The object of study is implementation of normalizing converters based on neural network methods to increase the accuracy of temperature measurements with Chromel-Alumel thermocouples (Type K). Detailed analysis of physical and technical limitations of Type K thermocouples is conducted including nonlinearity of characteristic curve, irreversible parameters drift during high-temperature exploitation, reversible instability (hysteresis), as well as influence of cold junction temperature. Traditional linearization and error compensation methods are compared with innovative approaches based on artificial neural networks. Multilayer perceptrons (MLPs) for static error compensation and recurrent networks with long short-term memory (LSTM) for dynamic effects accounting are validated as the most effective architectures for solving the stated problems. The study demonstrates that neural network methods enable complex adaptive error compensation that can not be achieved by traditional methods, which paves the way for the development of a new generation of intelligent temperature sensors. It is concluded that type K thermocouples are highly competitive and promising in modern industrial systems in an Industry 4.0 environment, provided they are equipped with intelligent neural network converters.
Maroz et al. (2026) studied this question.