Abstract:Due to the complex structure of the Construction Machinery Hydraulic System, failures are inevitable. In the past, manual inspection combined with downtime maintenance was a time-consuming and labor-intensive method. In this paper, we modeled and simplified the hydraulic system of construction machinerythe, which was combined with the BP neural network to learn the fault data. Pressure and flow monitors were installed in the pipeline system to track operational data. Fault conditions were simulated by modifying the parameters of selected components, and the data from these monitors were recorded. After organizing this data, principal component analysis (PCA) was employed to extract information and reduce dimensionality for use as input to neural networks. Additionally, the current operational status of the components, whether faulty or not, was manually labeled to serve as the ground truth labels. For each component, a separate neural network was constructed to learn the relationship between its input data and the corresponding labels. Results obtained on the test set show a high accuracy rate of 98.61% for valves and 96.52% for pumps.