The invention relates to a fault diagnosis method for an aero-engine control system sensor based on 3DCNN-JTFA, and the method comprises the steps: building a fault generator on the basis of an aero-engine-sensor mathematic model; randomly mixing aircraft engine flight data with output signals of the fault generator to obtain a sample data set; processing the sample data set through a JTFA methodto obtain a time-frequency graph; performing label training on the sample data set according to the fault type to obtain a 3DCNN model; inputting real-time measurement data of an aero-engine control system sensor into a 3DCNN-JTFA model, and completing real-time fault detection and isolation through the steps of preprocessing, JTFA analysis to obtain a time-frequency diagram, label setting, 3DCNNclassification and the like. According to the invention, JTFA, 3DCNN and multiple binary classification technologies are combined, an offline learning and online operation mode is adopted, the methodis used in a diagnosis experiment, the method is not sensitive to noise interference, and the diagnosis efficiency is improved. Meanwhile, a fault generator is designed, so that the problem of data mismatching is solved.