An uncontrolled self-learning algorithm is used for early detection of malfunctions in industrial equipment based on the combined algorithm of vibro-acoustic analysis of signal clustering and further recognition using the NM500 neuromorphic chip. At the stage of configuration and training, NT Adaptive.VAS studies data without any prior knowledge of failure modes, and when the studied modes of operation of nodes or aggregates converge to a certain number (as the input data is clustered, the number of new clusters stops growing), it is assumed that the device knows all the operating modes of the equipment. Then, during operation, the distances between the clusters of the input processed data are compared with the previously studied data (cluster centers) using the NM500 neuromorphic chip. If the numerical vector, composed of the features of the current cluster, characterizing the operating mode of the equipment does not coincide with the previously learned vectors compiled according to the same principle, a conclusion is made about the presence of a new, previously unknown operating mode of the equipment. Thus, the system is capable of detecting anomalous new data that was previously unknown. In this case, the assumption is made that the categories (clusters) discovered and learned at the “learning stage” are categories that can be attributed to the “conditionally good” state of the equipment. At the recognition stage, the emerging cluster belongs to the category of new and unknown, which is reported to the maintenance personnel, who must respond appropriately
