Neural intelligent control for a steel plant

Area of Science:

• Industrial Engineering
• Artificial Intelligence
• Materials Science

Background:

• Complex production processes require integrated approaches for performance improvement.
• Intelligent control, encompassing quality monitoring and optimization, enhances operations and product quality.
• Operator expertise is crucial for effective process management.

Purpose of the Study:

• To improve the performance of the Sollac hot-dip galvanizing line.
• To integrate operator skills into neural network models for enhanced process control.
• To optimize the thermal cycle and annealing furnace control.

Main Methods:

• Utilizing neural networks, including radial basis function and multilayer perceptron models.
• Employing fuzzy C-means clustering for weight learning.
• Developing a static inverse model for the annealing furnace combined with a regulation loop.
• Implementing robust learning criteria and neural network pruning for improved generalization.

Main Results:

• Successful determination of the optimal thermal alloying cycle using a radial basis function neural network.
• Effective control of the annealing furnace through a hybrid model.
• Enhanced generalization capabilities of neural networks through pruning.
• Demonstrated improvement in process performance and product quality.

Conclusions:

• Integrating operator knowledge into neural network models is effective for complex industrial process control.
• Hybrid control strategies combining neural networks and traditional methods yield superior results.
• The developed intelligent control system significantly enhances galvanizing line performance and product quality.