Application of adaptive control to gold processing plant

An adaptive control system utilizing neural network modeling and genetic algorithms has been successfully installed and tested in a gold processing plant located in Nevada. The principle design parameter of the system was feedrate improvement. After installation and commissioning, the system was able to pay for itself through increased feedrate within two weeks. The control system, utilizing a self-learning/adapting expert system shell, uses an expert system rule base written with both crisp and fuzzy logic rules complemented by continually updated predictors, and optimizers. Using a genetic algorithm-based selection process, the system is able to select among multiple competing neural network models and the expert system rules to provide optimum performance under continually varying input conditions. The paper describes the installation, rule, predictor, and optimizer configuration, and statistically verifiable testing of the control system, and the resultant performance improvements in an operating plant.