Detection of small gas leaks based on neural networks and D-S evidential theory using ultrasonics

This paper presents a method for solving the problem of recognising the leakage state of a measured object by means of a gas leak detection system, using an ultrasonic method based on data fusion and neural networks. The neural network is trained using cross-correlation information from a probe as the prior probability, combined with the Dempster-Shafer (D-S) evidential reasoning method, and then applied in the gas leak ultrasonic detection system. Experimental results show that recognition based on this combination is significantly better than with a single sensor. Consequently, the validity and correctness of this method have been verified.