Lightweight AI Models in Mineral Processing: Classifying Hydrocyclone Underflow With ResNet-18 and MobileViT-V2
This study investigates the use of lightweight deep learning models for classifying operational states of hydrocyclones in a copper mining process. A diverse data set was curated by capturing video footage of 10 hydrocyclones under varying operational conditions, including changes in lighting and un...
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/11075588/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | This study investigates the use of lightweight deep learning models for classifying operational states of hydrocyclones in a copper mining process. A diverse data set was curated by capturing video footage of 10 hydrocyclones under varying operational conditions, including changes in lighting and underflow dynamics. The data were processed into 1217 images labeled “fanning” or “roping”. The data set was augmented to enhance model generalization, and two architectures, ResNet-18 and MobileViT-V2, were trained using a Stratified K-Fold Cross-Validation approach to ensure reliable performance evaluation. Both models were evaluated using precision, recall and F1 scores, with results averaged across all folds. The models were also tested on an independent dataset to simulate real-world classification tasks. ResNet-18 achieved an F1 score of 87.3%, while MobileViT-V2 achieved a higher F1 score of 94.5%. These results are consistent with those reported in the existing literature. Notably, the approach does not rely on ensembling methods or fixed camera positions, demonstrating that lightweight deep learning models can effectively enhance the operational control of hydrocyclone processes in mineral processing plants. This study underscores the potential of deep learning for operational monitoring in industrial settings and highlights the critical role of model fine-tuning for optimising performance in specific applications. |
|---|---|
| ISSN: | 2169-3536 |