Artificial Intelligence for Estimating Laser Power from Temperature Distribution

We developed a laser heating test system called the selective laser thermoregulation (SLT) system using a fiber laser and a galvano mirror system to confirm the reliability of SiC/SiC CMCs, which are expected to improve the efficiency of aircraft engines. However, to realize the required temperature distribution using the SLT system, many parameters must be determined. For automatic determination of the parameters, we developed AI for estimating the laser power needed to reproduce the required temperature distribution. We created datasets for machine learning by numerical simulation of laser heating and compared three types of fully connected neural networks. The AI learned the relationship between laser power and the temperature distribution and then estimated the laser power from an untrained temperature distribution. The R 2 of the 2-layer neural network was -1.64, while the R 2 values of the 3-and 4-layer neural networks were 0.98 and 0.99, respectively. Thus, we found that three or four layers of fully connected neural network are sufficient for realizing AI which can estimate laser power from a temperature distribution.