ABSTRACT While machine learning has proven effective in Control Chart Pattern Recognition (CCPR), most models are designed for uncorrelated processes, failing to capture the complexities of real‐world applications. Convolutional Neural Network (CNN), a prominent deep learning approach, provides a promising alternative. This research comprehensively assesses the efficacy of pre‐trained architectural models in identifying six types of unnatural Control Chart Patterns (CCPs) within autocorrelated processes. Choosing an appropriate pre‐trained model is essential, particularly in light of the increasing demand for intelligent CCPR and the need for accurate classification. We performed an extensive evaluation of well‐known pre‐trained models, including AlexNet, ResNet‐18, VGG, and Efficient, using Monte Carlo simulations. Applying of transfer learning and fine‐tuning techniques, we evaluate the models' ability to generalize in the context of the new CCPR task. Our examination provides detailed insights into the performance of pre‐trained models, demonstrating their ability to recognize CCPs more efficiently than competing models. The proposed monitoring scheme offers an effective solution for CCPR across varying levels of autocorrelation. The study reveals that autocorrelation influences the performance of deep CNNs‐based CCPR. The best overall accuracy achieved across all autoregressive coefficients is 92.09%, with a peak accuracy of 96.25% when the autoregressive coefficient is within the range of 0.3, 0.5. The evaluation is extended to a random‐onset simulation setting that relaxes the static alignment assumption, enabling assessment of robustness to pattern misalignment and partial pattern visibility. Additionally, a real‐world case study demonstrates the best recognition model's applicability.
Sogandi et al. (Mon,) studied this question.