Monitoring variability in high‑dimensional processes remains a challenging problem in multivariate statistical process control, particularly when only a small subset of the covariance elements changes meaningfully. The Adaptive LASSO Thresholding (ALT‑norm) chart has shown strong performance in detecting sparse covariance shifts; however, as a Shewhart‑type statistic, it lacks the memory structure needed to enhance sensitivity to small, persistent changes. This study introduces an EWMA‑enhanced ALT‑norm framework that integrates exponential smoothing into the ALT‑norm statistic to improve detection of subtle deviations in the covariance matrix. The proposed EWMA‑ALT‑norm chart applies adaptive thresholding to the sample covariance difference matrix and updates the resulting monitoring statistic via an EWMA recursion, thereby accumulating information across successive samples. Extensive Monte Carlo simulations were conducted under seven covariance‑shift scenarios, multiple shift magnitudes, and various combinations of sample size and dimensionality. Results show that incorporating EWMA substantially improves detection performance, particularly for small and moderate shifts. Smoothing parameters ℝ?=0.20 and ℝ?=0.50 consistently provide the best trade‑off between memory and responsiveness, yielding lower ARL1 values than the original ALT‑norm chart across most scenarios. Overall, the proposed EWMA‑ALT‑norm chart offers a robust and computationally efficient approach for high‑dimensional covariance monitoring, extending the applicability of adaptive thresholding methods to settings where early detection of subtle changes in variability is critical.

Keywords

EWMA, adaptive thresholding, monitoring covariance matrix, multivariate statistical process control, average run length