A deep learning model for air leak detection from a pipe fitting using an accelerometer

Abstract

Gas leaks from fittings of a pneumatic pipe system result in the breakdown or failure of the system. It’s about half of the losses of output from production lines in the manufacturing sector. Deep learning (DL) methods can be used to detect gas leakage of the pneumatic pipe system. We propose the DL model for the detection of air leaks from pneumatic pipe system fittings using an accelerometer sensor system. We trained four models with four machine learning (ML) techniques with the data generated from our experimental pneumatic pipe. We augmented the collected data and used it to train all four models again and were able to mimic the natural behavior of the actual line and thereby augment the collected data, which was used in an enhanced training and testing process to create a better model. One of the trained models in which the augmented data was applied yielded the highly accurate result of 99.2%. Our main contribution to the field is our method of evaluating the accuracy of the model and the simple algorithm that one may use as a basis for building applications based on the model, together with the model's evaluation results. Our findings and contribution provide well-tested information to engineers and companies to avoid breakdowns in pneumatic pipe systems caused by air leaks. We claim that these contributions are new, and to the best of our knowledge have not previously been reported in the literature, thus are relevant and important contributions to the field.