To attain a successful width control for hot strip mill, it is important to control a width at each stand preventing excessive tension imbalance for some specified stands. Our research goal is to develop a multivariable control algorithm for the new width control system. In the proposed algorithm, the width is controlled by modifying the inter-stands tension, which value is derived from the measured width and the on-line calculated width. An optimal control principle is applied and the tension balance can be maintained by tuning the weighting matrix in the performance index. The on-line control algorithm is designed such that the tension modification value can be obtained in short time, compared with the optimal regulator. This calculation is realized using a formula manipulating procedure and the control performance is same to that by the optimal regulator. The control system based on this algorithm has been applied to the hot strip mill and the width deviation can be remarkably reduced than before.
A generalized predictive control (GPC) which can be designed in a state-space framework is proposed for single-input/single-output CARIMA systems. It is first shown that, in designing the steady-state 1-step Kalman predictor by applying the innovation model, there are two ways depending on whether the output value of the plant is directly used or not ; one is called here “direct output method” and another is called “output deviation method”. When solving the 1-step predicted estimate of the original output, the latter method can be further classified into two approaches which differ in respect of the interpretation of the predicted estimate for the output deviation. Thus, three different j-step ahead state-space output predictors can be designed for the GPC strategy. Finally, the performances of three GPCs are compared through some numerical simulations.
This paper presents a learning algorithm for neural networks having general feedback connections. Simulation results are presented to compare the algorithm with the existing backpropagation algorithm; the former is shown to have a greater capability in realizing complex tasks.
This paper deals with rule-based on-line scheduling of Flexible Manufacturing Systems (FMS). Operations research methods and simulation methods, which have been often used in off-line scheduling, cannot be applied to on-line scheduling, because of the computational complexity. To cope with this difficulty, we propose to apply a rule base to resolve the conflicts arised in machine tools or transports. To generate and evaluate the rule base for on-line schedulings, we construct an FMS simulation system to describe stochastic behavior, such as failures of machine tools, repair time and variations of processing times using hierarchical stochastic Petri nets. By using the FMS simulation system, we construct and evaluate the rule base for on-line scheduling.
The modality constrained programming (MCP) problem is previously proposed as the base model for unifying various fuzzy mathematical programming (FMP) problems. The relationships between MCP problems and six kinds of FMP problems have been discussed. However, some FMP problems, such as robust programming problem, FMP problem using fuzzy max, etc., did not have the relationships to MCP problems. In this paper, introducing the Gödel implication into modality measures, a fuzzy relation between elements is extended to eight kinds of relations between possibility distributions. MCP problems are reformulated using these eight kinds of relations. The relationships between reformulated MCP problems and four kinds of FMP problems, i. e., robust programming problem, FMP problem using fuzzy max, FMP problem proposed by Luhandjula and modality goal programming problem, are investigated. Itis emphasized that most of FMP problems are interpreted in the framework of MCP problems and formulated based on the modal concept.