This study presents an experimental study combining multiple sensor data and employing different deep learning algorithms to detect drone propeller damage. Under controlled conditions, vibration, acoustic and temperature data were simultaneously collected from quadcopter equipped with accelerometer, microphone and temperature sensors for a healthy propeller and 8 different faulty propellers. The data were collected for the quadcopter operating both indoors and outdoors, and external factors such as noise and wind were also included in the analysis. Short Time Fourier Transform, spectrograms and standard deviation, Kurtosis and skewness analyses were performed on the collected data in both frequency and time domains to improve the performance. After that, data was analyzed using deep learning architectures such as Convolutional Neural Networks, ResNet50, EfficientNet, MobileNetV2 and VGG16, to classify cases with damaged and intact propellers. The performance of the models was evaluated comparatively using accuracy, precision, sensitivity, and F1-score metrics. The results show that EfficientNet emerged as the most accurate and reliable, achieving an accuracy of 97.2% and the lowest log loss (0.0868), indicating both strong predictive capability and high confidence in its predictions. ResNet (87.5%) and VGG16 (89.2%) also performed well, with balanced precision and recall, making them suitable candidates for scenarios where accuracy is important, but resources are moderately constrained. Ablation and hypermeters analyses were applied to the EfficientNet architecture, which is the most successful model, and the results showed that the proposed approach should be carefully configured against modality deficiencies and that the evaluated architecture exhibits stable and reliable performance. These results demonstrate that multi-sensor fusion significantly increases detection accuracy, and deep learning models perform successfully in complex damage scenarios.