鏈條式抽油機(jī)的設(shè)計

鏈條式抽油機(jī)的設(shè)計,鏈條式抽油機(jī)的設(shè)計,鏈條,抽油機(jī),設(shè)計 6th World Congresses of Structural and Multidisciplinary Optimization Rio de Janeiro, 30 May - 03 June 2005, Brazil Optimisation of a CVT-Chain Lutz Neumann, Heinz Ulbrich Department of Mechanical Engineering, Technical University of Munich Boltzmannstrasse 15, 85478 Garching, Germany Email: neumannamm.mw.tum.de, ulbrichamm.mw.tum.de 1. Abstract In order to account for current and future requirements regarding the further improvement of the dynam- ical behaviour of a chain CVT (Continuous Variable Transmission) optimisation methods are substantial. The primary focus lies on the chain, which is the central part of the gear. In this work a concept to deal with this challenging task is introduced, leading to an optimisation tool for CVTs. First of all, a de- tailed dynamical model of the CVT is necessary to estimate the performance of the gear using numerical simulation techniques. The optimisation problem arising from the aim to optimise CVTs is identi ed in the next step. This includes the formulation of a target function quantifying the property which has to be optimised. For its evaluation simulated data is used. The optimisation problem is analysed in order to identify the requirements for a suitable optimisation algorithm. Implicit ltering will turn out to be the method of choice. For the implementation of complicated optimisation goals, a class of target functions is introduced. Therewith a tool for optimising CVTs is developed, which basically consists of the numerical simulation, the optimisation algorithm and the class of target functions. Its capability is shown by reducing the noise emission of a CVT which is achieved by obtaining optimal values for certain geometry parameters of the chain. 2. Keywords: CVT, optimisation, multibody dynamics, multibody simulation 3. Introduction Continuously variable transmissions are a well-established alternative to common gears such as manual or conventional stepped automatic transmissions. They allow to drive in the most e cient operating range of the engine, which can enhance both performance and fuel economy. As illustrated in Fig. 1, a continuous variable chain drive basically consists of the chain itself and PSfrag replacements plate rocker pin rocker pin chain Figure 1: CVT-drive (left), rocker pin chain (right) two pairs of cone discs, each with a xed and an axial moveable sheave. The pulleys are coupled by the chain, which is composed of rocker pins connected by plates (Fig. 1). The power is transmitted by frictional forces between the discs and the ends of the rocker pins. Hydraulic actuators apply forces onto the moveable sheaves. Thus the radius of the chain in the pulleys and as a consequence the transmission ratio of the gear can be adjusted and changed continuously. 1 Nowadays CVTs have to satisfy high requirements to compete with common gears and therefore an improvement of certain dynamic properties of a CVT is substantial. An important aspect is a preferably small noise emission of the gear which can be achieved by a reduction of the force amplitudes in the bearings of the pulleys in a given frequency domain. Another goal is to minimise the maximum tensile forces in the chain links to increase the lifetime of the chain. Furthermore a higher e ciency of the gear is desired. These improvements are realised by calculating optimal parameters of the chain, which are for example the length and the sti ness of the links or the geometry of the rocker pin joints. The model of the gear, which is used for the numerical simulation, is described in chapter 4. Next the optimisation problem is stated and analysed to nd an appropriate algorithm for its solution. Section 6 covers a description of the optimisation process and its practical implementation. Finally, by solving the starting problem to minimise the noise emission of the CVT, the high potential of the introduced optimisation environment is disscussed. 4. Simulation Model of the CVT A detailed dynamical model of the gear that covers all relevant e ects of the CVT is the basis for the optimisation of the gear. The CVT is modelled as a multibody system which is shown in Fig. 2. It contains three essential components, the pulley sets, the chain and the contact model, describing the connection between chain and the pulleys. Each of the pulley sets has two degrees of freedom: one PSfrag replacements Pulley A Pulley B ! M2 FP;1 FP;2 Figure 2: CVT-drive: Simulation model rotational degree of freedom around the pulley axis and one translational degree of freedom of its movable sheave. Here the sheaves are assumed to be rigid, models which take into account elastic deformations are described in Sedlmayr 2. Pulley A the driving pulley is kinematically excited by the angular velocity ! whereas an external torque M2 acts on the driven pulley B. The transmission ratio is adjusted by the forces FP;1 and FP;2 which act on the movable sheaves. In this work, a planar model of the chain is su cient. In order to consider the discrete structure of the chain which causes the polygonal excitation, every single link is regarded. Each link consists of a massless spring damper element, representing the link plates and connecting two neighbouring joints (sti ness c, damping coe cient d). A mass m is located in each joint and is composed of the mass of the rocker pin pair as well as the half mass of each of the adjacent link plates. The last component of the simulation model represents the frictional contact between the ends of the rocker pins and the sheaves. Neglecting its own dynamics, the pair of rocker pins can be modeled as one single, massless spring acting exclusively perpendicular to the model plane. Figure 3 shows the model of the bolt and the forces acting in the contact plane. For the derivation of the contact forces it is necessary to quantify the spring force FB of the bolt. It depends on the length lB and sti ness cB as 2 PSfrag replacements xB yB zB cB FB FB r FRr FRr FRtFRt FN FN # Figure 3: Model of a bolt well as on the local distance of the surfaces s of the sheaves (Fig. 4): FB = c B (lB s ) s lB 0 s lB : (1) The static equilibrium of forces perpendicular to the model plane provides a conditional equation for the normal force. Taking into account Eq. (1) it depends on the minimal coordinates of the according chain link and the contact force FRr in radial direction: FN = FRr tan # + FBcos# = FRr tan # + cB (lB s )cos# (2) To determine the remaining frictional forces as a function of the normal force, a continuous approximation of Coulombs friction law (Fig. 4) is used where _g is the vector of the relative velocity in the contact: F R = FN _gj_gj ; = 0 (1 e j_gj_gh ) (3) A detailed exposition of the model can be found in Srnik 1 and Sedlmayr 2. PSfrag replacements s s Coulomb continuous approximation j_gjlB FB _gh 0 Figure 4: Bolt force and friction characteristic Important for the further discussion of the optimisation task are the properties of the resulting equations of motion and the numerical simulation. As mentioned above, the model includes contacts with friction that may open and close during the simulation. This results in nonlinear equations of motion with a discontinuous right hand side in the form M q = h(q; _q; t): (4) Furthermore, the di erential equations are sti because of the sti ness of the springs representing the link plates. Both the discontinuities and the sti ness of Eq. (4) cause high simulation times. 3 5. Formulation and Analysis of the Optimisation Problem In this section the mathematical optimisation problem, that arises when a CVT is to be optimised, is speci ed. Based on this mathematical formulation an optimisation algorithm is selected, which is capable to solve the problem e ciently. A goal function f(p) = f(q(p); _q(p) (5) has to be de ned in dependence of given optimisation parameters p, that quanti es the optimisation target. The complete solution set (q(p); _q(p) := f(q(p; ti); _q(p; ti); i = 0; : : :; kg (6) of the initial value problem, which is based on the dynamical system (4), is needed for each evaluation of f. The data has to be computed on an equidistant grid, as some routines used for the evaluation of the target function require it, e.g. the fast fourier transformation. In order to provide more exibility to choose optimisation goals, arbitrarily complex target functions f must be allowed for evaluating the simulation data. Operations like min, max, least squares or even a FFT (Fast Fourier Transformation) may be used, as well as combinations of them. The objective is to minimise the target function f by nding optimal parameters p subject to upper and lower bounds for each component of p. This leads to a constrained optimisation problem min p2B f(q(p); _q(p); B = fp 2 IRm : li pi ui 8i = 1; : : :; mg: (7) When looking for a convenient optimisation algorithm, the actual problem has to be analysed under consideration of the following two aspects: rstly the properties of the objective function and secondly the constraints in the parameter space. For the problem at hand, the constraints are simple bound con- straints, which can be handled with a gradient projection method, see Kelley 3. Indeed the di culties lie in the objective function. As a time consuming simulation has to be performed for each evaluation, the desired method has to rely on as few function evaluations as possible. Therefore genetic algorithms are out of question. Former investigations have shown, that even optimisation problems concerning rather simple mechanical systems can result in target functions that are nonsmooth, have many local minima and may depend on the optimisation parameters chaotically. A possible optimisation surface for a mechanical problem may be shaped like the one in Fig. 5. PSfrag replacements f p1 p2 Figure 5: Possible optimisation surface Such a worst case has to be expected. Another problem arises from the fact, that a quite large sim- ulation model is implicitely integrated in the goal function f by the output of the numerical integration 4 process. As stated before arbitray methods processing the simulation data have to be accepted. Hence it comes clear that it is not possible to calculate analytical gradients nor hessians. For such a problem the the optimisation algorithm IFFCO (Choi, et. al. 4) seems to be very suitable. It is an implementation of the implicit ltering method for problems with bound contraints. The basic idea of implicit ltering is to utilise a sequence of di erence increments for gradient approximation, starting with a rather big value which is reduced while the iteration proceeds, instead of using very small increments right at the beginning. This provides the opportunity not to get stuck in the rst local minimum, but later when the increment gets small, a local minimum can be reached. A comprehensive description of implicit ltering is provided in Kelley 3. An important feature of this implementation is, that the approximation of gradients by central di erences is parallelised, what means a signi cant reduction of computation time for the whole optimisation process. 6. Optimisation Process In this section the simulation model of the CVT and the optimisation algorithm are combined to form the desired optimisation tool for CVTs. Practical aspects for its implementation are to be discussed. First some additional components and features have to be provided. For a more comfortable implementation of complicated optimisation goals, a library of target functions is developed. It contains several target function prototypes (Tab. 1), which can be used directly or in combination to evaluate the objective function. Table 1: Target function library. Target functions max min fast fourier transformation least squares : : : The class of target functions computes the goal function value f(p) directly from the simulation results and thus maintains an interface between the simulation and the optimisation algorithm. An extension of the simulation program is necessary, namely that the parameters p, which describe the property that is optimised, can be changed from outside of the simulation. Further the functionality to return the required values, e.g. forces in a certain component, has to be established. The combination of the numerical simulation of the CVT, the optimisation algorithm and the library for target functions results in a powerful tool to optimise CVTs. Fig. 6 illustrates the interaction between the components of the tool. The master process is the optimisation algorithm. It calls the CVT simulation and passes PSfrag replacements f(p) IFFCO simulation results target function library p CVT simulation Figure 6: Interaction of the components of the optimisation tool the parameter p whenever a function evaluation is needed. With this parameter set a complete simula- 5 tion of the CVT is executed. The target function value, requested by the optimiser, is calculated from the simulation results by using the target function library. This process iterates until the optimisation algorithm terminates and an optimal parameter set p is found. 8. Optimisation Results The high potential of the introduced tool for optimising CVTs is shown by solving an example problem. Noise emission is a signi cant problem of a CVT and it should be as small as possible. The noise is mainly generated by the polygonal e ect, which occurs where the chain is entering the pulleys, due to the discrete structure of the chain. Approaches to reduce these e ects are chains with links with di erent lengths or chains with an optimised rocker pin joint geometry. Here a chain with links having all the same length is investigated. An optimal link length shall be calculated and therefore the parameter vector is q = (alink): (8) A measure for the accoustical performance of the gear are the amplitudes of the forces in the bearings of pulley B. The aim is to reduce the force peaks in the frequency domain, which refer to the polygonal frequency and its multiples. The following objective function is chosen f = X 2fy;zg f1Z f0 FFT(F )d 2 fy; zg refers to the two components of the support forces of the pulleys in the plane which is perpendicular to the axes of the pulleys. FFT(s) is used as an abbreviation for the fast fourier transformation of the variable s. f0 and f1 are the lower and upper bound of the frequency domain in which the integral over the amplitudes of the forces shall be minimised. The target function f is evaluated numerically and includes two steps: rstly, the simulation of the entire CVT drive to get the forces F , which costs most of CPU-time and secondly, the calculation of the FFT and integration in Eq. (9), which is carried out automatically by the target function class. The starting point for the presented optimisation is a PIV chain with the link length alink = 9:85 mm. In the scope of this article only a special loadcase with the transmission ratio i = 1:0, the angular ve- locity ! = 104:7 rad/s of pulley A and the external torque M2 = 150 Nm is considered. Using of thePSfrag replacements alink mm target function f - start optimum2000 4000 6000 8000 9.6 9.8 10.0 Figure 7: In uence of the link length alink on the target function f. introduced optimisation tool, the target function was reduced to 50%. Figure 7 shows the correlation between the link length alink and the value of the target function f. It is obvious, that the starting con guration was not optimal with respect to the optimisation criterion at hand. The simulation results for the optimal link length alink = 9:93 mm and for the special loading case as mentioned above are compared to the initial con guration in Fig. 8. Each major peak of the amplitudes of the two compo- nents of the support forces can be reduced to one third for that loadcase. This signi cant improvement of the acoustical performance of the gear is achieved by changing the link length by 0.08 mm. 6 PSfrag replacements F y Fz Fy;optimised Fz;optimised 00 00 00 00 20 20 20 20 40 40 40 40 60 60 60 60 80 80 80 80 1000 1000 1000 1000 2000 2000 2000 2000 3000 3000 3000 3000 Frequency HzFrequency Hz N N Figure 8: Comparison of the FFT of the components of support forces before and after optimisation: the major peaks of Fz at 550 Hz and of Fy at 1100 Hz are reduced to about one third by optimisation. (i = 1:0, ! = 104:7 rad/s, M2 = 150 Nm). 9. Conclusions A tool for optimising CVTs has been introduced. It includes the detailed simulation model of the CVT, an optimisation algorithm and a library of target functions. With the aim to nd a suitable optimisation algorithm, the optimisation problem at hand is analysed and implicit ltering turns out to be a very good choice. The capability of the optimisation environment is illustrated by a practical example: the noise emission of the gear is reduced by an optimisation of the chain geometry. Complicated optimisation targets can be implemented quite comfortably and very good results are gained. The whole CVT will be optimised utilising this tool. 10. Acknowledgement The research work presented in this paper is supported by a contract with the DGF (Deutsche Forschungs Gemeinschaft, TB 43). 11. References 1 Srnik, J.; Pfei er, F., 1999. Dynamics of cvt chain drives. International Journal of Vehicle Design, Special Edition, 22(1/2), pp. 5472. 2 Sedlmayr, M., and Pfei er, F., 2001. Spatial contact mechanics of cvt chain drives. 18th ASME Bien. Conf. on Mech. Vibration and Noise DETC01/VIB. 3 Kelley, C. T., 1999. Iterative Methods for Optimization. Frontiers in Applied Mathematics. SIAM. 4 Choi, T. D., Eslinger, O. J., Gilmore, P. A., Kelley, C. T., and Patrick, H. A., 2001. Users Guide to IFFCO. Center for Research in Scienti c Computation, North Carolina State University. 7