ACOUSTIC TEMPERATURE TOMOGRAPHY USING A UNET BASED DEEP LEARNING APPROACH

The authors of this paper propose a tomographic reconstruction method based on a UNet deep learning model. Time-of-flight measurements of acoustic waves passing through a region are utilized to image the temperature distribution in the region of interest. While established methods based on a least-squares approach require a detailed and precise system model to achieve good results, the proposed method which is created by customizing UNet architecture requires no modelling. Furthermore, a synthetic acoustic time-of-flight dataset is created using finite element based software (COMSOL) for training and validating deep learning models. The created dataset is designed to emulate the top gas temperature distribution inside a blast furnace. Compared to the popular Tikhonov least-squares method, the proposed tomographic reconstruction method achieves better accuracy when tested on the created synthetic dataset. The proposed method provides an alternative and improved way of approaching temperature tomography and inverse problems in general.