Custom CNN for acoustic emission classification in gas pipelines

Siddiqui, Md Ismail Hossain and Sakib, Anamul Haque and Hossain, Amira and Fardin, Hasib and Pranta, Al Shahriar Uddin Khondakar (2025) Custom CNN for acoustic emission classification in gas pipelines. International Journal of Science and Research Archive, 15 (1). pp. 1760-1768. ISSN 2582-8185

This study introduces a basic Convolutional Neural Network (CNN) approach for classifying acoustic emissions in gas pipeline monitoring systems. By converting raw acoustic signals into spectrograms, we leverage the visual pattern recognition capabilities of CNNs to identify and categorize 12 different pipeline conditions from the GPLA-12 dataset. Our architecture consists of three convolutional layers with max pooling followed by fully connected layers, optimized for spectral feature extraction. Experimental results demonstrate that even this straightforward CNN implementation achieves superior classification accuracy compared to traditional machine learning methods. The model successfully distinguishes between normal operations, various leak types, and structural anomalies under different pressure conditions. This research provides a foundation for real-time gas pipeline monitoring systems that can detect potential failures before they escalate into costly and hazardous incidents, contributing to improved pipeline safety, reduced maintenance costs, and environmental protection.