Production Quality Modeling Based on Regression Rules Extracted from Trained Artificial Neural Networks

Although artificial neural network has been successfully applied to a variety of application problems, it is difficult to explain how the neural network achieves the goal. Yet in production quality modeling, the knowledge of how output characteristics varies with input attributes gives a great help to forecasting, monitoring and controlling in the production process. In this paper, a production quality modeling method based on regression rules extracted from artificial neural networks is proposed. Each rule corresponds to a subregion of the input space and a linear function that approximates the network output for all the samples in this subregion. Experiments on real industrial data demonstrate that the proposed approach not only can successfully extract simple and useful rules indicating important system information, but also have better performances than existing rule extraction methods and traditional statistical methods.