文献类型： 外文期刊

作者： Tang, Xiangning;Li, Wenlin;Wang, Yutang;Zhou, Qi

作者机构：

关键词： Cloud Edge Collaborative System;Edge devices;Embedded design;soft-sensing technology;oil plant production line status monitoring

期刊名称： 2025 5TH INTERNATIONAL CONFERENCE ON COMPUTER, CONTROL AND ROBOTICS, ICCCR

ISSN： -

年卷期： 2025 年

页码：

收录情况： 无来源刊（2025版）

摘要： In recent years, the automation level of oil plant processing production lines has been continuously improving, gradually replacing human physical activity. However, some mental labor that relies on the experience of operators cannot be replaced by machines, and these operations often have strong subjective intentions and lack sufficient data to support each other. Therefore, relevant successful experiences cannot be replicated, and targeted modifications cannot be made to the relevant operations. In addition, In the oil plant processing the lack of real-time online monitoring equipment for measuring the relevant components in the oil affects the digital transformation of oil plant processing. In response to the above issues, cloud edge collaboration system based on cloud servers, deep learning workstations, and embedded edge devices which was self-developed. The system takes relevant variables uploaded from oil plant production lines or laboratories as input, and through standardized and machine learning to obtains nonlinear mathematical model that can describe the relationship between relevant variables. Burn the program which contain the model into embedded edge devices that meet the environmental requirements of the production line, can prediction and monitoring the target objects of the production line. In order to predict the current of the squeezing machine, 22000 sets of process parameters were extracted from the oil production line process database established on Alibaba Cloud, such as squeezing machine current, water content of oil-seeds, squeezing chamber pressure and temperature as testing and training data for the neural network model. First using Visual software on a deep learning workstation to standardize the relevant data, and compare the validation and training losses of the relevant data under different neural network models such as MLP, LSTM, GRU, CNN to obtain the optimal nonlinear regression model. Further, using STM32CubeMx software to generate the program contain the nonlinear regression model, Last burned the program into the embedded edge device. Use this device as an edge controller to communicate with the Industrial Personal Computer through Google's Protocol Buffers protocol can get the predictive values. Comparing the predictive values of embedded edge devices, deep learning workstations, and actual value from production line. Two predicted results are basically consistent with high accuracy. It has been proven that edge devices based on embedded technology can effectively replace deep learning workstations in the application of miniaturized neural network model AI inference. This research had demonstrated the feasibility of using cloud edge collaborative systems based on embedded technology for state prediction and monitoring on digital agricultural processing production lines. It is also demonstrate this system have a broad application prospects on soft-sensing technology.

分类号：