To address the difficulty that local observation responses of different faults in servo mechanisms are similar, while mechanism differences are embedded in multi-source variable correlations and dynamic evolution processes, a servo mechanism fault diagnosis method with Mixture-of-Experts signal-semantic bridging is proposed. First, multi-source observation signals, including displacement, current, voltage and temperature rise, are uniformly represented, and semantic prototype alignment is used to establish the association between signal features and diagnostic semantics such as fault type, fault mechanism and fault location. Then, a fault-mechanism-constrained Mixture-of-Experts diagnosis model is introduced, in which local temporal context enhancement and gated routing are combined to form differentiated discriminative paths. Finally, hierarchical diagnosis results are organized into structured outputs. Simulation results show that the proposed method improves the recognition performance and hierarchical diagnosis consistency of complex fault modes in servo mechanisms, providing a new implementation approach for intelligent fault diagnosis and structured diagnostic expression.