Accounting for transient temperature measurement error with a high fidelity thermocouple model and application to metal/mold interfacial heat flux estimation

A high fidelity three-dimensional sensor model is developed to account for bias errors in solid-embedded thermocouples. The model yields unprecedented bias estimates. The detail of the model permits the representation of the three compositional regions of the thermocouple: the two wires and the weld. The improvement in the error estimates is demonstrated by contrasting the results with axisymmetric model results. A theory is developed to relate the sensed temperature within a thermocouple to the modeling of the thermocouple's sensed temperature.;The technique is applied to the problem of evaluating heat transfer at the cast metal/mold interface. The most expansive review ever published of metal/mold interfacial heat transfer literature is included. The methodology is demonstrated with a numerical aluminum sand casting experiment. Thermocouples configured parallel and perpendicular to the mold surface are considered. For both configurations, the RMS errors for the temperature histories improved by 93%. The RMS errors for the heat fluxes improved by 72% for the perpendicular case and by 57% for the parallel case.;The methods are applied to experimental data obtained from thermocouples installed perpendicular and parallel to the surface of aluminum sand castings. The temperature data was used to estimate the surface heat flux. The heat flux estimates increased by up to 85% for the perpendicular data and by up to 48% for the parallel data. For the perpendicular thermocouples, the RMS difference between the heat fluxes estimated with measured and corrected temperatures was 30.3 kW/m2 for the mold bottom and 56.6 kW/m2 for the top. For the parallel thermocouples, the RMS difference between the heat fluxes estimated with measured and corrected temperatures was 27.3 kW/m2 for the bottom and 18.8 kW/m2 for the top.;The kernel method for correcting thermocouple measurements is derived. This derivation yields a convolution which contains a kernel function. The kernel function can be expressed in terms of discrete numerical data in Laplace transform space. Methods for obtaining the values of the kernel in the time domain are introduced and evaluated. An adaptation of Beck's sequential function specification method is shown to be the most reliable technique for obtaining the kernel values.