This paper presents a geometric error compensation method for large five-axis machine tools. Compared to smaller machine tools, the longer axis travels and bigger structures of a large machine tool make them more susceptible to complicated, position-dependent geometric errors. The compensation method presented in this paper uses tool tip measurements recorded throughout the axis space to construct an explicit model of a machine tool's geometric errors from which a corresponding set of compensation tables are constructed. The measurements are taken using a laser tracker, permitting rapid error data gathering at most locations in the axis space. Two position-dependent geometric error models are considered in this paper. The first model utilizes a six degree-of-freedom kinematic error description at each axis. The second model is motivated by the structure of table compensation solutions and describes geometric errors as small perturbations to the axis commands. The parameters of both models are identified from the measurement data using a maximum likelihood estimator. Compensation tables are generated by projecting the error model onto the compensation space created by the compensation tables available in the machine tool controller. The first model provides a more intuitive accounting of simple geometric errors than the second; however, it also increases the complexity of projecting the errors onto compensation tables. Experimental results on a commercial five-axis machine tool are presented and analyzed. Despite significant differences in the machine tool error descriptions, both methods produce similar results, within the repeatability of the machine tool. Reasons for this result are discussed. Analysis of the models and compensation tables reveals significant complicated, and unexpected kinematic behavior in the experimental machine tool. A particular strength of the proposed methodology is the simultaneous generation of a complete set of compensation tables that accurately captures complicated kinematic errors independent of whether they arise from expected and unexpected sources.
Geometric error detection of curve-face gear pairs