In the working scenario of multi-physical parameter monitoring,the signal acquisition of hydraulic system often has the characteristic of multiple time scales,which leads to the loss of fault information and a decrease in accuracy during the diagnostic process.Therefore,a fault diagnosis method based on deep learning and multi-sensor information fusion was proposed.A multi-headed 1DCNN network was employed to perform parallel differential feature extraction on temperature,pressure,flow and other multi-sensor signals,and the subtraction-average-based optimizer was used to determine suitable convolution kernel sizes and sliding step hyperparameters for signal input with different sampling rates,thereby achieving further adaptation on the time scale and enhancing the convergence speed of the network.In the feature fusion stage,attention mechanism was introduced to dynamically allocate weights to reduce the overfitting risk of the fused multi-sensor data.A publicly available hydraulic dataset was utilized for analysis and validation,and compared with various methods.The results indicate that the proposed method can effectively extract and utilize multi-faceted fault information from multi-sensor signals for diagnosis,it requires no reliance on expert knowledge,and has high accuracy and stability.