Neural network CNC machining

Thermal error prediction compensation for numerical control machining based on neural network. The thermal error compensation principle and the structure sensor measure the temperature and thermal displacement values of the corresponding parts of the CNC machine tool, and then input them into the neural network prediction model through the time delay link TDL for one-step prediction of the thermal error, and perform the prediction value according to the current measurement result. Correction, feedback the result to the CNC machine controller, and the controller issues a corresponding compensation command to compensate for the thermal error that will occur.

The neural network structure and the model input and output quantity prediction model is the discrete input and output model of the controlled object. Since the CNC machine tool is a complex hot state system, the input quantity of the forecast model should include the temperature of the current and past key points. The rise and thermal errors are determined experimentally. The forecasting model of the system, which uses a three-layer BP network, has a structure of N3 (14, 30, 2), which has 14 inputs and two outputs. The choice of the number of nodes in the middle layer has a very important impact on the network learning and computing characteristics. According to the Kol-mogorov theorem, in order to theoretically accurately simulate a continuous function, the number of hidden layer nodes can be taken as 2M+1, where M is the number of input nodes, and the number of hidden layer nodes is 30.

The feedback correction forecast model of model prediction is the basis of forecast compensation technology, and the accuracy of forecast determines the accuracy of forecast compensation. Since the thermal process of CNC machine tools is complex, there are nonlinearities, uncertainties, time-varying and unmeasurable interference. In order to further improve the compensation accuracy, the forecast results are corrected online.