Abstract:

Objectives Acoustic tomography, as a non-invasive temperature detection technology, holds significant value in industrial process monitoring. However, it is constrained by insufficient spatial resolution caused by ill-posed inversion and sensitivity to noise. To address these issues, a acoustic tomography temperature distribution based on robust regularized extreme learning machine (ELM) is proposed. Methods A two-stage reconstruction framework is established. In the first stage, the network is trained using acoustic time of flight data and low-resolution temperature data to obtain a low-resolution temperature distribution on a coarse grid. In the second stage, the network is further trained using low-resolution and high-resolution temperature distribution, enabling high-resolution temperature reconstruction on a fine grid. Numerical simulations are conducted on typical temperature field models, and the proposed method is compared with traditional algorithms, including Tikhonov regularization, Landweber algorithm, algebraic reconstruction technique (ART), and ELM algorithm. Results The robust regularized ELM algorithm achieves an average relative error of 0.28% and a root mean square error of 0.38%, significantly outperforming the other algorithms in reconstruction quality. Conclusions The acoustic tomography temperature distribution based on robust regularized ELM balances computational efficiency and reconstruction accuracy, providing a new solution for high-resolution temperature monitoring in power plant boilers and similar equipment. It demonstrates significant engineering application value, particularly under harsh conditions such as high temperature and strong interference.

Key words: power plant, power station boiler, acoustic tomography, temperature distribution, high-resolution reconstruction, robust regularized, extreme learning machine, numerical simulation, reconstruction algorithm