An underwater crushing unit loaded on the underwater cleaning robot was intended to handle marine biofouling that adhered to the surface of the ship and the dam, and a prototype was initially built. A Computational Fluid Dynamics–Discrete Element Model (CFD-DEM) was created to boost the prototype’s crushing performance, and its rationale was validated by contrasting the simulation results with the results of experimental tests. Accordingly, the primary influences on crushing performance and the laws governing their influence were investigated. The Analytical Hierarchy Process (AHP) method was then used to establish a prediction model for the comprehensive evaluation indicator of crushing performance. The AHP was used, in this case, because of its ability to generate the weight of indicators. The prediction model was a quadratic polynomial function with the rotational speed, the normal velocity component at the outlet of the propeller, the mass flow rate of the particles at the inlet of the unit, and the thickness of the bushing as independent variables. The prediction model fitting effect met the requirements after the test. The primary elements influencing the underwater crushing unit’s performance were optimized using the prediction model. The average accumulation speed of particles in the crushing unit was reduced by 59.05%, and the mass flow rate of particles at the outlet was reduced by 11.93%. The maximum wear height of the bushing was reduced by 33.36%. The specific power was up 20.88%, and the overall crushing performance was up 9.87% when compared to before optimization.