Self-validating thermocouples are now mature but are held back by the inability of conventional software to identify the characteristic phase-change cell melting plateau in the temperature-time data. New techniques using the state-of-the-art in machine learning will be developed to locate any instance of the melting plateau and automate the corresponding in-situ recalibration, enabling truly driftless thermometry. The target uncertainty for the in-situ calibration is 1 °C.

In in-situ gas and combustion temperature measurements, because process conditions vary continuously, it is necessarily to develop robust measurement techniques that can provide a user with a real-time data flow. This is only possible with the use of AI approaches in a complex real-time data analysis. The target uncertainty for in-process gas temperature retrievals is 2 % which is aligned with the requirements for e.g. selective non-catalytic reduction (SNCR) process control (e.g. power plants and waste incineration).

Calibration surfaces generally have a large uncertainty in the areal temperature distribution due to the complicated heat exchange at the surface. Inserts for dry-block calibrators will be developed with thermometers distributed throughout the volume, to be operated in conjunction with a neural network and digital twin which uses the data from the sensor network to determine the temperature distribution on the surface plate. The target uncertainty on the plate temperature is 1 °C.

• The first task will investigate the use of machine learning for automation of self-validating thermometers.
• In the second task AI approaches to enable retrieval of gas temperatures using spectroscopic techniques will be developed.
• In the third task neural network approaches to thermometer networks used to retrieve the temperature distribution across low-cost dry-block calibrator inserts for calibration of thermal imagers and phosphors will be developed.