Nonlinear Model Order Reduction Research Articles

Lagrangian operator inference enhanced with structure-preserving machine learning for nonintrusive model reduction of mechanical systems

Complex mechanical systems often exhibit strongly nonlinear behavior due to the presence of nonlinearities in the energy dissipation mechanisms, material constitutive relationships, or geometric/connectivity mechanics. Numerical modeling of these systems leads to nonlinear full-order models that possess an underlying Lagrangian structure. This work proposes a Lagrangian operator inference method enhanced with structure-preserving machine learning to learn nonlinear reduced-order models (ROMs) of nonlinear mechanical systems. This two-step approach first learns the best-fit linear Lagrangian ROM via Lagrangian operator inference and then presents a structure-preserving machine learning method to learn nonlinearities in the reduced space. The proposed approach can learn a structure-preserving nonlinear ROM purely from data, unlike the existing operator inference approaches that require knowledge about the mathematical form of nonlinear terms. From a machine learning perspective, it accelerates the training of the structure-preserving neural network by providing an informed prior (i.e., the linear Lagrangian ROM structure), and it reduces the computational cost of the network training by operating on the reduced space. The method is first demonstrated on two simulated examples: a conservative nonlinear rod model and a two-dimensional nonlinear membrane with nonlinear internal damping. Finally, the method is demonstrated on an experimental dataset consisting of digital image correlation measurements taken from a lap-joint beam structure from which a predictive model is learned that captures amplitude-dependent frequency and damping characteristics accurately. The numerical results demonstrate that the proposed approach yields generalizable nonlinear ROMs that exhibit bounded energy error, capture the nonlinear characteristics reliably, and provide accurate long-time predictions outside the training data regime.

Analysis and control of large-signal stability with storage in small scale islanded AC microgrids

The analysis of small-signal stability in AC microgrids has yielded numerous findings. However, investigating large-signal stability remains a major difficulty and has received limited attention in research. The article presents the establishment of a non-linear reduced-order model for small-scale AC microgrids with storage. The model considers the changes in converter structure that occur during large disturbances. Next, we will analyse the system’s stability under large-signal conditions and identify the underlying unstable mechanism. The impact of limitation links on system stability is significant, as they result in structural changes. The proposal of “virtual synchronous damping control” (VSD) aims to improve the stability of AC microgrids by addressing the issues caused by an unstable mechanism. The validation of the proposed analysis and control methods are demonstrated by examining relevant case studies in MATLAB/Simulink.

Weakly nonlinear analysis of thermoacoustic oscillations in can-annular combustors

Can-annular combustors feature clusters of thermoacoustic eigenvalues, which originate from the weak acoustic coupling between $N$ identical cans at the downstream end. When instabilities occur, one needs to consider the nonlinear interaction between all $N$ modes in the unstable cluster in order to predict the steady-state behaviour. A nonlinear reduced-order model for the analysis of this phenomenon is developed, based on the balance equations for acoustic mass, momentum and energy. Its linearisation yields explicit expressions for the $N$ complex-valued eigenfrequencies that form a cluster. To treat the nonlinear equations semianalytically, a Galerkin projection is performed, resulting in a nonlinear system of $N$ coupled oscillators. Each oscillator represents the dynamics of a global mode that oscillates in the whole can-annular combustor. The analytical expressions of the equations reveal how the geometrical and thermofluid parameters affect the thermoacoustic response of the system. To gain further insights, the method of averaging is applied to obtain equations for the slow-time dynamics of the amplitude and phase of each mode. The averaged system, whose solutions compare very well with those of the full oscillator equations, is shown to be able to predict complex transient dynamics. A variety of dynamical states are identified in the steady-state oscillatory regime, including push–push (in-phase) and spinning oscillations. Notably, the averaged equations are able to predict the existence of synchronised states. These states occur when the frequencies of two (or more) unstable modes with nominally different frequencies lock onto a common frequency as a result of nonlinear interactions.

Non-intrusive, transferable model for coupled turbulent channel-porous media flow based upon neural networks

Turbulent flow over permeable interfaces is omnipresent featuring complex flow topology. In this work, a data-driven, end-to-end machine learning model has been developed to model the turbulent flow in porous media. For the same, we have derived a non-linear reduced order model (ROM) with a deep convolution autoencoder. This model can reduce highly resolved spatial dimensions, which is a prerequisite for direct numerical simulation, by 99%. A downstream recurrent neural network has been trained to capture the temporal trend of reduced modes; thus, it is able to provide future evolution of modes. We further evaluate the trained model's capability on a newer dataset with a different porosity. In such cases, fine-tuning could reduce the efforts (up to two-order of magnitude) to train a model with limited dataset (10%) and knowledge and still show a good agreement on the mean velocity profile. Especially, the fine-tuned model shows a better agreement in the porous domain than the channel and interface areas indicating the topological feature is less challenging for training than the multi-scale nature of the turbulent flows. Leveraging the current model, we find that even quick fine-tuning achieves an impressive order-of-magnitude reduction in training time by approximately O(102) and still results in effective flow predictions. This promising discovery encourages the fast development of a substantial amount of data-driven models tailored for various types of porous media. The diminished training time substantially lowers the computational cost when dealing with changing porous topologies, making it feasible to systematically explore interface engineering with different types of porous media.

Hyper-reduction for parametrized transport dominated problems via adaptive reduced meshes

We propose an efficient residual minimization technique for the nonlinear model-order reduction of parameterized hyperbolic partial differential equations. Our nonlinear approximation space is spanned by snapshots functions over spatial transformations, and we compute our reduced approximation via residual minimization. To speedup the residual minimization, we compute and minimize the residual on a (preferably small) subset of the mesh, the so-called reduced mesh. We show that, similar to the solution, the residual also exhibits transport-type behaviour. To account for this behaviour, we introduce adaptivity in the reduced mesh by “moving” it along the spatial domain depending on the parameter value. Numerical experiments showcase the effectiveness of our method and the inaccuracies resulting from a non-adaptive reduced mesh.

Domain Decomposition Strategy for Combining Nonlinear and Linear Reduced-Order Models

Increasingly ubiquitous reliance on expensive, high-fidelity numerical simulations has led to the emergence of reduced-order modeling as an effective method to predict high-dimensional, spatially discretized field outputs in an efficient manner. This study presents a method to construct parametric, nonintrusive reduced-order models (ROMs) by leveraging domain decomposition (DD) to enable using multiple dimension reduction (DR) techniques within different spatial subdomains of the field to construct a single predictive ROM. To enforce smoothness of solutions across subdomains, points at domain interfaces are reconstructed using the data-driven gappy proper orthogonal decomposition method. The method is assessed on three high-dimensional test cases with shocks, including one with two-dimensional turbulent flow around the RAE2822. Results show that fully nonlinear ROMs predict fields more accurately in the vicinity of discontinuous features compared to their linear counterparts, which are more accurate in regions that are far from these discontinuities. Furthermore, irrespective of the DR method, DD-based ROMs outperformed their global counterparts in all test cases. Finally, models that used a combination of DR methods simultaneously showed superior performance near shockwaves and a significant improvement in total error compared to existing approaches that employ a single DR method in all subdomains.

A graph convolutional autoencoder approach to model order reduction for parametrized PDEs

The present work proposes a framework for nonlinear model order reduction based on a Graph Convolutional Autoencoder (GCA-ROM). In the reduced order modeling (ROM) context, one is interested in obtaining real-time and many-query evaluations of parametric Partial Differential Equations (PDEs). Linear techniques such as Proper Orthogonal Decomposition (POD) and Greedy algorithms have been analyzed thoroughly, but they are more suitable when dealing with linear and affine models showing a fast decay of the Kolmogorov n-width. On one hand, the autoencoder architecture represents a nonlinear generalization of the POD compression procedure, allowing one to encode the main information in a latent set of variables while extracting their main features. On the other hand, Graph Neural Networks (GNNs) constitute a natural framework for studying PDE solutions defined on unstructured meshes. Here, we develop a non-intrusive and data-driven nonlinear reduction approach, exploiting GNNs to encode the reduced manifold and enable fast evaluations of parametrized PDEs. We show the capabilities of the methodology for several models: linear/nonlinear and scalar/vector problems with fast/slow decay in the physically and geometrically parametrized setting. The main properties of our approach consist of (i) high generalizability in the low-data regime even for complex behaviors, (ii) physical compliance with general unstructured grids, and (iii) exploitation of pooling and un-pooling operations to learn from scattered data.

Data-driven nonlinear parametric model order reduction framework using deep hierarchical variational autoencoder

AbstractA data-driven parametric model order reduction (MOR) method using a deep artificial neural network is proposed. The present network, a least-squares hierarchical variational autoencoder (LSH-VAE), is capable of performing nonlinear MOR for the parametric interpolation of a nonlinear dynamic system with a significant number of degrees of freedom. LSH-VAE differs from existing networks in two major respects: a deep hierarchical structure and a hybrid weighted, probabilistic loss function. The enhancements result in significantly improved accuracy and stability compared with conventional nonlinear MOR methods, autoencoders, and variational autoencoders. In LSH-VAE, the parametric MOR framework is based on the spherically linear interpolation of the latent manifold. The present framework is validated and evaluated on three nonlinear and multiphysics dynamic systems. First, the present framework is evaluated on the fluid–structure interaction benchmark problem to assess its efficiency and accuracy. Then, a highly nonlinear aeroelastic phenomenon, limit cycle oscillation, is analyzed. Finally, the framework is applied to three-dimensional fluid flow to demonstrate its capability to efficiently analyze a significantly large number of degrees of freedom. The superior performance of LSH-VAE is emphasized by comparing its results against those of widely used nonlinear MOR methods, a convolutional autoencoder, and $$\beta $$ β -VAE. The proposed framework exhibits significantly enhanced accuracy compared with that of conventional methods, while the computational efficiency continues to remain significantly higher.

Analysis of medium-frequency oscillation based on bifurcation theory and joint modelling of SVG and direct-drive wind turbine

With the expansion of wind power clusters in power system, medium and high frequency oscillations in wind farm area are preliminarily emerging. At present, linear analysis such as impedance analysis are mainly used to analyse the grid-connected oscillation of renewable energy generation. However, the explanation of the oscillation mechanism is not intuitive and does not take into account the nonlinearity of converter. Considering the influence of external loop control of SVG and direct-drive wind turbine on the medium-frequency characteristics of the parallel system, a reduced-order nonlinear model under voltage time scale is established. The conjoint analysis of Jacobi matrix, bifurcation diagram and phase space trajectory are applied to study the influence rule of machine-side output power and short-circuit ratio on the oscillation behaviour of the parallel system. In addition, the oscillation evolution process is analysed from the dynamic standpoint. It is found that the system operates on the limit-loop state when the medium-frequency oscillation occurs, which reveals the mapping relationship between the oscillation amplitude and the limit-loop trajectory. The experimental results ultimately validate the theoretical analysis conducted on the StarSim-HIL (hardware-in-the-loop) experiment platform.

Nonlinear model order reduction for problems with microstructure using mesh informed neural networks

Many applications in computational physics involve approximating problems with microstructure, characterized by multiple spatial scales in their data. However, these numerical solutions are often computationally expensive due to the need to capture fine details at small scales. As a result, simulating such phenomena becomes unaffordable for many-query applications, such as parametrized systems with multiple scale-dependent features. Traditional projection-based reduced order models (ROMs) fail to resolve these issues, even for second-order elliptic PDEs commonly found in engineering applications. To address this, we propose an alternative nonintrusive strategy to build a ROM, that combines classical proper orthogonal decomposition (POD) with a suitable neural network (NN) model to account for the small scales. Specifically, we employ sparse mesh-informed neural networks (MINNs), which handle both spatial dependencies in the solutions and model parameters simultaneously. We evaluate the performance of this strategy on benchmark problems and then apply it to approximate a real-life problem involving the impact of microcirculation in transport phenomena through the tissue microenvironment.

A novel geometric nonlinear reduced order modeling method using multi-fidelity surrogate for real-time structural analysis

A novel geometric nonlinear reduced order modeling method using multi-fidelity surrogate for real-time structural analysis

Multibody Constraints in the Geometrically Nonlinear Intrinsic Formulation

Abstract The intrinsic formulation for geometrically nonlinear beam dynamics provides a compact and versatile description of slender beam-like structures. With nonlinearities limited to second-order couplings in the formulation, it has been an attractive choice in formulating nonlinear reduced-order models for dynamic analysis and control design in aeroelasticity problems involving large displacements and rotations. Owing to its rotation-free formalism, the intrinsic formulation has not been formulated to accommodate multibody constraints, limiting its use against multibody structures with kinematic constraints. This work aims to address such weakness as we present developments in introducing multibody constraints into the full and reduced-order intrinsic equations while still preserving the beneficial traits of the method. We describe the resolution of displacement-level constraints using index-1 approach and adaptation of constraint stabilization strategies to the intrinsic formulation using state projection. The numerical behavior of the full- and reduced-order implementations are assessed using test cases with large static and dynamic deformations with time-domain simulations to demonstrate validity of the approach.

Theoretical dynamic reduction for steady-state nonlinear vibration responses of complex jointed structures having hysteretic contact behavior

A novel nonlinear dynamic reduction method was developed to determine the steady-state vibration responses of complex jointed structures having hysteretic contact behavior. By using the harmonic balance method to reformulate nonlinear dynamic equilibrium equations into a set of nonlinear algebraic ones, a dynamic reduction technique based on nonlinearity transformation was theoretically developed to iterate nonlinear vibration solutions in the local coordinate that was related only to the degree-of-freedoms of nonlinear joints. Only odd-order harmonic components were truncated to approximate the hysteretic nonlinear contact forces of joint interfaces, which was conducive to further dimension reduction of the nonlinear algebraic equations and iteration matrix. Then, a nonlinear dynamic reduction solver was developed to associate the steady-state nonlinear vibration responses of overall structures with the dynamic characteristics of underlying linear substructures, nonlinear joint models and external excitations. Combined with finite element analysis, the steady-state nonlinear vibration responses of a complex assembled structure with four reduced-order nonlinear joint models were numerically simulated to validate the proposed nonlinear dynamic reduction method by comparing with the literature. The comparative results of nonlinear frequency response functions exhibited good agreement, and the proposed method indicated higher computational efficiency. The experimental investigations of a rubber isolator system were also performed to validate the proposed method, which demonstrated good performance.

Model Order Reduction with Dynamically Transformed Modes for Electrophysiological Simulations

The numerical simulation of neuromuscular processes in skeletal muscles typically imposes high computational costs, resulting in the need for techniques to reduce these costs. One submodel in these types of multiphysics multiscale simulations is the electrophysiological model, describing the propagation of an action potential (AP) in a muscle fiber. Involved in the simulation of the propagation are two traveling waves, resulting in difficulties for classical model order reduction (MOR) techniques based on linear subspaces.
 Instead, we apply the nonlinear MOR method shifted Proper Orthogonal Decomposition (sPOD) to construct dynamically transformed reduced basis functions depending on time-dependent paths spanning the adaptive reduced ansatz space.
 Our numerical experiments demonstrate that the constructed reduced ansatz space can accurately capture the dynamics of two fully separated wavefronts and reduce the degrees of freedom of the whole simulation. However, it cannot represent the activation of the AP in the center of the fiber and overlapping wave parts. The constructed reduced order model outperforms the high-dimensional full order model in terms of the computational costs while the accuracy is maintained and reaches speedup factors between 2 and 73 depending on the time discretization.

Numerical accuracy of principal geodesic analysis on the sphere in director‐based dynamics of hybrid mechanical systems

AbstractWe aim at nonlinear model order reduction (MOR) in hybrid mechanical systems by means of Principal Geodesic Analysis (PGA) on the Riemannian manifolds S2 (the sphere) and SO(3) (the rotation group). MOR requires highly accurate and efficient implementations of the logarithm maps and the resulting lifts across multiple branches. However, in our cases these maps have singularities due to periodicity. In this work we focus on the logarithm and lift maps for the sphere S2. We conduct detailed numerical experiments on mechanical systems to achieve maximal accuracy in spite of the singularities.

Nonlinear WEC modeling using Sparse Identification of Nonlinear Dynamics (SINDy)

Modeling oscillating surge wave energy converter (OSWEC) systems to accurately predict their behavior has been a notoriously difficult challenge for the wave energy field. This is particularly challenging in realistic sea states where nonlinear WEC dynamics are common due to complex fluid-structure interaction, breaking waves, and other phenomena. Common modeling techniques for OSWECs include using potential flow theory to linearize the governing equations and ease computations, or using CFD to solve the full Navier-Stokes equations coupled with rigid body motion. However, both of these options have significant limitations. Potential flow theory breaks down in realistic sea conditions where nonlinear WEC dynamics are present, and CFD is often too computationally expensive for many applications such as real-time state prediction and optimal control, two areas of active research in the wave energy field. In particular, OSWEC dynamics are dominated by diffractive and viscous forces, often making common assumptions and linearization approximations (including small-body approximations) unreasonable, and CFD computationally intractable.
 To bridge this gap in modeling methods, we propose using Sparse Identification of Nonlinear Dynamics (SINDy) to build nonlinear reduced-order models (ROMs) that describe OSWEC behavior in response to large-amplitude regular waves. SINDy is an equation-free, data-driven algorithm that identifies dominant nonlinear functions present in system state dynamics using a library of nonlinear functions created from time series measurement data. The result is an ordinary differential equation (ODE) in time that can be solved from an initial condition to model and predict time behavior of the states. SINDy is parsimonious, meaning it uses a sparsity-promoting hyperparameter with the goal of only including the minimum number of terms to capture dominant dynamics, resulting in interpretable and generalizable results that are not overfit to the data. Using the discovered ROMs and integrating in time, not only can SINDy provide time series models and future state predictions of OSWEC dynamics, it can also give insights into which variables are critical in describing the underlying dynamics of the state. 
 In this study, we use SINDy to describe the nonlinear dynamics of a lab-scale OSWEC in a wave tank subjected to large-amplitude regular waves. We use nonlinear simulation data to generate kinematic, force, and torque data and use it as input to SINDy to identify ODEs that describe the measurement variables in time. We then integrate the ODEs to recreate the time series as well as predict future system behavior. We directly compare the resulting time series to the original data input to assess the accuracy of the SINDy model. We also interpret the dominant terms in the ODEs to gain insight on underlying mechanisms of the observed nonlinearity.
 Early results show SINDy is a promising tool for modeling nonlinear OSWEC dynamics. We are able to build ROMs for variables such as angular kinematics and the moment about the hinge that generate an accurate recreation of data measurements. We found strong dominance in cubic and quintic terms of the ROMs, suggesting higher-order nonlinearities in the system dynamics. These findings inspire future work in identifying underlying mechanisms driving nonlinearity.

Stochastic analysis of flexible rotor supported on hydrodynamic bearings

Rotating shaft instability leading to high amplitude vibration is a key problem with a wide range of industrial applications. The purpose of this study is to identify the response statistics for a conceptual model of a flexible shaft supported on hydrodynamic bearings (HDBs) and propose design strategies to reduce the response. A conceptual model of a flexible shaft with three discs supported on the HDBs is developed. A linearised model of the system is developed using the short bearing assumption. A Campbell diagram is developed for the conceptual model to identify the critical speeds wherein a high response is produced. These high responses occur at the rotor speeds corresponding to the synchronous forward whirl and the oil whip. The identified rotor speeds are validated by carrying out a run-up analysis on the test rig. Since the manufacturing process and operating conditions induce uncertainty by varying the eccentricity and the relative phase of the unbalance mass, the influence of the whirling modes on the response is investigated. The nominal positions of the unbalance masses are such that the predominant mode excited is the first mode. Experimentally a scatter plot on the maximum response at the bearing location is identified using six different positions of the eccentric mass. Further, a stochastic analysis using reduced order nonlinear model of the system is carried out. The probability density obtained is compared with a Monte Carlo simulation and the bounds of the experimental results. A control strategy is designed by altering the relative position of the unbalance masses to reduce the excitation of the first mode. The robustness of the strategy is verified by using a stochastic analysis, which indicates a reduction in the response magnitude.

Regression-based identification and order reduction method for nonlinear dynamic structural models

In this paper, we propose a regression-based nonlinear reduced-order model for nonlinear structural dynamics problems, called the Nonlinear Identification and Dimension-Order Reduction (NLIDOR) algorithm. We evaluate the algorithm using a simple toy model, a chain of coupled oscillators and an actual three-dimensional flat plate. The results show that NLIDOR can accurately identify the natural frequencies and modes of the system and capture the nonlinear dynamical features, while the linear Dynamic Mode Decomposition (DMD) method can only capture linear features and is influenced by nonlinear terms. Compared with the full-order model (FOM), NLIDOR can effectively reduce computational cost, while compared with DMD, NLIDOR significantly improves computational accuracy. The results demonstrate the effectiveness and potential of NLIDOR for solving nonlinear dynamic problems in various applications.

An analytical model for estimating the maximum energy harvesting efficiency from vortex-induced vibration

The present work presents an analytical model (AM) based on a nonlinear reduced-order model (ROM) for estimating the maximum efficiency (ηm) of energy harvesting from vortex-induced vibration (VIV) of a circular cylinder. The proposed AM was validated through comparisons against both the origin ROM and experiments over a wide range of parameters. Of interest, it provided a low-cost analytical approximation relating the maximum efficiency ηm to physical parameters. As inspired by the modified Griffin plot, the use of AM allowed the collapse of the maximum efficiency ηm as a function of the combined mass-damping parameter α. Of significance, the presented AM should be not only effective for estimating the maximum efficiency ηm but also useful to shed light on VIV dynamics.

Nonlinear reduced-order modeling for three-dimensional turbulent flow by large-scale machine learning

A large-scale machine learning-based nonlinear reduced-order modeling method was developed for a three-dimensional turbulent flow field (Re=1000) using a neural-network with unsupervised learning. First, a mode decomposition method was applied to three-dimensional flow field data using a convolutional-autoencoder-like neural network. Then, a reduced-order model (ROM) was constructed using long short-term memory neural networks (LSTMs). Consequently, it was demonstrated that the time evolution of the turbulent three-dimensional flow field can be simulated at a significantly lower cost (approximately three orders of magnitude) without a major loss in accuracy. However, neural-network-based mode decomposition for the three-dimensional flow field requires a huge computational cost in terms of calculation and memory usage. Therefore, a distributed machine-learning method was implemented using a hybrid parallelism scheme tailored to the network structure. Thus, it was possible to decompose 1.7 million cells of the three-dimensional flow field data into 64 modes and reproduce those with sufficient accuracy. In this study, a uniform flow around a circular cylinder model was used as a test case. To validate the method, the reduction performance of the proposed mode decomposition method was compared with the proper orthogonal decomposition (POD) method. Furthermore, the target flow field was reproduced using ROM, and the reconstruction accuracy was evaluated in terms of various criteria compared with the accuracy based on POD in conjunction with Galerkin projection method.