Estimation of thermal-deformation in machine tools using neural network technique

Abstract To reduce energy consumption in air-conditioned factories, a new method to compensate for thermal displacement using a neural network technique has been investigated. Three-layered feed forward neural networks have been trained using experimental data from a vertical milling machine. After confirming the potential of this method fundamentally, efforts have concentrated on compacting networks and reducing training data. Experimental results have shown that selection of optimal positions of temperature measurement, not using the spindle revolution speed as an input parameter, and preparation of training data sets under the same operation condition as the test data', are helpful in producing a compact and efficient neural network.