Inspired by the biological mechanism of spiderweb vibration perception, this study designs a spider-web topology triboelectric vibration sensor (SWT-TVS). Through bioinspired topological parameter space exploration and machine learning regression optimization, it achieves ultra-broadband effective perception (5–2000 Hz) and high-sensitivity response. A physical analytical model for eccentricity vectors was constructed through the differential electrode design. Combined with the adaptive decoupling mechanism of the ResNet-embedded dual-branch feature fusion network (DBFFN+ResNet), it eliminated the coupling effect of variable operational disturbances on fault characteristics, achieving 100% identification accuracy for shaft eccentricity parameters under strong disturbances.Through compound fault diagnosis tests on a gearbox, an overall recognition rate of 98.88% is achieved for eight types of compound bearing and gear faults. Furthermore, under multi-speed conditions (600–1600 rpm), the proposed method attains an average recognition rate of 98.6% for seven types of compound faults in full operational tasks, and maintains 92.1% accuracy in cross-speed tasks. This bioinspired structure establishes a physical foundation for high-sensitivity, broadband vibration perception. In synergy with intelligent diagnostic algorithms, it provides a highly reliable solution for vibration monitoring and fault diagnosis in high-end equipment such as aero-engines and Computer Numerical Control (CNC) machine tools.