Intelligent Modal Analysis System: Autonomous Identification and Tracking of Aircraft Modal Parameters

Kurzfassung

Efficient aircraft design necessitates lightweight constructions. However, this leads to aeroelastic challenges concerning loads and vibrations. One example is an unstable, self-excited vibration called flutter, which can result in structural failure. Flutter results from a coupling of one or more structural dynamic modes of the aircraft structure and the unsteady aerodynamics. This aeroelastic phenomenon happens at specific flight conditions, such as altitude and speed. The damping of one mode becomes negative at the flutter speed of a particular altitude. An infinitesimally small disturbance leads to elastic deformations of the aircraft, which increase until structural failure. It is essential to experimentally identify the modal parameters in a ground vibration test (GVT) and also in a flight vibration test (FVT) to predict and avoid flutter. In GVT, multiple experienced engineers analyze measurement data in parallel to achieve rapid results, as the first flight follows shortly after the test. However, engineers with varying experience levels can lead to inconsistencies, particularly in the identified damping values. In FVT, simple automation allows for real-time estimation and monitoring of eigenfrequencies and damping values. Nevertheless, only a limited number of vibration modes are reliably identified under changing flight conditions, resulting in excessive scatter in mode tracking. Existing automation methods for modal analysis rely on clustering techniques and focus on eigenfrequencies and modes. User-dependency is typically minimized by reducing the number of required hyperparameters, e.g., threshold values. However, this reduction often compromises the discrimination of spurious modes and the robust identification of all physical modes. Consequently, these automated methods have not yet achieved the high accuracy required for identifying critical aircraft modes. This dissertation presents a novel method for autonomous modal analysis (AMA). By combining state-of-the-art modal analysis methods (Stochastic Subspace Identification and Least-Squares Complex Frequency), a robust multi-tier clustering process, and adaptive hyperparameter optimization using Gaussian processes and Bayesian optimization, AMA achieves precise identification of all modal parameters while significantly reducing analysis time and userdependency. AMA enables real-time fusion of identification methods from both time- and frequency-domain to enhance reliability and reduce uncertainty. Additionally, a Kalman filter is integrated to decrease the identification scatter further while tracking modal parameters. AMA and the data fusion methods have been successfully extended into automated analysis chains for GVT and FVT. The functionality of the new system has been validated using simulation data and during actual GVTs and FVTs. In GVT analysis, the innovative system drastically reduces analysis time while identifying all aircraft modes that engineers have conventionally identified with significant effort. Furthermore, AMA enhances the accuracy of the identified damping values. For FVT applications, AMA is optimized for rapid analysis with a run time of under two seconds, enabling real-time monitoring of modal parameters under varying flight conditions. Through data fusion methods, the system significantly reduces uncertainty and ensures reliable tracking of eigenfrequencies and damping values, thereby improving the identification of flutter curves during flight tests. The validated capability to produce reproducible AMA results with high accuracy can help standardize structural dynamic and aeroelastic identification of aircraft. However, the learning process of AMA can also be transferred to other structures, such as bridges, buildings, or wind turbines.