Modern metallurgical production cannot be imagined without the use of Industry 4.0 technologies. Manufacturing today is being actively digitalized, and SUSU scientists have a significant role in this process.
Preventing accidents and unplanned equipment downtime in production is quite real today. This is possible thanks to the use of intelligent information processing methods. Data is collected from sensors that monitor heat flow from bearings under high load. Neural network acts as a tool for processing data from sensors.
Data is transmitted to the neural network via wireless communication channels. The neural network analyses the data, warning the machine operator about possible operational irregularities and the threat of equipment breakdown.
It is important that a technician can only monitor the operation of an already trained neural network. To train a neural network, you need a machine learning specialist.
An engineer at the Laboratory for Self-Monitoring and Self-Validating Sensors and Systems Denis Lebedev, within the project on the "Technologies for Diagnosing Elements of Automated Process Control Systems and Measuring Instruments", develops top-level software: systems for storing and visualizing data, implementing diagnostic algorithms; and also trains artificial intelligence models. The project is being implemented at the university as part of the Priority 2030 program.
"Neural network plays a key role in sensor diagnostics," notes Denis Lebedev. "And in general, the work of the Laboratory for Self-Monitoring and Self-Validating Sensors and Systems is aimed at using artificial intelligence and machine learning technologies in technical diagnostics tasks."
Neural network is one of the main blocks of the project. Without it, there will be no data mining or intelligent diagnostics. An important clarification: the quality of the neural network must be constantly assessed, external control must also be carried out.
Today, the laboratory engineers are at the stage of developing a neural network, designing an optimal model that will cope with the task. Next, a series of large-scale tests will be carried out to confirm the functionality of the system. The tests are planned to be carried out on the equipment of the Laboratory for Self-Monitoring and Self-Validating Sensors and Systems, the material and technical base of which has been improved within the framework of the Priority 2030 project.
A useful product of the project − a combined neural network data processing model − is planned to be used for a diagnostic system for machine tools in metallurgy and machine tool building.
If, according to the test results, the neural network successfully copes with the assigned tasks, then gradually part of the work of assessing the technical condition will fall on its "shoulders". In this case, an employee of the enterprise reads a status on the display, informing either that the equipment is working properly, or that some kind of defect has occurred. In the future, neural network can completely replace humans in tasks of determining the technical condition in this project. But to achieve this, there is still a lot of work to be done.
"Our goal is to develop data processing methods, including the use of neural network models, which in the future will independently cope with technical diagnostics tasks without the participation of a specialist in this field. These are the prospects for the development of our project," explains Denis Lebedev.
The technology has great potential for commercialization. Large industrial partners have already become interested in developments in this area. It is worth noting that the development of neural networks should be based on a specific customer since all enterprises have their own characteristics, and neural networks must be configured for a specific customer.
The project is being implemented at the Laboratory for Self-Monitoring and Self-Validating Sensors and Systems, operating at SUSU within the framework of the Priority 2030 program under the Science and Universities national project.