Three-dimensional temperature field inversion calculation based on an artificial intelligence algorithm
•The combination of finite element algorithms and machine learning algorithms.•A temperature field reconstruction model based on discrete boundary conditions.•A finite element program with modules applicable to the large datasets generating.•The optimal settings of hyperparameters under the general...
Uložené v:
| Vydané v: | Applied thermal engineering Ročník 225; s. 120237 |
|---|---|
| Hlavní autori: | , |
| Médium: | Journal Article |
| Jazyk: | English |
| Vydavateľské údaje: |
Elsevier Ltd
05.05.2023
|
| Predmet: | |
| ISSN: | 1359-4311 |
| On-line prístup: | Získať plný text |
| Tagy: |
Pridať tag
Žiadne tagy, Buďte prvý, kto otaguje tento záznam!
|
| Shrnutí: | •The combination of finite element algorithms and machine learning algorithms.•A temperature field reconstruction model based on discrete boundary conditions.•A finite element program with modules applicable to the large datasets generating.•The optimal settings of hyperparameters under the general model is summarised.•Data-driven temperature field calculations decoupled from the prior knowledges.
An increasing number of practical problems in heat transfer area are attributed to inverse heat transfer problems (IHTPs). One of the typical applications of the inverse problem is the prediction of the temperature field of an object from discrete temperature measurements of the surface. The study combines machine learning algorithms with numerical heat transfer methods to inversely predict the heat transfer boundary conditions of a hexahedral geometry from a finite number of discrete temperature measurements on its surface and then calculate the overall temperature field distribution. For the implementation of the numerical heat transfer forward problem, we first complete the coding of the finite element program to generate a training dataset for the inverse calculation by batch inputting the predefined boundary conditions. The inverse problem is modelled by constructing a neural network (NN) approach. The model is trained by calling data from the above dataset. The discrete temperature data are brought into the trained neural network for temperature field prediction. The results are tested for accuracy and generalisability. Finally, by comparing different hyperparameters and different training methods, a method of improving the efficiency and accuracy of the reconstruction results is proposed. The inversion calculation error is finally controlled to less than 0.1. In addition, the model implements the validation of commercial software simulation data and application to aero-engine turbine blades and vapour chambers. The approach can be extended as a generalised 3D temperature field reconstruction method. |
|---|---|
| ISSN: | 1359-4311 |
| DOI: | 10.1016/j.applthermaleng.2023.120237 |