Zastosowanie sztucznych sieci neuronowych do wyznaczania współczynnika dyfuzyjności cieplnej

摘要

The paper presents the results of research on methods for determining the coefficient of thermal diffusivity of thermal insulating material. This method uses periodic heating as an excitation and an infrared camera for measuring the temperature distribution on the tested material surface. A three-dimensional model of the phenomenon of heat diffusion in a sample of the tested material was used for simulation investigations. To solve the coefficient inverse problem, there is proposed an idea of using an artificial neural network. The periodic heating technique is based on diffusion of a heat wave in the specimen, which is a response to the temperature harmonic excitation - Section 2. The model of a nonstationary heat flow process in the sample of material is based on a three-dimensional heat-conduction model and includes the third kind Fourier boundary conditions. For solving the system of partial differential equations, used to describe the model, the finite element method was applied[1,2,10] The usability of the artificial neural network[7,8,9,] in solving the inverse heat transfer problem in a sample of heat insulation material is presented -Section 3. The network determines the value of effective thermal diffusivity on the basis of the amplitude decrease ΔA of the thermal wave on the specimen surface, and the time in which the thermal wave moves by the distance Δz on the specimen side surface. During selection of the optimal neural network architecture many configurations were assessed. Additionally, the influence of the input quantities errors on the estimated values of the thermal parameters was also analysed using the Monte Carlo technique[12]. Simulations of the heat flow process, network training and testing processes and analysis of the error sensitivity were carried out in the Matlab environment.%W artykule przedstawiono wyniki badań nad opracowywaną metodą wyznaczenia współczynnika dyfuzyjności cieplnej materiału termoizolacyjnego. W metodzie tej wykorzystano wymuszenie temperatury o charakterze harmonicznym oraz do pomiaru rozkładu temperatury planowane jest użycie kamery termowizyjnej. W badaniach symulacyjnych wykorzystano stworzony do tego celu trójwymiarowy model zjawiska dyfuzji ciepła w badanej próbce materiału. Do rozwiązania współczynnikowego zagadnienia odwrotnego zaproponowano koncepcję wykorzystania sztucznej sieci neuronowej.