Procedure for predicting the mechanical properties of pieces obtained by casting

Procedure that enables future mechanical properties to be determined based on real-time observation of the process variables; that allows an in-depth knowledge to be gained of the production process, based on the qualitative conclusions furnished by the analysis models to the factory’s engineering team and that moreover enables advance prediction of the presence of flaws or unacceptable mechanical properties in a manufacturing batch. It comprises the stages of:

a) Data gathering

b) Preprocessing, consisting of: a. Elimination of noise components, by means of Bayesian compression and singular value decomposition; b. Selection of variables by means of factorial analysis and expert knowledge

c) Cross-checking

d) Learning of analysis models in which a process of prediction is performed by sole classifier or by a combination of the classifiers

e) Tests of the model and measurements made.