Robot Calibration
As is generally known, the disadvantage of the serial kinematics is the insufficient absolute accuracy. If the robot axes are arranged in a row, their tolerances will be increased by the length of the axis connecting members. At the TCP (Tool Center Points) of the robot, these geometric errors react as positioning errors. In addition, compared to parallel kinematics, serial kinematics presents a much lower stiffness. The compliance of the gears and the axis connecting members lead by the torques, which are produced by the payload and the self-weight of the axis connecting members (links), to additional non geometric errors.
The purpose of the robot calibration is to determine and compensate these errors as best as possible.
From a scientific point of view, several different theories exist. The most successful one is the model-based theory. In this case, the model parameters are determined in order to describe the robot kinematics. Our products are based on this theory.
Geometric compensation
The calculated parameters are Denavit-Hartenberg parameters of the robot model.
In general, they are adjustment offsets and link lengths. Our software allows the direct transference of the parameter values into the robot control. The compensation of the robot errors is carried out by the robot control itself with the help of the standard inverse kinematics.
Non geometric compensation
In some special cases, the compensation can no longer be carried out with the help of the simple adjustment of machine parameters. For example in the case of resiliencies or if certain mathematic simplifications are broken (Parallelism of the axes 2 and 3, common intersection point of the wrist at the HWP).The reason for this is that there is no longer a closed representation of the inverse kinematics.
In this case, the calculated model for the compensation offers a mathematical copy of the real robot kinematics with the help of which the static behaviour of the kinematics can be forecast more or less exactly. Now, a numeric solution of the inverse kinematics has to be calculated. This solution has to be fast enough to find solutions of sufficient precision for every sampling point of a track application within an interpolation pulse of a few milliseconds.
