The growing competitiveness of markets and the increasing complexity of production systems pose new challenges to quality control, demanding more effective and adaptive tools. Statistical quality control (SQC), although established as a strategic resource for reducing variability and preventing defects, it shows limitations when dealing with non-normal data distributions or when faced with data with high dimensionality. In this context, machine learning (ML) techniques have emerged as complementary solutions to enhance the detection of out-of-control and abnormal patterns in control charts. Therefore, this study aims to map the current advances, emerging trends and critical knowledge gaps in the application of ML for SQC.
This article presents a combined bibliometric and systematic review of the literature. The bibliometric analysis is conducted on an initial set of 574 articles collected from scientific databases, while the systematic literature review is based on 80 studies published between 2010 and 2024, selected through the PRISMA protocol. The analysis aims to map the main techniques employed, the predominant types of tasks and learning, hyperparameter optimization techniques and data types employed, as well as the objectives proposed by the models.
The bibliometric analysis indicates an increasing trend in scientific publications, with research primarily concentrated in the fields of industrial engineering and applied computer science. China stands out as the leading contributor in terms of publication volume and the development of more robust co-authorship networks. Keyword analysis highlights the predominant use of artificial neural networks (ANN), support vector machines (SVM) and dimensionality reduction techniques in statistical process control, as well as the recent emergence of topics aligned with Industry 5.0 and human-centric manufacturing. The systematic literature review reinforces the predominance of ANN and SVM applied to SQC, mainly addressing classification tasks for detecting in-control and out-of-control process states. Most studies rely on simulated data and supervised learning approaches, with scarce use of real-world industrial data and a limited adoption of unsupervised and adaptive learning strategies, highlighting important gaps for future research. Future research should prioritize validation with real industrial datasets, the development of unsupervised and adaptive learning approaches capable of handling dynamic and non-stationary processes, and the integration of explainability, real-time analytics and human-centric principles in ML-based SQC systems.
This study identifies research gaps and proposes directions for future work, emphasizing the development of more robust, interpretable models suitable for connected industrial systems.
