Identifying the fault in propellers is important to keep quadrotors operating safely and efficiently. The simulation-to-reality (sim-to-real) UAV fault diagnosis methods provide a cost-effective and safe approach to detect the propeller faults. However, due to the gap between simulation and reality, classifiers trained with simulated data usually underperform in real flights. In this work, a new deep neural network (DNN) model is presented to address the above issue. It uses the difference features extracted by deep convolutional neural networks (DDCNN) to reduce the sim-to-real gap. Moreover, a new domain adaptation method is presented to further bring the distribution of the real-flight data closer to that of the simulation data. The experimental results show that the proposed approach can achieve an accuracy of 97.9\% in detecting propeller faults in real flight. Feature visualization was performed to help better understand our DDCNN model.