Three‐dimensional temperature distribution reconstruction using the extreme learning machine

The temperature distribution in real‐world industrial environments is often in a three‐dimensional (3D) space, and developing an efficient algorithm to reconstruct such volume information may be beneficial for increasing the system efficiency, improving the energy saving and reducing the pollutant emission. A new algorithm is put forward for 3D temperature distribution reconstruction (3DTDR) tasks according to the given finite temperature observation data. Owing to the distinct advantages, including fast learning speed, good generalisation ability etc. a new robust regularised extreme learning machine (RRELM) algorithm is developed for the 3DTDR task. Unlike existing tomography‐based measurement methods and local point measurement technologies, the proposed algorithm can reconstruct 3D temperature distributions on the basis of the given local temperature observation information. Furthermore, different from available inverse heat transfer problems, the proposed reconstruction method does not solve complicated partial differential equations. Numerical simulation results verify the effectiveness of the RRELM algorithm for 3DTDR problems. Furthermore, the proposed RRELM method has a satisfactory robustness for the noises in the measurement data and can achieve the high‐quality reconstruction when the sampling ratio is low. As a result, an effective algorithm is introduced for the 3DTDR problem.