Research On Fault Diagnosis Method Based On Multi-Sensors Fusion

Rolling bearings are a key component to support mechanical rotation.Once a fault occurs,it will directly affect the smooth operation of the entire system,and may even cause huge economic losses and safety accidents.The traditional mechanical fault repair method is scheduled to be overhauled,which is characterized by time-consuming and laborious costs.Currently,the state of repairing mechanical equipment based on the state of the components is the mainstream method for large-scale equipment maintenance in various countries.In recent years,with the establishment of China's heavy truck ground-to-vehicle safety monitoring and early warning system(4T)and the continuous development of big data and deep learning technology,data-driven fault diagnosis technology play an important role in repairing heavy trucks.The 4T system consists of four subsystems:truck rolling bearing vibration diagnosis system(TADS),truck infrared shaft temperature detection system(THDS),truck running status ground monitoring system(TPDS),and truck running fault image detection system(TFDS).The fault diagnosis model of multi-sensors characteristics is established by excavating the operating state data of key components of heavy trucks on 4T.After experimental analysis,this method can achieve accurate fault diagnosis of rolling bearings.The specific work of this paper is as follows:(1)Excavating the characteristics of temperature,pressure,offset and time domain signal of THDS,TPDS and TADS as the input of the model,and using feature preprocessing techniques such as missing value filling,feature deduplication and outlier correction to solve the missing data and noisy data.In this paper,GBDT is used to sort the importance of features,and combined with XGBoost model,the GD-XGBoost classification model is proposed to diagnose the rolling bearings.(2)Excavating the vibration signal of TADS,since the vibration signal of the rolling bearing has non-stationary and nonlinear characteristics,the Hilbert-Huang transform is used to obtain the Hilbert marginal spectrum of the signal as the input of the model.In this paper,the DenseNet network structure is improved,and the parameters of the convolution and pooling layers can be changed to process one-dimensional sequences.The Inception module is introduced to expand the network width,and the convolution kernels of different scales are used to extract features of different scales,and 1D-DenseNet neural network structure is proposed to diagnose the rolling bearing vibration signals.(3)For the uncertainty of the fault diagnosis of rolling bearings using single source model,this paper uses D-S evidence theory to integrate the decision layer results of the above two single source fault diagnosis models,which further improves the diagnosis accuracy and immunity of rolling bearing fault diagnosis model.Finally,the research is carried out systematically as part of the railway truck state repair diagnosis decision system.