Cost effective detection of uneven mounting fault in rotary wing drone motors with a CNN based method

Rotary wing drones stand out among Unmanned Aerial Vehicles with their vertical landing and take-off feature and are used in many industrial applications and different sectors. Ensuring the stability of motion in these vehicles is crucial. Errors in the motor assembly can disrupt the stability of the motion in rotary wing drones. Therefore, it is essential to detect these errors during the assembly phase. In this study, we propose a cost-effective method based on deep learning to detect assembly failure of brushless direct current motors, which are widely used in rotary wing drones. A test setup representing the motor assembly defects is created and vibration data for three different speeds of the motor are obtained through a low-cost vibration sensor. The combined one- and two-dimensional deep convolutional neural network (WDD-CNN), used to classify these data was trained with the Case Western Reserve University (CWRU) dataset and the data collected in this study. The hyper-parameter settings of the network were determined using the CWRU data set and the data obtained from the experimental setup described in the paper. The network parameters of the WDD-CNN network were transferred to the Raspberry Pi micro-controller with specialized software, and the classification process was performed there. The fact that the proposed method runs on a micro-controller reduces its cost. Because there is already a micro-controller card in drones. In addition, the selected sensor is cost-effective. Thanks to these features, the proposed method is cost effective. In this classification process performed on Raspberry Pi 5, assembly errors were detected with 97–100% accuracy.