Showing papers by "General Electric published in 2020"

Deep learning for tomographic image reconstruction

Abstract: Deep-learning-based tomographic imaging is an important application of artificial intelligence and a new frontier of machine learning. Deep learning has been widely used in computer vision and image analysis, which deal with existing images, improve these images, and produce features from them. Since 2016, deep learning techniques have been actively researched for tomographic imaging, especially in the context of biomedicine, with impressive results and great potential. Tomographic reconstruction produces images of multi-dimensional structures from externally measured ‘encoded’ data in the form of various tomographic transforms (integrals, harmonics, echoes and so on). In this Review, we provide a general background, highlight representative results with an emphasis on medical imaging, and discuss key issues that need to be addressed in this emerging field. In particular, tomographic imaging is an integral part of modern medicine, and will play a key role in personalized, preventive and precision medicine and make it intelligent, inexpensive and indiscriminate. The popularity of deep learning is leading to new areas in biomedical applications. Wang and colleagues summarize in this Review the recent development and future directions of deep neural networks for superior image quality in the tomographic imaging field.

The Cancer Microbiome: Distinguishing Direct and Indirect Effects Requires a Systemic View

Abstract: The collection of microbes that live in and on the human body - the human microbiome - can impact on cancer initiation, progression, and response to therapy, including cancer immunotherapy. The mechanisms by which microbiomes impact on cancers can yield new diagnostics and treatments, but much remains unknown. The interactions between microbes, diet, host factors, drugs, and cell-cell interactions within the cancer itself likely involve intricate feedbacks, and no single component can explain all the behavior of the system. Understanding the role of host-associated microbial communities in cancer systems will require a multidisciplinary approach combining microbial ecology, immunology, cancer cell biology, and computational biology - a systems biology approach.

Feature Space Augmentation for Long-Tailed Data

Abstract: Real-world data often follow a long-tailed distribution as the frequency of each class is typically different. For example, a dataset can have a large number of under-represented classes and a few classes with more than sufficient data. However, a model to represent the dataset is usually expected to have reasonably homogeneous performances across classes. Introducing class-balanced loss and advanced methods on data re-sampling and augmentation are among the best practices to alleviate the data imbalance problem. However, the other part of the problem about the under-represented classes will have to rely on additional knowledge to recover the missing information. In this work, we present a novel approach to address the long-tailed problem by augmenting the under-represented classes in the feature space with the features learned from the classes with ample samples. In particular, we decompose the features of each class into a class-generic component and a class-specific component using class activation maps. Novel samples of under-represented classes are then generated on the fly during training stages by fusing the class-specific features from the under-represented classes with the class-generic features from confusing classes. Our results on different datasets such as iNaturalist, ImageNet-LT, Places-LT and a long-tailed version of CIFAR have shown the state of the art performances.

Multi-Agent Based Attack-Resilient System Integrity Protection for Smart Grid

Abstract: Most System Integrity Protection (SIP) schemes deployed in smart gird today are centralized functions relying on wide-area communication. The highly centralized implementation makes SIP susceptible to the single point of failure induced by cyber attacks. In this paper, we present a novel multi-agent-based design to enhance the cyber resilience of SIP while focusing on augmenting its situational awareness and self-adaptiveness. Specifically, we have investigated data-driven anomaly detection and adaptive load rejection within the decentralized SIP set-up. After attaining a comprehensive taxonomy of operation states of a power grid as a cyber-physical system, we are able to convert the anomaly detection to a multi-class classification problem. A supervised learning algorithm, named as Support Vector Machine embedded Layered Decision Tree (SVMLDT), is proposed as a possible solution. Anomaly detection is carried out by every agent separately, but the final decision depends on the consensus among all interconnected agents. Besides, we propose an adaptive load rejection strategy to mitigate the Denial of Service (DoS) attacks targeting the load shedding scheme. A real load rejection SIP scheme adopted by Salt River Project is modified to fit in the IEEE 39-bus model as a study case. Experiment results show that the proposed SIP can detect anomalous grid operation states and then adjust its remedial actions accordingly to adapt to the under-attack situations.

Extraordinary performance of semiconducting metal oxide gas sensors using dielectric excitation

Abstract: Semiconducting metal oxides are widely used for gas sensors. The resulting chemiresistor devices, however, suffer from non-linear responses, signal fluctuations and gas cross-sensitivities, which limits their use in demanding applications of air-quality monitoring. Here, we show that conventional semiconducting metal oxide materials can provide high-performance sensors using an impedance measurement technique. Our approach is based on dielectric excitation measurements and yields sensors with a linear gas response (R2 > 0.99), broad dynamic range of gas detection (six decades of concentrations) and high baseline stability, as well as reduced humidity and ambient-temperature effects. We validated the technique using a range of commercial sensing elements and a range of gases in both laboratory and field conditions. Our approach can be applied to both n- and p-type semiconducting metal oxide materials, and we show that it can be used in wireless sensor networks, and drone-based and wearable environmental and industrial gas monitoring. Semiconducting metal oxide gas sensors with a linear response, broad dynamic range and high baseline stability can be created with the help of a dielectric excitation technique.

Feature Space Augmentation for Long-Tailed Data

Abstract: Real-world data often follow a long-tailed distribution as the frequency of each class is typically different. For example, a dataset can have a large number of under-represented classes and a few classes with more than sufficient data. However, a model to represent the dataset is usually expected to have reasonably homogeneous performances across classes. Introducing class-balanced loss and advanced methods on data re-sampling and augmentation are among the best practices to alleviate the data imbalance problem. However, the other part of the problem about the under-represented classes will have to rely on additional knowledge to recover the missing information.

The origin and formation of oxygen inclusions in austenitic stainless steels manufactured by laser powder bed fusion

Abstract: The origins of nano-scale oxide inclusions in 316L austenitic stainless steel (SS) manufactured by laser powder bed fusion (L-PBF) was investigated by quantifying the possible intrusion pathways of oxygen contained in the precursor powder, extraneous oxygen from the process environment during laser processing, and moisture contamination during powder handling and storage. When processing the fresh, as-received powder in a well-controlled environment, the oxide inclusions contained in the precursor powder were the primary contributors to the formation of nano-scale oxides in the final additive manufactured (AM) product. These oxide inclusions were found to be enriched with oxygen getter elements like Si and Mn. By controlling the extraneous oxygen level in the process environment, the oxygen level in AM produced parts was found to increase with the extraneous oxygen level. The intrusion pathway of this extra oxygen was found to be dominated by the incorporation of spatter particles into the build during processing. Moisture induced oxidation during powder storage was also found to result in a higher oxide density in the AM produced parts. SS 316L powder free of Si and Mn oxygen getters was processed in a well-controlled environment and resulted in a similar level of oxygen intrusion. Microhardness testing indicated that the oxide volume fraction increase from extraneous oxygen did not influence hardness values. However, a marked decrease in hardness was found for the humidified and Si-Mn free AM processed parts.

Highly efficient head-only magnetic field insert gradient coil for achieving simultaneous high gradient amplitude and slew rate at 3.0T (MAGNUS) for brain microstructure imaging.

Abstract: Purpose To develop a highly efficient magnetic field gradient coil for head imaging that achieves 200 mT/m and 500 T/m/s on each axis using a standard 1 MVA gradient driver in clinical whole-body 3.0T MR magnet. Methods A 42-cm inner diameter head-gradient used the available 89- to 91-cm warm bore space in a whole-body 3.0T magnet by increasing the radial separation between the primary and the shield coil windings to 18.6 cm. This required the removal of the standard whole-body gradient and radiofrequency coils. To achieve a coil efficiency ~4× that of whole-body gradients, a double-layer primary coil design with asymmetric x-y axes, and symmetric z-axis was used. The use of all-hollow conductor with direct fluid cooling of the gradient coil enabled ≥50 kW of total heat dissipation. Results This design achieved a coil efficiency of 0.32 mT/m/A, allowing 200 mT/m and 500 T/m/s for a 620 A/1500 V driver. The gradient coil yielded substantially reduced echo spacing, and minimum repetition time and echo time. In high b = 10,000 s/mm2 diffusion, echo time (TE) 50% reduction compared with whole-body gradients). The gradient coil passed the American College of Radiology tests for gradient linearity and distortion, and met acoustic requirements for nonsignificant risk operation. Conclusions Ultra-high gradient coil performance was achieved for head imaging without substantial increases in gradient driver power in a whole-body 3.0T magnet after removing the standard gradient coil. As such, any clinical whole-body 3.0T MR system could be upgraded with 3-4× improvement in gradient performance for brain imaging.

Data-Driven Fault Location of Electric Power Distribution Systems With Distributed Generation

Abstract: This paper is proposing a data-driven approach for fault location of distribution systems with distributed generations (DGs) by utilizing smart meters at low voltage (LV) networks and remote fault indicators (RFIs) at medium voltage (MV) networks. The determined fault location assists system operators with expedited service restoration, thus improving system reliability and resiliency. To quickly locate a fault, an enhanced escalation method is proposed to use outage reports from smart meters for prediction of the outage region. The determined outage region together with overcurrent notifications from RFIs with directional elements is jointly used to pinpoint the faulty line section. To this end, a new analytical model based on mixed integer linear programming (MILP) is proposed and each hypothetical fault location is modeled as decision variables. The result is an algorithm that is capable to support decision-making of single or multiple faulted line section(s) with incorrect and incomplete data from smart meters and RFIs for accurate fault location. In addition, an engineering way is presented to configure “power outage recognition time” of smart meters and logics for outage escalation are proposed in this paper. Simulation results based on a utility feeder validate the proposed methodology for fault location.

Deep-learning-based crack detection with applications for the structural health monitoring of gas turbines:

Abstract: Gas turbine maintenance requires consistent inspections of cracks and other structural anomalies. The inspections provide information regarding the overall condition of the structures and yield inf...

Feature learning for fault detection in high-dimensional condition-monitoring signals

Abstract: Complex industrial systems are continuously monitored by a large number of heterogeneous sensors. The diversity of their operating conditions and the possible fault types make it impossible to coll...

3-Tesla MRI of deep brain stimulation patients: safety assessment of coils and pulse sequences.

Abstract: Objective Physicians are more frequently encountering patients who are treated with deep brain stimulation (DBS), yet many MRI centers do not routinely perform MRI in this population. This warrants a safety assessment to improve DBS patients' accessibility to MRI, thereby improving their care while simultaneously providing a new tool for neuromodulation research. Methods A phantom simulating a patient with a DBS neuromodulation device (DBS lead model 3387 and IPG Activa PC model 37601) was constructed and used. Temperature changes at the most ventral DBS electrode contacts, implantable pulse generator (IPG) voltages, specific absorption rate (SAR), and B1+rms were recorded during 3-T MRI scanning. Safety data were acquired with a transmit body multi-array receive and quadrature transmit-receive head coil during various pulse sequences, using numerous DBS configurations from "the worst" to "the most common."In addition, 3-T MRI scanning (T1 and fMRI) was performed on 41 patients with fully internalized and active DBS using a quadrature transmit-receive head coil. MR images, neurological examination findings, and stability of the IPG impedances were assessed. Results In the phantom study, temperature rises at the DBS electrodes were less than 2°C for both coils during 3D SPGR, EPI, DTI, and SWI. Sequences with intense radiofrequency pulses such as T2-weighted sequences may cause higher heating (due to their higher SAR). The IPG did not power off and kept a constant firing rate, and its average voltage output was unchanged. The 41 DBS patients underwent 3-T MRI with no adverse event. Conclusions Under the experimental conditions used in this study, 3-T MRI scanning of DBS patients with selected pulse sequences appears to be safe. Generally, T2-weighted sequences (using routine protocols) should be avoided in DBS patients. Complementary 3-T MRI phantom safety data suggest that imaging conditions that are less restrictive than those used in the patients in this study, such as using transmit body multi-array receive coils, may also be safe. Given the interplay between the implanted DBS neuromodulation device and the MRI system, these findings are specific to the experimental conditions in this study.

Proliferation of Faulty Materials Data Analysis in the Literature.

Abstract: Matthew R. Linford1, Vincent S. Smentkowski2*, John T. Grant3, C. Richard Brundle4, Peter M.A. Sherwood5, Mark C. Biesinger6, Jeff Terry7, Kateryna Artyushkova8, Alberto Herrera-Gómez9, Sven Tougaard10, William Skinner11, Jean-Jacques Pireaux12, Christopher F. McConville13, Christopher D. Easton14, Thomas R. Gengenbach14, George H. Major1, Paul Dietrich15, Andreas Thissen15, Mark Engelhard16, Cedric J. Powell17, Karen J. Gaskell18 and Donald R. Baer16 Department of Chemistry and Biochemistry, Brigham Young University, Provo, UT 84602, USA; General Electric Research, Niskayuna, NY 12309, USA; Surface Analysis Consultant, Clearwater, FL 33767, USA; C.R. Brundle & Associates, Soquel, CA 95073, USA; University of Washington, Box 351700, Seattle, WA 98195, USA; Surface Science Western, University of Western Ontario, London, Ontario N6G 0J3, Canada; Department of Physics, Illinois Institute of Technology, Chicago, IL 60616, USA; Physical Electronics, Chanhassen, MN 55317, USA; CINVESTAV – Unidad Queretaro, Real de Juriquilla 76230, Mexico; Department of Physics, University of Southern Denmark, Odense 5230, Denmark; Future Industries Institute, University of South Australia, Mawson Lakes, SA 5095, Australia; University of Namur, Namur Institute of Structured Matter, B-5000 Namur, Belgium; College of Science, RMIT University, Melbourne, VIC 3001, Australia; CSIRO Manufacturing, Ian Wark Laboratories, Clayton, VIC 3168, Australia; SPECS Surface Nano Analysis GmbH, 13355 Berlin, Germany; Pacific Northwest National Laboratory, Richland, WA 99354, USA; National Institute of Standards and Technology, Gaithersburg, MD 20899, USA and University of Maryland, College Park, MD 20742, USA

Systematic review of pre-clinical and clinical devices for magnetic resonance-guided radiofrequency hyperthermia

Abstract: Clinical trials have demonstrated the therapeutic benefits of adding radiofrequency (RF) hyperthermia (HT) as an adjuvant to radio- and chemotherapy. However, maximum utilization of these benefits is hampered by the current inability to maintain the temperature within the desired range. RF HT treatment quality is usually monitored by invasive temperature sensors, which provide limited data sampling and are prone to infection risks. Magnetic resonance (MR) temperature imaging has been developed to overcome these hurdles by allowing noninvasive 3D temperature monitoring in the target and normal tissues. To exploit this feature, several approaches for inserting the RF heating devices into the MR scanner have been proposed over the years. In this review, we summarize the status quo in MR-guided RF HT devices and analyze trends in these hybrid hardware configurations. In addition, we discuss the various approaches, extract best practices and identify gaps regarding the experimental validation procedures for MR - RF HT, aimed at converging to a common standard in this process.

A System to Package Perspective on Transient Thermal Management of Electronics

Abstract: There are many applications throughout the military and commercial industries whose thermal profiles are dominated by intermittent and/or periodic pulsed thermal loads. Typical thermal solutions for transient applications focus on providing sufficient continuous cooling to address the peak thermal loads as if operating under steady-state conditions. Such a conservative approach guarantees satisfying the thermal challenge but can result in significant cooling overdesign, thus increasing the size, weight, and cost of the system. Confluent trends of increasing system complexity, component miniaturization, and increasing power density demands are further exacerbating the divergence of the optimal transient and steady-state solutions. Therefore, there needs to be a fundamental shift in the way thermal and packaging engineers approach design to focus on time domain heat transfer design and solutions. Due to the application-dependent nature of transient thermal solutions, it is essential to use a codesign approach such that the thermal and packaging engineers collaborate during the design phase with application and/or electronics engineers to ensure the solution meets the requirements. This paper will provide an overview of the types of transients to consider—from the transients that occur during switching at the chip surface all the way to the system-level transients which transfer heat to air. The paper will cover numerous ways of managing transient heat including phase change materials (PCMs), heat exchangers, advanced controls, and capacitance-based packaging. Moreover, synergies exist between approaches to include application of PCMs to increase thermal capacitance or active control mechanisms that are adapted and optimized for the time constants and needs of the specific application. It is the intent of this transient thermal management review to describe a wide range of areas in which transient thermal management for electronics is a factor of significance and to illustrate which specific implementations of transient thermal solutions are being explored for each area. The paper focuses on the needs and benefits of fundamentally shifting away from a steady-state thermal design mentality to one focused on transient thermal design through application-specific, codesigned approaches.

Image-driven discriminative and generative machine learning algorithms for establishing microstructure-processing relationships

Abstract: We investigate the methods of microstructure representation for the purpose of predicting processing condition from microstructure image data. A binary alloy (uranium–molybdenum) that is currently under development as a nuclear fuel was studied for the purpose of developing an improved machine learning approach to image recognition, characterization, and building predictive capabilities linking microstructure to processing conditions. Here, we test different microstructure representations and evaluate model performance based on the F1 score. A F1 score of 95.1% was achieved for distinguishing between micrographs corresponding to ten different thermo-mechanical material processing conditions. We find that our newly developed microstructure representation describes image data well, and the traditional approach of utilizing area fractions of different phases is insufficient for distinguishing between multiple classes using a relatively small, imbalanced original dataset of 272 images. To explore the applicability of generative methods for supplementing such limited datasets, generative adversarial networks were trained to generate artificial microstructure images. Two different generative networks were trained and tested to assess performance. Challenges and best practices associated with applying machine learning to limited microstructure image datasets are also discussed. Our work has implications for quantitative microstructure analysis and development of microstructure–processing relationships in limited datasets typical of metallurgical process design studies.

Radiomics study for predicting the expression of PD-L1 in non-small cell lung cancer based on CT images and clinicopathologic features.

Abstract: Purpose To predict programmed death-ligand 1 (PD-L1) expression of tumor cells in non-small cell lung cancer (NSCLC) patients by using a radiomics study based on CT images and clinicopathologic features. Materials and methods A total of 390 confirmed NSCLC patients who performed chest CT scan and immunohistochemistry (IHC) examination of PD-L1 of lung tumors with clinic data were collected in this retrospective study, which were divided into two cohorts namely, training (n = 260) and validation (n = 130) cohort. Clinicopathologic features were compared between two cohorts. Lung tumors were segmented by using ITK-snap kit on CT images. Total 200 radiomic features in the segmented images were calculated using in-house texture analysis software, then filtered and minimized by least absolute shrinkage and selection operator (LASSO) regression to select optimal radiomic features based on its relevance of PD-L1 expression status in IHC results and develop radiomics signature. Radiomics signature and clinicopathologic risk factors were incorporated to develop prediction model by using multivariable logistic regression analysis. The receiver operating characteristic (ROC) curves were generated and the areas under the curves (AUC) were reckoned to predict PD-L1 expression in both training and validation cohorts. Results In 200 extracted radiomic features, 9 were selected to develop radiomics signature. In univariate analysis, PD-L1 expression of lung tumors was significantly correlated with radiomics signature, histologic type, and histologic grade (p 0.05). For prediction of PD-L1 expression, the prediction model that combines radiomics signature and clinicopathologic features resulted in AUCs of 0.829 and 0.848 in the training and validation cohort, respectively. Conclusion The prediction model that incorporates the radiomics signature and clinical risk factors has potential to facilitate the individualized prediction of PD-L1 expression in NSCLC patients and identify patients who can benefit from anti-PD-L1 immunotherapy.

A Hybrid Approach for Fusing Physics and Data for Failure Prediction

Abstract: This work describes the architecture for developing physics of failure models, derived as a function of machine sensor data, and integrating with data pertaining to other relevant factors like geography, manufacturing, environment, customer and inspection information, that are not easily modeled using physics principles. The mechanics of the system is characterized using surrogate models for stress and metal temperature based on results from multiple non-linear finite element simulations. A cumulative damage index measure has been formulated that quantifies the health of the component. To address deficiencies in the simulation results, a model tuning framework is designed to improve the accuracy of the model. Despite the model tuning, un-modelled sources of variation can lead to insufficient model accuracy. It is required to incorporate these un-modelled effects so as to improve the model performance. A novel machine learning based model fusion approach has been presented that can combine physics model predictions with other data sources that are difficult to incorporate in a physics framework. This approach has been applied to a gas turbine hot section turbine blade failure prediction example.

Predictive modeling of laser and electron beam powder bed fusion additive manufacturing of metals at the mesoscale

Abstract: We present the results of 3D modeling of the laser and electron beam powder bed fusion process at the mesoscale with an in-house developed advanced multiphysical numerical tool. The hydrodynamics and thermal conductivity core of the tool is based on the lattice Boltzmann method. The numerical tool takes into account the random distributions of powder particles by size in a layer and the propagation of the laser (electron beam) with a full ray tracing (Monte Carlo) model that includes multiple reflections, phase transitions, thermal conductivity, and detailed liquid dynamics of the molten metal, influenced by evaporation of the metal and the recoil pressure. The model has been validated by a number of physical tests. We numerically demonstrate a strong dependence of the net energy absorption of the incoming heat source beam by the powder bed and melt pool on the beam power. We show the ability of our model to predict the measurable properties of a single track on a bare substrate as well as on a powder layer. We obtain good agreement with experimental data for the depth, width and shape of a track for a number of materials and a wide range of energy source parameters. We further apply our model to the simulation of the entire layer formation and demonstrate the strong dependence of the resulting layer morphology on the hatch spacing. The presented model could be very helpful for optimizing the additive process without carrying out a large number of experiments in a common trial-and-error method, developing process parameters for new materials, and assessing novel modalities of powder bed fusion additive manufacturing.

Effect of Particle Spreading Dynamics on Powder Bed Quality in Metal Additive Manufacturing

Abstract: Powder spreading precedes creation of every new layer in powder bed additive manufacturing (AM). The powder spreading process can lead to powder layer defects such as porosity, poor surface roughness and particle segregation. Therefore, the creation of homogeneous layers is the first task for optimal part printing. Discrete element methods (DEM) powder spreading simulations are typically limited to a single layer and/or small number of particles. Therefore, results from such model configurations may not be generalized to multiple layer processes. In this study, a computationally efficient multi-layer powder spreading DEM simulation model is proposed. The model is calibrated experimentally using static Angle of Repose measurements. The adhesion model parameter, cohesive energy density is related to adhesive surface energy and strain energy release rate parameters. The model results show that interaction between particle and the powder spreading rake leads to noticeable variation in packing density, surface roughness, dynamic angle of repose (AOR), particle size distribution, and particle segregation. The powder model is experimentally validated using a recoater spreading rig to measure the dynamic AOR at spreading speeds consistent with recoating speeds and layer heights used in AM processes.

Virtual Monoenergetic CT Imaging via Deep Learning.

Abstract: Summary Conventional single-spectrum computed tomography (CT) reconstructs a spectrally integrated attenuation image and reveals tissues morphology without any information about the elemental composition of the tissues. Dual-energy CT (DECT) acquires two spectrally distinct datasets and reconstructs energy-selective (virtual monoenergetic [VM]) and material-selective (material decomposition) images. However, DECT increases system complexity and radiation dose compared with single-spectrum CT. In this paper, a deep learning approach is presented to produce VM images from single-spectrum CT images. Specifically, a modified residual neural network (ResNet) model is developed to map single-spectrum CT images to VM images at pre-specified energy levels. This network is trained on clinical DECT data and shows excellent convergence behavior and image accuracy compared with VM images produced by DECT. The trained model produces high-quality approximations of VM images with a relative error of less than 2%. This method enables multi-material decomposition into three tissue classes, with accuracy comparable with DECT.

Review: Materials Ecosystem for Additive Manufacturing Powder Bed Fusion Processes

Abstract: Additive manufacturing technologies are revolutionizing modern component design across many industries, while leading to an evolution in materials science and engineering. Understanding and controlling the materials ecosystem in additive manufacturing is an essential factor for successful adoption. The relationships among materials chemistry, powder characteristics, processes and final part performance are key and crucial concepts in additive manufacturing technologies. Powder bed fusion (PBF) processes including laser and electron beam melting processes are fundamentally based on controlling the solid-to-liquid and liquid-to-solid phase transformations in each process layer. The powder characteristics, evolution of the microstructure through the additive manufacturing process and subsequent metallurgical post-processing are primarily responsible for material performance. A more comprehensive understanding of aspects such as powder characteristics, liquid- and solid-phase transformations, and the effects of repeated thermal cycling on metallurgical structure development will be required to effectively apply a design-for-additive approach. Numerical modeling and machine learning are among tools that can be used for developing such understanding. This article will provide a review and summary of the materials ecosystem for additive manufacturing powder bed fusion processes.