Diesel Engine Fault Detection using Deep Learning Based on LSTM

This work aims to create a deep learning model utilizing Long Short-Term Memory (LSTM) as a classification model to detect and diagnose potential problems in diesel engines. The default dataset comprises 3,500 data entries and four distinct dataset types. White noise in the dataset is attributed to additive Gaussian white noise (AWGN). The main contributions of this study comprise developing a classification model aimed at identifying damage in diesel engines, as well as evaluating the performance of the Long Short-Term Memory (LSTM) method. The K-Fold cross-validation technique was utilized, where the value of k was set at 5. This study examines the damage classification in diesel engines using four datasets: default 0dB Noise, 15dB Noise, 30dB Noise, and 60dB Noise. The K-Fold Cross Validation methodology was employed, utilizing a value of k=5. The training and validation processes were conducted iteratively, employing a learning rate of 0.00005 during the training phase. The training phase utilized a mini-batch size of 128, and the maximum number of epochs was set at 10,000. The results indicate that the dataset containing 60dB noise consistently demonstrates the highest level of accuracy, with an average accuracy of 97.91% across the five folds and the efficacy of Long Short-Term Memory (LSTM) models in classification tasks is improved in the presence of white noise.