Incipient Buckling Detection Using Nonlinear System Identification Algorithms: A Comparative Study

Document Type

Article

Publication Date

1-1-2026

Abstract

Structural health monitoring detects and characterizes damage to predict failure, but failures like buckling, which lack signs of damage, are harder to detect. Direct load measurement is costly and complex, especially for dead loads requiring copious sensors and data acquisition hardware. Vibration-based detection is a good alternative because it infers global structural characteristics with fewer sensors. This paper presents a structural health monitoring approach for the detection of incipient buckling in structures due to excessive load conditions. Existing nonlinear modeling algorithms available in commercial software platforms, such as MATLAB and other open-source toolboxes, are leveraged to make this method potentially available to practitioners. The proposed method attempts to infer the presence of incipient buckling behavior from the linearity of relatively low-order system identification models fitted to global vibration signals measured at low levels of ambient-scale vibration. Three widely available nonlinear system identification models are investigated in this proof-of-concept study: (1) a nonlinear autoregressive with an exogenous input (NLARX) algorithm with a wavelet function, (2) an NLARX algorithm with Tree Ensembles, and (3) a physics-informed dynamic mode decomposition for detecting incipient buckling. Synthesized ambient vibration data are obtained from numerical models of small-scale frame structures with varying gravity loads, incorporating nonlinearity via P-delta and large-displacement effects. For this proof-of-concept, variable control is vital, and using simulation data enables disabling material nonlinearity and connection slip effects to demonstrate the viability of nonlinear system identification for buckling alerts. Follow-up studies are planned to validate the results using physical models. The average norm of the nonlinear component of the output is chosen as the feature and compared at incremental loadings up to the incipient buckling load. Scaling issues, the effects of unmeasured disturbances, and unidentified sensor noise are also examined to highlight the limitations of the proposed methods in terms of excitation scale.

Publication Title

Structural Control and Health Monitoring

Link to Full Text

Share

COinS