Industrial pouring processes operate under highly dynamic conditions where small deviations can lead to defects, scrap, and production losses. Although modern foundries are equipped with multiple sensors and visual inspection systems, most monitoring approaches remain fragmented, unimodal, and difficult to interpret. Furthermore, annotated anomalous samples in industrial settings are scarce, hindering the development of traditional methods. As a result, many critical pouring anomalies are detected too late or lack sufficient contextual information for effective decision making. In this work, we propose a multimodal framework for industrial scene characterisation that combines visual information and process signals through an explainable Mixture-of-Experts (MoE)-style expert-fusion strategy. First, we deploy an ensemble of specialised modules that collaborate to identify regions of interest, assess pouring quality, and contextualise events within the production process, thereby generating an interpretable description of pouring events. Second, we introduce a novel anomaly detection method for multimodal video data, combining a self-supervised transformer with an outlier-aware clustering algorithm. Our approach effectively identifies rare anomalies without requiring extensive manual labelling. The resulting information is structured into a digital twin-ready representation, supporting synchronisation between the physical system and its virtual counterpart. This solution provides a scalable, deployable pathway to transform heterogeneous industrial data into actionable knowledge, supporting advanced monitoring, anomaly detection, and quality control in real foundry environments.

Industrial pouring processes operate under highly dynamic conditions where small deviations can lead to defects, scrap, and production losses. Although modern foundries are equipped with multiple sensors and visual inspection systems, most monitoring approaches remain fragmented, unimodal, and difficult to interpret. Furthermore, annotated anomalous samples in industrial settings are scarce, hindering the development of traditional methods. As a result, many critical pouring anomalies are detected too late or lack sufficient contextual information for effective decision making. In this work, we propose a multimodal framework for industrial scene characterization that unifies visual information and process signals through a Mixture of Experts (MoE) strategy. First, we deploy an ensemble of specialized modules that collaborate to identify regions of interest, assess pouring quality, and contextualize events within the production process, generating an interpretable description of pouring events. Second, we introduce a novel anomaly detection method for video multimodal data, combining a self-supervised transformer with an outlier-aware clustering algorithm. Our approach effectively identifies rare anomalies without requiring extensive manual labeling. The resulting information is structured into a digital-twin-ready representation, enabling seamless synchronization between the physical system and its virtual counterpart. This solution provides a scalable, deployable pathway to transform heterogeneous industrial data into actionable knowledge, supporting advanced monitoring, anomaly detection, and quality control in real foundry environments.

Metallographic evaluation of nodular cast iron is crucial for quality control in the foundry industry. Traditionally, this process relies on experts who visually interpret microscopic images. This study introduces HawkEye, a comprehensive software solution that automates metallographic analysis using advanced computer vision and deep learning models. Specifically, HawkEye software dynamically adapts its processing workflow based on the input image and its typological classification. The software supports both etched and non-etched specimens and automates the segmentation and classification of graphite nodules, gathering their morphological descriptors; it identifies microstructural phases and provides a global quality assessment. All these functions are embedded into a user-friendly interface designed for both laboratory and industrial use. Nevertheless, the key contribution of this work is the replacement of subjective evaluation with a reproducible, AI-driven approach, which significantly enhances the objectivity, traceability, and scalability of metallurgical analysis. In fact, the proposed approach achieves 99% accuracy in nodule classification compared to manual expert assessment, reduces manual image processing steps, and introduces a novel method for ferrite/perlite measurement in combination with carbide detection using YOLO and SAM models.