Prediction of Geometric-Thermal Machine Tool Errors by Artificial Neural Networks

摘要

In machining operations, the precision of the workpiece dimensions depends on theaccuracy of the relative position of the cutting tool and the workpiece. Among the key factors that affect the accuracy of this relative position are the geometric errors of the machine tool and the thermal effects on these geometric errors. Recent work on developing models to predict volumetric errors on NC lathes has led to a synthesis technique that combines the modeling of individual axis related geometric-thermal components by way of a rigid body kinematic model to produce predicted errors in the work volume of the machine. An alternative method of modeling these component errors is described in this report. Neural network computing is shown to be a viable technique for developing mappings between machine tool component error measurements and the vector consisting of both a component slide position and the temperature state of the machine as reported by the thermal sensors. The conjugate gradient algorithm, used to compute the optimum neural network weights for the machine tool error components, is described. A case study of the mapping results for one component error of an actual NC lathe is given. Finally, the source codes for the neural network algorithm and the conjugate gradient algorithm are given in FORTRAN.