Research on Intelligent Suppression for Torsional Vibration of Electric Vehicle Drivetrain

Publication Type:
Issue Date:
Full metadata record
Compared with the conventional fuel-consumption vehicle, electric vehicle does much better in controllability, regarding more accurate controlling of torque or rotation speed and more rapid response. The characteristic of rapid torque response will bring step motor torque output approximately when the drive tip in/out, because there exist torsional elastic damping parts such as tires and drive shaft in the drivetrain, moreover, there exists no flywheel to save the energy and buffer jerking, and then longitudinal vibration of vehicle that more sensitive to human will be introduced. In this thesis, by taking advantage of the excellent controllability of electric vehicle, active damping control of torsional vibration in vehicle drivetrain is researched, motor torque control in active damping for drivetrain torsional vibration is realized based on parameters intelligent estimation and motion state observation for the drivetrain. Improvement of electric vehicle drivability is realized focusing on driver’s tipping in/out and unloading motor torque before shifting. The main content of the thesis is as follows. Based on the drivetrain of electric drive logistics vehicle with two speeds automated manual transmission, multibody dynamic model of electric drivetrain is established, by Fourier transforming the multi-body dynamic model, frequency response function matrix of the system is obtained and frequency response analysis of the electric driving system can be carried on, and then natural frequencies, mode shapes and resonance characteristics of the drivetrain system can be revealed. In order to solve this problem nonlinear particle filter-based intelligent parameter estimation method is proposed in this thesis. As a critical parameter in dynamic control of powertrain, vehicle mass varies continuously in a wide range and large magnitude during operation, it is one of the most significant variables in control law constituting, but vehicle mass is difficult to be measured automatically in real-time. The particle filter is recursive filter based on Monte Carlo algorithm, using the processes of importance sampling and resampling and according to motor torque output and motion states of drivetrain, parameter estimation of vehicle mass can be realized. Besides, this vehicle mass intelligent estimation method is robust and statistical characteristics of disturbances and uncertainties in the powertrain system is unnecessary to be known. In this thesis accelerated adaptive second order super-twisting sliding mode observer (SMO) is proposed, it can remarkably attenuate “chattering”, the inherent drawback of sliding mode (SM) variable structure algorithm, and estimation error convergence is accelerated to a large extent by introducing of the “system damping”. Motion states are also necessary in the control law designing for the drivetrain system, aiming to get the state variables of the drivetrain that are not measured directly, Based on the proposed novel sliding mode observer, torque accumulated in the drive shaft is observed to provide information for selection of appropriate shifting time, so that torsional vibration and jerking in drivetrain caused by sudden releasing of torque accumulated in the drive shaft after shifting is avoided, meanwhile, it also provides state information for the active control algorithm in the following content. Based on the parameter estimation and state observation of the drivetrain, quantum genetics optimization and Linear Quadratic Gaussian joint algorithm is proposed to design the optimal control law to actively damp torsional vibration in drivetrain. According to the multi-feedbacks of the current motion states of drivetrain, optimal motor torque output command is calculated to compensate motion oscillations of drivetrain. The quantum genetic optimization unit utilizes qubits to replace binary encoding and quantum transformation is realized from quantum rotating gates so that the parameter in Linear Quadratic Gaussian to be optimized is searched faster to minimize the fitness function. By optimization of the parameter in Linear Quadratic Gaussian controller, more authentic and objective optimization of the controller performance is realized than the controller with subjective parameter selection from designer. Meanwhile, it will not increase the complexity and computational power consumption of the control law. Rapid prototyping experiment, test rig experiment and real vehicle experiment are carried on focusing on parameter estimation, state observation and torsional vibration damping control to test the performances of the proposed algorithms in this thesis.
Please use this identifier to cite or link to this item: