Effective Feature Extraction and Classification Method for Backlash Anomaly in Missiles via Machine Learning

Missiles can fail on many failures such as backlash, autopilot failure, and inertial navigation system saturation in every flight part. Some of them are predictable and avoidable, but most of them occur after firing. The main purpose is to create a perfect device. However, small or big errors generally come true. This article finds out why the backlash error occurs and the variables that are most affected by it. Backlash is one of the most important issues that can be encountered during the design of high-precision systems. This issue is more significant than the others because it directly affects the missile's rotation velocity. Thus, the system shows an anomaly which is the points different from the normal state of existence. These anomalies can affect other dynamics on missiles directly or indirectly. In this paper, a dataset was prepared with Monte Carlo simulations of flights and Random Forest Algorithms, Extreme Gradient Boosting, Light Gradient Boosting, Categorical Boosting, and our combined algorithm were performed for classification. The labeling of the dataset was done by assigning all samples of flights that had backlash anomalies and assigning all samples of flights that did not have backlash anomalies in different classes. Grid Search was used for all algorithms hyperparameters tuning. After all calculations, results are presented with a benchmark, and show that the developed model outperformed. Our state-of-the-art model also gives results about flight has backlash anomaly or not and also gives feature importance.