Dionisio P. Bernal · Northeastern University · PhdFit
Resume-aware faculty matching
Find professors who actually fit you
Upload your resume. PhdFit's six research agents compare your background with faculty profiles, recent publications, lab focus, and outreach opportunities, then rank professors with evidence you can review.
Resume-aware matchingEvidence from recent researchOutreach drafts included
Advisor Match Brief Resume uploaded
Resume signals5 found
NLPHCIFairnessGraph MLPyTorch
Six-agent analysisRunning
Axel · Candidate AnalystResume parsed
Nova · Professor ResearcherRecent papers
Echo · Match ExplainerEvidence linked
Atlas · Strategy AdvisorFit ranked
Lyra · Outreach WriterDraft queued
Juno · Meeting Prep CompilerNext step ready
SC
Dr. Sarah Chen
Stanford · NLP & interpretability
91
fit
Fairness + NLP projects overlap with her recent work.
Strong methods fit: model evaluation and interpretability.
Clear outreach angle for current trustworthy AI research.
Outreach angle
Ask how her lab is extending interpretability methods into fairness audits for real-world AI systems.
Dionisio P. Bernal
· ProfessorVerified
Northeastern University · Environmental Engineering
Dionisio P. Bernal is a Professor of Civil and Environmental Engineering at Northeastern University College of Engineering. His research focuses on system identification, fault detection and fault localization, earthquake engineering, soil structure interaction, and structural stability. He has received numerous honors and awards, including the 2020 Søren Buus Outstanding Research Award, the 2011 'Students Speak' Teaching Award, and the 2004 Martin W. Essigmann Award for Teaching Excellence from Northeastern University. Bernal earned his PhD, MS, and BS in Civil Engineering from the University of Tennessee in 1979, 1976, and 1975 respectively. He is a member of the American Society of Civil Engineers and has contributed significantly to the field through research, publications, and professional service.
Collapse during strong ground motions requires the effective tangent stiffness matrix to lose positive definiteness at some point in time. However, this loss alone does not guarantee collapse, since the velocity distribution at that instant rarely aligns with the unstable mode. Thus, inertia and, to a lesser degree, damping provide a temporary stabilizing effect, making a negative eigenvalue necessary but not sufficient for dynamic instability. In structural design, especially for multistory structures, constraints ensure the lowest eigenvalue of the second-order elastic stiffness remains substantially above zero. Under earthquake loads, plasticity may reduce this eigenvalue, potentially leading to statically unstable configurations and collapse. Assessing stability, therefore, requires understanding how a given distribution of plastic hinges alters tangent stiffness (for practicality, a lumped plasticity model is used). Thus, we explore how to raise the ground motion intensity required for instability by a scaling factor α. Simply increasing yield strength uniformly by α is possible but inefficient. Instead, strategically distributing strength increases—termed “local modifications”—can shift governing plasticity distributions toward more favorable ones. This approach is most relevant for tall buildings, which can reach unstable configurations without forming mechanisms, making dynamic instability critical. Effectiveness of local modifications is judged by tracking changes in the lowest eigenvalue of the effective tangent stiffness. Selected modifications are then validated through nonlinear second-order time history analyses, confirming their role in delaying collapse.
Localization of damage in a system is an issue that has been widely studied in structural health monitoring. The Dynamic Damage Locating Vector (DDLV) method localizes damage from information in the null space of the change in transfer matrices (∆G). A precisely known ∆G and an accurate model of the reference state represent ideal conditions in the DDLV approach. Since ideal conditions are not realized in practice, the question of robustness arises. This paper examines the performance of the DDLV approach under error in ∆G coming from truncation of the modes to a limited bandwidth. The numerical results are obtained using a rectangular thin plate model.
Journal of Wind Engineering and Industrial Aerodynamics · 2025-10-23 · 3 citations
articleOpen access
This study introduces the integration of Stochastic Subspace Identification (SSID) with the Kalman filter as a general-purpose method for mitigating ambient sensor noise in wind tunnel tests. Filter parameters are estimated using SSID, a robust data-driven technique for extracting system dynamics under unknown input conditions. The proposed SSID-Kalman filter is validated through numerical simulations and experimental data from three aeroelastic tests, including acceleration and strain measurements from free-standing bridge towers, as well as voltage signals from a PVDF piezoelectric film excited by the aeroelastic response of an airfoil-based wind energy harvester. Numerical validation confirms the filter's capability to recover ground-truth signals by suppressing the noise. In wind tunnel experiments, the filter reduces noise across both low- and high-frequency ranges while preserving resonant components of the structural response. The signal-to-noise ratio gains reach 98.4 % for acceleration, 79.4 % for strain, and over 600 % for voltage. Compared with spectral subtraction, a method recently proposed by the first author, the SSID-Kalman filter demonstrates superior performance, particularly when dominant vibrational modes have only marginal energy above the noise floor. The method provides wind tunnel practitioners with a flexible and robust tool for sensor denoising and is broadly applicable to dynamic measurements representable within a state-space framework. • This study integrates Stochastic Subspace Identification (SSID) with the Kalman filter as a general-purpose method for reducing ambient sensor noise. • The method was validated using acceleration and strain measurements from bridge towers and extended to voltage signals from a PVDF piezoelectric film.The method was validated using acceleration and strain signals from bridge towers, voltage signals from a piezoelectric film. • The SSID-Kalman filter effectively suppresses noise, reduces signal standard deviation and improves signal-to-noise ratio. • The filter outperforms spectral subtraction when dominant vibrational mode energy is only slightly higher than ambient noise.
Journal of Vibration and Control · 2025-02-19 · 1 citations
articleSenior author
A scheme is presented for damage localization in linear and time-invariant (LTI) systems using static output feedback eigenstructure assignment. The feedback is realized through signal processing of open-loop input–output data, and the assignment is formulated to render an eigenvalue subset invariant to a number of perturbations that define the spatial domain of the plausible damage distribution. For each perturbation, a feedback gain is designed to realize the eigenvalue invariance, whereby the damage location can be inferred from the shifts in the assigned eigenvalues. The interrogation is posed as a statistical hypothesis test with eigenvalue uncertainty bounds derived from an asymptotic analysis of the assignment formulation. Numerical and experimental examples are presented to validate the theoretical developments and convey the details of the damage localization application.
Detection and identification of nonlinearity is a task of high importance for structural dynamics.On the one hand, identifying nonlinearity in a structure would allow one to build more accurate models of the structure.On the other hand, detecting nonlinearity in a structure, which has been designed to operate in its linear region, might indicate the existence of damage within the structure.Common damage cases which cause nonlinear behaviour are breathing cracks and points where some material may have reached its plastic region.Therefore, it is important, even for safety reasons, to detect when a structure exhibits nonlinear behaviour.In the current work, a method to detect nonlinearity is proposed, based on the distribution of the gradients of a data-driven model, which is fitted on data acquired from the structure of interest.The data-driven model selected for the current application is a neural network.The selection of such a type of model was done in order to not allow the user to decide how linear or nonlinear the model shall be, but to let the training algorithm of the neural network shape the level of nonlinearity according to the training data.The neural network is trained to predict the accelerations of the structure for a time-instant using as input accelerations of previous time-instants, i.e. one-step-ahead predictions.Afterwards, the gradients of the output of the neural network with respect to its inputs are calculated.Given that the structure is linear, the distribution of the aforementioned gradients should be unimodal and quite peaked, while in the case of a structure with nonlinearities, the distribution of the gradients shall be more spread and, potentially, multimodal.To test the above assumption, data from an experimental structure are considered.The structure is tested under different scenarios, some of which are linear and some of which are nonlinear.More specifically, the nonlinearity is introduced as a column-bumper nonlinearity, aimed at simulating the effects of a breathing crack and at different levels, i.e. different values of the initial gap between the bumper and the column.Following the proposed method, the statistics of the distributions of the gradients for the different scenarios can indeed be used to identify cases where nonlinearity is present.Moreover, via the proposed method one is able to quantify the nonlinearity by observing higher values of standard deviation of the distribution of the gradients for lower values of the initial column-bumper gap, i.e. for "more nonlinear" scenarios.
A widely used approach for virtual sensing takes a set of eigenvectors not larger (in cardinality) than the number of measurements, selects the partition associated with the measurements, and computes the modal amplitudes that justify the measurements. The full response is subsequently estimated by using the computed modal amplitudes and the truncated basis over the full domain. Since the approach projects the measured response, which has contributions from all the modes, on a truncated basis, the obtained result is not an approximation of the truncated modal series, nor is it an approximation of the response from all the modes. This paper puts forth a modification based on appropriate filtering that improves consistency. It is shown that uniqueness does not restrict the number of modes in the projection basis to the number of sensors but only the number of modes that can be simultaneously considered in an approach that can be applied sequentially to include as many modes as desired.
Sensitivity-based model updating entails a computational parameter estimation problem, in which a set of model parameters is adjusted to minimize the discrepancy between identified system features and the corresponding model predictions. In order to promote the posedness and conditioning of the estimation problem, it is often necessary to exclude some of the uncertain parameters and accept the errors induced by treating them at their nominal values. The present paper proposes a parameter rejection scheme that seeks to mitigate the noted errors by minimizing the sensitivity of the estimation problem to changes in the excluded parameters. The sensitivity minimization is realized in a closed-loop setting with output feedback eigenstructure assignment, which, given that the operation is offline, can be implemented through processing of open-loop input-output data. Consequently, the scheme rests on the assumption that open-loop input-output data can be collected while the considered system is linear and time-invariant. The implementation and validity of the scheme are demonstrated in the context of numerical and experimental examples with vibrating systems. • A scheme is presented for parameter rejection in sensitivity-based model updating. • The sensitivity of the updating solution to the rejected parameters is minimized. • The rejection formulation uses static output feedback eigenstructure assignment. • The feedback is realized virtually by processing of available input-output data. • The eigenstructure assignment is posed as a homogeneous system of equations.
The utility of complex perturbations to estimate the sensitivity of modal parameters in self-adjoint conservative models is examined. In the Complex Step Derivative (CSD) approach the parameter for which the derivative is required is perturbed in the direction of the imaginary axis and the eigenvalue problem for the perturbed matrices is solved; the real part has the solution of the unperturbed eigenvalue problem and the imaginary the sensitivities times the perturbation magnitude. This paper examines the utility of the CSD approach as a method to compute sensitivities in structural dynamics.
The Kalman filter gain can be extracted from output signals but the covariance of the state error cannot be evaluated without knowledge of the covariance of the process and measurement noise Q and R. Among the methods that have been developed to estimate Q and R from measurements the two that have received most attention are based on linear relations between these matrices and: 1) the covariance function of the innovations from any stable filter or 2) the covariance function of the output measurements. This paper reviews the two approaches and offers some observations regarding how the initial estimate of the gain in the innovations approach may affect accuracy.
