Supervisión inteligente de un proceso complejo basada en un modelo neuronal. Un caso real de aplicación.

Abstract

An intelligent supervisory system inspired on a Neural Network Output Error model is presented herein. The application for predicting tool wear in a milling process is selected as a case study. The supervision block consists of a neural model, the weighted sum of squared residuals method and the tool condition index for a decision-making. This work shows the combined use of residual vector norm and the norm of the residual vector derivative to compute adaptive thresholds. The study analyses the influence of infinity and Euclidean norm on the results. Experimental tests are run in a professional machining centre under different cutting conditions using real-time data and new, half-worn and worn tools. The results show this supervisory system’s suitability.