Objective. To increase the quality of work of tensometric systems, using data obtained from the strain gauge system in the form of maps of the passage of moving objects, it is necessary to conduct research on the data of the strain gauge system and develop a set of algorithms, thanks to which it is possible to obtain the characteristics of railway cars in rolling stock and carry out identification of railway objects objects with a minimum error for further use of the obtained results in the construction of a system of identification and weighing of moving objects through tensometric systems. Research methods. To build a system for identifying different types of moving objects through single-platform railway scales in motion, it is proposed to use machine learning methods, namely neural networks and clustering algorithms. The software implemented as part of scientific research is written in the Python programming language using numPy, sklearn, statistics, and other libraries. Findings. Using such methods of machine learning as convolutional neural networks, clustering, perceptron and relying on the reference data of railway objects that can be used on the territory of Ukraine, a number of algorithmic solutions were obtained and implemented in the form of software, which identify the type of car by such characteristics such as the axle of the cart, the axle of the wagon, the ratio of the base of the wagon to the length of the wagon between the couplings, the weight of the axles. Using the weight coefitient for a specific tensometric system, during the calibration of the scales, the dependence of the weight of the car on its type and the mass of each of the axles was obtained. The originality. After conducting a study of the data on the passage of railway carriages and auto couplings through single-platform scales, it was established that the types of wagons can be categorized by such characteristics as the ratio of the base of the wagon to the length of the wagon between the auto couplings, the axle, weight. To obtain the ratio of the wagon base to the length of the wagon between autocouplings, it is necessary to perform data segmentation and clustering as follows - the wagon base is found as the distance between the middle of two bogies, and the length of the wagon between autocouplings as the middle of the distance between the bogies to the middle of the autocoupling. Practical implications. Using such methods of machine learning as convolutional neural networks, clustering, perceptron and others, an algorithmic solution for data processing of strain gauge systems was obtained, which allows to increase the accuracy of the identification of wagons, while reducing the dependence of the results on the speed of the wagons, which allows to increase the capacity of weighing systems of enterprises
Read full abstract