A real-time neural network estimator for workpiece thermal expansion error

This thesis presents a real-time method for estimation of thermal expansion error in workpieces undergoing machining processes. Thermal expansion errors have plagued and will continue to plague machine tool users and manufacturers. Current research has not undertaken the task of developing a compensation method for workpiece thermal expansion errors because of the difficulty in measuring expansions in real-time. This research describes a method for accomplishing this measurement in real-time, which could potentially be used in a compensation method for thermal expansion errors in workpieces. A mathematical model is developed to predict thermal expansions for a cylinder undergoing a turning process. Simplifying assumptions are made and the model is solved using finite element methods. Sets of cutting conditions, not including the actual cutting conditions, are chosen using a design of experiments approach, and the mathematical model is solved at these conditions. This approach is intended to allow the neural network to predict expansions over a range of boundary conditions. The neural network is trained, using solutions from the mathematical model, to predict the expansions in real-time from physically measurable inputs. The neural network estimator is compared to the mathematical model and found to reproduce its function accurately. An experimental test bed has been constructed and experiments have been performed to verify the neural network estimator. Experiments show that the estimator follows the correct trend of the expansion and performs well overall. The use of the estimator in a compensation method could potentially reduce thermal expansion errors and make the manufacture of precision parts more accurate.