Fault Diagnosis Method Based On Convolutional Neural Network And Multi-source Information Fusion

With the advancement of manufacturing technology,the structure of precision equipment become larger and more and more complex,and the correlation between the structure of the transmission system is growing.Catastrophic accidents and economic losses caused by mechanical faults happens sometimes.Fault detection of unmanned aerial vehicle(UAV)transmission system is a hot topic in the research field at present.The comprehensive detection and diagnosis of multi-sensor networks can avoid the limitations of a single information source.However,data analysis of multiple sensing parameters leads to the inability of traditional analytical models to meet the needs of accurate troubleshooting analysis and fast online condition monitoring.Convolutional Neural Network(CNN)uses powerful representation capabilities to efficiently and accurately mine equipment health from data,and to get rid of the dependence of traditional fault diagnosis methods on signal processing technology and expert experience.It has good analytical capabilities for multi-feature input parameters.Compared with the fully connected deep network,CNN has fewer training parameters,higher training efficiency and higher generalization ability.Therefore,based on CNN,this paper studies the fault diagnosis of UAV transmission system.At present,most fault diagnosis methods use time-frequency transform to convert the vibration signal into a time-frequency map for processing,artificially participate in feature extraction.And if the signal don't have a periodic law,the method is invalid.the method fails.Firstly this paper improves the fault diagnosis method based on CNN.The data processing of the time domain signal to the image is completed in a simple processing manner,and the adaptive extraction of the fault feature and fault diagnosis are completed by optimizing the CNN network structure,which MlpConv and global average pooling layer are used to replace the typical full connection layer.The accuracy of the network structure after optimization is verified to be 99.90%.The recognition accuracy is high and the robustness is good.Secondly,based on the requirements for the real-time online fault detection and the multi-attribute detection of the UAV transmission system,the one-dimensional vibration signal is input into the CNN for training in this paper,and no feature extraction is needed.A generic solution is built that can be used for multiple fault data.The method can quickly and accurately identify the type and degree of faults,eliminates the need for feature extraction algorithms,and is more adaptive to engineering use in terms of speed and hardware requirements.Then the fault diagnosis method based on multi-sensor information fusion and CNN is studied.Solving the problem that a single source's information reflecting the incompleteness of the device's operating status.In this paper,by optimizing the data-level information fusion method and taking advantages of HMM in the dynamic modeling of non-stationary time series,proposes a CNN-HMM fault diagnosis method to realize fault diagnosis of multi-sensor information fusion.Compared with the traditional method,the accuracy and robustness of the method were improved.At the end of this paper,a fault diagnosis software system is developed based on C# programming language.The diagnosis algorithm proposed in this paper is integrated.It is designed the data collection and analysis function to meet the engineering requirements,and has good man-machine interaction,can meet the algorithm integration and sharing.