Measurement method of metal surface absorptivity based on physics-informed neural network

Abstract

Characterizing the absorptivity of a rough metal surface is a difficult but important task. The uncertainty will be enlarged by using the indirect method, i.e. 1 – reflectance measurement. In contrast, the calorimetric method is of high fidelity. However, it is difficult to extract the absorptivity. The variation of temperature follows the heat conduction equation which is a differential equation. Therefore, a method based on physics-informed neural networks (PINNs) is proposed. In this method, the temperature rising curve is fitted to the differential equation by the neural network. The differential equation is incorporated into the network through the loss function. When the training is done, the absorptivity can be extracted. For demonstration, the numerical test and experimental test are performed. A set of temperature profiles with different absorptivity values is generated numerically. Then the absorptivity is extracted by PINN. The numerical results show that this method is able to determine the absorptivity and possesses the advantages of strong anti-interference capability and high accuracy. The maximum absolute error is 0.00092 in the range of 0.05 to 0.2. In the experiment, sand-blasted gold coated aluminum plates are used as the test objects, and they are heated by a continuous wave infrared laser. The temperature is measured by a K thermocouple. Then the absorptivity values of different samples are determined by the PINN, ranging from 2% to 10% because of the differences in roughness and electroplating process. The measurement repeatability is < 1%. The proposed method is very promising to become a powerful tool for measuring the absorptivity of rough metal surface.