Capacitive sensor-based fluid level measurement in a dynamic environment using neural network

A measurement system has been developed using a single tube capacitive sensor to accurately determine the fluid level in non-stationary tanks, namely automotive fuel tanks. The system determines the fluid level in the presence of dynamic slosh. A neural network-based approach is used to process the sensor signal and achieve substantial accuracy compared with the averaging method, which is normally used under such conditions. The sensor readings were obtained by experimentation carried out under various dynamic conditions. The sensor response was recorded at various slosh frequencies and fuel volumes; which was then used to train three different neural network topologies. Field trials were carried out to obtain the actual driving data for the purpose of testing the neural networks using MATLAB software. One static neural network topology, namely Feed-forward Backpropagation Neural Network, and two dynamic neural network topologies, namely Distributed Time Delay Neural Network and NARX Neural Network, have been investigated in this work. The developed fluid level measurement system is capable of determining the fluid level in a dynamic environment with a maximum error of 8.7% by using the two dynamic neural networks, and 0.11% using the static feed-forward backpropagation neural network.