This study is grounded in the development of a laboratory-based simulated gearbox fault experimentation platform. Different types of gearbox malfunctions were simulated, and pertinent data on hydraulic fluid characteristics were meticulously gathered. Leveraging the BP neural network, a remote online fault diagnosis model for hydraulic fluids was meticulously crafted. The model was trained using the dataset encompassing the acquired hydraulic fluid feature information. Subsequently, the fault diagnosis model underwent meticulous optimization based on the outcomes of the training process. Experimental results unequivocally demonstrated a substantial enhancement in diagnostic accuracy when compared to conventional models post-optimization.