New Approaches to Compensating for the Effect of Platinum-Rhodium Thermocouple Drift

Thermocouples made with thermoelements of platinum (Pt) and its alloys with rhodium (Pt-Rh) are widely used in industrial applications. A well-known problem of Pt-Rh thermocouples is the potential drift in temperature measurement (i.e. calibration drift) due to oxidation and oxide vaporization. In this paper, in-situ calibration/correction methods are considered based on a multi-wire thermocouple structure. A multi-wire thermocouple has multiple wires joined together forming a common measurement end and can provide multiple temperature readings at any time. Hence, any two wires of the multi-wire assembly can form an individual thermocouple. In this study, multiple in-situ methods have been proposed for true temperature measurement (i.e. correcting for the drift) by exploiting the multiple temperature measurements. These are: a physical modelling method, a semi-physical method (data-driven but informed by physical modelling) and a purely data-driven method. The physical modelling method studies the link between the oxide vapor pressure and change of rhodium mass fraction due to differential vaporisation of platinum and rhodium. Based on the experimental observation that the temperature drift ratios between different individual thermocouples can be approximated by constants, true temperature can be retrieved by simple linear operations. The semi-physical approach is also informed by the physical model. It exploits the structure of a multi-wire thermocouple and estimates the true temperature as the solution to an optimization problem. The data-driven method, on the other hand, removes the need for physical modelling. It aims to cancel the thermocouple drift by exploiting the linearity between different thermocouple pairs in the multi-wire thermocouple. Statistical tests such as the co-integration test are proposed for the selection of thermocouple pairs satisfying the linear assumption. The proposed methods are tested with a five-wire thermocouple dataset. The results show that a significantly reduced error in temperature measurement has been achieved with the proposed methods.