Digital Twin-based Gearbox Fault Diagnosis Methodology Research

As an important part of the mechanical transmission system,planetary gearboxes are widely used in major mechanical equipment because of their large transmission ratio,small size,smooth transmission and large load capacity.However,as planetary gearboxes usually operate in complex environments such as high altitude,high speed and high temperature,resulting in their internal structures being highly susceptible to failure.Digital twin technology provides a new idea to solve the lack of sufficient fault samples for training deep neural networks in practical engineering scenarios,but the key technologies such as the construction and evaluation methods of high-fidelity digital twin models,the lightweight strategy of digital twin models,and the evaluation of virtual-real consistency of digital twin are the urgent technical bottlenecks to be broken in this field.This paper focuses on the fault diagnosis method of planetary gearbox based on digital twin,and the specific research contents are as follows.1.The construction of high-fidelity digital twin model for complex equipment and the model evaluation method are studied.First,a digital twin initial model with a 1:1ratio to the research object is constructed in UG(Unigraphics NX)with the gearbox test bench as the research object;then,digital twin model evaluation indexes are established and the constructed initial model is evaluated by quantitative methods,and the results show that the accuracy and intuitiveness are satisfied,but the validity is not sufficient;finally,in order to improve the gearbox test bench digital twin model Finally,in order to improve the effectiveness of the digital twin model of gearbox test bench,a light-weighting method of digital twin model based on point cloud streamlining is proposed.2.A lightweight digital twin method for planetary gearbox fault diagnosis is proposed.First,the initial digital twin model of the gearbox test bench is converted into point cloud data and simplified by the streamlining algorithm,and then the lightened digital twin model is obtained by repackaging with 3D reconstruction technology,and the validity of the lightened model is 86.3%,which meets the validity requirement.Finally,the streamlined point cloud data are input to the Point Net++ model to complete fault diagnosis,and the experiments show that the model is superior in the accuracy of planetary gearbox fault diagnosis.3.The interaction and fusion between physical space and lightweight digital twin data and experimental research were carried out.Firstly,the digital twin model of the lightweight gearbox test bench was imported into ADMAS software for dynamics simulation to obtain simulation data and provide a virtual data source;then,the planetary gearbox fault diagnosis test bench was built and the signal acquisition system was used to obtain real measurement data;finally,the comparison study of simulation data and real measurement data under different health states,the frequency of theoretical calculation and the frequency of fault characteristics,the error was Finally,the error is less than 2%,which proves the initial consistency of the virtual and real data.4.A cross-domain data fusion and real-time data-driven digital twin method is proposed for planetary gearbox fault diagnosis.First,the simulation training convolutional neural network and the real measurement data are used as the test set to achieve fault diagnosis,and then,a new training data set is constructed by mixing the simulation data and the real measurement data in the same health state,and the results show that the diagnosis accuracy gradually improves as the percentage of the real measurement data in the training data set increases.Then,a study on fault diagnosis by training convolutional neural networks with simulation data generated by real-time sensor data-driven digital twins is carried out,and it is shown that the fault diagnosis accuracy of the method will be greatly improved.5.A digital twin virtual-real consistency discrimination method for intelligent identification decision of service health state is summarized.The method is based on health state recognition accuracy calculation,and provides a discriminative method for digital twin consistency by constructing a four-dimensional structure evaluation system of initial consistency,value consistency,trend consistency,and engineering consistency,which provides a basis for achieving evolutionary iteration of digital twin models.