Thermocouples data linearization using neural network

Abstract EN

Thermocouples are usually used for measuring temperatures in steel industry, gas turbine, diesel engine and many industrial processes.

Thermocouple usually have nonlinear Temperature-Voltage relationship (mV = f(T˚)).

However, on the monitoring side, it is required to have the inverse relationship [T˚ = f-1(mV)] to determined the actual temperature sensed by the thermocouple.

In this work the neural network is fully utilized to represent the required inverse nonlinear relationship of different and most popular thermocouples (K, J, B) Types.

Levenberg Marquardt is used as learning process to find these neural networks.

It is found that each type of thermocouples under test can be represented by a single neural network structure.

Moreover, the obtained results show the power of neural network in representing the inverse static relationship of each thermocouple that gives less than 1% of the actual measured temperature in the whole temperature range in comparison to polynomial fitting method.