금형의 절삭가공에서 표면거칠기 예측 수식의 유사-실험 데이터 활용을 위한 인공신경망 구축에 대한 연구

A Study on the Establishment of Artificial Neural Networks for the Use of Similar-Experimental Transition Data of Surface Roughness Prediction Equation in Mold Machining

Surface roughness is one of the quality factors of molds that significantly influences the quality and performance of the final product, so it should be carefully considered during mold processing. To achieve the targeted surface roughness in mold machining, it typically relies on the utilization of cutting models for predicting cutting forces and experimental studies to optimize machining conditions. Because it is difficult to intuitively deduce the correlation between cutting variables and actual surface roughness, experiments are necessary in various machining conditions to adapt to changing machining environments. Furthermore, in micro-machining environments like in this study, various factors such as the difficulty of detecting micro-cutting signals, the lack of established standard models for predicting micro-cutting forces, and increased machining costs make it challenging to secure surface roughness through interpretation models and experiments. Moreover, although the trend of utilizing artificial intelligence in industries is increasing, there exist limitations in applying the technology due to the extensive time, manpower, and costs involved in collecting high-quality data for constructing artificial neural networks. In this study, to overcome these limitations and supplement experimental data necessary for AI learning, a neural network conversion model was proposed to convert surface roughness prediction equations into experimental data. Then, by using the converted formula data as similar-experimental data along with actual experimental data, an artificial neural network model for predicting surface roughness was constructed. The predicted surface roughness data obtained through the proposed method was compared and analyzed against actual surface roughness data. As a result, the prediction model incorporating similar-experimental data achieved an accuracy improvement of over 36% compared to models using only experimental data. The surface roughness prediction model with similar-experimental data is expected to replace labor-intensive and costly activities of collecting experimental data in various machining environments.