Vehicle Diagnostics using Artificial Intelligence and Digital Signal Processing Methods

Description

Modern industrial, energy, and transportation sectors require advanced diagnostic technologies to detect early-stage faults under real-world conditions. This dissertation develops a comprehensive, low-cost, and scalable framework integrating high-bandwidth MEMS sensing, digital signal processing (DSP), and deep learning to monitor engine misfires and rotating machinery defects.

The core hardware utilizes an ADXL1002 MEMS accelerometer integrated with a BeagleBone Black embedded platform. This setup enables high-frequency vibration acquisition from vehicle engines under varying speeds and loads. To ensure generalizability, the study incorporates datasets for rotating element bearings—critical components in industrial drives and generators. To navigate non-stationary and noise vibration data, the work employs advanced DSP techniques:

• Empirical Mode Decomposition (EMD): Isolates transient components.
• Spectral Kurtosis (SK): Enhances fault-induced modulations.
• Short-Time Fourier Transform (STFT): Provides interpretable time-frequency signatures.

These methods allow for the isolation of specific signatures associated with misfire dynamics and bearing anomalies, significantly improving sensitivity to early-stage defects. The framework evaluates several sophisticated models including Deep CNN-LSTM & Multi-Head DCNN-LSTM for complex temporal dependencies and multi-scale features from raw signals.

The results demonstrate high diagnostic accuracy across various scenarios, proving the feasibility of deploying deep learning-based fault detection on edge-level platforms with the following key contributions including validated MEMS module, high-resolution misfire vibration dataset, enhanced DSP methodology, and optimized high-performance AI models.

While primarily focused on automotive engines, these methods are broadly applicable to electric motors, drivetrains, and renewable energy assets, advancing the future of sustainable industrial monitoring.