Research On Condition Monitoring And Fault Diagnosis For EAST Molecular Pump Based On Machine Learning And Compressed Sensing

EAST superconducting tokamak device is designed to provide fundamental research for building safe and stable fusion reactors in the future.High vacuum environment is essential for plasma physics experiments,the molecular pump is one of the main equipment for EAST to obtain high vacuum environment.As a kind of precision machinery,molecular pump usually needs to work at a high speed to ensure the efficiency and accuracy.The molecular pump is conducive to improve the vacuum degree and other key performance indexes of EAST and ensure the accuracy and reliability of the experimental results.The molecular pump of EAST will be failed due to vacuum leakage,impurities and other reasons.If the molecular pump is not maintained in time,it may eventually be damaged.The molecular pump fault will lead to the interruption of experiment and safety accident.To avoid the occurrence of the above problems,it is critical to study the condition monitoring and fault diagnosis of EAST molecular pump.This thesis is mainly divided into the following parts:Aiming at the problem of overfitting and low accuracy in the unbalanced dataset of EAST molecular pump fault detection,this thesis proposed a new method for molecular pump fault detection.This approach combined with time domain analysis,frequency domain analysis and cost-sensitive Light GBM algorithm.Firstly,we collected the normal state and fault state vibration signal of EAST molecular pump,then extracted the time domain and frequency domain features of the original signal.Next,we established a cost-sensitive Light GBM fault detection framework by optimizing the misclassification cost function.Finally,the features are used as the input of cost-sensitive Light GBM algorithm to realize the molecular pump fault detection.In addition,the EAST molecular pump fault detection experimental system was built up to verify the proposed method,the result show that the accuracy of the proposed approach is 99.4%,the proposed method also can consistently outperform traditional Light GBM,KNN and Logistic Regression algorithm in terms of false alarm rate and miss detection rate.This method can effectively solve the problem of overfitting and low accuracy and realized the fault detection of molecular pump with high accuracy.To deal with the problem of the vibration signal is easy to be affected by noise and the characteristic parameters are difficult to select,a fault diagnosis approach based on compressed sensing sparse theory and a convolution neural network fusion 1D CNN and 2D CNN is proposed.Firstly,the public data set is used to verify the fusion convolution neural network,the results show that fusion convolution neural network has more advantages than traditional methods in feature extraction.The self-made data set is used to verify the method based on compressed sensing sparse theory and fusion convolution neural network.Specifically,the original signal is processed by the discrete cosine transform and the value of discrete cosine transform coefficient less than 2% of peak-to-peak value is changed to zero.The signal after denoising is obtained by the inverse discrete cosine transform.Secondly,according to the compressed sensing sparse theory,the signal after noise reduction is trained by K singular value decomposition algorithm to obtain the corresponding over-complete dictionary.Then,the signal is decomposed on the corresponding over-complete dictionary to obtain the sparse representation coefficient.The sparse representation coefficient is used as the input of fusion convolution neural network to realize the molecular pump fault diagnosis.The vacuum leakage fault experiment of molecular pump is designed to verify the effectiveness of the proposed method,the results show that the accuracy is 99.75%,which is higher than the traditional method.To address the problem of application of EAST molecular pump condition monitoring,a vibration signal acquisition and analysis system of EAST molecular pump is installed.The signal acquisition and condition monitoring software of EAST molecular pump is designed by Lab VIEW and Python.The software functions include: data acquisition,real-time condition monitoring,signal analysis,data storage and alarm.The software runs under the WINDOWS system,the design of the software is easy to understand and solve the application problem of EAST molecular pump condition monitoring.