Data-Driven Integrated Analysis in Monitoring Manufacturing Processes Using Principal Component Analysis and Multivariate Control Charts (Case Study: Mobarakeh Steel Company, Isfahan)

This research aims to design and implement a novel and integrated framework for effective monitoring and control of complex production processes in the steel industry. Given the inherent characteristics of such processes—namely their multivariate nature, strong interdependencies among variables, natural process fluctuations, and the high volume of data collected from sensors and equipment—traditional quality monitoring approaches such as I-MR and EWMA control charts often fall short. These univariate methods are typically incapable of simultaneously capturing variable interactions and frequently result in false alarms or delayed detections of subtle but significant process shifts. Consequently, leveraging multivariate and hybrid statistical approaches has become a fundamental necessity to enhance the precision, speed, and efficiency of quality monitoring systems in modern manufacturing environments.



To address these challenges, the study employed Principal Component Analysis (PCA) as a data reduction technique to transform the raw high-dimensional data into a smaller set of uncorrelated variables (principal components) that retain the majority of the data’s variability. The analysis revealed that the first three principal components (PC1, PC2, and PC3) explained most of the variance in the original dataset. These components were subsequently used as key monitoring variables and analyzed using a Multivariate Exponentially Weighted Moving Average (MEWMA) control chart. MEWMA, due to its exponential weighting mechanism and sensitivity to gradual shifts, proved to be an appropriate tool for monitoring the behavior of principal components derived from PCA.



To validate the effectiveness of the proposed hybrid model, the MEWMA outputs were compared with those of traditional univariate control charts, specifically EWMA and I-MR. Real-world process data from a major steel production line were utilized for this purpose. The comparative analysis showed that over 95% of the out-of-control points identified in the EWMA charts were also captured by the MEWMA chart, highlighting the accuracy and reliability of the hybrid approach. Furthermore, after optimizing the λ sensitivity parameter within the MEWMA model, the overlap with I-MR chart results increased to over 82%, further supporting the model’s robustness in detecting both multivariate and univariate anomalies.



One of the key innovations of this study lies in significantly reducing the number of control charts required to effectively monitor the process without compromising detection accuracy. Whereas traditional methods often necessitate the monitoring of each individual variable separately, the proposed model enables comprehensive process monitoring using only four charts—three univariate charts for the principal components and one multivariate MEWMA chart. This simplification reduces system complexity, computational burden, and analysis time for quality control teams, ultimately lowering the operational costs associated with process supervision. Additionally, the PCA component acts as a noise filter, focusing analysis on critical process features and improving the signal-to-noise ratio in monitoring systems.



From an industrial application perspective, the proposed model was implemented on a pilot basis in one of the major steel manufacturing plants. Feedback collected from production managers and quality control experts indicated a high level of satisfaction with the model’s performance. They reported reduced reaction times to process changes, fewer unnecessary production halts due to false alarms, and improved decision-making in predictive maintenance activities. The combination of early detection and actionable insight provided by the model contributed to tangible gains in process stability and product quality.



In summary, this research demonstrates that integrating advanced multivariate statistical techniques such as PCA and MEWMA with conventional univariate monitoring tools can lead to a highly effective and scalable framework for process quality control. The hybrid system not only improves monitoring precision and reduces false alarms but also streamlines data analysis and enhances the responsiveness of quality control operations. The proposed framework is not limited to the steel industry; it is highly transferable to other sectors that manage complex, data-intensive production environments—such as chemical processing, automotive manufacturing, and electronics assembly. Its adoption could pave the way for the next generation of smart quality monitoring systems, fostering higher productivity, reduced waste, and more agile industrial operations.