Towards Intelligent Control Loop Fault Diagnosis in Process Industries

Abstract

This thesis presents data-driven and computationally efficient methodologies for intelligent fault diagnosis in industrial process control loops. The work focuses on three major aspects: oscillation detection and characterization, valve stiction detection, and root cause analysis (RCA) in interconnected process systems. A neural network-based approach using FFT and autocorrelation-derived features is proposed for oscillation detection, achieving high accuracy with significantly reduced computational complexity. Methods for estimating oscillation period and amplitude are also developed and validated using industrial data.