Signal conditioning of thermocouple using intelligent technique

Abstract

This paper implies using a neural network based technique for dispensation of a thermocouple signal [1]. At the stage of linearization, the thermocouple cold junction compensation and sensor transfer curve are in the same stride. For sensing the reference junction temperature a thermistor has been used and whose temperature tables are taken into consideration [2][3]. A multilayered Artificial Neural Network (ANN) has been trained using Levendberg-Marquadt algorithm [4]. For initializing the biases and weight of the ANN, to minimize measurement errors, and linearize the thermocouple by simulation methods this algorithm is as well very constructive [4]. Data of thermocouple for different material types may approach from the standard tables and must be interpolated for any readings not directly contained in these tables. The fluctuation in the hotness of the reference junction of the thermocouple have an effect on the repeatability of the thermocouple hence the reaction of the thermocouple is studied at different ambient temperatures from 0-45°C and the error due to increase in this temperature is also minimized simultaneously. As training of data increases the network becomes more capable of reducing the error close to zero. So a new technique is utilized which trains the data more and more along with increasing temperature leading to less errors than conventional or previous instances of the same program[5].