Time Delay Neural Network Applied for Compensation of Temperature Sensors Nonlinear Dynamics on Hysteresis Measurements

There exists many kinds of technologies and sensors for measuring temperature, however, two of the most common types of sensors are thermocouples and thermistors. These sensors have dynamical response to a temperature change in the surroundings, and also have nonlinearities in the mapping between temperature and electrical signal at their output. When dealing with the characterization of thermal hysteretic devices, the nonlinear dynamical response of the temperature sensor may cause a severe distortion in the shape of the hysteresis curve. In this paper is presented a nonlinear inverse filtering structure using an artificial neural network for distortion correction in vanadium dioxide films thermal hysteresis measurements. A time-delay neural network is applied to obtain an approximation of the true hysteretic characteristics, removing the undesired distortion. The network training is performed using experimentally measured data. A comparison is presented between the neural method and a parametric estimation.