Optimal Rocker Arm Design in High Speed Internal Combustion Engines

J. W. David and Yimin Wei Norih Carolina State Univ.

J. A. Covey General Motors

ABSTRACT

One of the primary objectives of building an automotive engine is to produce suflicient power throughout its most common operating range. The objective of this study is to determine how maximum engine speed may be increased through rocker arm modification.

Current knowledge suggests that there are two primary factors in the design of rocker arms that will effect the engine's operatirlg speed: the mass moment of inertia and the stfiess. Experimental and computational methods were used to investigate the influence of these two factors on valve train performance.

The ANSYS Finite Element package Design Optimization Routine was used to optimize the design of a typical Chevrolet NASCAR rocker arm and one used in the Buick V6 Indy Engine. Also investigated was the use of various materials for rocker arm construction.

INTRODUCTION

Many racing engines are built using components similar to those used in production c a ~ ~ . As engine speed is a primary concern in racing, many of these components must be redesigned to enable the engine to have a sufficiently wide rpm band without sanificing reliability. The primary objective of this study is to determine how maximum engine speed may be increased through rocker arm modification.

Figure 1 shows the schematic of a typical pushrod type valve train toward which this study is directed. While there are many factors that determine the maximum speed of an engine, one of the most critical is the valve bounce phenomenon. Valve bounce occun when the valve impacts the valve seat upon closing and opens again, breaking the seal of the combustion chamber. This reduces cylinder pressures and horsepower and leads to premature failure of valve train components. Valve bounce can also reach an amplitude such that the valves and pistons collide, resulting in the immediate destruction of the engine. Figure 2 shows a typical plot of valve bounce amplitude versus engine speed. Internal vibrations in the mechanism cause the amplitude to increase

and decrease through the speed range. Finally, the amplitude increases explosively at a certain speed, which in this study is called the limit speed. This is the speed at which unstable, chaotic valve motions start and is definitely the upper limit of engine operating speed. This is the criteria used to define limit speed throughout this study.

Plvot Rocker Arm f i : j Retainer ash

Valve Spring

yllnder Head I I valve I I

,Y Cam Llfter

Figure 1: Typical Pushrod Valve Train

EXPERIMENTAL STUDY

Current knowledge suggest that there are two primary rocker arm characteristics which affect valve train performance: mass moment of inertia and stiffness. Thus, a series of experiments

were conducted to determine the relative influence of these two parameters on valve train performance.

Figure 2: Typical Valve Bounce Curve

Eleven rocker arms, of identical ratio and motion geometry, were tested. The variations were in width and location of lightening slots, producing 11 different combinations of stiffness and moment of inertia. All other valve train components were identical. For this part of the study, the overall system compliance was measured, the mass moment of inertia was measured and a limit speed test was conducted The results of the test appear in Table 1.

Table 1. Results of Inertia, Deflection and Limit Speed Tests

As can be seen, the rocker arm with the smallest inertia (1 1) was not the best, nor was the one with the greatest inertia (3) the worst. The one with the least deflection performed well, as did the one that had a slightly higher inertia and deflection. Thus, the trend seems to be one of decreasing moment of inertia and decreasing system deflection (increasing rocker arm stiffness), but some irregularities exist in the data. The best rocker arms (6 and 7) tended to have the best combination of low inertia and low system detlection. This suggests that a suitable objective function for the design optimization may be the ratio of stiffness to moment of inertia.

VALVE TRAIN SIMULATIONS

Many models have been developed to simulate valve trains. The model developed by Kim [l] and Kim and David [2] is used here as it has proven to be accurate at engine speeds above 9,000 rpm with pushrod type valve trains. A schematic of the model appears in Figure 3. The impoltant aspect of this model is that it is wmputationally el5cient as well as being accurate. In addition, it models valve spring surge and incorporates the gap nonlinearitia found in racing valve trains. A system identification procedure developed by Liu [3] was used to determine the stiffness and damping coefficients, except for K,, which represents the rocker arm stiffness. Mass coefficien& were determined by direct measurement.

C U M FRICTION AT ROUSE PIVOT

V Y * L A 9 1

Ks tOULCne FRICTIOU c. AT V N V E STEP: V N V E SPRING

Figure 3: Schematic of Valve Train Model [I]

Simulations were performed with various combinations of I, and K,, producing three dimensional performance surfaces. As there were two valve trains of interest, those of the Chevrolet NASCAR engine and the Buick V6 Indy engine, simulations were petformed on both systems. The results for the Chevrolet NASCAR engine appear in Figure 4 and those for the Buick V6 Indy engine in Figure 5. For both systems, the dominant trend is to increase the ratio of stiffness to moment of inertia. Local variances are due to system nonlinearities and appear to be of secondary importance for these two systems.

Figure 4: Design Surface for the Chevrolet NASCAR Engine

98

Figure 5: Design Surfaces for the Buick V6 Indy Engine

The simulation and experimental study indicate that optimal rocker a m design for these systems can be performed by maximizing the ratio of stiffness to moment of inertia. This is beneficial in that a complex nonlinear valve train model does not need to be included in an optimization loop to assess candidate designs, greatly reducing the computational effort required.

OPTIMAL DESIGN PROCEDURE

Engineering is often an iterative process that strives to obtain a best or optimal design. The optimum design is usually one that is as efficient and effective as possible. The usual path to the optimum is a traditional one. The desired function and performance of the design are first defined. Then, a manual design process is usually pursued in which the engineer:

1) Develops an initial design 2) Performs and evaluates an

analysis of the design 3) Modifies the design 4) Repeats steps 2 and 3 until an

optimal design is obtained. This procedure is often costly and time consuming, often requiring the use of "less than optimal" designs.

With the development of the digital computer, it is now possible to integrate the design cycle into a progammed mathematical technique. The analysis, evaluation and mcdification tasks are all performed automatically, making it possible to obtain optimal designs at lower costs. In this study, the commercial finite element program ANSYS [4] is used to perform the design optimization. As the objective function has been define& the procedure now is to:

1) Parametrically define the problem 2) Initialize design variables 3) Perform finite element analysis 4) Extract desired information from the FEA to

evaluate the objective function 5) Establish 3 and 4 in the ANSYS design

optimization routine.

The ANSYS program was used to generate a solid model of the initial mker arm. In all, 137 parameters were used to define the geometry of the mker ann. Of these, 10 were selected as design variables as shown in Figure 6. Upper and lower bounds of these variables were specified as design

Figure 6: Design Variables for the Optimization Process

constraints. It is here that the design engineer most influences the process in how many and which variables are optimized. Obviously, other "optimal" designs can be obtained by choosing different variables. In addition, a constraint was applied to the maximum equivalent Von Mieses stress during each optimization loop to minimize potential fatigue problems with the new designs. Aluminum, steel and a metal-matrix composite were evaluated for their effectiveness as materials. The metal-matrix material used was a particulate composite of aluminum and 25% silicon- carbide.

The finite element model used in this study is shown if Figure 7. Eight-node isoparametric brick elements were used The pushrod end was assumed fixed over four nodes and a force was applied to the valve end and also distributed over four nodes. The bearine bore was constrained so as to allow rotation. The rocker arm stiffness was calculated as the ratio of applied force to deflection at the valve end. The moment of inertia was calculated using the solid modeler. With appropriate units, this yields an objective function on the order of lo-'', so the objective function was multiplied by 10 '~ to avoid numerical problems.

The improvements for this case are not dmnatic, but this is not surprising in that the manufacturer had already performed some optimization on the product. Notice that of the candidate materials, both steel and aluminum produced similar results. This is because the ratio of elastic modulus to density is similar for these materials. The metal-matrix results appear promising due to the low value of the objective function. However, in testing, the metal-matrix rocker arm was only slightly better than an aluminum one. This is thought to be due the metal-matrix material exhibiting a bilinear elastic modulus and at the low amin seen in a rocker arm, the material behaves basically like aluminum. More testing is required in order to confirm this.

Table 4 shows the optimization results for the Buick V6 Indy Engine. Only an aluminum design was attempted for this problem. Here, the results are more pronounced in that no optimization work had been performed on the initial design. Figure 9 shows the initial and optimal designs. Also, Figure 10 shows the simulation results for the original and optimal designs. In this case, an improvement in valve train

limit speed of 300 rpm was obtained only with improvements in the rocker arm.

Table 4. Optimization Results for the Buick V6 Indy Engine

Figure 7: Finite Element Discritization and Boundary Conditions

Initial Design Final Design

The optimization process was allowed to run up to 100 iterations for each design. In most cases, however, less than 30 iterations were required to obtain convergence. Each design took approximately one hour to complete on an JBM RS6000 workstation.

The results for the Chevrolet NASCAR engine appear in Table 2. Note that in this table, the inertia is only that of the rocker arm body and doesn't include the roller tip, pushrcd seat adjusting mew or bearings.

K,(~o'N I m)

3.42

3.89

Table 2. Optimization Result for the Chevrolet NASCAR Engine

Figure 9: Initial and Optimal Rocker Arm for the Buick V6 Indy Engine

~,(lO'kg - m') 12.35

9.61

7- 72m 7.m 7- =:h. s;d

- - - m

,o+&.l ,Oplii..d

Figure 10: Dynamic Performance of Original and Optimal Designs for the Buick V6 Indy Engine

100

Limit Speed (mm)

8700

8900

CONCLUSIONS

Rocker arm design was used to improve the performance of high-speed engines. By using a nonlinear simulation model, an objective function for optimal rocker arm design was defined as the ratio of mass moment+f-inertia to stiffness. Using this within the ANSYS FEA package produced designs which improved engine limiting speed.

ACKNOWLEDGMENTS

The authors would l i e to express their appreciation to Mr. Dan Jesel of Jesel, Inc. for producing the rocker arms used in this study.

REFERENCES

[l] D. Kim, Dynamics and Optimal Design of High Speed Automotive Valve Train Systems, Ph.D thesis, Mechanical and Aerospace Engineering, North Carolina State University, Raleigh, NC, April, 1990.

[2] D. Kim and J.W. David, A CombinedModel for High Speed Valve Train Dynamics. Partly Linear and Partly Nonlinear, SAE Technical Paper Series 901726.

131 A. Liy System Identification and Optimal Design ofHigh Speed Valve Train Systems, Ph. D thesis, Mechanical and Aerospace Engineering, North Carolina State University, Raleigh, NC, April, 1994.

[4] ANSYS 4.4A1 Finite Element Package, Swanson Analysis, Inc