Multi-layer Perceptron Diagnosis Method of Strip Surface Defects Based on Biogeography-Based Optimization Algorithm.

Detecting surface defects in strip steel is an essential step in the production process, and it has consistently held a prominent position in both domestic and international contexts. Stripping surface defects has a significant impact on the product's overall appearance. Moreover, it plays a crucial role in maintaining the strip product's wear resistance, corrosion resistance and fatigue strength. Failing to address these defects would inevitably result in a reduced service life for the strip product. This paper introduces a classification and diagnosis approach using the Multi-Layer Perceptron, optimized by the Bio geography-Based Optimization algorithm (BBO), for the purpose of diagnosing strip surface defects. The multi-layer perceptron is trained by using the BBO algorithm to find the best connection weight and bias value, so that it can identify the training set and test set, and make adjustments according to different needs. This paper carried out simulation experiments on strip surface defect data set in UC I data set. In this research, we evaluated the effectiveness of the proposed approach by benchmarking it against five alternative optimization algorithms. These included particle swarm optimization, ant colony optimization, distribution estimation algorithm, genetic algorithm, and extreme value search algorithm. It becomes evident that the accuracy and speed of the proposed method have experienced substantial enhancements. The application of the BBO algorithm in optimizing the multi-layer perceptron has been demonstrated to effectively address the challenge of diagnosing strip defects. [ABSTRACT FROM AUTHOR]