With the continuous application of industrial Internet and software defined Internet of Things in equipment operation and maintenance scenarios, motor bearing fault diagnosis is gradually shifting from single-source monitoring to multi-source collaboration and intelligent recognition. In this paper, for heterogeneous data such as vibration, current, temperature, acoustic emission and speed, a fault diagnosis method including time alignment, multi-source representation, deep adaptive extraction, multi-scale feature fusion and dynamic weighting is constructed, and channel response calibration, context aggregation and joint loss optimization are used to improve the feature expression ability under complex conditions. The experimental results show that the inter-class distance of the proposed method reaches 3.47, the contour coefficient is 0.72, and the stability index is 0.93. The overall recognition accuracy is 97.3%, the F1 value is 96.9%, the average inference time is only 11.3 ms, and the performance of 95.9% is still maintained under the noise disturbance condition. Research shows that this method can provide effective technical support for online monitoring and intelligent operation and maintenance of motor bearings in the software-defined Internet of things environment.
