By the advancement to the 4th Industrial Revolution, the application of the Information technology and the drones with enhanced equipment and devices attached is drastically increasing nowadays. Especially, the drones for photographing/filming and for...
By the advancement to the 4th Industrial Revolution, the application of the Information technology and the drones with enhanced equipment and devices attached is drastically increasing nowadays. Especially, the drones for photographing/filming and for agricultural purposes are commonly seen around us and they are applied intimately within our daily lives. Moreover, the Ministry of Land, Infrastructure and Transport announced the plan for commercialization of the drone taxis until 2025 and they are estimated to transport the passengers from Incheon International Airport to Yeouido in only 20 minutes of travel time. Accordingly, by the increase of the use and application of drones generally increase in our society, at the same time so does the risk of fall/crash. The drones are composed by relatively simple structure when compared to airplanes and therefore their durability is also relatively lower and the risk of fall/crash is higher. Thus, in the operation of drones the management of risk of fall/crash is a very crucial factor. In this study there has been researched the methodology to forecast through real time monitoring, any situation which might produce the motor failure, which is the most frequent cause of drone failure. The existing studies were mostly limited in the area of electric locomotives or motors within factories.
In this thesis, there has been studied the motor vibration anomality by using the LSTM with the purpose of preventing the risk of fall/crash of the drone on the basis of the point that also in case of the drone, the risk of the fall/crash was the greatest when the motor anomality occurred. In this study, the drone was rotated in constant speed of 1,205rpm and the samples of vibration data were collected from the drone by using the acceleration sensor with 100msec cycle. The data of abnormal vibration of drone motor was saved through the FTP server, the data was learned through pre-processing process and it was composed to prove as stated in the test data. In addition, the 6 classifications of abnormal vibration of drone motor for this study were selected from the most frequent failures which occur in drone motors, and they were verified as: “Broken Rotor Bars”, “Broken Rotor”, “Shaft Unbalance”, “Faulted Bearings”, “Voltage Unbalance”, “Propeller Crack”. For the enhancement of the determination of abnormal vibration in motor, there has been selected and applied as the forecasting model, the LSTM (Long Short Term Memory) which is an artificial Neural Network based on Deep Learning. Th Tensor Flow has been applied to produce the learning model and the detected abnormal vibration of the motor was analyzed with the LSTM neural network. Moreover, by extracting the characteristics of the anomality data and through the standardization of the average values and standard deviation, the method for collection of vibration data has been reexamined. The proposed method has detected the various abnormal vibration status of the motor such as Broken Rotor Bars, Broken Rotor, Shaft Unbalance, Faulted Bearings, Voltage Unbalance, Propeller Crack, and therefore there could be proposed the method of adoption of a system for detection of abnormal vibration of the drone. Additionally, there has been compared the forecasting rate of the RNN and the LSTM with the samples of the Three (3) representative failures of the drone motor such as Broken Rotor Bars, Broken Rotor and Shaft Unbalance, and it has been verified that the forecasting rate of the LSTM was higher in R-Square value averagely 5.28% higher than that of the RNN. Accordingly, when forecasted the abnormal vibration of the motor by using the LSTM with high forecasting rate, among the Six (6) classifications of failure, in the Voltage Unbalance the forecasting rate reached 99.84%, and the failure classification with the lowest accuracy has been verified as the Broken Rotor with the forecasting rate accuracy of 97.69%. Therefore, all the Six (6) failure classifications were verified with high forecasting rate accuracy of 97% or higher.
This study on detection of drone motor anomality by using the LSTM has been based on empirical data which was collected through direct driving operation by the researcher, and therefore it can be applied in practice, and the researcher hereby expects that, if the studies on application of LSTM were advanced even more hereafter it could contribute significantly for the progress of the drone industry.