2022 Volume 55 Issue 1 Pages 38-50
This paper proposes a hybrid online batch fault monitoring method that combines the signed digraph (SDG) with the k-nearest neighbors classifier (kNN) using dynamic time warping (DTW) distance. The sum of the k smallest DTW distances between the ongoing batch and normal-operational batch references is calculated for online detection. If the increase in the sum is greater than the predefined detection control limits, the sample is labeled as “abnormal” and an SDG diagnosis is made. The k corresponding normal samples from the k nearest reference batches are retrieved to calculate the upper and lower variable control limits for each variable online. Quantitative values are transformed into qualitative ones using these control limits, and the variable nodes with non-zero signs are diagnosed through SDG. The signs of a specified portion of causality arcs in the SDG are updated with calculations using online measurements. Each diagnostic route is given a weight determined by both the normal historical behavior and ongoing behavior of its root variable, and the diagnostic routes with the highest weights are considered to be the root causes of the occurring fault. The proposed DTW-kNN-SDG method was validated using data from a simulated batch production of penicillin with a variety of fault types, magnitudes, and fault duration times, and novel diagnosis results were subsequently achieved.
