Gas leakage fault detection of pneumatic pipe system using neural networks

Abstract

This paper is concerned with gas leakage fault detection of a pneumatic pipe system using a neural network filter. In modern plants, the ability to detect and identify gas leakage faults is becoming increasingly important. The main difficulty to detect gas leakage faults by sound signals lies in that the practical plants are usually very noisy. In this research, neural networks are used as nonlinear filters to eliminate noise and raise the signal noise ratio of the sound signal. Neural networks have a self-learning ability. This ability can be utilized to build a self-adaptive nonlinear filter. By learning from practical samples of noise and adjusting coefficients of connection weights of neural networks, a neural network filter can predict the future sample values of noise based on present and previous sample values. The predicted error between predicted values and practical ones constitutes output of the filter. If the predicted error is zero, then there is no leakage. If the predicted error is greater than a certain value, then there is a leakage fault. Through application to practical pneumatic systems, it is verified that nonlinear filters with neural networks was effective in gas leakage detection from noise background.