ML-driven unity/WebGL digital twin with RGB sensing for bioleaching process control
Industrial bioleaching processes often suffer from suboptimal yields and monitoring gaps due to the extreme acidity and corrosiveness of the environment. This article presents a lightweight, web-based digital twin (DT) framework for semi-industrial bioleaching optimization, in tegrating low-cost RGB...
| Autores: | , , |
|---|---|
| Tipo de recurso: | artículo |
| Fecha de publicación: | 2026 |
| País: | España |
| Institución: | Universitat Politècnica de Catalunya (UPC) |
| Repositorio: | UPCommons. Portal del coneixement obert de la UPC |
| Idioma: | inglés |
| OAI Identifier: | oai:dnet:upcommonspor::9430cc375e30c050552ba6ced90cb8d0 |
| Acceso en línea: | https://hdl.handle.net/2117/460917 https://dx.doi.org/10.1109/TII.2026.3679379 |
| Access Level: | acceso abierto |
| Palabra clave: | Bioleaching Digital twin (DT) Industry IoT Multimodal sensing Optical sensing Support vector machines (SVM) Unity-WebGL Àrees temàtiques de la UPC::Enginyeria química::Biotecnologia |
| Sumario: | Industrial bioleaching processes often suffer from suboptimal yields and monitoring gaps due to the extreme acidity and corrosiveness of the environment. This article presents a lightweight, web-based digital twin (DT) framework for semi-industrial bioleaching optimization, in tegrating low-cost RGB sensing (transformed to hue, saturation, and value space for robustness), IoT connectivity, and support vector machine (SVM) regression within a Unity-WebGL platform. To ensure industrial-grade reliability, the predictive pipeline utilizes a group-based three way split strategy to eliminate data leakage and ensure generalization to unseen experimental batches. While traditional random-split approaches often yield overfit results,our SVM-based regressorachievesageneralized R2 =0.55,providing stable and physically consistent predictions of copper concentration. A targeted ablation study demonstrates that noncontact optical sensing independently outperforms traditional pH probes (R2 = 0.52 vs. R2 = 0.42),offering critical operational resilience in corrosive media(pH <2.0). Model transparency is further validated through residual diagnostics and response surface analysis, confirming homoscedastic behavior and chemical consistency. Furthermore, a pilot evaluation indicates that the DT enables 57.3% faster anomaly detection than legacy supervisory control and data acquisition systems, facilitating proactive intervention. The framework’s decoupled WebGL architecture ensures zero-install deployment, offering a scalable blueprint for real-time bioprocess monitoring in data-scarce industrial environments. |
|---|