Research On Load Identification Based On Parameter Optimized BP Neural Network Of Double Span Rotor System

The large rotating machinery plays a crucial role in the national infrastructure construction projects,such as mine construction,railway construction and port development.And the rotor system is the most important part of the large rotating machinery.With rotating machinery developing towards high speed,overload and intelligent,the rotor system must develop from single span and single disk to long,thin,more spans and disks,so the rotor system must be more security and stable.During the operation of rotating machinery,there will always be loose,incorrect,touching,unbalanced and other faults,which correspond to different vibration information of the rotor system.And the rotor system shows different operation states under different loads,which also correspond to different vibration information of the rotor system.This will affect the working state of the whole machine and possibly safety accidents will occur.Therefore,it is of great significance to study the vibration characteristics of double span rotor system under different types and sizes of loads.Firstly,the double span rotor system is simplified according to the rotor dynamics theory,its mathematical model is established,and the force analysis iscarried out based on the model.Then the differential equation of vibration of the system is solved with Lagrange equation.In the mathematical software MATLAB/simulink environment,a simulation model of double span rotor system is build,different sizes and types of loads are applied to the double span rotor system by signal loader,and the vibration displacement response signals of double span rotor system are obtained.By analyzing the time domain signal,the characteristics of the excitation load are consistent with the variation characteristics of the vibration displacement signal.Secondly,the method of BP neural network with parameter optimization is proposed for the study on the load identification of double span rotor system.When designing the standard BP neural network structure,the four parameters of transfer function,training function,learning function and hidden layer node are studied deeply,and the best network structure is finalized.The weight and threshold of the network are not updated in a timely manner so that the performance of the whole network is influenced when the network is trained.Therefore,the method of optimizing the network's weights and thresholds by genetic algorithm and particle swarm algorithm is proposed.The results show that the load identification error of BP neural network optimized with particle swarm algorithm is minimal by using these three methods to carry out load identification research on the double span rotor system.Finally,the double span rotor test bed system is build.According to the experimental scheme,different sizes and types of loads are applied to the doublespan rotor system and the vibration response signals of the system are obtained.Using these three methods to carry out load identification research,the experimental results show the correctness of the simulation results and verify the feasibility and effectiveness of the BP neural network with parameter optimization in the research on load identification of double span rotor system.