【作者】

Guofeng Zhao

【摘要】

该论文已在芬兰赫尔辛基举行的第28届CIMAC大会上发表。论文的版权归CIMAC所有。Diesel is a complex nonlinear device. This is a challenge to the control system.MAP is widely used to achieve proper performance. MAP calibration is time-consuming and even special skill needing. Intelligent control algorithm has been proposed to improve this situation. Intelligent control theory has been discussed many times so the focus of this paper is on engineering realization. Application target is diesel speed control. Three parts is included in this research: principle design and simulation validation, controller development and software framework, and the last part bench test and result analysis. A self-adaptive diesel speed governor strategy based on BP neural network and PID is designed. BP neural network is used to modify the speed-loop PID parameters, while the rack-loop is controlled by traditional PID. To validate the feasibility of the new control strategy a MATLAB/SIMULINK mean value model is built. A physics model of proportion electro-magnet actuator and an experiment data based oil pump mode are included in the model. Loading experiment confirms the self-adaptive competence of designed BP-PID control algorithm. Neural network can be used to learn any nonlinear system, but the fact is that neural trend to learn the non-existed correlation during the transient process of diesel engine, it is also called over-learning. To avoid this phenomenon, learning method with supervision is developed. Base on the characteristic of neural network calculation requirement analysis is discussed in detail, especially the demand for MCU. A neural network controller is designed based on STM32F103 micro control unit. To meet real time requirement of diesel control system software framework of key function should be specially built. Foreground and background, time sharing concept of OS has been adopted. Finally, the last part is performance contrast experiment. Bench experiment is carried out on a D6114 diesel to verify the governing performance of the system. Start experiment, steady state experiment and load sudden change experiment are conducted. Attention is paid on starting overshot, setting time and overshot during load sudden change, system robustness in the whole operating range. The experiment results are compared with that of traditional PID. The data analysis suggests that BP-PID control strategy may have similar performance with tradition PID but the improvement of robustness and idle speed performance is obvious. Consideration can be given to high speed, low speed, high temperature and low temperature. Speed fluctuation in idling speed is improved. The most important the system stabilize is reliable enough. STM32F103 MCU is capable of running a neural network control system.

【会议名称】

第28届CIMAC会议

【会议地点】

芬兰 赫尔辛基

【下载次数】

2

返回