CLASSIFICATION OF THE TECHNICAL CONDITION OF WATER PUMPING EQUIPMENT USING CONVOLUTIONAL NEURAL NETWORKS

Abstract

The problem of equipment repair management in water supply systems is considered. The purpose of the study is to study the issues of the possibility of classifying the technical condition of pumping equipment through the use of modern machine learning models. Classification is carried out on the data that were obtained by vibration diagnostics during the maintenance of pumping units. The initial data is converted into chalk spectrograms, the results of the conversion are presented as images. The model classifies time series according to one of three states: satisfactory, acceptable, not acceptable. Accuracy assessment is carried out by calculating indicators such as: precision, recall, F1-score. Since each class has a different number of measurements in the source data, the accuracy is determined by calculating each of the indicators for each class separately and then averaging them. Convolutional neural networks are used to classify images. To solve the problem, a network with a deep ResNet-50 architecture was chosen. The accuracy indicators that the network showed in other classification tasks and the speed of network learning determined the choice of this configuration. The average value of the precision indicator reaches 0.81, recall 0.90, and F1-score 0.85, which is a good result for the top-1 criterion. Classification of chalk spectrograms obtained from the initial data shows good results, despite the fact that the data has not been cleared of noise. The proposed methods and models have been tested on real data, which confirms the possibility of their use in the development of an intelligent information system for managing the technical condition of pumping equipment.