Online Classification of Machine Operation Modes Based on Information Compression and Fuzzy Similarity Analysis

This paper proposes a computational scheme for online classification of process data sets that represent different operation modes of machines and other systems by use of fuzzy similarity analysis. The classification procedure starts with a preliminary given small number of known operation modes (data sets) which constitute the initial size of the Knowledge Base (KB). During the online classification, the dissimilarity (difference) degree between the newly submitted operation mode and each of the current modes stored in the KB is computed. As a result, depending on the preliminary given threshold for classification, the newly submitted mode is classified as belonging to a certain class (operation mode) from the KB or is considered as a new (unknown and different) mode. In the latter case, this mode is added as a new entry of the KB and used for the further online classification. An unsupervised learning algorithm (a modification of the Neural-Gas leaning algorithm) is used in the paper for compression of the original “raw” data from each operation mode into Compressed Information Model (CIM) with a smaller number of neurons. Then the similarity analysis is performed as a two-input fuzzy inference procedure that uses the Center-of-Gravity Distance and the Weighted Average Size Difference between two CIMs. The membership functions and the singletons of the fuzzy inference procedure are tuned in the paper by using a given set of test operation modes. The whole computational scheme is illustrated and discussed on a real example for classification of 5 main operation modes of a hydraulic excavator. Keywords— Fuzzy Similarity, Information Compression, Online Classification, Operation Modes, Unsupervised Learning