Experimental investigation and performance prediction of cryogenic temperature sensor for cryogenic rocket engine using artificial intelligence techniques

Abstract Cryogenic temperature sensors are widely adopted in cryogenic and semi-cryogenic rocket engines to measure the temperature values. Those sensors are in-house developed resistance sensors and are calibrated in between 4.2 K to 300 K to ensure the desired redundancy, accuracy, and repeatability before their authentic use in space missions. With our in-house developed wet cryogenic calibration facility, these sensors are calibrated and post-processed for plotting graphs of resistance versus temperature in appropriate ranges, and developing fitting equations. In this paper, an attempt has been made to use artificial intelligence techniques to predict the resistance as a function of temperature for the first time. The raw data set of a typical cryogenic sensor is collected from the experimental investigation and has been categorized into three sub-categories; about 70% of data is used for training, 15% data is used for testing, and the remaining 15% is used for validation purposes. Various types of back propagation schemes have been selected during the training of neural network and the effect of each of them on the prediction value has been compared. It is noticed that the accuracy of the prediction of resistance as a function of temperature is in excellent agreement with experimental data and fitting equations.