A Baseline Modeling and Anomaly Detection Method for Inertial Navigation System Based on Integrated 1DCNN

Flight test is an essential step before aircraft delivery. Safe flight test can not only reduce the cost of aircraft research, but also has great significance for the study of aircraft performance. In order to ensure the safety of flight test, it is necessary to implement anomaly detection for aircraft systems. Inertial navigation system, as a complex system can provide accurate position, velocity, and attitude information for aircraft during flight test. It is very important to monitor the state of inertial navigation system for ensuring the safety of flight test. However, traditional single anomaly detection models perform poorly on different test data and have poor generalization capabilities. In order to improve the generalization ability of the anomaly detection model with inertial navigation data, this paper proposes a prediction-based anomaly detection method based on an integrated one-dimensional convolutional neural network (1DCNN). Firstly, multiple training sets are divided based on the bootstrap method, and then integrated 1DCNN prediction models are built for corresponding training sets. Finally, a static threshold judgment method based on the PauTa criterion is used to detect whether any abnormality occurs. This paper verifies the effectiveness of the proposed method through experiments with real inertial navigation data. The results show that compared with the traditional 1DCNN method, the proposed method significantly improves the performance of anomaly detection and has good generalization ability on different test conditions. The TPR of anomaly data detection reaches more than 97%, which proves that the method has effective anomaly data detection for inertial data.