Multimodal computer vision and machine learning for predictive monitoring of planar-flow casting of amorphous steel foils
- 1CSIRO Technology, Hobart, Tasmania, Australia
- 2CSIRO Technology, Clayton South, Victoria, Australia
Abstract
Background: Planar-flow casting (PFC) is widely used for manufacturing amorphous steel foils, where small variations in process conditions can rapidly affect foil quality. Existing monitoring approaches primarily rely on heuristic measurements or offline inspection, providing limited capability for quantitative process prediction and early identification of unstable operating conditions. Methodology/Principal Findings: We present a multimodal computer vision and machine learning framework for predictive monitoring of industrial PFC using synchronised visible and infrared imagery acquired during full-scale production. More than 300,000 RGB and thermal image frames collected from approximately 200 production runs were processed to extract process-informed geometric, thermal and texture descriptors, including ridge density, ridge width, reflection characteristics and thermal-gradient features. Conventional image-processing baselines were compared with gradient-boosted decision trees (XGBoost) and a multimodal deep-learning fusion model combining RGB, infrared and process metadata. Both learning-based approaches substantially improved prediction of the nozzle-to-wheel gap compared with conventional computer-vision heuristics, achieving mean absolute errors of approximately 0.01 mm and R2 values of approximately 0.97. In contrast, prediction of the optical-flow-based melt-flow proxy remained substantially more challenging, indicating that this variable is influenced by factors not fully represented in the available imaging modalities. Feature-importance, calibration and ablation analyses demonstrated that ridge-density and thermal-gradient descriptors contributed most strongly to predictive performance, while complementary analyses showed that unstable production exhibited broader ridge-width distributions and larger edge-to-centre thermal gradients than stable operation. Conclusions/Significance: The proposed framework provides accurate, interpretable and non-contact predictive monitoring for industrial planar-flow casting while offering quantitative insights into process behaviour beyond conventional image-based inspection. The results demonstrate the value of combining multimodal imaging with machine learning for intelligent monitoring of advanced manufacturing processes and establish a foundation for future real-time process control and digital manufacturing applications.
Related
- Reactive and predictive processes during unpredictable driving hazards in virtual reality: an exploratory brain and body study with multimodal neurophysiological monitoring
- Metabolite-centric Knowledge Graph Unveils Predictive and Mechanistic Insights through Self-supervised Multimodal Learning
- A multimodal deep learning for rice crop health monitoring using vegetation index feature selection
- Real-time noise cancellation for EEG monitoring and brain-computer interfaces
- A multimodal computer-vision based pipeline for eye-hand coordination quantification during the Nine-Hole Peg Test
BibTeX
@article{dutta2026multimodal,
title = {{Multimodal computer vision and machine learning for predictive monitoring of planar-flow casting of amorphous steel foils}},
author = {Ritaban Dutta and Michael Kellam and Daniel Liang},
journal = {PLOS One},
year = {2026},
month = nov,
doi = {10.1371/journal.pone.0347230},
url = {https://doi.org/10.1371/journal.pone.0347230},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Multimodal computer vision and machine learning for predictive monitoring of planar-flow casting of amorphous steel foils
AU - Ritaban Dutta
AU - Michael Kellam
AU - Daniel Liang
PY - 2026
DA - 2026/11/03
JO - PLOS One
DO - 10.1371/journal.pone.0347230
UR - https://doi.org/10.1371/journal.pone.0347230
AB - Background: Planar-flow casting (PFC) is widely used for manufacturing amorphous steel foils, where small variations in process conditions can rapidly affect foil quality. Existing monitoring approaches primarily rely on heuristic measurements or offline inspection, providing limited capability for quantitative process prediction and early identification of unstable operating conditions. Methodology/Principal Findings: We present a multimodal computer vision and machine learning framework for predictive monitoring of industrial PFC using synchronised visible and infrared imagery acquired during full-scale production. More than 300,000 RGB and thermal image frames collected from approximately 200 production runs were processed to extract process-informed geometric, thermal and texture descriptors, including ridge density, ridge width, reflection characteristics and thermal-gradient features. Conventional image-processing baselines were compared with gradient-boosted decision trees (XGBoost) and a multimodal deep-learning fusion model combining RGB, infrared and process metadata. Both learning-based approaches substantially improved prediction of the nozzle-to-wheel gap compared with conventional computer-vision heuristics, achieving mean absolute errors of approximately 0.01 mm and R2 values of approximately 0.97. In contrast, prediction of the optical-flow-based melt-flow proxy remained substantially more challenging, indicating that this variable is influenced by factors not fully represented in the available imaging modalities. Feature-importance, calibration and ablation analyses demonstrated that ridge-density and thermal-gradient descriptors contributed most strongly to predictive performance, while complementary analyses showed that unstable production exhibited broader ridge-width distributions and larger edge-to-centre thermal gradients than stable operation. Conclusions/Significance: The proposed framework provides accurate, interpretable and non-contact predictive monitoring for industrial planar-flow casting while offering quantitative insights into process behaviour beyond conventional image-based inspection. The results demonstrate the value of combining multimodal imaging with machine learning for intelligent monitoring of advanced manufacturing processes and establish a foundation for future real-time process control and digital manufacturing applications.
SN - 1932-6203
ER -