This paper presents an interpretable diagnostic method for rail fastening systems based on the fusion of local Higuchi fractal dimension and continuous wavelet transform energy calculated from the derivative of a transverse differential rail-profile signal. Normalized features extracted from the fastening zone are combined using a conservative AND rule and a weighted logistic indicator with a Youden threshold. The validation was performed on 1100 measurements covering K, SKL and SB fastening systems, concrete and wooden sleepers, different railway-line categories and measuring-draisine speeds of 60-110 km/h. The dataset includes varied defect types, borderline cases, signal disturbances and normal fastenings recorded under difficult conditions. The results show that the weighted fractal-wavelet fusion increases defect sensitivity compared with the AND rule, while retaining a transparent interpretation of false alarms and missed detections. Across all 1,100 measurements, weighted fusion at Z ≥ 0.43 increased TPR from 0.427 to 0.845 and F1 from 0.568 to 0.767 relative to the AND rule, with accuracy rising from 0.740 to 0.795 and AUC reaching 0.909, although FPR increased from 0.052 to 0.239; McNemar’s test confirmed a significant change in the error profile (p = 0.0008). The method is intended as a decision-support layer for targeted inspection rather than as a fully autonomous replacement for field verification.