Search for: All records

The most common eye infection in people with diabetes is diabetic retinopathy (DR). It might cause blurred vision or even total blindness. Therefore, it is essential to promote early detection to prevent or alleviate the impact of DR. However, due to the possibility that symptoms may not be noticeable in the early stages of DR, it is difficult for doctors to identify them. Therefore, numerous predictive models based on machine learning (ML) and deep learning (DL) have been developed to determine all stages of DR. However, existing DR classification models cannot classify every DR stage or use a computationally heavy approach. Common metrics such as accuracy, F1 score, precision, recall, and AUC-ROC score are not reliable for assessing DR grading. This is because they do not account for two key factors: the severity of the discrepancy between the assigned and predicted grades and the ordered nature of the DR grading scale. 

This research proposes computationally efficient ensemble methods for the classification of DR. These methods leverage pre-trained model weights, reducing training time and resource requirements. In addition, data augmentation techniques are used to address data limitations, improve features, and improve generalization. This combination offers a promising approach for accurate and robust DR grading. In particular, we take advantage of transfer learning using models trained on DR data and employ CLAHE for image enhancement and Gaussian blur for noise reduction. We propose a three-layer classifier that incorporates dropout and ReLU activation. This design aims to minimize overfitting while effectively extracting features and assigning DR grades. We prioritize the Quadratic Weighted Kappa (QWK) metric due to its sensitivity to label discrepancies, which is crucial for an accurate diagnosis of DR. This combined approach achieves state-of-the-art QWK scores (0.901, 0.967 and 0.944) in the Eyepacs, Aptos, and Messidor datasets.

Herein, a new lightweight syntactic foam is reported with strong mechanical properties, unique multifunctionalities, and recyclability. Multifunctionality of materials and structures has gained ever‐increasing interest as an excellent approach to designing minimalistic systems. Inspired by nature, these materials can perform multiple functions besides bearing a load. Due to their shape‐changing and damage‐healing property, shape memory vitrimers (SMVs) are a great example of multifunctional materials readily exploited for many applications. Using nickel and silver‐plated hollow glass microbubbles (HGMs), an SMV‐based syntactic foam is introduced here that supplements the multifunctionality of SMVs with electrical conductivity and ferromagnetism, which enables a series of additional potentials such as strain sensing, damage monitoring, Joule heating, and electromagnetic interference shielding. Despite its low density and outstanding mechanical properties, this foam exhibits shape memory behavior, which can be triggered by an electrical current, and damage healing capability due to its reversible dynamic covalent bonds. Especially its recyclability makes recycling the expensive silver‐coated and nickel‐coated HGMs feasible, making this foam cost‐effective and environmentally sustainable. With its many features and economical manufacturability, this syntactic foam has a potential to be utilized in many applications, ranging from aerospace structures to biomedical devices to household items.

Frontal polymerization (FP) of epoxy monomers is typically achieved with a radical‐induced cationic frontal polymerization (RICFP) process that combines a thermal radical initiator with an onium salt superacid generator. In this paper, we show that both thermal and UV‐initiated cationic frontal polymerizations are possible for common epoxy and vinyl ether monomers with only an iodonium superacid generator in the absence of a standalone thermal radical initiator. Increasing superacid generator concentration resulted in an increase in front velocity, as did the addition of vinyl ether to epoxies. The front velocity is reduced by the addition of 4‐methoxyphenol (MeHQ), indicating free‐radicals play a significant role.

Cultivated natural fibers have a huge possibility for green and sustainable reinforcement for polymers, but their limited load-bearing ability and flammability prevent them from wide applications in composites. According to the beam theory, normal stress is the maximum at the outermost layers but zero at the mid-plane under bending (with (non)linear strain distribution). Shear stress is the maximum at the mid-plane but manageable for most polymers. Accordingly, a laminated composite made of hybrid fiber-reinforced shape memory photopolymer was developed, incorporating strong synthetic glass fibers over a weak core of natural hemp fibers. Even with a significant proportion of natural hemp fibers, the mechanical properties of the hybrid composites were close to those reinforced solely with glass fibers. The composites exhibited good shape memory properties, with at least 52% shape fixity ratio and 71% shape recovery ratio, and 24 MPa recovery stress. After 40 s burning, a hybrid composite still maintained 83.53% of its load carrying capacity. Therefore, in addition to largely maintaining the load carrying capacity through the hybrid reinforcement design, the use of shape memory photopolymer endowed a couple of new functionalities to the composites: the plastically deformed laminated composite beam can largely return to its original shape due to the shape memory effect of the polymer matrix, and the flame retardancy of the polymer matrix makes the flammable hemp fiber survive the fire hazard. The findings of this study present exciting prospects for utilizing low-strength and flammable natural fibers in multifunctional load-bearing composites that possess both flame retardancy and shape memory properties.

We report progress towards development of a cyber-physical trust anchor for additive manufacturing systems. The additive manufacturing commercial sector needs cyber-physical trust anchors to establish a secure supply chain, to detect counterfeiting and to ensure part provenance. However, the underlying technology of cyber-physical trust anchors requires optimization and spans several sectors ranging from mathematics, additive manufacturing, materials science, nondestructive evaluation, to cyber science. The fast and effective deployment of cyber-physical trust anchors requires an educational component. This project present a novel method for authenticating additively manufactured parts. Features are extracted using advanced X-ray imaging, transformed into unique identifiers, and bound with security features for cloud-based blockchain authentication. A plan for the low-cost and safe incorporation of cyber-physical trust anchor research in education is included. The anticipated outcome is an optimized trust anchor prototype and educational product suitable for interdisciplinary research and coursework to develop the workforce needed for cyber-secured physical supply chainsd.

Damage healing in fiber reinforced thermoset polymer composites has been generally divided into intrinsic healing by the polymer itself and extrinsic healing by incorporation of external healing agent. In this study, we propose to use a hybrid extrinsic-intrinsic self-healing strategy to heal delamination in laminated composite induced by low velocity impact. Especially, we propose to use an intrinsic self-healing thermoset vitrimer as an external healing agent, to heal delamination in laminated thermoset polymer composites. To this purpose, we designed and synthesized a new vitrimer, machined it into powders, and strategically sprayed a layer of vitrimer powders at the interface between the laminas during manufacturing. Also, a thermoset shape memory polymer with fire-proof property was used as the matrix. As a result, incorporation of about 3% by volume of vitrimer powders made the laminate exhibit multifunctionalities such as repeated delamination healing, excellent shape memory effect, improved toughness and impact tolerance, and decent fire-proof properties. In particular, the novel vitrimer powder imparted the laminate with first cycle and second cycle delamination healing efficiencies of 98.06% and 85.93%, respectively. The laminate also exhibited high recovery stress of 65.6 MPa. This multifunctional composite laminate has a great potential in various engineering applications, for example, actuators, robotics, deployable structures, and smart fire-proof structures.

We analyzed the effects of crosslinking fraction and number of functional sites per hardener molecule on the stress recovery and topology of thermoset shape memory polymers (TSMPs) via coarse-grained molecular dynamics simulations. After systematically varying the quality of the crosslinked network by manipulating the number of unique epoxies reacted with each hardener, we found that two fingerprints correlate well with stress recovery of TSMPs. These fingerprints are the fraction of epoxy molecules connected to two distinct hardener molecules, and the fraction of molecules that are part of the largest or main network in the system. Their product can be used as a topological score ($S_{\text{topo}}$) to quantify the topological feature of the network. When analyzing stress recovery as a function of$S_{\text{topo}}$, we found a strong correlation between$S_{\text{topo}}$and recovery stress. Moreover, we observed that while a higher crosslinking fraction did frequently lead to a higher stress recovery, many exceptions existed. High functionality hardeners tend to exhibit higher stress recovery at similar$S_{\text{topo}}$, especially at high (>0.65)$S_{\text{topo}}$. These results suggest that increasing the number of functional sites per hardener molecule combined with improving the topology of the network with a method such as semi batch monomer addition can lead to an improvement in the stress recovery of TSMPs.

A catalytic surface should be stable under reaction conditions to be effective. However, it takes significant effort to screen many surfaces for their stability, as this requires intensive quantum chemical calculations. To more efficiently estimate stability, we provide a general and data-efficient machine learning (ML) approach to accurately and efficiently predict the surface energies of metal alloy surfaces. Our ML approach introduces an element-centered fingerprint (ECFP) which was used as a vector representation for fitting models for predicting surface formation energies. The ECFP is significantly more accurate than several existing feature sets when applied to dilute alloy surfaces and is competitive with existing feature sets when applied to bulk alloy surfaces or gas-phase molecules. Models using the ECFP as input can be quite general, as we created models with good accuracy over a broad set of bimetallic surfaces including most d-block metals, even with relatively small datasets. For example, using the ECFP, we developed a kernel ridge regression ML model which is able to predict the surface energies of alloys of diverse metal combinations with a mean absolute error of 0.017 eV atom⁻¹. Combining this model with an existing model for predicting adsorption energies, we estimated segregation trends of 596 single-atom alloys (SAAs)with and without CO adsorbed on these surfaces. As a simple test of the approach, we identify specific cases where CO does not induce segregation in these SAAs.

This paper examined the effect of Si addition on the cracking resistance of Inconel 939 alloy after laser additive manufacturing (AM) process. With the help of CALculation of PHAse Diagrams (CALPHAD) software Thermo-Calc, the amounts of specific elements (C, B, and Zr) in liquid phase during solidification, cracking susceptibility coefficients (CSC) and cracking criterion based on$$\left| {{\text{d}}T/{\text{d}}f_{{\text{s}}}^{1/2} } \right|$$$\left(\text{d}T/\text{d}{f}_{\text{s}}^{1/2}\right)$values (T: solidification temperature,f_s: mass fraction of solid during solidification) were evaluated as the indicators for composition optimization. It was found that CSC together with$$\left| {{\text{d}}T/{\text{d}}f_{{\text{s}}}^{1/2} } \right|$$$\left(\text{d}T/\text{d}{f}_{\text{s}}^{1/2}\right)$values provided a better prediction for cracking resistance.

Graphical abstract

Dissociation of CO₂on iron clusters was studied by using semilocal density functional theory and basis sets of triple‐zeta quality. Fe₂, Fe₄, and Fe₁₆clusters were selected as the representative host clusters. When searching for isomers of Fe_nCO₂,n=2, 4 and 16 corresponding to carbon dioxide attachment to the host clusters, its reduction to O and CO, and to the complete dissociation, it was found that the total spin magnetic moments of the lowest energy states of the isomers are often quenched with respect to those of initial reagents Fe_n+CO₂. Dissociation pathways of the Fe₂+CO₂, Fe₄+CO₂, and Fe₁₆+CO₂reactions contain several transition states separated by the local minima states; therefore, a natural question is where do the spin flips occur? Since lifetimes of magnetically excited states were shown to be of the order of 100 fs, the search for the CO₂dissociation pathways was performed under the assumption that magnetic deexcitation may occur at the intermediate local minima. Two dissociation pathways were obtained for each Fe_n+CO₂reaction using the gradient‐based methods. It was found that the Fe₂+CO₂reaction is endothermic with respect to both reduction and complete dissociation of CO₂, whereas the Fe₄+CO₂and Fe₁₆+CO₂reactions are exothermic to both reduction and complete dissociation of carbon dioxide. The CO₂reduction was found to be more favorable than its complete dissociation in the Fe₄case.