A Deep Autoencoder-based Adaptive Control Chart Framework under Weak Supervision for Industrial Anomaly Detection

Nowadays, industrial anomaly detection (IAD) is challenged by multi-dimensional data, nonlinear process behavior, and the lack of reliable fault labels. To address these challenges, this paper proposes a Synergistic Machine Learning–Adaptive Control Chart (SML-ACC) framework under a weakly supervised setting for IAD. The framework combines data-driven feature learning with probabilistic supervision and adaptive statistical decision rules. Furthermore, this allows a reliable and interpretable process monitoring without requiring extensively labeled fault data. It is noted that normal process behavior was learned using only fault-free samples. In addition, deviations from normal operation were measured through fused anomaly scores, which are monitored using ACC with dynamically updated control limits. The proposed framework was evaluated using the three public industrial benchmark datasets namely Tennessee Eastman Process, Steel Plates, and SECOM dataset. Thereafter, the performance was analyzed through comprehensive benchmarking against classical statistical process control methods and unsupervised baselines. Notably, systematic ablation studies were conducted to evaluate the contribution of individual framework components. The results indicate that SML-ACC outperforms the tested classical SPC and unsupervised baseline methods in terms of F1 score and AUC value on the TEP and Steel Plates datasets. Its performance with the SECOM dataset is similar to the evaluated baselines. SML-ACC has an F1 score of 0.604 and an AUC of 0.762 on the Tennessee Eastman Process dataset. The F1 score for cross domain evaluation without retraining is around 69.2% of the value of the source domain. This performance drop is a result of the domain shift, and a motivation for future work on domain adaptation. Collectively, the findings suggest that a hybrid of deep representation learning and adaptive SPC can be an effective and scalable solution for anomaly detection in the context of Industry 4.0.