| Nowadays permanent magnet(PM)machines are widely used in different industrial applications.With the pursuit of high-power density,high reliability and low cost,high performance thermal management has become a hot research area.The online thermal model is a key factor in realizing thermal management,and it has attracted more and more attention.Considering all these aspects,the establishment,calculation,and application related to the PM machine online thermal model are explored in this thesis,based on a 10 k W double six phase open-end PM machine.The main content is as follows.(1)The working point-based PM machine online loss model is researched.Machine losses are the direct reason for the temperature rise.The speed and torque are the typical parameters to describe the working point of the machine.To predict machine thermal condition under different operating point,the relationship between machine losses and operating points need to be established.Currently,the electrical machine is usually driven through the voltage source inverter and Pulse Width Modulation(PWM)technique.However,conventional simplified models ignore the influences of PWM and only take the fundament component into consideration.To solve this problem,the online loss model based on the working point of the machine and considers the influences of PWM highfrequency harmonics is researched.First,based on the machine operating point,equivalent circuit model,and control strategy,the analytical expression of the winding current is obtained through the Fourier transform,and the high-frequency current harmonics are obtained.Then,the AC copper losses which are composed of the skin effect and proximity effect are calculated through the squaredfiled-derivation and Ferreira model,respectively.The relationship between frequency,AC copper loss,and temperature is also clarified.After that,the connection between harmonic current and PM loss is studied through magnetomotive force,and a simplified PM loss model is proposed.Last,the impact of fundamental current and PWM high-frequency components on core losses is analyzed,and a simplified online core loss model which describes the relationship between core losses,current,and speed is proposed.(2)The PM machine online thermal model is researched.Based on the machine structure and the principle of lumped parameter thermal network(LPTN),an online LPTN model which contains unknown thermal parameters is established through the combination of prior knowledge and datadriven idea.The corresponding unknown parameters are obtained through the genetic algorithm and sequential quadratic programming(SQP)optimization method.The experimental verification demonstrates that the average error of the obtained model is 1.2℃ and the average relative error is2.2%.To further realize the temperature prediction in the circumferential direction,a typical slotbased multi-slots model is proposed.The one-slot LPTN model is extended to typical slots,the structure and major parameters of the model corresponding to different slots are identical,and the only variation lies in the external heat exchange-related parameters,namely,housing heat capacity,housing-environment heat transfer resistance,and stator-housing equivalent contact air gap length.The multi-slots model is also verified based on the corresponding experimental data.In addition,based on the research results,a kind of experimental data-driven online thermal model is proposed.The requirement for machine structural and material information is eliminated,which makes the machine temperature prediction can be achieved more conveniently and efficiently.The proposed method is also compared with a data-driven Long Short Term Memory neural network,and the results demonstrate that the proposed method has a lower calculation burden and stronger generalization ability.(3)The revision technique of the online LPTN model is investigated.Due to the aging of insulation,the thermal performance of the machine in their different life stages are not exactly the same.Once the thermal characteristics of the material change,the thermal performance of the machine also varies,and the prediction error of the established online LPTN model may become significant.To solve this problem,an online adaptive revision method based on the SQP algorithm is proposed.And a fast calculation method of the Jacobi matrix of the prediction error about the thermal parameters is proposed to improve the calculation efficiency,which makes the gradient descent principle-based model parameter optimization can be realized efficiently online.The comparison with the conventional method,which calculates the Jacobian matrix through the numerical approach,proves that the proposed method reduces the computation time by 76.7%,and the average relative error is reduced from 10.8% to 2.3%.The Extended Kalman Filter algorithm has high efficiency,however,the model parameters obtained by the proposed method are more accurate,and the average relative error is reduced from 8.8% to 1.1%.The experimental data-driven approach is used to obtain some thermal parameters of the online LPTN model,thus the model prediction error may also be large under new operating conditions that differ significantly from the experiments.To solve this problem,a model updating method based on the idea of linear parameter varying model is proposed to extend the range of model applicable working conditions.First,a numerical model is obtained for the new operating condition based on the corresponding measured data.Then,the numerical model and the original model are transformed into a coherent global model.The experimental results show that the obtained global model can achieve decent performance under both new and original working conditions.Thus,the applicable working point range of the model is extended.(4)The application of the established online LPTN model is researched.The initial temperature,working time,thermal condition,and operating point are related through the established model.This enables the real-time analysis and calculation of the machine capability,such as the available operating point range and remaining operating time.For frequent start-stop applications,the temperature of different parts at the initial time may vary greatly.Since the initial temperature information of some parts cannot be obtained through sensors,the temperature prediction cannot be processed directly through the online LPTN model.To solve this problem,a kind of method based on the Kalman filter algorithm and limited measured data is proposed to observe the full machine temperature,thus the temperature prediction can be processed without complete initial temperature information.To reduce the calculation burden under non-constant operating conditions,an equivalent temperature operating point calculation method is proposed.Based on limited measured temperature data,the calculated equivalent working point can be used to simplify the long-time temperature prediction under variable operating conditions.Finally,based on the principle of model predictive control,a control method that dynamically regulates machine output according to the instantaneous thermal conditions is proposed.The performance of the proposed method is assessed based on standard operating conditions,and the results prove that the proposed method can improve the dynamic performance of the machine without spoiling thermal restrictions. |
PDF Full Text Request