You are on page 1of 6

12th IFToMM World Congress, Besanon (France), June18-21, 2007

Gear vibration reduction using genetic algorithms


M. Barbieri G. Bonori G. Scagliarini University of Modena and Reggio Emilia Modena, Italy F. Pellicano

Abstract In the present paper, two different techniques are proposed to design spur gear profile modifications for noise reduction: an iterative Heuristic approach and an optimization approach based on Genetic Algorithms. Different objective functions are investigated and dynamic simulations are carried out to check the effectiveness of the new design solutions.

Keywords: Profile modification, gear, optimization, genetic algorithm.

I. Introduction The reduction of vibration and noise is nowadays a very important issue in the design of gear power transmissions. The main reasons are the increase of performance, standard noise restrictions and market competitors. Experimental tests [1] proved that several nonlinear phenomena, such as sub and super harmonic resonances and chaotic motions, occur when the dynamic transmission error is present. The literature offers a large number of paper regarding the simulation of the dynamic behaviour of gears: Kahraman and Singh [2] focused the attention on the effect of time varying mesh stiffness using the harmonic balance method for a gear system with clearance; Amabili and Rivola [3] proposed a different approach based on Hill infinite determinant to predict transition curves and stability regions; Parker et al. [4] used a semi-analytical technique to capture the non linear dynamic response of a gear pair. Many authors underlined the effect of modifications on the dynamics of gears: Kahraman and Blankenship [5] experimentally measured the effect of the involute tip relief on the dynamic response, while Bonori et al. [6] simulated the effect of different profile modifications and manufacturing errors on gear noise. A key point on the design process of spur gears is the choice of the best profile modification; it is not trivial difficult to define design guidelines for profile modifications. In 1999, Smith [7] suggested the application of Harris maps on tip and root relief, improving the work of Harris [8] and Niemann [9]. New approaches where proposed by Beghini et al. [10] and Fonseca et al. [11], they used iterative and genetic approaches respectively; such papers did not provide the effect on the dynamics.

In the present work two optimization techniques are presented to find suitable profile modifications for a spur gear pair. Both optimization approaches are based on the minimization of the peak to peak of the static transmission error (PPTE). The first method searches the optimum solution by means of iterative calculations of the PPTE, using many combinations of the parameters, which define the profile modifications. The second method uses a genetic algorithm approach, able to provide strong reduction of the PPTE, with a higher computational efficiency. The genetic algorithm approach is also used with a different objective function, i.e. the averaging of mesh stiffness harmonic content. In order to evaluate the effectiveness of the optimization techniques, an actual gear pair is considered as case study and dynamic simulations are performed. For this purpose a simple spur gear dynamic model is developed, including a smoothing technique for the clearance type nonlinearity. Numerical integrations allow to obtain the whole dynamic scenario. II. Analytical model In the present section the general equation of motion of a spur gear system is described. Details of the smoothing technique, used to approximate clearance, and mesh stiffness calculation are given. A. Equation of motion In the present model each gear is represented by a rigid disk, coupled along the line of action through a time varying mesh stiffness k(t) and a constant mesh damping c (see Fig. 1). dg1 and dg2 are the gear base diameters; g1 and g2 are angular position of the pinion and gear respectively; Ig1 and Ig2 are rotary inertia; Tg1(t) and Tg2(t) are the driver and breaking torques respectively. Shafts and bearings elasticity are neglected.
Pinion e(t) k(t) dg2 Gear

c Tg1 g1 dg1 Ig1 Ig2

g2 Tg2

Fig.1. Gear model.

12th IFToMM World Congress, Besanon (France), June18-21, 2007

k(t) is evaluated through a nonlinear finite element model, which takes into account standard profile modifications (linear and parabolic relief, see Fig. 2). The relative dynamics along the line of action, Fig. 1, is described by the following equation: (1) me x (t ) + c ( x (t )) + k (t ) f ( x (t )) = Tg (t ) where over-dot means time derivative and f(t) is the backlash function used to simulate clearance (see next sections for details).
TIP RELIEF magt
Magnitude at tip

f1 (t ) =

1 ( x(t ) b) {1 + tanh[ ( x(t ) b)]} 2 1 ( x(t ) + b) {1 + tanh[ ( x(t ) + b)]} 2

(5)

where is the shape parameter and 2b is the total clearance along the line of action. In the following, =108 will be used to assure enough accuracy and computational efficiency. D. Mesh stiffness The evaluation of the mesh stiffness within a mesh cycle is obtained by means of a commercial software package CALYX that uses a combined surface integral approach and a finite element (FEM) solution, [15,16]. An important feature of this technique is that no mesh refinement is required, close to the contact zone, to capture the effect of profile modifications (see Fig. 3).

Start roll angle at tip

ts

Start roll angle at root End roll angle at root

rs re ROOT RELIEF

Magnitude at root

magr

Fig. 2.Profile modifications (root and tip linear relief).

In equation (1), x(t) is the dynamic transmission error along the line of action:
x (t ) = d g1 2

g1 (t )

dg2 2

g 2 (t )

(2) (3)

me is the equivalent mass:


me =

{( d

2 g1

2 4 I g1 + d g 2 4 I g 2

) (

)}

Tg is the equivalent applied preload: d g1Tg1 (t ) d g 2Tg 2 (t ) Tg (t ) = me + 2 I g1 2I g 2

(4)

Fig. 3.Detail of the gears meshes on the contact area.

where Tg 2 (t ) = Tg1d g 2 / d g1 and Tg equal to constant. It is to note that Tg is expressed in Newton [N], i.e. it represents a force on the line of action; conversely, Tg1 and Tg2 are torques expressed in [Nm]. B .Backlash According to literature, many non-linear phenomena may occur in gear systems due to clearance between the meshing teeth [1]. For this reason many authors [12-14], suggest the use of a non-linear displacement function f(t) to describe the change of stiffness, which is related to the loosing of contact and the back side low impacts. In the present model the following smoothing function is used to approximate the non-smooth system and improve the computational efficiency:

The approximation of the mesh stiffness is provided by a Fourier expansion which is obtained numerically using 20 samples within a mesh cycle:
k (t ) = k0 + i ki cos(imt i )

(6)

Figure 4 shows a comparison between the mesh stiffness evaluated form FE analysis and its expression obtained using equation (6). D. Normalization A dimensionless form of equation (1) is obtained by letting:
n =
Tg k0 c x ; = ; = n t ; Tg = ; x= 2 me bmen b 2men

(7)

and:

12th IFToMM World Congress, Besanon (France), June18-21, 2007

ki =

ki ; k (t ) = 1 + i ki cos i m i ; 2 men n

(8)

Equation (1) assumes the following form:


x( ) + 2 ( x( )) + k ( ) f ( x ( )) = Tg

(9)

where ( ) = d( )/d. Equation (14) is solved by means of an adaptive step-size Gear Algorithm according to literature [17]. Note that using the smoothing function allows to solve equation (9) with a standard numerical integrator.

experimental results is performed: in 1997 Kahraman and Blankenship [1] presented a number of experiments on a physical system with clearance combining parametric and external forcing excitation. One of the tests concerned a spur gear set with geometrical and physical parameters reported in Table I . A damping ratio =0.01 and an external torque Tg1=340Nm are used. The natural frequency of the system is n=1.983104 rad/s. Using 20 positions within a mesh cycle the peak to peak value of the mesh stiffness is 6.121107N/m. Five harmonics are considered in the Fourier expansion of the stiffness. In Fig. 5 the comparison between the present model and Ref. [1] shows that the present theory is accurate in reproducing the gear vibration.

m t
Fig. 4.Mesh stiffness: (blue) CALYX; (red) Fourier series expansion.

Teeth number Module [mm] Pressure angle [Deg] Base diameter [mm] Tooth thickness at pitch diameter [mm] Outer diameter [mm] Root diameter [mm] Face width [mm] Mass [kg] Inertia [kg m2] Youngs modulus [MPa] Poissons coefficient Center distance [mm] Backlash [mm] Profile modifications

Pinion 50 3 20 140.95 4.64 156 140.68 20 2.5161 0.0074 206000 0.3 150 0.1447 None

Gear 50 3 20 140.95 4.64 156 140.68 20 2.5161 0.0074 206000 0.3

Fig. 5.Comparisons with literature: * Ref. [1]; present model.

IV. Optimization It is well known that the minimization of the amplitude of oscillation (peak to peak) of the static transmission error (PPTE) yields to a reduction of gear noise excitation. The present optimization approaches search for the best set of profile modifications able to minimize the peak to peak value of the static transmission error (Tg/k(t)). In this sense the main objective function of the analysis is the PPTE. The two main characteristics of the present multivariable optimization problem are: 1. Each set of profile modifications depends on 8-10 different parameters, which vary within a large domain; 2. The objective function can be evaluated only numerically. For this reason two optimization strategies are proposed: i) an Heuristic approach that searches for the optimum by partially sweeping the parameter space; ii) Genetic Algorithms. In the following, both approaches are tested on the same case study; the genetic algorithm technique is also applied

Table I. Geometrical data of a spur gear set from Ref. [1].

III. Comparison with literature In order to check the accuracy of the model with respect to an actual system, a comparison with some

12th IFToMM World Congress, Besanon (France), June18-21, 2007

using the average of the amplitude of the first 7 harmonics of equation (6) as objective function. A. Case study Table II shows geometrical data and material properties of the case study. The profile modifications set represents the original design of a noisily gear pair. The purpose of the analysis is to verify if the design parameters can be improved in order to reduce vibration.
SYSTEM DATA Module Pressure angle Center distance Nominal Torque (pinion) Tg1 GEOMETRIC PROPERTIES Number of teeth Face width Tooth thickness (on pitch circle) Outer diameter Root diameter Inner diameter Hob tip radius Addendum modification Crowning 3 mm 20 deg 111 mm 470720 Nmm PINION 28 27 mm 6.115 mm 93.1 mm 79.1 mm 40 mm 0.9 mm 1.927 mm 0.015 mm MATERIALS PROPERTIES Young Modulus Poisson's ratio Density 206000 MPa 0.3 7.85 10 kg/mm
-6 3

GEAR 43 22.5 mm 6.713 mm 139.7 mm 126.2 mm 40 mm 0.9 mm 2.748 mm 0.015 mm

PROFILE MODIFICATIONS Type of modification 1 1=linear; 2=parabolic Start roll angle at tip Magnitude at tip Start roll angle at root End roll angle at root Magnitude at root 30.16 deg 0.016 mm 23.47 deg 14.43 deg 0.016 mm

1 29.21 deg 0.018 mm 25.21 deg 20.58 deg 0.018 mm

Table II. Geometrical data of the spur gear pair.

B. Heuristic optimization In the heuristic method the following profile modification parameters are considered (see Figure 2): Start roll angles at tip on pinion and gear, both are varied between the pitch diameter and the outer diameter; Magnitude of modifications at tip on pinion and gear, both are varied between 0 and 40 m;

Start roll angles at root on pinion and gear, both are varied between the begin of the involute and the pitch diameter; The magnitude of profile modifications at root on pinion and gear, both are varied between 0 and 40 m. The end roll angle at root of the pinion and gear are not considered as optimization parameters, because they are equal to the start of active involute profiles. A total amount of eight parameters are involved in the heuristic optimization process. Such parameters will not be considered simultaneously. Indeed the main idea is to sweep the full parameter domain by means of iteration steps considering the variation of only two parameters at time. Therefore, four steps are required to cover all variables domain. The result of each step is a surface which shows the effect of the variation of couples of parameters on the PPTE. Figure 6 illustrates the results of each single iteration on the case study: Figure 6a shows the PPTE versus start roll angles at tip on pinion and gear, a 5050 grid in the parameter space is considered, while the other parameters are left constant; Fig. 6b shows the effect of the modification magnitude at tip both for pinion and gear (4141 grid); Fig. 6c shows the effect of start roll angles at root on pinion and gear (5050 grid); Fig. 6d shows the effect of the modification magnitude at root (4141 grid). The result of the optimization process is a new configuration of profile modifications, which reduce the PPTE from 0.0051 mm to 0.0017 mm. Note that in all cases the PPTE is obtained from the analysis of k(t), which is sampled with 15 samples within a mesh period; therefore, at the end of the optimization process 125430 finite element computations have been carried out. Although the heuristic technique analyzes a large number of different profile modification sets, it is an approximate approach, because it does not lead to a minimum in a mathematical sense. In order to check the effectiveness of the optimization on gear vibration a dynamic analysis is performed according the model described in the previous sections. Figure 7 represents results of a simulation in terms of the amplitude-frequency curve; the rotation speed is varied within the range 500-22000 rpm. The initial set of gears (called CNH modifications) presents large amplitude of vibration close to the natural frequency of the system; the response is nonlinear (softening) and loosing of contact is present, as proven from the softening behaviour of the curve. After the optimization, the amplitude of the response is significantly smaller than the initial one; the response is linear. At low frequencies the optimization does not necessarily reduce the amplitude of the response, because minimizing the PPTE does not imply a reduction of higher harmonics of k(t), see equation (6).

12th IFToMM World Congress, Besanon (France), June18-21, 2007

a)

grid 5050

b)

grid 4141

higher probability to be chosen. Then, a string of bits is assigned to each solution; the string contains a unequivocal representation of the point in a binary coding. In the present case, since the objective function depends upon eight variables, the domain is a subset of 8 and the string is a juxtaposition of eight substrings (see Fig. 8). Each substring represents the value of a profile modification parameter set by means of a Gray code [18].
Pinion
ts
11 bits mag t

Gear
rs
mag r 6 bits ts 11 bits

mag

6 bits

11 bits

6 bits

rs 11 bits

mag

6 bits

c)

grid 5050

d)

grid 4141

0110...

01..
Fig. 8.Composition of the string.

...01

Fig. 6. Minimum of PPTE vs.: a) start roll angles at tip on pinion and gear; b) magnitude of modifications at tip on pinion and gear; c) start roll angles at root on pinion and gear; d) magnitude of modifications at root.

Crossover defines a new population by randomly coupling strings, from the previous population and cutting both strings of a couple, with a probability pc, at a randomly chosen point. Then, the second part of the second string is joined with the first part of the first one and viceversa. In order to increase the variability and to avoid skipping some regions of the domain, mutation changes the values of each bit in the population of strings with a given probability pm. In the present case, the genetic analysis was performed using the following parameters: - Number of points in the population, n = 50 - Number of bits in each strings, m = 68 - Probability of crossover, pc=0.6 - Probability of mutation pm = 0.033

Fig. 7. Amplitude-frequency diagram: initial modifications; - present optimization.

C. Optimization with genetic algorithm The main idea of genetic algorithms is learning, through iterations, what regions of the domain contains better solutions and focusing the search in those regions. Despite other optimization methods, which take a solution (point) and improve it by means of iterative transformations (e.g. steepest gradient), genetic algorithms start from a population of randomly chosen points and transform them similarly to Natural Genetics. Three are the main transformations: selection, crossover and mutation. Selection consists in extracting a new population of solutions from the previous one (or from the initial points) with a sort of cheating roulette: the best solutions have a
Fig. 9.Amplitude-frequency diagram: heuristic optimization; - - genetic optimization

The results obtained after 50 iterations (73500 FEM computations) is PPTE=0.0016323 mm. Figure 9 shows the dynamic behaviour of the system with the proposed profile modifications in comparison with the corresponding modifications obtained with the previous Heuristic method. Despite the lower peak to peak, the amplitude of the main resonance is higher using the new modifications: the relationship between peak to peak and

12th IFToMM World Congress, Besanon (France), June18-21, 2007

resonances is weak. All dynamic simulations show a direct relationship among the harmonic content of the static transmission error (STE) and the corresponding resonance, i.e. the i-th harmonic of the STE is related to resonance at 1 of the mesh frequency. This behaviour i suggests the use of a new objective function, which takes into account the harmonic content of the STE.
0.12 /4
n

Acknowledgements The authors thanks SiMech Lab and Case New Holland S.p.A. that supported the research. References
[1] Kahraman A., Blankenship G.W. Experiments on nonlinear dynamic behavior of an oscillator with clearance and periodically time-varying parameters. Journal of Applied Mechanics 64 (1997) 217-226. [2] Kahraman A., Singh R. Non-Linear Dynamics of a Spur Gear Pair. Journal of Sound and Vibration, 142(1) (1990) 49-75. [3] Amabili M., Rivola A. Dynamic analysis of spur gear pairs: steadystate response and stability of sdof model with time-varying meshing damping. Mechanical systems and signal processing, 11(3), 375-390, 1997. [4] Parker R.G., Vijayakar S.M., Imajo T. Nonlinear dynamic response of a spur gear pair: modeling and experimental comparisons. Journal of Sound and Vibration, 237, 435-455, 2000. [5] Kahraman A., Blankenship G.W. Effect of involute tip relief on dynamic response of spur gear pairs, Journal of Mechanical Design. 121, 313-315, 1999. [6] Bonori G., Andrisano A.O., Pellicano F. Stiffness evaluation and vibration in a tractor gear. Proceedings of IMECE 2004, ASME International Mechanical Engineering Congress, Anaheim, CA, 2004. [7] Smith J.D. Gear noise and vibration. Marcel Dekker Inc., New York, USA, 1999. [8] Harris S.L. Dynamic loads on the teeth of spur gears. Proceedings of the Institution of Mechanical Engineers, 172(2), 87-100, 1958. [9] Niemann G., Baethge J. Transmission error, tooth stiffness, and noise of parallel axis gears. VDI-Z, 2(4), 2(8), 1970. [10] Beghini M., Presicce F., Santus C. A method to define profile modification of spur gear and minimize the transmission error. AGMA Technical Paper, 04FTM3, 2004. [11] Fonseca D.J., Shishoo S., Lim T.C., Chen D.S. A genetic algorithm approach to minimize transmission error of automotive spur gears sets. Applied artificial Intelligence, 19(2), 153-179, 2005. [12] Azar R.C., Crossley F.R.E. Digital simulation of impact phenomenon in spur gear system. Journal of Engineering for Industry, (1977) 792-798. [13] Tomlinson G.R., Lam J. Frequency response characteristics of structures with single and multiple clearance-type non-linearity. Journal of Sound and Vibration, 96(1) (1984) 111-125. [14] Blankenship G.W., Kahraman A. Steady state forced response of a mechanical oscillator with combined parametric excitation and clearance type non-linearity. Journal of Sound and Vibration, 185(5) (1995) 743-765. [15] Vijayakar S.M. A combined surface integral and finite element solution for a three dimensional contact problem. Int. J. Numer. Methods Eng., 31 (1991) 525-545 [16] Parker R., Vijayakar S.M., Imajo T. Nonlinear dynamic response of a spur gear pair: modeling and experimental comparisons. Journal of Sound and Vibration, 237(3) (2000) 435-455.. [17] IMSL. IMSL Fortran Library Users Guide, MATH/LIBRARY Volume 1 and 2, Visual Numerics Inc., USA, 2003. [18] Bck T., Fogel D. B., Michalewicz Z. Handbook of Evolutionary Computation. Ringbound edition 1997.

/2 n

Harmonic content of mesh stiffness


0.01

0.1

0.008

0.006

Non dimensional semiamplitude

0.004

0.08
0.002

0.06

n 0.04 n/6 n/5 0.02 /3


n

n/7

0 0

0.2

0.4

0.6 m/n

0.8

1.2

Fig. 10.Amplitude frequency response using new objective function and harmonic content of related STE.

Figure 10 shows dynamic results after minimizing the average amplitude of the harmonics used in Fourier approximation of STE. It is clear that the use of Genetic Algorithms in conjunction with a suitable objective function can lead to a great improving of the dynamic performances of the gear pair, in terms of noise and vibration, with an acceptable computational cost. V. Conclusion Two optimization techniques are developed in order to reduce vibrations generated by gear pair meshing. Nonlinear finite elements are used in order to evaluate the meshing stiffness gear pairs. Dynamic analyses are carried out in order to check optimizations obtained on static basis. Results show that both approaches greatly improve the dynamic performances of gears in terms of vibration amplitude over wide range of excitation frequency; Genetic Algoritms behave better, the peak to peak of the static transmission error is slightly reduced with respect to the Heuristic method and the computational cost is reduced. It is to stress that in both cases, the dynamic behaviour, after optimization, drastically improves: no loosing of contact is found anymore.

You might also like