An image reconstruction technique based on best-first search for electrical capacitance tomography in the lost foam casting process

The lost foam casting (LFC) process is one of the most energy efficient casting methods used in the production of complex metal castings. The metal fill profile is an important indicator of casting quality in LFC. Hence, monitoring that profile is important to quickly discover failures in the process. In this research, an array of metallic electrodes is mounted around a target area and capacitance measuring circuits are used to measure the mutual capacitances between these electrodes. Measuring the change in capacitance values between the electrodes as grounded molten metal progresses through the foam pattern provides a simple nondestructive method of acquiring filling information.;The online monitoring of the molten metal characteristics while replacing the foam pattern is important to minimize the associated defects. X-ray imaging tomography techniques are currently being used for assessing the filling characteristics of the molten metal. However, they suffer from the natural hazards of radiation and are relatively expensive. The use of electrical capacitive tomography (ECT) techniques offers a less accurate, but cheaper and safer imaging solution. This is identified as an essential tool to enhance quality and productivity, while reducing the adverse effects of changing or altering the object in any way during the casting process.;The research work presented in this thesis is concerned with the development of a novel approach based on Best-First Search (BFS) algorithm to produce ECT images for conducting materials. In the BFS technique, a root pixel is grown until the goal distribution is reached and the distribution path is determined by minimizing an objective function at each execution level for selecting an optimal path. This solution is suited for more accurate reconstruction of ECT images during the LFC process over those obtained with conventional ECT algorithms such as iterative linear back projection (ILBP). Application of Genetic Algorithms to the BFS nonlinear solver is also explored in this work in order to steer the iterative process to obtain an optimal solution. This method produces images similar to those obtained by using only the BFS technique but requires less number of iterations. Accuracy of the inverse problem solution critically depends on the accuracy of the forward problem solution. The forward problem solution while running the BFS algorithm, which is the determination of inter-electrode capacitances given a metal distribution, is based on Artificial Neural Network (ANN). This method is relatively fast and accurate when compared to linear forward projection (LFP), a common technique used in ECT systems. Practical foundry tests are conducted using the wide-frequency capacitance measuring circuit developed at TTU. Test results obtained show that the algorithm is suitable for estimating the fill pattern during the LFC process.