Enhancing the Accuracy of a CNC Machine using Artificial Neural Networks

Traditionally, geometric errors of CNC machine axes are rectified by means of calibration with subsequent calculations of compensation values applied directly in a machine controller or with the aid of parameterised kinematic error models.  The main drawback of these methods is that they require a calibration process, which can be not economically viable for some machine applications. In this research, the proposed approach is based on software compensation with the aid of Artificial Intelligence. The developed methodology allows for compensation of the displacement errors of the linear axes of a special purpose CNC machine using artificial neural networks. Two artificial neural networks, namely the Global Neural Network and the Global Vector Neural Network were applied. The artificial neural networks were derived from the data obtained in the drilling of a large number of holes, in a random sequence, in test plates located in the X-Y plane. The coordinates of the drilled holes were then measured with the aid of a CMM to determine geometric errors associated with the particular machine axis. A genetic algorithm was used to train the neural networks on previously drilled holes. The network generated predicted errors, based on which compensation values were then determined and applied in subsequent drilling tests.  As a result, the Global Neural Network and Global Vector Neural Network reduced the average machine displacement error by 55% and 66% respectively for the X axis, and 65% and 55% respectively for the Y axis if compared with uncompensated position values. Thus it is demonstrated that the developed methodology can be applied for compensation of machine axes errors using a relatively simple machining procedure and evolutionary computing.