This paper proposes an optimised data acquisition and transmission method based on cloud-edge collaboration. By constructing a real-time data processing framework that integrates CNN algorithms, optimising the protocol conversion mechanism for heterogeneous networks, and designing a data interaction control system with business prioritisation and dynamic bandwidth allocation capabilities, the processing efficiency and transmission reliability of manufacturing data are significantly improved. In metal coating thickness detection, the CNN-based cloud-edge collaborative fusion algorithm achieved a fusion result of 61.1146 µm (reference value: 61.11 µm), with a relative error of only 0.0088%, outperforming the arithmetic mean method (0.1353%) and the evidence theory method (0.0362%). The fusion process took 0.12 ms, representing an over 80% speedup compared to traditional methods. In the 10Mb candy packaging recognition task, the cloud-edge collaborative model demonstrated comprehensive performance leadership, with a latency of only 4.36 seconds, which is 51.6% of the fog computing FC’s 8.45 seconds and 26.3% of the local computing LC’s 16.57 seconds. The energy consumption of the cloud computing CC algorithm is 329.41 J, which is 49.1% more energy-efficient than FC’s 646.61 J and 78.6% lower than LC’s 1540.72 J. The reliability task success rate is 95.11%, significantly higher than FC’s 83.23% and LC’s 65.26%. This study validated the significant advantages of the cloud-edge collaboration architecture in terms of data real-time performance, energy efficiency, and reliability, providing an effective solution for optimising intelligent manufacturing systems.
