Research On Ultrasonic Detection Method Of Compressed Gas Leak

Compressed gas is playing an important role in the fields of textile,chemical production,biopharmaceuticals,aerospace,etc.But gas leakage caused by corrosion,aging,and manmade damage causes huge economic losses and resources waste,and threaten personal safety to staff as well.Gas leakage has seriously affected the application of compressed gas.Therefore,it is of great significance to study the detection method of compressed gas leak.At present,the main detection method is to utilize the characteristic that the spectral energy of leak signal and environmental noise have the largest difference at 40 k Hz,using ultrasonic detector to detect.But there exist disadvantages such as single feature selection for judgment and susceptible to noise.At the same time,the way of manual inspection is time-consuming and labor-intensive,and the detection accuracy is limited by human subjective ability.For the detection of some flammable and explosive compressed gases,it will also threaten the personal safety of inspectors.In order to solve the shortcomings of the existing methods,by combing feature extraction and classifier recognition,a new leak detection mode is proposed based on ultrasound characteristics of leak signal.The main method ideas and working contents are as follows:(1)By analyzing the generation principle of ultrasonic signal,and time-frequency domain characteristics of leak signal,the influence of low-frequency noise and the rationality of feature extraction for ultrasonic frequency bands were determined,and the importance of preprocessing was clarified.(2)In order to extract the ultrasonic frequency band of the signal,empirical mode decomposition(EMD)was used for preprocessing.First,the principle and limitations of the EMD algorithm were analyzed.To solve the problem that spectrum aliasing makes low frequency interference and ultrasonic signals mixed to affect the preprocessing procedure,the entropy theory was introduced for optimization.Some common entropy algorithms were tested to determine the ability of entropy to measure the complexity of frequency domain.Considering the requirements of leak detection,EMD algorithm was proposed by introducing entropy into the iterative conditions.Experiments show that the improved EMD can effectively alleviate the problem of spectral aliasing.(3)In view of the wide spectrum and complex distribution of the ultrasonic frequency band,Mel-frequency cepstral coefficients(MFCC)were selected as features.Because of the problem that the Mel transform function is a fit to human ear's hearing mechanism,the filter distribution is difficult to directly apply to leak detection.An improved MFCC was proposed.By increasing the form of the transformation function and modifying the function parameters,the filter distribution was more suitable for extracting the characteristic of the ultrasonic frequency band.At the same time,the MFCC extraction process was optimized to match the leak detection requirements.Experiments prove that the improved MFCC has higher detection accuracy and stronger robustness.Finally,dimension reduction algorithm was introduced to optimize the feature vector and reduce the time consumption for classification detection.(4)Based on the requirements of leak detection,the rationality and advantages of the mode which pairs feature extraction and classifier detection were proved.By testing the performance of different classifiers,select Support Vector Machine(SVM)as the optimal classifier and determine the model parameters.By collecting data through experimental simulation of leakage environment,a data set was constructed to test the performance of the leak detection method.Some relevant results were obtained: Compared with MFCC,the improved MFCC promotes the accuracy rate by more than 7%.At the same time,due to the unbalanced performance of MFCC on training set and test set,it is confirmed that the improved MFCC is more robust.Comparing the performance of the paper's feature extraction method with the feature of ultrasonic detection and speech signal processing on data set,the detection accuracy of the proposed algorithm can reach more than 95%,which is far more than other methods.The time consumption also meets the requirements for realtime performance.The influence of the selection of key parameters such as the number of filters,dimensionality reduction,and target frequency range on the detection method is determined.It provides a reference for the application of detection method in actual leakage scenarios.