Infrared thermography of extreme temperature surfaces

Many industrial processes still cannot be carried out with optimal accuracy because there do not exist accurate enough online real-time measurements to enable effective process control. This particularly applies to high-temperature processes characteristic of the metallurgical industry, but also several other industries. Today, measurements at extreme temperatures are often done only periodically and manually, close to the process, which presents an HSE challenge. Commercially available infrared cameras often operate at long wavelengths and equipment with tailor-made, expensive optics. During use in real process conditions, such equipment often generates false alarms due to unknown changes in surface emissivity or varying visibility conditions. If we succeed in research and development of custom-made measurement systems that generate large amounts of high-quality data and reliably estimate the desired process parameters in real time, these could be used for the development of digital twins that can be dynamically updated as an important step on the way to successful digitization of the metallurgical industry.

The main goal of the pre-project was to find out whether a multispectral camera operating at VIS/NIR wavelengths can provide sufficient data to enable reliable and accurate thermography of surfaces with varying emissivity at temperatures higher than 1800 °C. The pre-project goals have been achieved in their entirety through several tests carried out in close collaboration with Elkem Technology. The pre-project has demonstrated that the multispectral camera provides a sufficient data basis and identified practical challenges, accuracy and resolution requirements for selected cases, the minimum number of wavelengths at which the measurement must be made, as well as which algorithms and models are best suited to transform the raw data into the desired information. Knowledge gaps have been identified which must be addressed through subsequent main projects.

For diffusely reflective surfaces, the surface's emissivity does not vary with wavelength. However, when the measurement is made in a high-temperature environment, temperature gradients occur which require further optical modeling to determine how these affect the temperature measurement. Also, in the case of molten metal, the multispectral camera provides sufficient data to enable reliable and accurate thermography. However, the emissivity of molten metal is very low and depends on both wavelength, surface chemical composition, physical properties and turbulence. Optical modeling is required to determine how the measurement is affected by other nearby high temperature surfaces. In order to reliably convert the data into accurate and reliable information, the data must be combined with expert knowledge in radiation thermometry. This can be done in an efficient way by using knowledge-based algorithms. These algorithms form the basis for the establishment of so-called expert systems, which were one of the first successful forms of artificial intelligence software.