114333 research outputs found
Sort by
Multivariate logistic wind fragility functions of overhead distribution poles
No full textOverhead distribution lines are highly vulnerable to extreme wind events, yet fragility modeling remains challenging due to multiple interacting failure mechanisms and strong dependence on asset and site characteristics. This study presents a physics-informed and scalable framework for modeling multivariate wind-induced fragility of overhead utility networks. The approach integrates stochastic wind simulation, generated to match the Kaimal turbulence spectrum and validated using spectral error metrics, with finite-element–based structural analysis to capture governing failure mechanisms, including pole rupture, pole overturning, and conductor tensile breakage. Mode-specific multivariate logistic regression models are calibrated as parameterized, closed-form surrogate fragility functions based on measurable asset, geotechnical, and hazard descriptors, allowing for direct application without the need for repeated finite-element simulations or statistical analysis. The resulting models demonstrate strong and stable predictive performance across pole classes and failure modes, with AUC values exceeding 0.97. Joint failure probability is quantified using both independence-based aggregation and a copula-based formulation that captures dependence between competing failure modes, enabling consistent risk aggregation from individual spans to feeder-level applications. By combining physics-based modeling with computational efficiency and closed-form fragility expressions, the proposed framework supports practical, risk-informed screening and planning of power distribution infrastructure under extreme wind hazards.This article is published as Qudaisat, Muneer, and Alice Alipour. "Multivariate Logistic Wind Fragility Functions of Overhead Distribution Poles." International Journal of Disaster Risk Reduction (2026): 105996. doi: https://doi.org/10.1016/j.ijdrr.2026.105996.This paper is based upon work supported by the National Science Foundation under Award No. 2429602
Bridging the real-synthetic gap in microscopy with inversion-based diffusion
No full textAccurate cell counting in fluorescence microscopy images is a fundamental task in biomedical research and clinical diagnostics, supporting applications such as cancer monitoring and stem cell therapy. However, the effectiveness of deep learning approaches for automated cell counting is often limited by the scarcity of large, well-annotated microscopy datasets. One promising strategy to address this limitation is the generation of synthetic training data. Yet, a significant performance gap persists between models trained on synthetic data and those evaluated on real images, primarily due to differences in visual appearance and structural complexity—a challenge known as the domain gap.
This thesis proposes a novel Sim2Real framework based on Inversion-Based Style Transfer (InST) with diffusion models to bridge the real-synthetic gap in microscopy. Specifically, the method leverages latent-space Adaptive Instance Normalization (AdaIN) and stochastic inversion within a diffusion-based generative model to transfer the visual style of real microscopy images onto synthetically generated images, while weakly preserving the underlying cell structure. This approach enables the synthesis of realistic, structure-aware microscopy images that can supplement limited annotated data for model training.
Comprehensive experiments were conducted to evaluate the effectiveness of the proposed InST-based framework for downstream cell counting tasks. By pre-training and fine-tuning EfficientNet-B0 models on various data sources—including real images, hard-coded synthetic data, and the publicly available Cell200-s dataset—our results demonstrate that models trained with InST-synthesized images achieve up to a 37\% reduction in Mean Absolute Error (MAE) compared to models trained on hard-coded synthetic data, and a 52\% reduction compared to Cell200-s. Remarkably, this approach also outperforms models trained solely on real data. Further improvements were achieved when combining InST-synthesized data with lightweight domain adaptation techniques such as DACS with CutMix.
The findings of this thesis provide strong evidence that InST-based style transfer can substantially reduce the domain gap between synthetic and real microscopy data, offering a scalable and effective pathway for improving automated cell counting performance while minimizing the need for manual annotation
Aberrant skeletal muscle morphogenesis and myofiber differentiation characterize equine myotonic dystrophy
No full textEquine myotonic dystrophy (eMD) is a rare neuromuscular disorder of undetermined origin marked by muscle hypertrophy and stiffness, dystrophic muscle histopathology, and myotonic discharges. In humans, myotonic dystrophy (DM) arises from trinucleotide repeat expansions in dystrophia myotonica protein kinase (DMPK) (DM1) or tetranucleotide expansions in cellular nucleic acid-binding protein (CNBP) (DM2), which disrupt mRNA processing and induce embryonic splicing patterns across multiple genes. In 6 eMD Quarter Horse types, (2–36 months-of-age) and 8 control Quarter Horses we determined: (1) fiber type composition of triceps, gluteal, and semimembranosus muscles; (2) differential gene (DEG) and protein (DEP) expression using transcriptomic and proteomic analyses; (3) presence of repeat expansions in transcripts of DMPK or CNBP and (4) exon 7 retention in CLCN1 or exon 22 splicing in ATP2A1. Predominance and clustering of type 1 fibers, expression of embryonic myosin, and upregulated mitochondrial and sarcomeric DEPs characterized eMD hindlimb musculature. Gene ontology (GO) analysis of 730 upregulated DEGs identified numerous GO terms related to morphogenesis of mesoderm-derived tissues and upregulated genes impacting myoD expression in eMD muscle. Top upregulated DEG involved myogenesis (MYOZ2, SBK2, SBK3, PAMR1), neurons, transcription/translation, cytoskeleton, basement/plasma membranes, and calcium binding/transport. Top upregulated proteins also impacted muscle morphogenesis (MUSTN1, CSRP3, TMSBX4, PDLIM, CALD1) as well as categories of mitochondria, sarcomere, extracellular matrix/ basement membrane, transcription, translation, cell cycle regulation, neurons amongst others. Downregulated DEP primarily impacted mitochondria, the sarcomere and glycogen metabolism. Notably, unlike human myotonic dystrophy, trinucleotide repeat expansions were not found in the DMPK 3’UTR (CTG)n nor tetranucleotide repeat expansions (CCTG)n in intron 1 of CNBP. Isoforms of CLCN1 containing fetal exon 7 were detected in equal frequency in eMD and control muscle and exon 22 was not alternatively spliced in ATP2A1 as has been found in DM1. Thus, distinct from DM1 and DM2, eMD is driven by unique molecular mechanisms impacting skeletal muscle morphogenesis, neurons and regulation of gene transcription/translation that alter fiber type composition, distribution and morphology. The origin of myotonia does not appear to be driven by a mutation in CLCN1 or retention of exon CLCN 7. Expanded splice site analysis and further research is warranted to elucidate the cause of myotonia and the distinct etiology of eMD.This article is published as Valberg SJ, Williams ZJ, Ames EG, Mickelson JR, Nout-Lomas YS, Landolt G, et al. (2026) Aberrant skeletal muscle morphogenesis and myofiber differentiation characterize equine myotonic dystrophy. PLoS One 21(1): e0341655. doi: https://doi.org/10.1371/journal.pone.0341655.Internal Funding from the College of Veterinary Medicine, Michigan State University Freeman Fund and Mary Anne McPhail Endowment, College of Veterinary Medicine, Michigan State University
Analysis and development of gas atomization for high molecular weight polymers
No full textAdditive manufacturing (AM) using polymer powders is a rapidly expanding field, yet its growth is constrained by the limited availability of suitable polymer feedstocks. Current powder production methods—such as cryogenic milling, solvent deposition, and emulsion polymerization—are often limited in the polymers that can be used, or inefficient, producing irregular particle morphologies that hinder flowability and process reliability. These limitations are particularly pronounced for commodity polymers like polyolefins, which account for the majority of commercial polymer production but remain underrepresented in AM powder libraries. Gas atomization, widely employed for metal powders, offers a promising alternative for generating spherical, free-flowing particles. However, its application to polymers presents unique challenges due to their high viscosity and non-Newtonian melt behavior.
This work investigates the adaptation of gas atomization for polymer systems, focusing on polyethylene glycol (PEG, 20,000 g/mol) as a model material. A novel gas die geometry was developed to form the polymer melt into a sheet prior to atomization, enabling an expedited breakup mechanism distinct from conventional droplet formation models. Additionally, heated atomization gas was used to reduce fibrous byproducts and promote spherical particle formation, with minimum effective gas temperatures correlated to polymer thermal transitions. Experimental observations confirm the presence of the proposed sheet-based breakup mode and demonstrate improved powder morphology under optimized conditions.
These findings are then applied to high-density polyethylene (HDPE) to explore the translation of parameters from lower molecular weight polymers to commercially significant high molecular weight polymers. This work establishes critical design and process parameters for polymer gas atomization (PGA), providing a foundation for extending this technique to other high molecular weight polymers. By addressing key barriers in polymer powder production, this research advances the feasibility of PGA as a scalable, efficient method for producing high-quality powders which could expand material options for additive manufacturing and bridge the gap between commercial polymer production and AM feedstock availability
Parallel, lock-free framework for decision diagram (DD) based search
No full textThis thesis presents the development of a parallel Decision Diagram-based Benders Decomposition (DD-BD) solver for stochastic mixed-integer programming and introduces a new lock-free work-stealing algorithm designed to overcome its scheduling bottlenecks. Existing work-stealing queues performed inefficiently for solver workloads involving bulk node generation and single-stealer concurrency. To address this, an unbounded queue supporting native bulk push and steal operations was developed, achieving lock-free progress and constant-latency performance. Benchmarks show significant improvements over state-of-the-art implementations such as C++ Taskflow. The algorithm has been integrated into the solver's parallel framework, forming the foundation for further work on adaptive load balancing and distributed scalability in large-scale stochastic optimization
Epigenetic regulation during soybean Phytophthora sojae interactions: Evolutionary and functional analysis of demethylase genes
No full textDNA methylation, an epigenetic modification, has established roles in regulating transposable element suppression, gene expression, genome stability, and response to stress factors. DNA demethylation antagonizes DNA methylation through the removal of methyl groups from DNA by enzymes (DNA demethylase). DNA demethylation is a conserved biological process that has been linked to the regulation of plant growth, development, and stress responses. Dynamic changes in methylation patterns, influenced by DNA demethylase (dMTase), have been reported to be essential in plant-pathogen interactions. However, in soybean, very little is known about the function of dMTase during plant-pathogen interactions. In this study, we characterize five soybean dMTases (GmROS2-1, GmROS2-2, GmROS1, GmDME1-1, and GmDME1-2) and elucidate their functions. To examine the evolutionary relationships between these genes, a phylogenetic and dN/dS analysis was performed, revealing recent paralogous relationships between GmROS2-1 and GmROS2-2, and between GmDME1-1 and GmDME1-2. Furthermore, our analysis determined that GmROS1 was the only dMTase under positive selection. While genic and domain annotations were relatively conserved across the five soybean dMTases, motif patterns in the protein sequences revealed unique patterns between the two GmDME1, two GmROS2, and GmROS1. These patterns are consistent with how soybean dMTase clustered in the phylogenetic tree. Furthermore, the tertiary protein structure between soybean and Arabidopsis known dMTase revealed high conservation of the DNA glycosylase region, implying, soybean dMTase shares the core catalytic characteristics of demethylation. The distribution of TF binding sites across the promoter and UTR regions of soybean dMTase genes indicates that, although each gene possesses a largely unique regulatory signature, shared TF-binding motifs across several soybean dMTases highlight partially conserved regulatory roles. Consistent with the expression data of the soybean demethylase, it was revealed that ROS-like genes are more highly expressed than DME-like genes in root and shoot tissues. This is consistent with known data in Arabidopsis, where AtROS1 is more expressed in vegetative tissue and AtDME is more expressed in the endosperm and embryo. To further elucidate the role of soybean dMTases in plant immunity, we generated CRISPR-Cas9 knockout mutants of gmros2-1(1) from the Williams 82 (Wm82) variety. Successful knockout lines were then infected with the Phytophthora sojae isolate, P6497, to see if resistance in the empty vector (EV) could be overcome in the gmros2-1 knockout lines. Tissue integrity analysis and qPCR revealed diseased tissue in the mutants and in the EV. While we noticed that the EVs trended similarly to our control Wm82 seedlings, the difference was not statistically significant. Given the varying data, further optimization of the inoculation and evaluation process is necessary. Collectively, these results are a step forward in understanding how soybean dMTase contributes to plant immunity against P. sojae
Saturated buffer performance under alternative weir settings: Implications for design and management
No full textSaturated buffers are important edge-of-field conservation practices to reduce nitrate-nitrogen (NO3-N) loading from subsurface (tile) drainage systems to downstream waters. The impact of seasonal management of weir elevations in the water control structure on NO3-N removal has not been well studied. This study evaluated the effect of control box weir elevation management on nitrate removal and compared in situ flow treatment with design predictions from the USDA Natural Resources Conservation Service conservation practice standard 604. A 253 m long saturated buffer draining approximately 6 ha was monitored for flow and NO3-N load from 2022 to 2024. The weir elevation was adjusted to “full drainage” (no treatment), “growing season” (reduced treatment capacity), and “fallow season” (full treatment capacity) settings according to weather conditions and field operations. During a 29-day full drainage period in 2022, the saturated buffer bypassed 25% of the annual drainage flow and 28% of the annual NO3-N load. However, the fraction of flow treated and NO3-N load removal efficiency was greater in 2022 than 2023 and 2024. Treated flow within the saturated buffer was greater than predicted, while peak drainage system flow was less, resulting in a greater percentage of drainage system capacity treated by the saturated buffer than designed. These discrepancies suggest that alternative design methods should be explored. While the saturated buffer removed substantial NO3-N in the year with alternative weir management, careful consideration should be given for potential sites that may require extended full drainage periods, as large NO3-N losses can bypass during such conditions.This article is published as Johnson, Gabriel M., Thomas M. Isenhart, Christopher Hay, and Andrew J. Craig. Saturated buffer performance under alternative weir settings: Implications for design and management. Journal of Environmental Quality 55 (2026}: e70136. https://doi.org/10.1002/jeq2.70136Natural Resources Conservation Service, Grant/Award Number: Conservation Innovation Grant NR213A750013G03
Quantifying and interpreting swine co-Infections using standardized diagnostic data and field-based pathogen interaction assessment
No full textThe complexity and the need for integrated diagnostic interpretation are increasingly challenging swine disease surveillance, particularly in cases of co-infection. This dissertation addresses critical gaps in understanding swine disease co-detection dynamics by leveraging confirmed tissue diagnoses, developing composite disease metrics, and evaluating field-level pathogen interactions. Considering (a) the diagnostic complexity of multifactorial swine diseases, (b) the need for scalable tools to monitor endemic pathogen activity, and (c) the importance of optimizing sampling strategies for surveillance, this dissertation explores diagnostic integration, disease indexing, and co-detection impacts on performance.
The overarching objectives were: (a) to characterize co-diagnosis patterns in swine using confirmed tissue diagnoses across multiple veterinary diagnostic laboratories, (b) to develop a disease index quantifying endemic pathogen activity, and (c) to evaluate oral fluid sampling as a diagnostic tool for detecting co-infections and assessing performance impacts.
Chapter 2 examined co-diagnosis trends using over 45,000 confirmed tissue diagnoses from two U.S. veterinary diagnostic laboratories. Co-diagnosis was reported in over half of the cases, especially during the wean-to-finish phase. Statistical modeling using COM-Poisson regression revealed that cases involving bacterial and viral insults were significantly associated with more distinct etiologies, underscoring the complexity of swine disease interactions.
Chapter 3 introduced a novel disease index to quantify endemic pathogen activity, integrating four variables—disease occurrence, co-diagnosis, state occurrence, and syndromic surveillance alarms—using weighted normalization and bootstrap modeling. The index demonstrated strong temporal consistency and sensitivity to emerging pathogen trends, such as porcine sapovirus and astrovirus, while consistently ranking PRRSV and Streptococcus suis as dominant endemic pathogens.
Chapter 4 evaluated the diagnostic performance of oral fluids and pooled pen fecal samples tested by PCR for detecting Lawsonia intracellularis DNA using Bayesian latent class analysis. Oral fluids showed higher sensitivity in high-prevalence scenarios, validating their utility for herd-level surveillance.
Chapter 5 investigated the co-detection dynamics of Lawsonia intracellularis, PCV2, and PRRSV in wean-to-finish pig groups. Groups with elevated detection of Lawsonia intracellularis and PCV2 in the presence of PRRSV exhibited significantly higher mortality and reduced growth performance, highlighting the impact of pathogen interactions on productivity.
This dissertation advances swine disease surveillance, primarily at the co-diagnosis level, by integrating diagnostic data across laboratories and also assessing field-level co-detections through field studies conducted on commercial farms. The findings provide actionable insights for veterinarians, diagnosticians, and producers to enhance disease monitoring, interpretation, and herd health management in the U.S. swine industry
Molecular mechanisms and therapeutic strategies in organophosphate neurotoxicity
No full textOrganophosphates (OPs) are chemical nerve agents that propagate a prolonged state of seizure, known as status epilepticus, due to overstimulation of the cholinergic system. While medical countermeasures such as oximes, atropine, and benzodiazepines reduce mortality rates, they are unable to rescue the long-term consequences of OP exposure. Survivors experience persistent spontaneous recurrent seizures, behavioral comorbidities, including anxiety and memory loss, neurodegeneration, and neuroinflammation. Furthermore, those who suffer from OP intoxication have a poor prognosis for long-term quality of life. A key driver of lasting brain injury is oxidative stress, a state in which the overproduction of reactive oxygen species alters DNA, triggers neuronal cell death, and aberrantly activates inflammatory brain cells, including microglia and astrocytes. A significant source of reactive oxygen species is the enzyme NADPH oxidase (NOX). Our previous work demonstrated an upregulation of NOX and other oxidative stress markers following exposure to the OP diisopropylfluorophosphate (DFP) in rodent models. Here, we tested the mitochondrial-targeted NOX inhibitor mitoapocynin (MPO) to determine its efficacy in reducing oxidative stress and brain injury in rats exposed to DFP. We found that while MPO reduced OP-induced inflammation and oxidative stress markers in the serum, it had no effect on neuroinflammatory or neurodegenerative markers. Based on previous concerns about the bioavailability of apocynin and its derivatives, we speculated that the dosing regimen required optimization to achieve maximum efficacy in the brain. Therefore, we tested an increased dose of MPO by the oral route of administration, as well as MPO-encapsulated polyanhydride nanoparticles (NPs) via the intramuscular route in the DFP model. Both the increased dose of free drug MPO and MPO-NP treatments reduced reactive astrogliosis in the brain 8 days after DFP. These findings demonstrate that MPO attenuates signs of neuroinflammation historically associated with maladaptive long-term outcomes. To evaluate the safety and efficacy of nanoparticle carriers, we tested two formulations of NPs and a cocktail of the two. Liver, kidney, and brain function in animals treated with polyanhydride NPs remained normal, indicating that polyanhydride NPs are well tolerated in rats via the intramuscular route of administration. In the same study, we investigated the combination of MPO-NP (i.m.) and MPO oral dosing following OP exposure. Surprisingly, animals exposed to DFP and treated with MPO showed signs of liver dysfunction and weight loss, whereas non-exposed animals given MPO did not. In the final study, we investigated novel pathological mechanisms of excitotoxicity and neuroinflammation. We found that Src family kinases, enzymes that mediate glutaminergic receptor signaling and microglia reactivity, were maladaptively upregulated following DFP exposure. Overall, our findings show that MPO attenuates signs of neuroinflammation associated with long-term symptoms of OP poisoning, though its efficacy and tolerability are dependent upon the dosing regimen. Future studies will focus on optimizing MPO dosing and conducting a thorough investigation of other NOX isoforms in the DFP model. Additionally, we identified Src family kinases as a contributing pathological mechanism in OP-induced neurotoxicity. Together, the outcomes of this dissertation advance our understanding of oxidative and excitotoxic mechanisms and identify therapeutic strategies to improve long-term outcomes after OP poisoning and acquired epilepsy
DURACIM: Durable compute-in-memory (CiM) inference for resource-limited environments
No full textDeep neural networks (DNNs) often encounter a performance bottleneck known as the memory wall, where the compute units are faster and consume less energy than data movement from memory. Compute-in-Memory (CIM) architectures can support MAC operations within memory arrays, helping to resolve the memory wall. However, they suffer from a hardware trade-off. High-density RRAM crossbars are well-suited for efficient analog computation; however, they tend to wear out quickly under repeated writes. In contrast, SRAM arrays have high endurance but require a significantly larger area. The following thesis proposes a heterogeneous CIM mapping approach that supports memory technology matching with the data-access and write behavior of the network. Stable, high-read-load early layers are mapped into dense RRAM (as dynamic tiles) to utilize parallel analog computation, while deeper, write-heavy layers are mapped out to SRAM (as static tiles) for long-term durability. This design approach yields a fast and energy-efficient solution.
Early-exit classifiers serve as a mechanism to minimize the number of samples that enter deep layers, lowering the write stress on the endurance-constrained memory regions. The proposed heterogeneous mapping, evaluated on ResNet-50 (CIFAR-100), provides 86-88% relative energy savings compared to a full-depth baseline while preserving accuracy within approximately one percentage point.
Through the transition from deep-layer computation to high-endurance SRAM static tiles, the system is able to increase the effective RRAM (dynamic tiles) lifespan to approximately 1.02 x10^7 years, essentially removing endurance as a practical constraint. Even when using RRAM for both dynamic and static tiles, the PPO-discovered early exits increase the lifespan by approximately 1.9 times compared to the baseline. In general, these findings suggest that the intelligent design of distributed DNN workloads, which are distributed across varied CIM memories rather than treating the architecture as a unified one, is essential to achieve high performance, high energy efficiency, and durable CIM accelerators