Intelligent Diagnosis of Gasoline Engine Faults using Acoustic Features

Automatic detection and diagnosis of various types of aircraft engine faults is essential for ensuring flight safety and reliability. Such fault detection methods can be performed on large volumes of datasets with high accuracy and require low computational time. In contrast, manual fault detection is not only prone to human errors but is also time-consuming. Here, we proposed a machine learning algorithm to detect and diagnose 27 types of engine faults. Firstly, we employed a low pass filter to eliminate the unwanted noise present within the acquired raw audio signals. Then, two cepstral features were extracted from the pre-processed signals. Next, we classified the faults, first by providing two cepstral features individually and then by feeding the fusion of these cepstral features as input to different classifiers. Cubic Support Vector Machines outperform all other classifiers for all three approaches with an accuracy of 95.40%, 93.20% and 97.40% respectively. Lastly, to confirm the reliability and robustness of our approach, we compared the performance of our system by using cross-validation to that of the hold-out validation technique. For both cases, our model demonstrated promising results with an accuracy of 97.4% and 97.3% respectively. We expect that our proposed approach will be beneficial not only for aircraft engine faults but also for different machinery faults in various industries.