Analysis of heavy vehicle air compressor failures using machine learning methods

In spite of being a small component, air compressors help the heavy vehicle to keep its motion safely on long journeys. In the study, as a result of the literature and technical service interviews, the most common air compressor failure cases were determined. The determined failure types were evaluated within themselves and test preparations were made according to the failure degrees. For the experiments, the test setup was created, software and sensor hardware, and 19 different operating conditions data were recorded. With 80% of the recorded 23,987 data, Support Vector Machines (DVM), K-Nearest Neighbor (K-nn), Naive Bayes Classifier, Random Forest Algorithm and Artificial Neural Networks algorithm models were created in the PYTHON program and these models were tested with the remaining 20% of data. Models were subjected to 10-fold cross validation. Then, the accuracy rates of the models were determined according to the test data. Support Vector Machines Radial Based Function Kernel 100%, K-Nearest Neighbor algorithm Manhattan Distance Criteria 99.50%, Gaussian Naive Bayes Classifier 94.60%, Random Forest Algorithm 99.30%, Artificial Neural Networks gave 99.80% accuracy. Kappa and F1 score values of the models were examined, and complexity matrices were created for training and test data. In the data that the models never encountered, the estimated class was determined by percentage and the results obtained were evaluated. This study shows that machine learning algorithm models can be effective in predicting heavy vehicle air compressor failures.