High-dimensional process monitoring and change point detection using embedding distributions in reproducing kernel Hilbert space

High-dimensional process monitoring has become ubiquitous in many domains, which creates tremendous challenges for conventional process monitoring methods. This article proposes a novel Reproducing Kernel Hilbert Space (RKHS)-based control chart that can be applied to high-dimensional processes with sophisticated process distributions to detect a wide range of process changes beyond the ones that are detected by traditional statistical process control methods. Through extensive experiments on both simulated and real-world processes and various kinds of process change patterns, it is shown that the RKHS-based control chart leads to improved statistical stability, fault detection power, and robustness to non-normality as compared with existing methods such as T2 and MEWMA control charts.