Abstract

In laser drilling, the appropriate laser parameters must be specified to obtain the desired machined hole. Therefore, the effects of process variables on the machined-hole parameters must be investigated and a method for constructing the optimum irradiation condition search must be devised. In this study, we establish a method to predict machined-hole parameters based on artificial intelligence (AI) using the process variables after investigating the parameter distribution of machined holes created via laser drilling. Using this method, the shape of the resulting machined hole can be determined prior to machining. Furthermore, applying the results of this study allows one to determine the irradiation conditions, which consequently enables the identification of process variables that satisfy the required hole parameters and machining quality.

First, the effects of the pulse width and pulse spacing of multiple pulses on machined-hole parameters were investigated. The parameter distribution of the machined-hole diameter was evaluated using the response-surface method by varying the pulse width of a CO2 laser. The result shows that the parameter distribution of the machined-hole diameter varied significantly under different pulse widths. Next, we propose an AI-based method for predicting machined-hole parameters using process variables on a small dataset. The proposed method reduces the prediction error of machined-hole parameters by feeding back the prediction results to the hole-opening process based on laser irradiation via stacking, which is one of the ensemble methods of AI models in the prediction model. By introducing this method, we confirmed that the prediction error was smaller compared with the case when AI was designed to perform prediction using process variables. Furthermore, the stacking was validated using a model comprising ideal data, which is not suitable for actual operation as the measured machined-hole parameters are used as input data during the validation.

This content is only available via PDF.
You do not currently have access to this content.