A phase resonance approach for modal testing of structures with nonlinear dissipation

Abstract

The concept of nonlinear modes is useful for the dynamical characterization of nonlinear mechanical systems. While efficient and broadly applicable methods are now available for the computation of nonlinear modes, nonlinear modal testing is still in its infancy. The purpose of this work is to overcome its present limitation to conservative nonlinearities. Our approach relies on the recently extended periodic motion concept, according to which nonlinear modes of damped systems are defined as family of periodic motions induced by an appropriate artificial excitation that compensates the natural dissipation. The particularly simple experimental implementation with only a single-point, single-frequency, phase resonant forcing is analyzed in detail. The method permits the experimental extraction of natural frequencies, modal damping ratios and deflection shapes (including harmonics), for each mode of interest, as function of the vibration level. The accuracy, robustness and current limitations of the method are first demonstrated numerically. The method is then verified experimentally for a friction-damped system. Moreover, a self-contained measure for estimating the quality of the extracted modal properties is investigated. The primary advantages over alternative vibration testing methods are noise robustness, broad applicability and short measurement duration. The central limitation of the identified modal quantities is that they only characterize the system in the regime near isolated resonances.

Introduction

The technical relevance of nonlinear vibrations is increasing for a number of reasons: The ever-growing demand for energy and material efficiency leads to lightweight design. Slender structures are more prone to large deformation nonlinearities and lightweight materials commonly exhibit nonlinear behavior. These structures are more likely to be driven into nonlinear self-excited vibrations, as in the case of modern turbine blades where this is a dominant design limitation [1]. Finally, novel technologies such as nonlinear vibration absorbers take advantage of nonlinearities to achieve substantially improved performance.

Since the pioneering work of H. Poincaré in 1905, a solid theoretical framework for nonlinear vibrations has been developed. A versatile toolbox of robust and efficient computational methods for the analysis of nonlinear vibrations has been established within the last three decades. In contrast, by far most experimental methods are based on linear theory and fail in the presence of nonlinearities. For example, conventional frequency response functions (FRFs) no longer provide a complete picture of the dynamic behavior, since the frequency response depends on the excitation level in the nonlinear case. Moreover, the emergence of new (e.g. sub- or super-harmonic) resonances or the coexistence of steady vibration states have no counterparts in linear theory.

In the following, the experimental methods for the analysis of nonlinear vibrations is briefly addressed. Only the most important methods, in the opinion of the authors, are mentioned, and the presentation is limited to their main idea and current limitations. For a more comprehensive review of methods for nonlinear system identification, we refer to [2].

Nonlinear FRFs are a straight-forward extension of their linear counterpart. They rely on the measurement of the frequency response to harmonic excitation. In contrast to the linear case, the FRF is determined for different excitation levels of interest. The primary deficiencies of this method are twofold. First, it is practically impossible to realize a purely harmonic excitation due to the interaction between specimen and excitation system [3,4], or to keep the level constant throughout resonances. Second, the fine variation of both frequency and level leads to high testing effort and exposing the structure to nonlinear, typically high, vibrations for long times. This can cause severe fatigue and wear damage (destructive testing).

The central idea of the Restoring Force Surface Method (RFSM) is to determine local nonlinear restoring forces indirectly using the dynamic force equilibrium and measuring mass and deformation quantities [[5], [6], [7]]. An important deficiency of RFSM in its conventional form is that one of the subsystems connected to the nonlinearity must act as a rigid body. Moreover, RFSM is limited to single-valued nonlinear forces, making it unsuitable for hysteretic forces as in the case of dry friction.

The purpose of nonlinear subspace identification (NSID) is to derive a nonlinear differential equation system reproducing the response in the tested dynamic regime. The central idea of the method is to interpret the nonlinear forces as internal feedback forces [8]. The linear system matrices and coefficients matrices associated to nonlinear forces of an a priori assumed form are then determined by minimizing the deviation between measured and simulated response [9,10]. An advantage of NSID is that no specific input signal to the system is required, such that typically random broadband excitation signals are used. This class of excitation signals can also be utilized to identify a polynomial nonlinear state-space model [11], where a linear state-space model is extended by multivariate polynomials and identified based on measured data only. The deficiency is the method's immaturity: The applicability to e.g. hysteretic nonlinearities has so far only been demonstrated for a rather idealized numerical example [12]. Perhaps more severely, it appears still to be an unresolved problem to choose the excitation level of a broadband signal for deriving a nonlinear model that is valid for a range of amplitudes of harmonic excitation signals [12]. This is especially relevant in case of forced dynamics around a resonance. To enrich the model, several excitation levels could be included in the training data. This would, however, raise the required measurement duration with the aforementioned drawbacks.

The purpose of nonlinear modal analysis is to characterize the dynamical behavior of mechanical systems in terms of natural frequencies, modal damping ratios and vibrational deflection shapes, as a function of the vibration level. These quantities determine at which excitation frequencies resonances are expected, how the resonant vibration energy is distributed within the system, how well the resonances are damped, and how well the system resists dynamic instabilities. The recent literature is particularly rich of contributions on the theoretical framework of nonlinear modes, numerical computation techniques and applications to the analysis and design of nonlinear mechanical systems. On the other hand, studies on experimental nonlinear modal analysis, i.e. Nonlinear Modal Testing (NMT) are scarce. Peeters et al. [13] propose a two-step procedure where they first attempt to isolate a nonlinear mode by well-appropriated external forcing, and then to stop the excitation to extract the modal characteristics during the free decay (presuming light damping). For the isolation, they suggest to control the force in such a way, that at every excitation point the different harmonics of external force and response displacement have a 90° phase lag. For lightly-damped structures with smooth stiffness nonlinearity, already a single-point, single-harmonic forcing can provide a satisfying isolation of the nonlinear mode if the modes are well-spaced [[13], [14], [15], [16]]. For systems with nonlinear damping, it is straight-forward to augment this procedure by a damping quantification using time-frequency-analysis of the free decay response [15,17,18], although the term NMT is not used in the literature for these methods. An important benefit of NMT is the comparatively short measurement duration, since the system's response is only tested once for each vibration level. The primary deficiency of NMT in its present form is its inaccuracy due to two reasons: First, the transient character of the analyzed response makes the method highly sensitive to noise, which particularly limits the quality of the extracted damping measures. Second, switching-off the excitation usually leads to a finite impulsive loading of the structure. As a consequence, the free decay is spectrally distorted and may contain multiple significant frequency components, so that the extracted damping measure cannot be directly attributed to a specific mode.

The goal of the present work is to extend NMT by a substantially more accurate damping quantification. The key idea is to analyze stabilized time series under sustained excitation, as opposed to transient free decays, by relying on the recently proposed extended periodic motion definition of damped nonlinear modes [19]. The experimental method is theoretically derived in Section 2. To thoroughly assess its accuracy and robustness against noise and imperfect isolation, the method is first applied to a virtual experiment (Section 3). In Section 4, the results of an experimental verification for a friction-damped system are presented. This paper ends with conclusions and directions of future work (Section 5).