V. COMPARISON OF STATIC AND DYNAMIC HYSTERESIS CURVES
Four of the balls were measured under static load conditions using commercial test equipment in the Faculty of Engineering at Sydney University. The results are shown in Fig. 5. Each ball was compressed between parallel steel plates at a uniform rate over a period of one minute, held at this compression for one minute and then allowed to expand at a uniform rate, over a period of one minute, back to its original shape. The break in the curve at maximum compression is due to relaxation of internal stress in the ball during the one minute pause between the compression and expansion cycles. The static and dynamic hysteresis curves cannot be compared directly since ~a! the dynamic curve is plotted as a function of the y displacement of the center of mass, and the static curve is plotted as a function of the ball compression, x; and ~b! both sides of the ball were compressed equally in the static test, whereas only the contact side of the ball is compressed in a dynamic bounce. If it is assumed that y 5x/2 for a static compression and that y5x for a dynamic compression then the dynamic and static curves yield similar values for the effective spring constant k5F/y at maximum compression. Alternatively, the dynamic value of F/x at maximum compression is about twice that of the static value. The area enclosed by a static hysteresis curve is less than that of the corresponding dynamic curve for the same compression. This is particularly evident for the superball, where the energy loss is almost negligible during a static compression and expansion. The effect is less pronounced for a baseball since the static and dynamic hysteresis losses are both relatively large. The superball tested did not bounce particularly well, a result that could possibly be attributed to microscopic cracks in the ball. Old superballs, with visible cracks in their surface, bounce even worse. The dynamic tennis ball results are unusual in that the ball is much stiffer during the initial impact than at later times, resulting in a pronounced kink in the force wave form and in the dynamic hysteresis curve. The kink was also observed with other tennis balls, old and new, pressurized and unpressurized. Brody8 has also measured the static hysteresis curve for a tennis ball, using a hemispherical cap to avoid static compression of the upper surface. His results are qualitatively similar to those shown in Fig. 5~a! and indicate that the ball tested by Brody was slightly stiffer and probably newer. The tennis ball used throughout this experiment was an old, relatively soft ball. The static hysteresis curve shown by Brody, as well as the static curve shown in Fig. 5~a!, both enclose an area that is only about 50% of the actual energy loss when a tennis ball bounces off a rigid surface. The dynamic curves in Fig. 3 account for 100% of the energy loss since the y displacement has been calibrated from measurements of v1 and v2 . Part of the discrepancy between the actual loss and the loss estimated from the static curves can be attributed to the increased initial stiffness of the ball during an actual bounce. Part of the discrepancy is also due to losses in the cloth cover. It is known that the cloth cover on a tennis ball contributes significantly to the energy loss, since a rubber ball without a cloth cover bounces better than one with a cloth cover. The effect of the cloth would not be apparent in a static compression test if the cloth recovers elastically from a compression during the test, but not during the short period of the impact. A similar relaxation effect is commonly observed with paper, since paper unfolds very slowly after bending or folding. The increased stiffness of rubber for a high-speed compression can be modeled approximately by the relation m d2y/dt252ky2g dy/dt, where k is the effective spring constant and g dy/dt is a velocity-dependent force term related to the viscosity of the rubber.7,10 Such a model results in a hysteresis curve of finite area since the model equation describes damped harmonic motion. The hysteresis curve in this case commences with y50 and F52gv1 at t50. The model hysteresis curve bears a resemblance to the tennis ball data, for an appropriate choice of g, but it does not give a good fit and is not relevant to any of the other balls. There is no evidence of any velocity-dependent force acting on any of the other balls, since F50 at t50 for all of the balls. Consequently, the energy loss in all cases appears to be due to a time-dependent relaxation of the internal stresses in the ball. Such an effect is referred to simply as an ‘‘elastic aftereffect’’ in the rheology literature.10 The effect is complicated by the fact that a spectrum of different time constants is usually required to describe the relaxation. In the case of the steel ball, losses in the ceramic piezo and the brass rod may account for almost all of the energy loss.