![]() | Only 14 pages are availabe for public view |
Abstract Work on artificial neural networks has been motivated right from its inception by the recognition that the human brain computes in an entirely different way from the conventional digital computer. The brain is a highly complex, nonlinear, and parallel computer (information - processing system). A neural network is a massively parallel distributed processor made up of simple processing units, which has a natural propensity for storing experiential knowledge and making it available for use. It resembles the brain in two respects: 1. Knowledge is acquired by the network from its environment through a learning process. 2. Interneuron connection strengths, known as synaptic weights, are used to store the acquired knowledge. The manufacturing systems developed considerably during the last decade. They became extremely complex and difficult to control. In today’s highly competitive market, quality and productivity have become the pnmary goals for which manufacturers strive vigorously. To answer these requirements, it is necessary to have powerful tools for modeling, analysis, and control. The intent of thesis is to develop a neural network based - control schema to model and control, complex and continuous manufacturing systems using artificial neural networks. Modeling any complex system of manufacturing systems by neural networks starts by modeling separately the basic elements of these systems by neural models. These different models are then connected together to form the overall studied system. This . approach is illustrated using an example of open / closed loop modeling manufacturing systems, model that is consisting of four stations; each one is made up of a stock with limited capacity and a machine. The thesis develops an integration activation function in neural networks to model continuous manufacturing systems, develop a min (x, y) function and a learning algorithm to control modeled manufacturing systems. The learning algorithm depends on backpropagation learning algorithm with some modifications to be effective only on some weights. The simulation results for the modeled manufacturing systems show also the control applied to the model. Neural network controller has the better minimum energy and smoothing with changing the speed of machines than the petri net controller. The thesis presents two other applications. The first application is to develop an intelligent controller to control ship direction. This requires a procedure to acquire the control rules for a moving ship to avoid collision with another moving object and then steer back to reach a certain destination. This objective can be achieved by applying the proposed modeling and control algorithms. The second application is to develop an intelligent controller for missile tracking. The proposed neural network controller has the best maximum proximity, and smoothing in direction compared with expert and fuzzy controllers, but low sum of energy. Simulation results for the motion of the controlled ship using the neural network controller showed improved performance of the motion control and the potential of the proposed control schema of the ship. |