Combining the Taguchi method with artificial neural network to construct a prediction model of a CO2 laser cutting experiment

When using the Taguchi method, an L18 or L27 orthogonal array is usually adopted. However, this requires many experiments (18 or 27 runs, respectively), consuming time, and resources. This study proposes a progressive Taguchi neural network model, which combines the Taguchi method with the artificial neural network to construct a prediction model for a CO2 laser cutting experiment. During CO2 laser cutting, energy from the moving laser is accumulative. The paper develops an integral equation of energy density during laser beam movement and lets it determine the sliding level of control factor. Meanwhile, the paper proposes that in Stage 1, only less number of experiments is required to be conducted by L9 orthogonal array. After the crucial supplementary experimental training samples proposed in Stage 2 are also included, high-accuracy prediction of artificial neural network can be completed. Based on analysis from the progressive Taguchi neural network, the Stage 1 preliminary network-with only a few available experimental examples-has achieved good predictive ability from regions near the Taguchi control points. For regions further out, the predictions have been increasingly unreliable. Nevertheless, the high precision of Stage 2 Taguchi network has good predictive results for all regions.