چكيده به لاتين
Nowadays, due to the increasing growth of the railway industry and the increase in the speed of trains, the need for maintenance and repair of railway systems has become the main need of the railway industry. In recent years, data-driven fault diagnosis techniques have received much attention due to their speed, accuracy, and lack of the system model. Therefore, in various industries, especially in the railway industry, the use of machine learning has become the main subject of research. Data-based fault diagnosis techniques generally consist of 3 main parts: data acquisition, feature extraction, and classification. Diesel locomotives are known as the heart of traction engines, and a fault in these engines can completely disrupt the performance of the locomotive and the train. One of the most important parts of diesel locomotives are starter motors (a type of direct current motor). which are responsible for moving and turning on the locomotives. To monitor these motors, current and speed signals are received from these motors. To use these signals in classifiers and to identify the pattern of healthy and defective signals, the characteristics of these signals are extracted by wavelet transformation. The discrete wavelet transformation method was used in 9 levels, which provides time-frequency features. Then, from the detail levels of these signals, statistical features such as skewness, elongation, etc. were extracted. This greatly increased the separability of the data, so that relatively simpler classifiers could be used to separate the data from each other. 80% of the data is used as training data and 20% of the remaining data is used as test data. Then, using the cross validation method, the hyperparameters of the support vector machine are updated and the best hyperparameters are selected by greedy search. Finally, after applying SVM on the test data, we reach the results of FAR = 0.2 and MAR = 0.0. Considering that the MAR criterion is a very important criterion in the railway industry, it is a favorable result compared to the volume of data.
