This work systematically examines the generation and cumulative evolution of the quasi-static component (QSC) induced by nonlinear self-interaction of the primary Lamb wave in variable-thickness plates. Existing QSC investigations are predominantly restricted to uniform-thickness waveguides, where the saturated QSC peak amplitude remains nearly constant under group-velocity mismatch and waveform broadening takes place. In engineering components, continuously varying thickness alters local dispersion characteristics, modal wave structures and energy-transmission conditions, yet the underlying QSC physical mechanisms have remained rarely explored. In this paper, a segmented discretization theoretical model, rooted in second-order perturbation and adiabatic approximation, was developed to describe the displacement-field superposition of QSC across thickness-varying segments. Simulations and experiments consistently verify that the generated QSC corresponds to the zero-frequency S0 mode, based on group velocity and wave structure analysis. Distinct from uniform-thickness plates, a unique cumulative transition was found in variable-thickness waveguides. The QSC amplitude grows monotonically with a declining growth rate up to 140 mm, governed by coherent superposition of the incident and newly generated QSCs. Beyond that, increasing group velocity mismatch, dispersion attenuation and thickness-dependent transmission loss result in progressive peak-amplitude decay alongside substantial temporal broadening. The acoustic nonlinearity parameter βQSC rises steadily with propagation distance, whereas the second harmonic (SH) counterpart βSH oscillates periodically and saturates within only 30 mm, owing to continuously broken phase-velocity matching. These findings elucidate the unique QSC evolution physics in thickness-varying waveguides and demonstrate the superior robustness over conventional SH for early damage assessment, paving the way for nondestructive evaluation and health monitoring of practical variable-thickness components.
Access to design information like the geometry and material data of an aged building structure has been a practical challenge in estimating the influences of to-be-constructed railways on existing buildings. The measurement-based building modeling method utilizes the modal properties identified from ambient vibration testing to construct the modal equations of motion directly. However, the applicability of using identified modal properties in limited modes at limited measurement locations for the response predictions has not been examined in practical situations. Therefore, this paper examines this prediction method through shaking-table tests with a four-story steel specimen. Three recorded railway-induced ground motions with different dynamic characteristics were utilized as inputs. The mean absolute prediction error across the floor centers was below 1.60 dB on average for each of the three events, provided that the foundation input is accurately known. The experiment revealed that the identified mode shapes can be linearly dependent when only limited locations can be measured. The prediction accuracy is improved by removing the linearly dependent modes, though multiple modes are found predominant in the responses. Moreover, the predominant modes depend on the dynamic characteristics of the inputs and change with the floor number. In addition, the modeling method estimates the vibrations where the measurements are taken, but the prediction accuracy is reduced when measurements are only taken at the centers of floors, compared with measurements at locations with large mode-shape amplitudes like the midpoints of edge beams.
Quality-related fault detection, as an essential component of process monitoring, is of great significance for ensuring the safe and efficient operation of industrial processes. However, existing quality-related fault detection methods rely heavily on quality variables (QVs). When these QVs are either unavailable or unreliable, these methods become unserviceable. To address these challenges, a fault detection method based on a quality-associated virtual variable (Qav) is proposed in this paper. By constructing a Qav for linear systems, we can indirectly capture the quality characteristics of the system, thereby alleviating the reliance on QVs in traditional methods. In the proposed method, process variables are categorized into quality-related and quality-unrelated groups using lasso regression. These categorizations are then integrated into fault detection based on independent component analysis (ICA), referred to as Qav-based ICA (QavICA). Furthermore, the proposed QavICA is extended to nonlinear systems, a Qav-based kernel ICA (QavKICA) is proposed, wherein mutual information serves as criteria for variable categorization. Additionally, the sufficient conditions for fault detection in both linear and nonlinear systems are theoretically derived, respectively. Finally, several experiments on a numerical example, Tennessee Eastman Process (TEP) and electro-hydraulic servo mechanism are conducted to demonstrate the feasibility and effectiveness of the proposed method.
Distributed acoustic sensing (DAS) measures differential optical phase over finite apertures, so phase amplitude and signal-to-noise ratio (SNR) depend on both the wavefield and the measurement configuration. We develop a measurement-chain framework that separates directional strain projection, finite-gauge averaging, phase accumulation, and strain calibration. We evaluate this framework with conditional simulations, an independent paired DAS–geophone data set, and separate field trials of four cable configurations (Lr=1,3,5,and8). For a fixed 5 m axial model gauge, pitch-resolved integration gives the largest 30 Hz peak differential phase at Lr=8: 1.478rad, or 1.437 times the straight-fiber value. Under the nominal 40 Hz noise model, Lr=8 also has the highest mean PSD-SNR, 59.38±2.23dB, exceeding the Lr=1 mean by +3.31dB. Each mean is calculated by averaging repeatwise SNR values in dB. In the full noise sensitivity grid, Lr=8 has the highest point estimate of SNR in 59 of 64 frequency-condition cells when additive-noise standard deviation is independent of Lr; Lr=1 leads all 64 cells in each case when the standard deviation scales as Lr0.5 or Lr. In the independent paired data, increasing the numerical processing aperture from 2.04 to 10.20 m raises median DAS–geophone correlation from 0.046 to 0.342. The descriptive rankings from the helical-cable field trials vary with frequency. Modeled SNR gains depend on how noise scales with fiber path length, and the independent paired records show improved agreement with a larger processing aperture.
Precise road roughness recognition plays an important role in providing priori road information for active suspension control, thereby contributing to promoting the ride comfort and handling stability for intelligent vehicles. Nevertheless, the existing road roughness recognition approaches based on vehicle dynamical responses and neural networks still suffer from shortcomings of poor internal physical interpretability and high data demand. To tackle this challenge, this paper presents a novel physics-informed neural network (PINN) framework integrating the dynamic equations of suspension systems for robust and adaptive road roughness recognition. This proposed PINN framework can be regarded as a unified framework for quarter-car suspension systems, which imposes constraints on both the forward and backward propagation processes by embedding suspension dynamics into the network architecture and loss function, respectively. Besides, three variants of the proposed unified PINN framework are developed, aiming to customize three commonly encountered scenarios of the quarter-car suspension models involving linear parameters, nonlinear dynamic parameters and measurement noises, with comprehensive simulation verifications conducted for each variant of PINNs. Finally, bench test verifications have been performed for validating the applicability and effectiveness of our proposed PINN framework. The advantages of our proposed PINN framework for robust road roughness recognition have been demonstrated by comparing with one-dimensional convolutional neural network, long short-term memory and multilayer perceptron, especially under small-sample training conditions. Our proposed unified PINN framework demonstrates great application prospects in road environment perception for intelligent vehicles and transportation systems.
