Measurements Data Analysis and Defect Classification Using NN

This paper presents an analysis of measurement data of turbine housing workpieces, collected with coordinate measuring machine (CMM), so it can be used to train neural network to identify defect workpieces. Software used to create and train neural network are Matlab and NeuroSolutions. Comparison of minimum training mean square error (EMS) for different number of neurons in hidden layer is given and the best number of neurons is chosen. Methodology for reducing the number of necessary dimensions to efficiently classify defects is researched.