared to complete linkage. Therefore, the application of the linkage functions presented in this study has led to an enhancement in the performance of the cluster-based method.