Russian Researchers Learn to Predict Malfunction of Critical Equipment
Neural network
Scientists from Volgograd State Technical University (VolgSTU) have developed a unique model for estimating the remaining operating life of generating equipment, which allows for accurately predicting the time of its failure.According to the creators of the development, the specific feature of their model is a hybrid structure built on the basis of deep neural networks, a powerful tool of artificial intelligence (AI). The results of their study were published in the ACM Transactions on Cyber-Physical Systems journal.Power generating equipment is used to generate energy that is used in the organization of autonomous, emergency, or backup power supply. Specialists explain that the issue of reliability of such equipment is important for almost all enterprises of the fuel and energy complex. The uninterrupted supply of electricity and heat to city residents and industrial enterprises depends on it.Despite preventive maintenance (PM), sudden equipment failure cannot be ruled out. The use of an accurate failure prediction tool can change approaches to maintenance and repair. It also significantly reduces the cost of equipment downtime.”Reliability-centered maintenance is a worldwide trend. The main idea of this approach is to determine the optimal set of maintenance operations and their frequency of application, taking into account the probabilities and consequences of equipment failures,” Maxim Shcherbakov, head of the Computer-Aided Design and Exploratory Construction Department, explained.In study, the prediction accuracy of the remaining life increased by 1.5 times due to the allocation of equipment life cycle intervals and application of the proposed model, Shcherbakov noted.It is based on a combination of mathematical models known as deep neural networks. This technology mimics the operation of the central nervous system and solves complex technical problems, such as pattern recognition or, as in this case, the prediction of equipment failures. In their paper, the scientific team proposed a new configuration based on a convolutional neural network (CNN) and a long short-term memory (LSTM) neural network.The next task of the research team is to create a new technology that allows not only to predict equipment failures, but also to form optimal solutions – what exactly should be implemented to prolong the service life of the equipment.The research is part of a VolgSTU strategic project within the framework of the “Priority 2030” program.The university development program includes four strategic projects, including the “Center for Digital Scientific and Educational Projects and Developments” and “Technologies for Industrial Innovation Cluster.”