A DNN for small leakage detection of positive pressure gas pipelines in the semiconductor manufacturing

To solve the small leakage detection of positive pressure gas pipelines in the semiconductor manufacturing, a deep neural network (DNN) was proposed based on sound signal. For dealing with sound signals as suitable inputs, a pre-process model was designed for de-noising and features extraction. And a DNN with four layers was built for training and learning the sound signals, to precisely and quickly detect the sound of small leakage of pipelines. By the simulated results, they show the presented DNN gained a good percentage of detection than normal neural network (NN). And the learning rate and numbers of neurons in each layer are significant parameters for the model. It is so important that our method provides a novel way for any small leakage detection of positive pressure gas pipelines by involving sound signals.