The dynamic behavior of both servohydraulic and electromechanical materials-testing machines is affected by the stiffness of the test specimen. This sensitivity is different in load and strain control. In load control the controller gain has to be set higher for soft specimens than for stiff specimens. In strain control, the reverse is true; the controller gain has to be set lower for soft specimens. The machine controller therefore has to be tuned to suit each type of test specimen. Tuning until recently has been conducted manually. It requires a fair degree of skill and often is not done properly.
An incorrectly tuned controller can seriously affect the quality of the materials test. If the proportional gain is set too low the resulting bandwidth reduction impairs the ability of the machine to closely follow the demand signal. Too high a gain can result in closed-loop instability that can rapidly destroy the test specimen.
The tuning problem is compounded by the fact that, in the majority of materials tests, the stiffness of the test piece alters as the test proceeds. Stiffness decreases with the propagation of fatigue cracks or the onset of plasticity or can increase if the test material cyclic-hardens. Some components have an inherent nonlinear stiffness characteristic. They become more or less stiff as the component is strained. In some tests, stiffness changes gradually while in others it fluctuates rapidly during each loading cycle. Varying stiffness means that even if the machine controller is optimally tuned to start with, it is unlikely to remain so throughout the test.
This paper describes an adaptive control system that overcomes these difficulties. It removes sensitivity to stiffness by continually updating the PID controller terms according to real-time stiffness measurements. There is no longer any need to conduct a tuning experiment every time a different type of specimen is installed and stiffness changes during a test are automatically accommodated.
A particularly demanding problem occurs in tests like low-cycle fatigue where the test specimen is exercised beyond its elastic limit. Sudden and significant stiffness changes that occur at each strain reversal are difficult to detect accurately because rates of loading and straining are near zero. This can be overcome by remembering how stiffness changed during the last cycle and this is demonstrated in a real low-cycle fatigue test.