外文翻譯--汽車主動懸架系統(tǒng)的神經(jīng)網(wǎng)絡(luò)控制運算法則的研究（譯文）

1、Study on Neural Networks Control Algorithms for Automotive Adaptive Suspension SystemsL.J.Fu, J.G.Cao School ofAutomobile Engineering, Chongqing Institute ofTechnology, Xingsheng Road No.04 Yangjiaping, Chongqing, China

2、400050 E-mail: flj@cqit.edu.cnAbstract-The semi-active suspension, which consists of passive spring and active shock absorber in the light of different road conditions and automobile running conditions, is the most popul

3、ar automotive suspension because active suspension is complicated in structure and passive suspension cannot meet the demands of various road conditions and automobile running conditions. In this paper, a neurofuzzy adap

4、tive control controller via modeling of recurrent neural networks of automotive suspension is presented. The modeling of neural networks has identified automotive suspension dynamic parameters and provided learning signa

5、ls to neurofuzzy adaptive control controller. In order to verify control results, a mini-bus fitted with magnetorheological fluid shock absorber and neurofuzzy control system based on DSP microprocessor has been experime

6、nted with various velocity and road surfaces. The control results have been compared with those of open loop passive suspension system. These results show that neural control algorithm exhibits good performance to reduct

7、ion of mini-bus vibration.I. INTRODUCTIONThe main functions of automotive suspension system areto provide support the weight of automobile, to provide stability and direction control during handling maneuvers and to prov

8、ide effective isolation from road disturbances. These different tasks lead to conflicting design requirements. The semi-active suspension, which consists of passive spring and active shock absorber with controllable damp

9、ing force in the light of different road conditions and automobile running conditions, is the most popular automotive suspension because the active suspension is complicated instructure and conventional passive suspensio

10、n cannot meet the demands of different road conditions and automobile running conditions. Simi-active suspension with variable magnetorheological(MR) fluid shock absorbers has some advantages in reducing automobile vibra

11、tion at relative lowcast and power. So far, there are a number of control methods that have been developed for semi-active suspension, start with skyhook method described by Karnoopp, et al.[l] This method attempts to ma

12、ke the shock absorber exert a force that is proportional to the absolute velocity between sprung masses. Some investigations useC. R. Liao, B. Chen School ofAutomobile Engineering, Chongqing Institute ofTechnology, Xings

13、heng Road No.04 Yangjiaping, Chongqing, China 400050 E-mail: chenbao(cqit.edu.cnlinear suspension model, which is linearized around the operational points, and control algorithm are derived using linear models, such as L

14、QG and robust control [2,3]. These control methods cannot make a full exploitation of semi-active suspension resources because of automotive suspension is inherent non-linear performance. In order to improve performance

15、of nonlinear suspension system, some intelligent control techniques, such as fuzzy logic control, neural networks control and neurofuzzy control, have been recently applied to nonlinear suspension control by researchers

16、[4,5]. In this paper, a neurofuzzy adaptive control controller is applied to control suspension vibration via modeling of recurrent neural networks of automotive suspension and continuously variable MR shock absorbers. T

17、he controller structures design and neurofuzzy control algorithms are presented in section 2. A recurrent neural networks dynamics modeling of suspension are shown respectively in section3. The control system experimenta

18、tions are given in section 4 and some conclusions are finally drawn in section5.HI. NEUROFUZZY ADAPTIVE CONTROL ALGORITHMS FOR AUTOMOTIVE SUSPENSIONSThe neurofuzzy control system presented in this paper, shown in Figure

19、1, is composed of a neurofuzzy network and a recurrent neural network modeling of mini-bus suspension. The neurofuzzy network is defined as adaptive controller, which has function of learning and control. The function of

20、 recurrent neural network is to identify mini-bus suspension model parameters. y(t) and yd(t) are systemactual output and system desire output respectively in Figure1. xl(t) is system error of system actual output betwee

21、nsystem desire output, x2(t) is system error rate of systemactual output between system desire output. xi (t)and x2 (t)are defined as fellows:xI (t) e(t)= y(t)- Yd (t) (1)X2 (t)= e(t)= e(t + 1)- e(t) (2)0-7803-9422-4/05/

22、$20.00 C2005 IEEE 1795where w'I , , w° are weight of the recurrent neuralnetwork, Xj(t)is output of neuron with local feedbackloop neuron in the hidden layer, p , q are input neuronnumber and feedback neuron num

23、ber respectively. The activation function for both input neurons and output neurons is linear function, while the activation for neurons in the hidden layer is sigmoid function.he objective function E(t) can be defmed in

24、 the terms ofthe error signal e(t) as:E(t) = _[y(t) - .y(t)]2 = 1e2(t) (7) 2 2That is, E(t) is the instantaneous value of the errorenergy. The step-by-step adjustments to the synaptic weights of neuron are continued unti

25、l the system reach steady state,i.e. the synaptic weights are essentially stabilized.Differentiating E(t) with respect to weight vector wyields. aE(t) __ 8 =-e(t) 0Y() (8)From expression (1), (2) and (3), differentiating

26、 A(t)0 D I with respect to the weight vector w1 w,- , w,-Y respectivelyyields.aS(t) = x (t)As(t) wo ax1Q)-( W aXI (t) aWj J aWjFrom (4), (5) and (6), analyzingvalue of synaptic weight is determined byw(t + 1) = w(t)+ q *

27、 e(t) 89(t) (12)where q the leaning-rate parameter, A detailedconvergence analysis of the recurrent training algorithm is rather complicated to acquire the leaning-rate parameter value. According to expression (13), the

28、weight vector wfor recurrent neural network can be adjusted. We establish athe Lyapunov function as follows V(t) = 1/2 * e2 (t),whose change value AV(t) can be determined after somet iterations, in the sense that(13)We h

29、ave noticed that the error signal e(t) after some titerations can be expressed as follows from expression (13) and(14),ae(t) ao(t) ae(t) ae(t) -- ~,Aw=-qe(t)~ =77e(t) ~,the aw “O“w aw 'O“w Lyapunov function increment

30、 can determined after some titerations as follows(14) Mtt)= --q- &(t) +v2.e(t)-- =-V(t)where^(t) 2 2jt 1> 6(t) 2 A = 1 0'()lp q 2-5l 0(t)ll >] 2 ql[2-77'] >O ° 2 2 0w(9) ?7 maxa~(t) 29 if <q&

31、#39;f< 2, then AV(t)< O, wax1(t)D and aWjx1 (t)uxiyields respectively recurrent formulas. ax1(t) a-f[S (t)] FX. x(tt 1) 1ax1 (O) =,WjD=axi (t)aNiaf S(t)] [ +w a t- i)&4 L aN'iax1(o) (11) avn =0Having comput

32、ed the synaptic adjustment, the updatednamely the recurrent training algorithm is convergent.IV. ROAD TEST AND RESULTS ANALYSESTo make a demonstration the validity of neural control algorithm proposed in the paper, an ex

33、perimental mini-bus suspension with MR fluid shock absorber has been manufactured in China. The mini-bus adaptive suspension system consists of a DSP microprocessor, 8 accelerationsensors, 4 MR fluid shock absorbers, and

34、 1 controllableelectric current power with input voltage 12V. The DSP microprocessor receives suspension vibration signal input from accelerometers mounted respectively sprung mass and un-sprung mass. In accordance with

35、vibration signal and control scheme in this paper, the DSP microprocessor adjusts damping of adaptive suspension by application control signal to the controllable electric current power connected to electromagnetic coil

36、in MR fluid shock absorbers. Magnetic field produced by the electromagnetic coil in MR fluid shock absorbers can d vary damping forcein both compression and rebound by adjustment of flow1797I I,&V(t) = 1 2 (t +1)-e 2