Abstract: Through group explicit numerical solution approach of unidimensional heat transferring, the existence and law of thermoelastic effect are analyzed. With the increase of time, the relationship between temperature rise and thermal deformation tends to steady state gradually, but absolute steady state cannot be reached. In the process of heat transfer, with the increase of distance, the temperature decreases rapidly, and the more distant a point from the heat source is, the narrower the thermoelastic effect loop of the point will be. A model which combines nonlinear time series models with neural network models (NARMAX-NN) is put forward to identify the thermoelastic effect. By using the model, the thermal dynamic model of high-speed feed system is built. Compared with multi-variable regression model, feedback neural network model and generalized least squares-output error model, this model has better accuracy and robustness and can accurately carry out modeling and prediction for time-varying and nonlinear thermal error characteristics under machining condition of complex structure and multi-heat sources.

Key words: Feed system, Modeling of thermal error, Nonlinear auto-regressive moving average neural network with exogenous inputs, System identification, Thermoelastic domino effect