An Intelligent Quadcopter Unbalance Classification Method Based on Stochastic Gradient Descent Logistic Regression

As the global use of Unmanned Aerial Vehicles (UAVs) expands rapidly across a vast array of applications, it becomes imperative to identify their flaws and imbalances. This study presents and demonstrates the efficacy of an Optimized Stochastic Gradient Descent Logistic Regression (OSGD-LR) machine learning algorithm for classifying unbalances in quadrotor UAVs. Utilizing vibration-based signal processing characteristics in the time domain necessitated the use of a 3-axis accelerometer sensor and a data acquisition device. As the proposed model is tested with the quadcopter in hover mode at a fixed motor speed, minor and major imbalance conditions are detected. The proposed model tests, trains, and accurately predicts the flight's status as either balanced or unbalanced. In addition, diverse unbalanced blade scenarios were used to demonstrate the model's reliability in identifying the locations of imbalanced blades. In terms of classifying unbalanced UAV blades, the OSGD-LR is significantly more precise than a standard logistic regression method, as demonstrated by experimental results.