Dynamic Simulation And Deep Transfer Diagnosis Method Of Connecting Rod With Crack

Reciprocating compressor is an important power equipment in the production process of large petrochemical industry.As the main part of transmission mechanism,connecting rod plays an important role in transmitting the movement of crankshaft and piston,and its safety state directly affects the production and operation of the industry.At the same time,the connecting rod is easy to crack due to long-term alternating load.The location of the on-site sensor is far away from the vibration source,and the weak signal when the connecting rod cracks is easy to be covered by other excitation sources,which makes it difficult to accurately obtain its label data and extract features using traditional machine learning.Moreover,the operation state of the equipment is complex,and the original depth model cannot adapt to the new data samples after the working conditions change.Based on this,taking the connecting rod of reciprocating compressor as the research object,this thesis obtains the labeled sample data through static analysis,rigid-flexible coupling dynamics simulation and crack fault simulation experiment.Then,this thesis puts forward a fault diagnosis method of cracked connecting rod of reciprocating compressor based on deep transfer learning,which solves the problem that the accuracy of model diagnosis decreases due to the distribution difference of sample data across scenes.The main work is:(1)In order to get the crack initiation position of the connecting rod,the static model of the connecting rod of reciprocating compressor is established.Through the analysis of the kinematics and dynamics of the transmission mechanism of the reciprocating compressor,taking the reciprocating compressor test-bed as the object,the static model of the connecting rod is established by using ANSYS Workbench.Referring to the experiment,the stress of the connecting rod is calculated and defined as the load and constraint at both ends.The maximum stress and deformation of the connecting rod are analyzed to determine the crack position of the connecting rod.(2)Aiming at the problem of lack of tag data,this thesis obtains the vibration signals of experiment and simulation through simulation in test-bed and dynamic simulation,and analyzes the similarities between them.By customizing the connecting rod,this thesis completes the experimental simulation of different cracks and working conditions,and analyzes the vibration response law.Through the three-dimensional solid model of the cracked connecting rod established by Solid Works and the rigid flexible coupling models which are established by Adams and ANSYS,the kinematic and dynamic characteristic parameters and fault vibration signals of the connecting rod and the sample data are obtained.By comparing the simulation signal with the experimental signal,and further quantifying the time interval between the two adjacent shocks,it is proved that the simulation is feasible to migrate to the experiment.(3)Aiming at the problem that the distribution of data samples changes under variable working conditions,this thesis proposes method named MTF-Re CORAL for fault diagnosis under experimental cross working conditions and simulation to experimental scenarios.Taking Markov transition field(MTF)as data preprocessing means,CORAL is introduced into the deep learning model to reduce the distribution difference of sample data.Compared with the traditional time-frequency domain image processing method(STFT),the diagnosis accuracy of the proposed method is improved by 8%.The experimental data of different working conditions are used as samples of source domain and target domain.Compared with Deep CORAL,the accuracy of this method is improved by 11% and 2% under variable load and variable speed,which proves the robustness of this method.The simulation and experimental signals are defined as source domain and target domain respectively,the diagnostic accuracy is improved by15% and 30% respectively compared with Deep CORAL and TCA,which realizes the diagnosis across scene from simulation to experimental,and proves that it is feasible.