Development of a Multi-Agent System for Intelligent Diagnostics of Industrial Equipment

Modern industrial automation systems generate a large volume of production data during operation, the processing of which by modern artificial intelligence methods allows for timely diagnostics of the condition and prediction of wear of expensive equipment. The article develops an innovative multi-agent system based on neuroimmune-endocrine interaction for diagnostics of industrial equipment. The system consists of agents specializing in data reduction, based on an artificial neural network (ANN), an artificial endocrine algorithm (AEA) and an artificial immune system (AIS). The task of these agents is to reduce the size of the database without losing its information content. Also, predictive agents based on AIS and AEA have been developed, which solve the problem of classifying the condition of equipment based on information obtained after data reduction and predict equipment wear. Experiments were carried out on real industrial data of the oil refinery TengizChevroil LLC. The modeling results showed the prospects of using this approach. The following values of the AUC (Area Under ROC Curve) metric were obtained from 0.86 to 0.90, the throughput of the multi-agent system is 1,000 tasks per second, the prediction time is 1 ms, and the fault tolerance is 100%.