An Edge-Intelligent Flotation Monitoring and Control System based on 8051 MCU and Offline EfficientNet-B5 Inference on Raspberry Pi

Boyu Yang; Fangyuan Yao; Wenqi Bai; Yuxin Zhang; Hang Zhu; Zhaoning Yin

doi:10.6919/ICJE.202506_11(6).0046

Authors

Boyu Yang
Fangyuan Yao
Wenqi Bai
Yuxin Zhang
Hang Zhu
Zhaoning Yin

DOI:

https://doi.org/10.6919/ICJE.202506_11(6).0046

Keywords:

Edge Computing; 8051 MCU; Raspberry Pi; EfficientNet-B5; Flotation Monitoring; Intelligent Control; Embedded System; Offline Inference.

Abstract

To address the limitations of traditional flotation processes such as delayed feedback and low detection precision, this paper proposes an edge-intelligent flotation monitoring and control system. The system integrates a low-power 8051 microcontroller unit (MCU) for real-time multimodal sensor data acquisition with a Raspberry Pi for local image-based state recognition. An EfficientNet-B5 model, pre-trained and quantized via TensorFlow Lite, is deployed offline on the Raspberry Pi to enable high-accuracy flotation froth classification with an average inference latency of 250–350 ms. Through attention-based feature refinement and float32-quantized model optimization, the proposed system achieves a 67.5% reduction in memory usage and an 18.4% decrease in average inference time compared to the original model. In addition, the mean and peak power consumption are reduced by 18.3% and 9.5%, respectively, ensuring stable performance in low-power embedded environments. A Tkinter-based GUI further supports both manual and automatic control modes, enabling intelligent adjustments in flotation operations. The proposed architecture realizes a closed-loop control chain from sensing to decision-making and actuation, offering a scalable and energy-efficient solution for intelligent mineral processing at the edge.

Downloads

Download data is not yet available.

References

[1] Chen L L, Han J W, Qin W Q, Liu W. Comprehensive utilization of lead-zinc smelting slag: A review[J]. Nonferrous Metals (Mineral Processing Section), 2021, (3): 49–55. DOI:10.13779/j.cnki.issn1001-0076.2021.03.007.

[2] Peng Y, Bradshaw D. Froth flotation: A century of innovation and its recent trends in machine vision and intelligent control[J]. Minerals Engineering, 2020, 146: 106139.

[3] Zhang H, Liu Q, Zhou D. Adaptive reagent control in flotation using real-time image analysis and probabilistic models[J]. Journal of Cleaner Production, 2022, 363: 132543.

[4] Yang Z, Zhang X, Wang Y, et al. Flotation froth image classification using convolutional neural networks[J]. Minerals Engineering, 2020, 157: 106530.

[5] Shi F, Xiao W, Hu Y, et al. Intelligent recognition of flotation working conditions based on froth images and deep CNNs[J]. Advanced Powder Technology, 2021, 32(3): 964–973.

[6] Liu J, Zhou L, Zhang C. Challenges in deploying deep learning for flotation control in mining environments[J]. Journal of Intelligent Manufacturing, 2022, 33(1): 157–170.

[7] Huang D, Zhao Y, Chen Y. Domain adaptation for robust froth recognition in real-world flotation scenarios[J]. Expert Systems with Applications, 2023, 213: 119034.

[8] Wang M, Sun Q, Liu Z. Lightweight CNN deployment on embedded platforms for industrial visual tasks[J]. IEEE Transactions on Industrial Informatics, 2021, 17(10): 6812–6821.

[9] Banerjee A, Panigrahi S, Kar A, et al. Edge computing in mining: Opportunities, challenges and application potential[J]. IEEE Access, 2021, 9: 104276–104288.

[10] Kumar A, Singh S, Chauhan H. Embedded deep learning for fault detection using Raspberry Pi and OpenCV[J]. Procedia Computer Science, 2020, 171: 2057–2064.

[11] Mahmoud M, Al-Ayyoub M, Jararweh Y. Deploying CNNs on low-resource hardware: An empirical study with Raspberry Pi and Arduino platforms[J]. Future Generation Computer Systems, 2022, 128: 313–324.

[12] Tang J, Xiao T, Duan Z, et al. Towards real-time object detection on low-power devices: Quantization-aware training and model optimization[J]. IEEE Transactions on Industrial Informatics, 2021, 17(12): 8402–8413.

[13] Tan M, Le Q V. EfficientNet: Rethinking model scaling for convolutional neural networks[J]. International Conference on Machine Learning (ICML), 2020: 6105–6114.

[14] Tan M, Le Q V. EfficientNet: Improving accuracy and efficiency with a new family of models[J]. arXiv preprint, arXiv:1905.11946, 2020.

[15] Chowdhery A, Warden P, Shlens J, et al. Visual wake words dataset and efficient models for small-footprint keyword spotting[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2021, 43(5): 1747–1759.

[16] Zhang Y, Wang H, Xie M. Lightweight and efficient CNN deployment on Raspberry Pi using TensorFlow Lite[J]. Embedded Systems Letters, 2022, 14(2): 56–60.

[17] Yanten C, Kracht W, Diaz G, Lois-Morales P, Egana A. Froth images from flotation laboratory test in Magotteaux cell[J]. Data, 2023, 8(4): 69. https://doi.org/10.3390/data8040069