AI-Driven Acoustic Emission System for Real-Time Ammonia Leak Detection in Industrial Pipelines

Abstract

One of the latest developments in non-destructive evaluation is Acoustic Emission (AE). The use of high-frequency AE systems for pipeline monitoring has been increasing. Leak-related information can be detected in a pipeline’s AE signal; however, reducing noise often requires both signal processing and redundant measurements. In this work, AE testing was employed to detect ammonia leakage in pipelines by systematically analyzing signal properties. The leaks were simulated under various conditions, including pressure release, corrosion, and applied strains. Machine learning based on Support Vector Machines (SVM), a type of artificial intelligence (AI), was used to classify faults and assess detection accuracy. Key characteristics from AE, flow rate, and gas leakage sensors placed in the prototype setup were analyzed to diagnose leaks effectively. Before training the machine learning model, the sensor data from the ammonia pipeline were pre-processed and organized by frequency bands. The processed acoustic data were then input into the SVM classifier to extract leak-related discriminant features. Datasets generated under different conditions such as pressure drops, stress distributions, and leak sizes were evaluated using the SVM model in Python. The results achieved high accuracy, producing reliable and meaningful predictions suitable for the proposed platform implementation.