Inverse estimation of heat flux using linear artificial neural networks

Abstract

A modified approach is presented for estimating heat flux from an interiorly embedded thermocouple junction in the context of linear dynamical systems. Owing to the similarity of the convolution expressions describing one-dimensional feedforward heat conduction and the impulse response truncation (IRT) filter, the response of a fixed-length, linear forward heat conduction model is mathematically given based on an IRT filter. Motivated by the calibration methods proposed by Stefan Löhle, Majid Keyhani, and Jay I. Frankel et al., an alternative method is proposed for identification of the finite impulse response based on linear artificial neural networks (ANN) and calibration data from experiments. The heat flux is estimated based on the weights of the linear ANN and Beck's future-time regularization method without the knowledge of the thermophysical properties of the material, depth and inertia of the thermocouple junction, and heat losses of the sensor. Results of tests performed for validation using the short arc Xenon lamp calibration system show that, in general, the heat flux estimated based on the proposed method agree well with the known calibrated values.