Executive Summary : | Pipelines are crucial for transporting liquid and gas resources, but they are also susceptible to corrosion, erosion, and external intrusions, leading to leakages that can cause significant economic losses and environmental threats. Detecting leaks and locating their positions is crucial, and developing a reliable leak diagnosis system remains a significant challenge. Acoustic Emission (AE) technology has gained significant attention for pipeline leak diagnosis, as it detects stress waves generated by leaking fluids and captures leak attributes. Artificial intelligence (AI) and data analytics tools have led to the development of various data-driven approaches for AE-based pipeline leak diagnosis. Machine learning algorithms like artificial neural network and support vector machine have been extensively applied to analyze AE leak signals, while cross-correlation analysis and signal processing methods like wavelet transform have been utilized for precise leak localization.
However, existing techniques have several shortcomings, such as requiring large amounts of prior labelled leak data for training and being ineffective for automated real-time leak detection. One-class classification-ML techniques (OCC-ML) can resolve this issue by training fault detection models using the system's healthy condition data only. This project proposes developing leak detection and localization models based on OCC-ML and VMD techniques, with an experimental setup simulating pipeline leaks and AE signals acquired for further analysis. If successful, this research could lead to the widespread adoption of intelligent leak diagnosis systems in various industries, ensuring minimal financial losses and safety. |