Real-time monitoring of batch processes using the fast k-nearest neighbor rule

In recent years, fault detection using the k-nearest neighbor rule (FD-kNN) has been proposed and successfully applied to monitor batch processes. In FD-kNN, the data from a whole batch process are considered as a high-dimensional sample, and only when the ongoing process is finished, the FD-kNN can execute the anomaly detection. In order to detect the anomalies timely, the improved FD-kNN (IFD-kNN) has been proposed to monitor batch processes in real time. However, in IFD-kNN, the k nearest neighbors can be obtained by calculating the distances between the incoming sample and all other samples in the training set. Obviously, the process of near neighbor searching is highly time-consuming, especially for the large-scale processes, which may cause significant delay in providing the detection results of IFD-kNN. To address this difficulty, this paper uses exact euclidean locality sensitive hashing (E²LSH) to accelerate the near neighbor searching algorithm in order that IFD-kNN can be used to monitor the large-scale batch processes in real time. Finally, a numerical simulation example is used to demonstrate the effectiveness of the proposed method.