| The focus of the research work was on designing and optimizing a tungsten metal powder production process based on a mechanical alloying (MA) displacement reaction. In the preferred version of the process, wolframite concentrates would be used, bypassing costly chemical processing used conventionally, and is initiated at room temperature:{dollar}{dollar}rm 2(FeO)WOsb3(wolframite)+5Mg=5MgO+Fesb2Osb3+2W{dollar}{dollar} To develop the MA process, six major objectives were chosen: (1) reduce tungsten from tungsten oxide; (2) produce tungsten directly from wolframite; (3) measure the reaction kinetics; (4) develop a dynamic model for the reactions; (5) computer simulate a continuous operating system with a neural network controller for tungsten production; and (6) evaluate neural network image analysis as a method of feed back control.; The procedure was to combine reactants into SPEX or ball mills, and MA from two to forty minutes, measuring temperatures to determine reaction kinetics. Products were analyzed using SEM, EDAX and ICP chemical analysis, X-ray diffraction, and optical microscopy. For the dynamic model we included event probabilities of particle welding, fracturing, and chemical conversion and an activated complex, and numerically solved the resulting set of highly nonlinear differential equations. The model was verified using published copper oxide rate data and our tungsten kinetics. Our next procedure was to simulate a continuous process that was controlled by an advanced neural network-based model predictive controller. The procedure for image analysis was to take two dimensional Fourier transforms of optical microscope photographs of reaction products, then to extract pattern recognition features from the Fourier coefficients, and then to train Kohonen type neural networks.; Results were summarized as follows: a high tungsten content powder was successfully produced, not only from pure WO{dollar}sb3{dollar}, but more importantly, from wolframite mineral samples and flotation concentrates. Chemical reduction was successful in laboratory SPEX mills, a small planetary ball mill, and in a large, three foot internal diameter, pilot-scale ball mill. Reaction kinetics depended on reactant ratios. A thermite-reaction based explosion occurred in the large ball mill with one set of reactant ratios. The dynamic model successfully described both published copper oxide reduction data and our own kinetic results for tungsten. Computer simulations showed that a neural networks control algorithm was capable of preventing explosions in a continuous operation. Kohonen neural networks were very effective at measuring reaction conditions from images of reaction products off-line, but the implementation of an on-line image sensor was too complicated.; It was concluded that a continuous process for tungsten powder production would be more efficient and competitive than current processes, especially for emerging sporting goods markets where lead is being replaced in conventional shotgun ammunition. Future work should be directed at pilot plant testing of the continuous MA process. |
