Numerical simulations were performed to study the feasibility of erosion detection in hydraulic tubes and hoses using fluid dynamic pressure response analysis. Reflected pressure signals caused by wall thinning were studied to locate and quantify pipe defects. Simulations were conducted for steel pipes as well as hoses. Results showed that for a steel pipe, since the stiffness of the fluid is much less than the pipe material’s, a very big change of wall thickness is needed to have a meaningful change in wave propagation speed and therefore the dynamic pressure response. For hoses, the wall stiffness is much less than steel pipes, hence it is more feasible to detect changes in stiffness. A dataset of 10 000 dynamic pressure impulse responses from samples with randomly generated eroded geometries was calculated to train a gated recurrent unit (GRU) neural network. Results showed that under perfect conditions (no noise), we are able to detect an eroded section’s location, length, and change in wave propagation speed with relative errors of 2.69%, 4.88%, and 3.79%, respectively. The changes in the wave propagation speed was also categorized into three classes of low, mild, and severe erosion with the accuracy of 97.3%. Under more practical conditions including sensor noise, the accuracy of erosion detection is degraded, especially in the case of steel tubing. By retraining the model with noisy data, the drop in the accuracy is compensated to about 96%.