Preflight Diagnosis of Multicopter Thrust Abnormalities Using Disturbance Observer and Gaussian Process Regression

This paper presents a preflight diagnosis method for detecting multicopter’s motor abnormalities using jig equipment data. While operating multicopters on a regular basis, determining whether it can perform the flight or not is important. For this, we use disturbance observer’s output as a feature for detecting degree of the abnormality by Gaussian process regression. During the ground inspection test where most of the disturbances are under control, motor degradation and disturbances are significantly correlated. Then, motor degradation can be estimated using the Gaussian process regression. To create multivariate output models against different degrees of motor abnormalities, we use multitask a Gaussian process regression model. To verify the performance of the proposed approach, actual preflight tests on a ground jig device developed in-house were performed with an actual quadcopter drone.