A feature based solution to Forward Problem in Electrical Capacitance Tomography

A new feature based technique is introduced to solve the forward problem (FP) in electrical capacitance tomography with a target application of monitoring the metal-fill profile in lost foam casting (LFC) process. The new technique for solving FP is based on extracting key features from given metal distributions and then training a neural network with these features. The output of neural network is a scaling factor that modifies the linear sensitivity matrix traditionally used in the solution of the FP. The training and testing data is generated through ANSYS and MATLAB simulations. This approach shows promising results. The neural network was able to learn the effect of these features on scaling factor. The RMS error for training distribution was 1.94% and for test distribution, it was in between 2% to 15% depending on the electrode pair with an average of 5%.