Abstract: The high-intensity continuous operation of agricultural machinery under complex working conditions is prone to multiple types of failures. Traditional warning systems have problems such as high response delay, insufficient feature representation, and poor resource adaptability, which seriously affect the continuity of operation and equipment life. Research edge computing oriented architecture, build a four tiered topology system of "acquisition fusion diagnosis decision-making", integrate multi-source sensor distributed acquisition, multi-mode feature hierarchical fusion, lightweight hybrid model diagnosis and adaptive early warning decision-making mechanism, and achieve near source processing and early warning of fault signals. The system adapts to low-power edge device deployment through topology optimization and model compression technology, and completes 200 hours of continuous operation testing under multiple operating conditions. The results show that the system has an average inference delay of 0.068 seconds, a fault diagnosis accuracy rate of 97.2%, a misjudgment rate and a missed detection rate controlled at 2.1% and 2.7% respectively, an average warning advance of 8.4 seconds, a memory occupancy of ≤ 31.5 MB, and is suitable for various work scenarios such as rotary tillage, sowing, harvesting, and transportation. It has high real-time performance, high reliability, and engineering practicality.

Keywords: Edge computing, Agricultural machinery, Intelligent fault warning, Topology optimization, Multimodal fusion

Pages: 57-68