Michael D. Todd · University of California, San Diego · 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.
Michael D. Todd
· Distinguished Professor and ChairVerified
University of California, San Diego · Structural Engineering
Professor Michael D. Todd is associated with the Center for Extreme Events Research at UC San Diego. His work involves advancing engineering research essential to protecting critical infrastructure and bio-systems from extreme hazardous events. The center's researchers, including Professor Todd, are recognized as world-renowned experts in experimental and computational methods for extreme events research. Their efforts focus on developing better ways to safeguard built infrastructures and bio-systems from various extreme events such as terrorist attacks, mining explosions, car crashes, sports collisions, and natural disasters like landslides. After extreme events, the center provides rapid damage and vulnerability assessments to first responders, aiding them in making informed decisions under challenging circumstances. Professor Todd's research is informed by real-world challenges and involves close collaboration with industry partners, contributing to fundamental advancements in the field of extreme events protection.
Structural Health Monitoring · 2025-09-02 · 1 citations
article
Reliable lock operation is crucial for uninterrupted waterway traffic, but failures in fill/empty Tainter valves can hinder lock management. To address this, an in-service lock was instrumented with an array of sensors spanning the motor to the final gear shaft. Unexpected impulses were observed in the jack shaft and sector gear bearing block accelerometer signals during certain valve opening events, coinciding with anomalies in sector gear angular displacement signals. To detect these impulses, a single-channel hierarchical Bayesian framework was employed, utilizing sector gear bearing block acceleration signals. A binary hierarchical Bayesian hypothesis testing approach was developed, assuming Gaussian noise versus Gaussian-distributed signals conditioned on the mean and standard deviation (STD) of the impulses. Unlike traditional Bayesian detection methods that pool all datasets/events together or analyze single events/datasets in isolation, this approach captures aleatory and epistemic uncertainties by modeling the impulse signal mean and dispersion as random variables (RVs). The relationships among different detection models (Neyman–Pearson detection, matched filter, Bayesian detection, and hierarchical Bayesian detection) are mathematically demonstrated, showing how they are interconnected. To address signal non-stationarity within events and signal magnitude across multiple events, optimal windowing and signal normalization were applied to ensure statistical reliability and computational efficiency. A local sensitivity analysis was performed to determine the effect of the decision threshold in this binary hypothesis testing to the hyperparameters. The results provide critical insights into the operational health of the Tainter valve system, enabling more reliable diagnostics and predictive maintenance for lock machinery, although the approach could be applied to detecting impulsive signals in any time series application.
Journal of Mechanical Design · 2025-10-23 · 1 citations
articleSenior author
Abstract Finite element (FE) model updating is essential for design, analysis, or response prediction of engineering systems. However, uncertainties from various sources often lead to discrepancies between model predictions and actual observations. Bayesian parameter updating and state estimation provide a probabilistic framework by estimating posterior distributions of parameters or states that define the FE model. Yet, traditional simulation methods like Markov Chain Monte Carlo (MCMC) face significant challenges in high-dimensional, multimodal spaces or when priors differ greatly from posteriors. This study introduces the Normalizing flow Enhanced GlObal and Local sAmpler (NEGOLA) that integrates normalizing flow, a machine learning technique for transforming simple distributions into complex ones, with a Gaussian kernel to generate new samples in Bayesian model updating. By concurrently running multiple chains and alternating between normalizing flow (global sampler) and Gaussian kernels (local sampler), NEGOLA achieves faster convergence and requires fewer FE model evaluations compared to conventional methods. The effects of the number of Markov chains and the switching steps in the NEGOLA algorithm are studied to provide insight into an optimal configuration of the algorithm. The algorithm was validated using single- and multiple-degree-of-freedom (SDOF and MDOF) dynamic systems and outperformed traditional approaches such as the unscented Kalman filter (UKF) and transitional MCMC (TMCMC) in scenarios involving time-varying parameters and abrupt changes. It was found that NEGOLA converges 20.3 times faster than the UKF in the SDOF system and requires 59.37% fewer FE model evaluations than TMCMC in the MDOF system. Results show that NEGOLA enhances convergence rates and accuracy, making it a promising tool for Bayesian model updating under uncertainty.
Journal of Computing in Civil Engineering · 2025-02-27 · 9 citations
articleSenior author
Significant discrepancies between the actual and designed dimensions of precast concrete (PC) components can result in construction delays and extensive rework, highlighting the imperative for dimensional quality assessment of these components. Point cloud technology can restore the three-dimensional (3D) information of objects, serving as an effective tool for dimensional quality assessment. The original point clouds often contain substantial nonessential data and outliers, necessitating the automatic extraction of relevant PC components from these complex point clouds. Existing point cloud segmentation methods suffer from insufficient accuracy and high cost. This study proposes a point cloud segmentation method employing multiview fusion for the dimensional quality assessment of PC components. First, an improved structure from motion (SfM) method is used to reconstruct the point cloud of PC components from multiview images. Second, an improved DeepLabv3 model is used to segment the multiview images and generate the corresponding masks. Third, a point cloud segmentation method using multiview fusion is proposed to extract the target PC component point cloud, facilitating the dimensional quality assessment. Compared to traditional point cloud segmentation methods (i.e., DBSCAN, K-means, mean shift, and region growing), the proposed method achieves the highest F-score exceeding 99.5%, and the relative errors between the calculated dimensions and the actual dimensions measured manually are below 1.3%. The results demonstrate the effectiveness of the proposed method in segmenting the point clouds of PC components and facilitating the dimensional quality assessment.
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.
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.
Autonomous monitoring strategies are becoming an increasingly popular application modality due to their adaptable deployment and reconfigurability. Motivated by low-cost robotic swarms conveying ultrasonic transducers that are maneuverable into arrays for pipe inspections, this paper aims to develop an initial framework for predicting such an array’s performance and understanding the effects of target defect scattering properties and array topology parameters on that performance. Assuming uniformly omnidirectional transducers, we first develop a predictive spatial probability of detection (POD) model that depends on target defect scattering and distance and verify it against experiment. We then synthesize these single-transducer models into arrays and derive global POD performance metrics that are parameterized by defect scattering properties (via the single transducer model) and array topology (geometric arrangement and transducer pitch). We perform a performance evaluation in the case of a highly directional scatterer (e.g., a crack) over the global parameter space to make suggestions about array design.
