The invention discloses an anomaly detection edge computing system, relates to the field of edge computing. The system comprises a hardware subsystem and a software subsystem. The hardware subsystem comprises a data acquisition module and a computing platform. A detected signal is digitally transmitted to the computing platform through the data acquisition module, and the computing platform carries out high-complexity operation and supports data integration, a data communication interface and network service. The software subsystem runs on the computing platform, processes and analyzes the digitized signal and presents the digitized signal to the outside, and comprises a model online training module, a model execution module and a Web service. The system supports online training of the self-adaptive abnormity detection model, when industrial equipment of different types is detected or an original diagnosis model of certain equipment fails, online training can be carried out again, model parameters can be updated again, and universality and applicability requirements are met.