Application of time series classification methods in the operational control of the progress of technological operations

The work discusses the practical testing of machine learning methods for the implementation of automated operational control of technological operation. The technological operation of tightening the screw connection was chosen as the object of research. To implement automatic control, the solution of the problem of classifying the sequence of data obtained during the execution of operations is considered. The distinctive feature of the task is a limited set of data used in training and testing neural networks. The authors have performed a review of sources and a comparative analysis of the effectiveness of training and the use of neural networks with recurrent and convolutional architectures. The experimental studies were carried out on a robotic stand using a prototype of a screw-down device of the authors’ design. Based on the test results of the considered models, a neural network architecture is proposed that provides an optimal ratio between accuracy and speed of its operation.