| The object-oriented approach is both feasible and useful when applied to design an optimized self-learning process controller for process engineering. The conventional process controllers either become inefficient or even fail simply because they cannot adapt themselves to the new environment when the operating conditions change. Because neural networks can emulate many logical and intelligent functions existing in biological systems, this new technology holds a great promise for providing better control of engineering processes.; Although neural networks offer a very promising alternative, the lack of knowledge to design an efficient and compact architecture, when given a specific control problem, remains an obstacle to the advance of neural network research work. When a neural network is over-designed with an unnecessarily huge number of connections, it makes the structure unmanageable from a researcher's point of view. In addition, the excessive number of unknown weights also makes the training too time-consuming to be practical.; In this dissertation work, an object-oriented approach is proposed to conduct the design and research because our brains are wired to operate more efficiently in an object-oriented world. This “minor” emphasis shift will change the seemingly meaningless, massively interconnected neural network into meaningful patterns and objects that we can easily work with. A liquid level control system is implemented to demonstrate how this object-oriented approach works. |
