Industrial Internet of Things (IIoT) environments generate high-dimensional and continuously drifting sensor data that challenge the stability and reliability of traditional PCA-based monitoring. This study introduces a unified Drift-Aware Whitening Residual Monitoring (DAWRM) framework that reformulates the classical SPE/ Q statistic into an adaptive self-calibrating residual model. The framework integrates covariance whitening through Ledoit–Wolf shrinkage, distribution-free conformal calibration, and a recursively defined Adaptive Drift-Sensitivity Index (ADSI) to maintain valid decision thresholds even when the underlying measurement distribution evolves. In addition, the proposed architecture supports the digitalization of measurement processes by enabling autonomous drift-aware calibration and real-time uncertainty control within IIoT-based monitoring pipelines. In addition,to support decision-level discrimination, the system is complemented by a Drift-Weighted k -Nearest Neighbor (DW-KNN) classifier that incorporates drift-modulated distance metrics for online fault classification. Experimental evaluation using two heterogeneous datasets—a multi-sensor Industrial Fault Detection benchmark and a Smartphone human-activity stream—demonstrates that the hybrid DAWRM-DW-KNN architecture sustains high detection stability under both abrupt and gradual drift. Across both datasets, the method maintains mean accuracies above 0.95, low false-alarm rates, and consistent Quality-of-Adaptation (QoA ≥ 0.88), confirming that drift-aware whitening and adaptive calibration operate cohesively in real time. The results verify the thesis that integrating covariance whitening, distribution-free calibration, and drift-sensitivity modeling yields a robust, distribution-agnostic monitoring framework capable of supporting digitalized measurement systems in industrial and human-centric IIoT applications.
Jawad et al. (Sun,) studied this question.