Deep Learning with LPC and Wavelet Algorithms for Driving Fault Diagnosis.

Area of Science:

• Automotive Engineering
• Machine Learning
• Signal Processing

Background:

• Vehicle fault detection and diagnosis (VFDD) and predictive maintenance (PdM) are crucial for preventing accidents caused by mechanical failures.
• Existing research has focused less on using voiceprint features for identifying specific vehicle faults.

Purpose of the Study:

• To propose an early voiceprint driving fault identification system using machine learning algorithms.
• To investigate the effectiveness of different signal processing techniques and deep learning architectures for vehicle fault classification.

Main Methods:

• Constructed a dataset of 43 common vehicle mechanical malfunction voiceprint signals.
• Filtered voiceprint data using Linear Predictive Coefficient (LPC) and Wavelet Transform (WT).
• Applied Deep Neural Network (DNN), Convolutional Neural Network (CNN), and Long Short-Term Memory (LSTM) for fault identification.

Main Results:

• CNN achieved the best accuracy on the LPC dataset.
• DNN demonstrated superior performance and training time on the Wavelet Transform dataset.
• The Wavelet Transform-DNN combination improved accuracy by up to 16.57% and reduced training time by up to 21.5% compared to other methods.

Conclusions:

• The proposed voiceprint-based system enables early detection of vehicle mechanical faults.
• Combining Wavelet Transform with DNN offers a highly accurate and efficient method for vehicle fault diagnosis.
• This technology facilitates proactive driver alerts for potential vehicle machinery failures.