Artificial Intelligence (AI) based diagnosis and analysis for roto-dynamic machinery fault recognition and detection

Abstract

The reliability of rotating machinery is critical in desalination plants, where pumps are essential for continuous operation. Conventional vibration-based predictive maintenance methods remain limited by manual feature extraction, noise sensitivity, and difficulty in detecting complex fault patterns. To address these limitations, this study proposes an Artificial Intelligence Diagnosis and Analysis (AIDA) system that integrates deep learning with high-resolution vibration signal processing to classify faults and predict failures in centrifugal pumps within a Reverse Osmosis (RO) desalination plant. The methodology combines real-time data acquisition from IEPE sensors, Fast Fourier Transform (FFT) analysis, and AI-based pattern recognition for early fault detection. The proposed system was validated on an operational centrifugal pump, where it successfully identified a bearing defect, assessed its severity, and estimated remaining operational life. The results show that the AI-based approach improves accuracy, automation, and early fault detection compared with traditional vibration analysis methods, highlighting its potential for predictive maintenance in desalination and other industrial sectors.