Reconstruction method for inversion problems in an acoustic tomography based temperature distribution measurement

Abstract

In industrial areas, temperature distribution information provides a powerful data support for improving system efficiency, reducing pollutant emission, ensuring safety operation, etc. As a noninvasive measurement technology, acoustic tomography (AT) has been widely used to measure temperature distribution where the efficiency of the reconstruction algorithm is crucial for the reliability of the measurement results. Different from traditional reconstruction techniques, in this paper a two-phase reconstruction method is proposed to ameliorate the reconstruction accuracy (RA). In the first phase, the measurement domain is discretized by a coarse square grid to reduce the number of unknown variables to mitigate the ill-posed nature of the AT inverse problem. By taking into consideration the inaccuracy of the measured time-of-flight data, a new cost function is constructed to improve the robustness of the estimation, and a grey wolf optimizer is used to solve the proposed cost function to obtain the temperature distribution on the coarse grid. In the second phase, the Adaboost.RT based BP neural network algorithm is developed for predicting the temperature distribution on the refined grid in accordance with the temperature distribution data estimated in the first phase. Numerical simulations and experiment measurement results validate the superiority of the proposed reconstruction algorithm in improving the robustness and RA.

Temperature Distribution Reconstruction Method for Acoustic Tomography Based on Compressed Sensing

“…During the measurement period, one transceiver acts as a transmitter and other transceivers work as receivers. Because the transceivers on the same edge of the domain to be measured cannot produce an effective signal, 14 54 independent sound paths are formed, that is M = 54. The domain to be measured is evenly divided into N = 9 × 9 = 81 cells, which means the length of slowness signal is 81.…”

“…The average relative error E ave and root mean square error E rms 14 were used to evaluate the performance of the reconstruction algorithms, which are defined by…”

“…Typical reconstruction algorithms include the least square algorithm (LSA), the algebraic reconstruction technique (ART), the simultaneous iterative reconstruction technique (SIRT), and the simultaneous algebraic reconstruction technique (SART). 5,12,14,15 To mitigate the ill-posed nature of the inverse problem, these algorithms require the number of cells contained in the measured area to be less than the number of the sound paths. Thus, these methods can hardly offer entire and exact information of temperature distribution because of their limited interpolation scope.…”

“…Liu et al 5 first reconstructed the temperature distribution using the SART on a coarse grid, and then they predicted the temperature distribution on a refined grid by the Gaussian process regression method. Liu et al 14 constructed a new cost function by employing the Fair function of M -estimation and solved the cost function by a grey wolf optimizer to obtain the temperature distribution on a coarse grid, and then they predicted the temperature distribution on a refined grid by using a BP neural network incorporating the improved Adaboost RT algorithm. For online reconstruction, Bao et al 16 proposed an offline iteration method based on the SIRT and a smooth filter.…”

Measurement of indoor air temperature distribution using acoustic travel-time tomography: Optimization of transducers location and sound-ray coverage of the room

“…ATOM utilizes the dependency of the speed of sound on air temperature and flow velocity in the medium. The average speed of sound, along the propagation paths, can be determined by traveltimes estimation of a defined acoustic signal between transducers [6,7,8]. For discretization of the temperature and flow field, the tomographic domain is divided into volumetric grid cells called voxels.…”