Detecting Incipient Fault Using Wasserstein Distance

This article develops a novel process monitoring method based on the Wasserstein distance for incipient fault detection. The core idea is to measure the difference between the normal data and the faulty data. For Gaussian distributed process variables, the paper proved that the difference measured by the Wasserstein distance is more sensitive than the Hotelling¡
⁻
s
$T^{2}$
and the Squared Prediction Error (SPE) in the Principal Component Analysis (PCA) framework. For non-Gaussian distributed data, a Project Robust Wasserstein distance (PRW) model under the PCA framework is proposed and an algorithm called Riemannian Block Coordinate Descent (RBCD) algorithm is used to solve this model, which is fast when the number of sampled data is large. An application study to a glass melter demonstrate the effectiveness of the proposed method.