Online detection of deviation in performance of multichannel dynamical processes

In this paper a computational method for detection of deviation in performance of multiple parallel working channels in a dynamical process is proposed and discussed. The method consists of the following computation steps: one-dimensional multi-agent clustering of the outputs of the channels; similarity analysis of all pairs of channels; calculating the weighted global distance for each channel; static and dynamic ranking of the deviation of all channels. The proposed method is intended to work in online mode by continuously processing Data Blocks with fixed length and the respective results are the form of the ranking positions of the channels. The process with Rank 1 is the most deviated process channel. An experimental example is given in the paper for detection and analysis of the deviation in performance of six batteries used for driving a small electric vehicle. Other possible applications for online monitoring and performance evaluation of dynamical processes are also discussed in the paper.