Cord-rubber composite (CRC) hoses in hose pumps used for fertigation in arid regions like Xinjiang often fail prematurely due to interfacial debonding under cyclic compression. To investigate this failure, a multiscale analysis framework was developed. The macroscopic CRC hose compression model was established using the rebar element method to represent cord reinforcement, coupled with a full-cycle fluid-structure interaction (FSI) simulation based on the Arbitrary Lagrangian-Eulerian (ALE) formulation to capture realistic pump loading and identify critical zones. A submodeling technique incorporating a cohesive zone model (CZM) with fatigue degradation rules was then applied to analyze interfacial damage evolution. Results reveal that fatigue-driven debonding is predominantly governed by shear strain in the x-z plane. Based on entropy-weighted fatigue damage indices, a machine learning-assisted multi-objective optimization was conducted to refine the hose cross-section. The optimized design reduces total interfacial fatigue damage by 2.75% and increases stiffness by 0.26%, effectively extending service life and suppressing pulsation amplification. This work provides a practical failure diagnosis methodology and an engineering-oriented redesign strategy for improving the reliability of the CRC hoses in agricultural pumping systems.