Abstract
For future space exploration programs such as NASA's Constellation Program, it is essential to develop a technique to measure the amount of fuel inside cryogenic tanks. Under a low gravity environment, common fuel gauging methods used on the Earth environment such as using a float sensor do not work. The Radio Frequency (RF) mass gauging is an attempt of measuring an amount of fuel inside a cryogenic tank from a RF tank resonance spectrum. This research demonstrates an application of Adaptive Neuro-Fuzzy Inference System (ANFIS) that constructs a Fuzzy inference system whose membership functions were adjusted by the relationship between inputs (sets of RF mode frequencies) and outputs (corresponding fuel amounts). In addition to the application of ANFIS, raw RF spectrum data were smoothed out by a moving average filter and RF mode frequencies obtained from the smoothed signals were used as the inputs for predicting the amount of fuel with a programmed algorithm built together with the ANFIS engine. Out of 56 cases tested, the ANFIS engine produced only 8 cases where the errors were exceeded over 25 pounds of the 311-lb full capacity tank. The maximum error was 94.78 lbs but the minimum error was only 1.31-lbs which is 312.21-lbs (the actual output) out of 311-lbs (the desired output). This result indicates that the proposed Neuro-Fuzzy inference engine using 2 eigen modes (1st eigen mode and 2nd mode) as 2 inputs and 4 membership functions can perform the prediction of fuel amount very accurately.