Practical Process History-Based Fault Detection and Diagnosis Algorithms Using Multivariate Statistical Process Control, Frequency, and Time-Frequency Approaches Addressing Variability in Datasets

Abstract

This thesis addresses the development of innovative techniques for fault detection and diagnosis for rotary machines, explicitly focusing on Belt Starter Generators (BSGs) and Internal Combustion Engines (ICEs). The primary objective is to create methodologies that enhance the accuracy and reliability of fault detection systems by utilizing advanced signal processing and machine learning approaches. Three key contributions are made in this research. First is the integration of Short-Time Fourier Transform (STFT) with Principal Component Analysis (PCA)-based Multivariate Statistical Process Control (MSPC). This novel approach addresses the challenges posed by data variability in industrial environments. It significantly improves the robustness of fault detection algorithms. Second, the research develops the Adapted Local Binary Pattern (ALBP) method for bearing fault detection in BSG systems. This method combines signal processing techniques with computer vision algorithms to achieve superior accuracy without increasing computational complexity. Additionally, innovative strategies for handling domain shifts, such as multi-sensory multiblock analysis and domain adaptation techniques, are proposed to ensure consistent fault detection performance across different operating conditions. Third, the practical applicability of the proposed methodologies is thoroughly evaluated through comprehensive validation experiments using real-world industrial data. The results demonstrate the effectiveness and reliability of these methods, highlighting their potential for real-world implementation. These contributions enhance the state-of-the-art in fault detection and diagnosis, significantly improving operational reliability and efficiency for industrial machinery. This thesis presents substantial advancements in fault detection and diagnosis, providing robust solutions to address variability and domain shift challenges in industrial data. The findings contribute to safer and more efficient industrial operations, reinforcing the importance of advanced analytical techniques in the automotive industry.