Abstract:We introduce a method for constructing reduced-order models directly from videos of dynamical systems. The method uses a non-intrusive tracking to isolate the motion of a user-selected part in the video of an autonomous dynamical system. In the space of delayed observations of this motion, we reconstruct a low-dimensional attracting spectral submanifold (SSM) whose internal dynamics serves as a mathematically justified reduced-order model for nearby motions of the full system. We obtain this model in a simple polynomial form that allows explicit identification of important physical system parameters, such as natural frequencies, linear and nonlinear damping and nonlinear stiffness. Beyond faithfully reproducing attracting steady states and limit cycles, our SSM-reduced models can also uncover hidden motion not seen in the video, such as unstable fixed points and unstable limit cycles forming basin boundaries. We demonstrate all these features on experimental videos of five physical systems: a double pendulum, an inverted flag in counter-flow, water sloshing in tank, a wing exhibiting aeroelastic flutter and a shimmying wheel.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is how to directly construct a reduced - order model of a nonlinear dynamical system from a video. Specifically, the author proposes a method to isolate the motion of a user - selected part from the video through non - invasive tracking techniques and reconstruct a low - dimensional attracting spectral submanifold (SSM) in the delay - observation space. The internal dynamics of this submanifold can serve as a mathematically reasonable reduced - order model for the motion near the entire system. This model is given in a simple polynomial form and can clearly identify important physical system parameters, such as natural frequencies, linear and nonlinear damping, and nonlinear stiffness.
### Background and Motivation of the Paper
Nonlinear dynamical systems are ubiquitous in many fields in nature and engineering, such as sloshing in truck tanks, turbulence, spacecraft motion, and vibrations in joint structures. Full - order modeling of these systems is usually challenging because they have a large number of degrees of freedom and uncertain physical parameters. Therefore, reduced - order modeling of nonlinear mechanical systems has become an increasingly active research area, which is of great significance in system identification, design optimization, and model - predictive control.
Traditional reduced - order methods such as proper orthogonal decomposition (POD) combined with Galerkin projection, and dynamic mode decomposition (DMD) and its variants, although effective in capturing linearized dynamics, are unable to model essentially nonlinear phenomena, such as co - existing isolated attracting fixed points, limit cycles, or transitions between states. Machine - learning methods, such as neural - network - based training, although they can provide data - driven reduced - order models, often lack physical interpretability, are prone to overfitting, and require a large amount of data and extensive parameter tuning.
In recent years, reduced - order methods based on spectral submanifolds (SSM) have emerged as a new approach to construct reduced - order models of nonlinear dynamical systems from data. SSM is the unique smoothest invariant manifold of a dynamical system tangent to the non - resonant spectral subspace near the equilibrium state. The internal dynamics of these manifolds provide mathematically exact reduced - order models, and all nearby trajectories will synchronize to these manifolds exponentially fast.
### Main Contributions of the Paper
1. **Utilization of Video Data**: The paper proposes a method for extracting trajectories from video data. This method is based on template - matching techniques and can accurately recover the trajectories of a dynamical system without changing the system behavior.
2. **Modeling of Nonlinear Dynamics**: Using the extracted trajectory data, a reduced - order model is constructed through SSM theory. This model can not only faithfully reproduce the attracting steady - states and limit cycles but also reveal hidden motions not observed in the video, such as unstable fixed points and the basin boundaries formed by unstable limit cycles.
3. **Application Examples**: The paper verifies the effectiveness of this method on experimental videos of five physical systems, including double pendulums, flags in counter - current, sloshing in water tanks, aeroelastic flutter of spreading wings, and wheel rocking.
### Method Overview
1. **Template - Matching Tracking Algorithm**: The user selects a small area in the video as a template. The algorithm rotates and translates the template to match the best position in the target frame, thereby extracting the motion trajectory of the object.
2. **Background Removal**: The background is removed by the frame - averaging method to highlight the foreground object and improve the tracking accuracy.
3. **Non - Maximum Suppression**: The non - maximum suppression (NMS) algorithm is used to remove redundant detection boxes to ensure that only the most accurate detection results are retained.
4. **SSM Reduced - Order Model Construction**: Using the extracted trajectory data, an SSM reduced - order model is constructed in the delay - embedding space. The manifold parameterization coefficients and the reduced - order dynamics coefficients are solved by minimizing the cost function.
### Conclusion
The method proposed in this paper provides a new approach for constructing reduced - order models of nonlinear dynamical systems from video data and has broad application prospects, especially in scenarios where it is difficult to install sensors, such as high - temperature, high - pressure, or high - humidity environments.