TECHNICAL ASPECTS RELATED TO THE DEVELOPMENT OF A MONITORING SYSTEM FOR METAL-CUTTING EQUIPMENT USING MACHINE LEARNING

The purpose of this article is to examine the features of developing a monitoring system for metal-cutting machines from the point of view of data collection and model training. This paper examined the economic feasibility of a process monitoring system with machine learning based algorithms. The development of such a monitoring system can reduce equipment downtime, reduce production costs, and increase the productivity of plants. As a result of analysing previous work in the field of smart manufacturing, the requirements for data acquisition sensors were formulated and described – high sensitivity, fast response, accuracy, providing high signal-to-noise ratio, versatility, robustness to the manufacturing environment, low cost, rapid system deployment and modularity. The paper reviewed and described the main data parameters that need to be considered in model selection and training, including nonlinearity, non-Gaussianity, noisiness, high dimensionality, multiclustering and dynamism. The results of modelling the behaviour of faulty process variables were visually demonstrated. The modelling was carried out in order to show that the process variables are not fixed values, therefore, it is not possible to directly compare the process variable with some fixed threshold value, i.e. it is necessary to take into consideration the dynamics of changes in the variable even in the threshold values of normal behaviour. The machine learning models complexity in terms of computing power requirements was also discussed.