Research On The Abnormal Sound Recognition Method Of Aircraft Engine

Abnormal sound event detection is a new research hotpot in the field of voice recognition and has a wide range of applications.Aeroengine as the heart of the aircraft,the reliability of its performance is an important guarantee for aircraft performance and safe flight.Usually,when an aircraft engine malfunctions,it can be attributed to an abnormal sound produced by the engine.If the engine running state can be detected by recognizing an abnormal sound of the engine,not only can the cost of operation and maintenance be reduced,but also the number of accidents occur.However,there are few researches on abnormal sound detection of aircraft engines at home and abroad.So we put forward the construction method of aircraft engine abnormal sound recognition system and which combines the new theories and new applications of signal processing,which can effectively promote the intelligent development of aviation safety technology.The main research contents are as follows:(1)We establish an aircraft engine sound library,and preprocess the sound signal in the sound library,then extract the MFCC or GFCC parameters characterizing the sound features,and last form the sound feature vector of the aircraft engine.(2)The abnormal sound recognition method of aircraft engine based on GMM-UBM is researched.Firstly,training the UBM model use the feature parameters obtained before,then getting GMM-UBM model use the MAP adaptive algorithm,and detecting the sound of engine use GMM-UBM model.Finally,through the MATLAB simulation experiments,the effects of characteristic parameters,degrees of GMM mixture,dimensions of feature parameters,and feature orders on the abnormal sound recognition system are verified.The experimental results show that the aircraft engine abnormal sound recognition system based on GMM-UBM works best when the GMM mixture degree is 128 and the engine sound features are 12 dimensional MFCC parameters.(3)The application of deep learning method in aircraft engine abnormal sound recognition is researched.Through the analysis and study of the basic knowledge of deep learning,we present the abnormal engine sound recognition systems based on MFCCCNN,MFCC-LSTM-RNN and MFCC-FNN.The system is set up under the TensorFlow framework,and use the python language programming operation,the effects of the time of test data,the training times of training data and the learning rate of neural network on the neural network system are verified.The experimental results show that when the test time is half of the training time,the Epoch is 30 and the learning rate of neural network is 0.001,the neural network system constructed in this subject has an average recognition rate of more than 95% and has a very good roubustness.In summary,the abnormal sound recognition system of aircraft engine built in this subject can get a good recognition effect and provides a reference for the application of sound event monitoring technology in the aviation fields.