Optimizing quality control through the integration of neural networks and Lean Six Sigma: a case study in the steel industry

Abstract

Steelmaking processes, due to their continuous and highly automated nature, generate a large amount of data with exponential growth, making the use of advanced tools essential for its analysis and utilization. In this context, this research examines the implementation of artificial intelligence (AI) through neural networks in an industrial process within the steelmaking sector, utilizing a hybrid approach that combines Lean Manufacturing and Six Sigma to detect defects during the rolling process at an international structural steel manufacturing company. By applying a structured diagnostic approach using DMAIC (Define, Measure, Analyze, Improve, and Control), the critical defects with the most significant impact were identified. This finding guided the design of an AI-based solution aimed at modeling the high variability inherent in the steelmaking process. The proposal evaluates the use of the Sigmoid and ReLU activation functions for binary defect classification, testing different types of optimizers such as Adam and SGD, and training different network models by evaluating network architectures and configurations. The results demonstrate that integrating AI with continuous improvement methodologies can optimize critical processes in complex industrial environments. The proposed model reflects its adaptability and scalability for integration into real-time monitoring systems, offering optimal efficiencies for identifying defective products. The methodological approach of this proposal represents a significant contribution to the body of technical knowledge, particularly in improving critical processes with high variability and continuous production, as well as complex and costly production lines, which are common in the steel sector.