Fault Classification in Diesel Engines Based on Time-Domain Responses through Signal Processing and Convolutional Neural Network

In today's interconnected industrial landscape, the ability to predict and monitor the operational status of equipment is crucial for maintaining efficiency and safety. Diesel engines, which are integral to numerous industrial applications, require reliable fault detection mechanisms to reduce operational costs, prevent unplanned downtime, and extend equipment lifespan. Traditional anomaly detection methods, such as thermometry, wear indicators, and radiography, often necessitate significant expertise, involve costly equipment shutdowns, and are limited by high usage costs and accessibility. Addressing these challenges, this study introduces a novel approach for fault detection in diesel engines by analyzing torsional vibration data in the time domain. The proposed method leverages short-term Fourier transform (STFT) and continuous wavelet transform (CWT) techniques, integrated with a convolutional neural network (CNN) to identify hidden patterns and diagnose engine conditions accurately. The method achieved a detection accuracy of 96.5% with STFT and 92.2% with CWT. To ensure robustness, the model was tested under various noise conditions, maintaining accuracies above 70% for noise levels up to 40%. This research provides a practical and efficient solution for real-time fault detection in diesel engines, offering a significant improvement over traditional methods in terms of cost, accessibility, and ease of implementation.