Machine tool thermal errors are one of the most significant factors affecting the accuracy of machine tools. They can be compensated for through a model-based approach. However, the lack of accuracy and robustness of thermal error models prevents the thermal error compensation from achieving greater success. The objective of this research is to develop a new dynamic modeling methodology to improve the accuracy and robustness of the thermal error models. This research consists of the following components: (1) The analysis of the dynamics of the thermo-elastic systems, (2) non-stationary multivariable system identification, (3) part-oriented machine-error calibration, (4) on-line model adaptation, and (5) the identification of nonlinear thermo-elastic systems.; The analysis of the dynamic characteristics of the machine thermo-elastic systems reveals that the pseudo-hysteresis effect is a major factor causing poor robustness of the conventional static models. A dynamic modeling methodology is therefore developed, based on the system identification theory, in order to capture the system dynamics. The procedure for the dynamic modeling includes input variable screening, model structure determination and model validation. In addition, due to the non-stationary nature of the thermo-elastic systems, a time-varying system identification methodology is developed. Based on a part-oriented thermal error calibration methodology developed for fulfilling the fast measurement requirement of the dynamic thermal error modeling, time-varying thermal errors of a face-milling process with large process variations are successfully tracked and modeled.; Since the pre-process trained thermal-error model may not be accurate and robust enough in long term, a model self-adaptation system is developed in order to continuously update the model, using process-intermittent measurements. The recursive model adaptation, based on the Kalman filter and multiple-sampling horizons, minimizes intrusion to production while maintaining good model adaptation capability. Nonlinearity is another major factor influencing model estimation accuracy. A new neural network modeling strategy, called the Integrated Recurrent Neural Network, is applied to identify the dynamics, nonlinearity and non-stationarity of the thermo-elastic systems. Experimental results prove that the dynamic models thus developed show great advantages over the conventional static models in terms of model accuracy and robustness for different working conditions.
展开▼
