Segmentation of Additive Manufacturing Defects Using U-Net

Additive manufacturing (AM), also known as “3D printing,” has transitioned from concepts and prototypes to part-for-part substitution—and now to the creation of part geometries that can only be made using AM. As a wide range of mobility OEMs begin to introduce AM parts into their products, the question between insourcing and outsourcing the manufacturing of AM parts has surfaced. Just like parts made using other technologies, AM parts can require significant post-processing operations. Therefore, as AM supply chains begin to develop, the sourcing of AM part building and their post-processing becomes an unsettled and important issue. Unsettled Aspects of Insourcing and Outsourcing Additive Manufacturing discusses the approaches and trade-offs of the different sourcing options for production hardware for multiple scenarios, including both metallic and polymer technologies and components.

Additive Manufacturing (AM) technology is considered as one of the most promising manufacturing technologies in the areas of aerospace and defense industries. However, a lack of assurance of quality with AM parts is a key technological barrier that prevents manufacturers from adopting AM technologies, especially for high value added applications. Therefore, it is critically important to monitor the quality of products during the AM process. In this paper, current process monitoring especially the defects measurement for metal AM in Powder Bed Fusion (PBF) was firstly reviewed. And then, an optical in-situ inspection method based on multi-spectrum is proposed. The optical measuring system with infrared and white light imaging system is designed and optimized. Imaging data fusion algorithms is proposed to obtain the enhanced measuring results from infrared and white light imaging system. Simulation studies were undertaken to verify the validity of the proposed monitoring system. The research work is helpful for the optimization of process parameters so as to control the quality in the AM process.

Additive manufacturing (AM) of short-fiber reinforced composites are actively being considered for construction of low-cost, weight reducing alternatives to non-structural metal components. In addition AM parts are being used for rapid manufacturing of composite tooling. AM is notorious for generating parts with higher porosity than more traditional manufacturing technologies. Due to the aerospace industry’s low risk tolerance, novel material systems require comprehensive characterization of their associated manufacturing defects and the impact of defects on performance in order to receive certification. AM short fiber composites still require this analysis. Although traditional composite manufacturing methods, such as an autoclave or VARTM can produce porosity, the origin and transport of porosity has been thoroughly studied and acceptable limits for part qualification have been established to minimize the effect on performance. In AM components studies establishing the origination of porosity origin and processing defect related minimization strategies are lacking. A preliminary study on the impacts of porosity caused by print parameters such print speed, layer height, first layer height, and step-over distance has been undertaken. Direct Ink Writing (DIW) was selected for this study using an epoxy-based ink filled with clay and chopped carbon fiber. To understand performance, fracture and shear specimens were fabricated with different test parameters based on the assumed levels of porosity from screening tests. Samples were then mechanically tested using a novel in-plane shear test method and a single edge notch tension test. This study explores the processing parameter’s contribution to porosity and establishes general trends to understand the influence of porosity on performance for additively manufactured chopped fiber composites.

Additive manufacturing (AM) is a class of advanced manufacturing technologies that facilitates fabrication of geometrically complex, high-performance functional parts. AM processes often produce a “near-net” shape comprising the intended design and sacrificial support structures to be removed in postprocessing. The amount of extra material in the near-net shape directly increases fabrication cost and clean-up time. This paper presents a shape modification method for AM parts to minimize the need for sacrificial material in the near-net shape. Given an existing triangulated boundary representation of the part, we formulate the problem as a generic shape optimization that minimizes the changes made to the original geometry by locally modifying vertex coordinates while both maintaining the manifold nature of the representation and satisfying the overhang angle constraints, when possible. Specifically, we augment a pseudo-energy regularization term to our primary objective function for self-supporting to ensure that the optimized shape is valid and suitable for AM. We propose two mesh regularization methods, using virtual springs and sphere packing, and provide differentiable expressions of the objective function and constraints to enable efficient gradient-based optimization. We show that the sphere packing is more robust and retains the quality of the mesh during optimization. Finally, we demonstrate the effectiveness and efficiency of our method in modifying realistic models in 3D.

Automated image segmentation for cardiac septal defects based on contour region with convolutional neural networks: A preliminary study

Additive manufacturing (AM), also known as 3D printing, is a rapidly developing technology with tremendous potential in both developmental and production applications. Solar Turbines Incorporated is committed to AM technology for gas turbine applications. The ability to metal 3D print novel designs of turbine blades capable of actual turbine engine operation would effectively reduce design validation cycle time, and allow acquisition of key performance data early in a design campaign. In support of Solar’s advanced manufacturing development and ongoing engine efficiency improvement goals, Solar initiated a project to print a full set of Mercury™ 50 stage 2 turbine blades to be run in a development engine. Solar leveraged years of experience with design and serial production of AM components in support of this project. A significant challenge faced when printing turbine blades is producing metal with mechanical properties sufficient to withstand the rigors of engine operation. As a rotating component within the hot section of the engine, turbine blades experience high centrifugal and pressure loads at elevated temperatures. After investigation of possible alloys capable of meeting the requirements of the Mercury™ 50 design envelope, the gamma prime (γ’) strengthened nickel superalloy Inconel™ 738LC was selected to provide the best opportunity for successful development engine testing. Solar partnered with Oak Ridge National Laboratory (ORNL) to produce the Inconel™ 738LC blades with Electron Beam Melting (EBM) powder bed fusion process. Once a rough blade shape was printed, the fir-tree attachment, blade tip shroud, and air flow path surfaces were finished using both conventional and non-conventional machining processes. In-process monitoring, metallurgical evaluations, mechanical testing, and non-destructive inspection techniques were used to validate the printed blade material integrity and conformance to geometric design intent. Planned future activities include assembly of the AM blades onto a disk for spin pit testing to validate the mechanical integrity and design margin of the blades. The final phase of the project will be to install the bladed disk assembly into a Mercury™ 50 engine at Solar Turbines to conduct a series of hot-fired engine performance tests.

Segmentation of medical images has been the most demanding and growing area currently for analysis of medical images. Segmentation of polyp images is a huge challenge because of the variability of color depth and morphology in polyps throughout colonoscopy imaging. For segmentation, in this work, we have used a dataset of images of the gastrointestinal polyp. The algorithms used in this paper for segmentation of gastrointestinal polyp images depend on profound deep convolutional neural network architectures: FCN, Dual U-net with Resnet Encoder, U-net, and Unet_Resnet. To improve the performance, data augmentation is performed on the dataset. The efficiency of the algorithms is measured by using metrics such as Dice Similarity Coefficient (DSC) and Intersection Over Union (IOU). The algorithm Dual U-net with Resnet Encoder obtains a higher DSC of 0.87 and IOU of 0.80 and beats the other algorithms U-net, FCN, and Unet_Resnet in segmentation of gastrointestinal polyp images.

Interstitial lung disease (ILD) refers to a group of various abnormal inflammations of lung tissues and early diagnosis of these disease patterns is crucial for the treatment. Yet it is difficult to make an accurate diagnosis due to the similarity among the clinical manifestations of these diseases. In order to assist the radiologists, computer-aided diagnosis systems have been developed. Besides, the potential of deep convolutional neural networks (CNNs) is also expected to exert on the medical image analysis in recent years. In this paper, we design a new deep convolutional neural network (CNN) architecture to achieve the classification task of ILD patterns. Furthermore, we also propose a novel two-stage transfer learning (TSTL) method to deal with the problem of the lack of training data, which leverages the knowledge learned from sufficient textural source data and auxiliary unlabeled lung CT data to the target domain. We adopt the unsupervised manner to learn the unlabeled data, by which the objective function composed of the prediction confidence and mutual information are optimized. The experimental results show that our proposed CNN architecture achieves desirable performance and outperforms most of the state-of-the-art ones. The comparative analysis demonstrates the promising feasibility and advantages of the proposed two-stage transfer learning strategy as well as the potential of the knowledge learning from lung CT data. Graphical Abstract The framework of the proposed two-stage transfer learning method.

Diabetic Retinopathy (DR) is a highly consequential condition that significantly impacts individuals with diabetes, often resulting in permanent vision loss if left undetected. The accurate assessment of DR severity heavily relies on the precise segmentation of lesions from fundus images. However, manual segmentation processes are complex and time-consuming. In order to address these limitations, this research introduces a novel approach namely Deep Feature Fused Residual with U-Net (DFFR-U-NET) for the segmentation of DR lesions, specifically Haemorrhage (HM), Hard Exudates (HE), and Optic Disc (OD). The proposed method employs a Convolutional Neural Network (CNN) architecture, which incorporates the U-Net model enhanced with a modified bottleneck using residual blocks. This modified U-Net model, composed of robust convolutional blocks and a modified bottleneck, yields optimal results in the segmentation of DR lesions. The training and validation of the proposed method are conducted using the IDRiD dataset, focusing on HM, HE, and OD segmentation. The experimental results demonstrate that the proposed method achieves a high level of accuracy in the segmentation of HM, HE, and OD lesions. Model performance is assessed using key metrics such as accuracy, precision, Mean Intersection Over Union (IOU), Mean Dice Coefficient, and Average Hausdorff distance. The analysis reveals that the proposed model attains an accuracy of 98% for HM, 99% for HE, and 99% for OD. Precision scores are 98% for HM, 99% for HE, and 99% for OD. Moreover, the Mean IOU values are 0.91 for HM, 0.95 for HE, and 0.99 for OD. The Mean Dice Coefficient for HM is 0.91, for HE is 0.95, and for OD is 0.99. Additionally, the Average Hausdorff distance for the background is 2.84 for HM, 1.24 for HE, and 0.38 for OD, while for lesions, it is 2.84 for HM, 1.24 for HE, and 0.38 for OD. When compared to existing studies, the proposed approach establishes a state-of-the-art performance in retinal lesion segmentation, highlighting its superiority in this domain.

A statistical description of X-ray CT (computerized tomography) imaging, from the projection data to the reconstructed image, is presented. The Gaussianity of the pixel image generated by the convolution (image reconstruction) algorithm is justified. The conditions for two pixel images to be statistically independent (for a given probability) and the conditions for a group of pixel images to be a spatial stationary random process and ergodic in mean and autocorrelations are derived. These properties provide the basis for establishing the stochastic image model and conducting the statistical image analysis of X-ray CT images.

While X-ray computed tomography (XCT) is pushed further into the micro- and nanoscale, the limitations of various tool components and object motion become more apparent. For high-resolution XCT, it is necessary but practically difficult to align these tool components with sub-micron precision. The aim is to develop a novel reconstruction methodology that considers unavoidable misalignment and object motion during the data acquisition in order to obtain high-quality three-dimensional images and that is applicable for data recovery from incomplete datasets. A reconstruction software empowered by sophisticated correction modules that autonomously estimates and compensates artefacts using gradient descent and deep learning algorithms has been developed and applied. For motion estimation, a novel computer vision methodology coupled with a deep convolutional neural network approach provides estimates for the object motion by tracking features throughout the adjacent projections. The model is trained using the forward projections of simulated phantoms that consist of several simple geometrical features such as sphere, triangle and rectangular. The feature maps extracted by a neural network are used to detect and to classify features done by a support vector machine. For missing data recovery, a novel deep convolutional neural network is used to infer high-quality reconstruction data from incomplete sets of projections. The forward and back projections of simulated geometric shapes from a range of angular ranges are used to train the model. The model is able to learn the angular dependency based on a limited angle coverage and to propose a new set of projections to suppress artefacts. High-quality three-dimensional images demonstrate that it is possible to effectively suppress artefacts caused by thermomechanical instability of tool components and objects resulting in motion, by center of rotation misalignment and by inaccuracy in the detector position without additional computational efforts. Data recovery from incomplete sets of projections result in directly corrected projections instead of suppressing artefacts in the final reconstructed images. The proposed methodology has been proven and is demonstrated for a ball bearing sample. The reconstruction results are compared to prior corrections and benchmarked with a commercially available reconstruction software. Compared to conventional approaches in XCT imaging and data analysis, the proposed methodology for the generation of high-quality three-dimensional X-ray images is fully autonomous. The methodology presented here has been proven for high-resolution micro-XCT and nano-XCT, however, is applicable for all length scales.

Remote sensing, particularly satellite-based, can play a valuable role in monitoring areas prone to geohazards. The high spatial and temporal coverage provided by satellite data can be used to reconstruct past events and continuously monitor sensitive areas for potential hazards. This paper presents a range of techniques and methods that were applied for in-depth analysis and utilization of Earth observation data, with a particular emphasis on: (1) detecting mining subsidence, where a novel approach is proposed by combining an improved U-Net model and Interferometry Synthetic Aperture Radar (InSAR) technology. The results showed that the Efficient Channel Attention (ECA) U-Net model performed better than the U-Net (baseline) model in terms of Mean Intersection over Union (MIoU) and Intersection over Union (IoU) indicators; (2) monitoring water conservancy and hydropower engineering. The Xiaolangdi multipurpose dam complex was monitored using Small BAsline Subsets (SBAS) InSAR method on Sentinel-1 time series data and four small regions with high deformation rates were identified on the slope of the reservoir bank on the north side. The dam body also showed obvious deformation with a velocity exceeding 60 mm/a; (3) the evaluation of the potential of InSAR results to integrate monitoring and warning systems for valuable heritage and architectural preservation. The overall outcome of these methods showed that the use of Artificial Intelligence (AI) techniques in combination with InSAR data leads to more efficient analysis and interpretation, resulting in improved accuracy and prompt identification of potential hazards; and (4) finally, this study also presents a method for detecting landslides in mountainous regions, using optical imagery. The new temporal landslide detection method is evaluated over a 7-year analysis period and unlike conventional bi-temporal change detection methods, this approach does not depend on any prior-knowledge and can potentially detect landslides over extended periods of time such as decades.

The relationship between the behavior of water in snow and its microstructure is crucial to improve the prediction of wet snow disasters. X-ray computed tomography (X-ray CT) is frequently used to observe snow microscopically. However, distinguishing between ice and water in the X-ray images is difficult because ice exhibits an X-ray absorption coefficient similar to that of water. In contrast, magnetic resonance imaging (MRI) acquires nuclear magnetic resonance (NMR) signals of protons in a liquid and visualizes the NMR signal intensity, enabling discrimination between water and ice signals. However, snow grains and pore spaces cannot be distinguished in MRI images because they do not generate NMR signals. To investigate the relationship between the microstructure of snow and the distribution of liquids in snow, we developed a novel method that combines X-ray CT and MRI images to compensate for the disadvantages associated with each method. Using this method, we successfully visualized where liquid (C12H24) occupied pore spaces. We also showed the possibility of using C12H24 instead of water to obtain water retention curve of snow cover, which is a fundamental aspect of hydraulic properties. Although there is room for improvement in the visualization of water in snow, such as shortening the imaging time to escape snow metamorphosis and image superimposition methods, this method is expected to effectively elucidate the behavior of water in snow and clarify the characteristics of wet snow.

This paper proposes a Deep Convolutional Neural Network (DCNN) architecture for biomedical image segmentation. The proposed work explores the role of separable convolution within the context of biomedical image segmentation. The proposed architecture, XceptionUnetV1, is an amalgam of the Xception and U-Net models. Additionally, the model uses dilated convolution to increase the field view of the filters. The proposed model achieves better results than various versions of U-Net. The proposed DCNN architecture requires much lesser parameters, approximately \(1/4^{th}\) of the U-Net model. The proposed model has been tested on chest X-rays for lung segmentation.

Chaos Game Representation (CGR) is an emerging means of visualising and representing genomic and proteomic sequences. There exist many open questions related to its effective application to various computational tasks. In this work, we begin to address some of these questions by comparing four variants of the Chaos Game to generate CGR imagery as part of a multi-class classification task to identify the source organism for a given protein. We propose a novel nodal configuration for icosagon and 20-flake CGRs. Using two datasets, we performed fine-tuning using seven deep convolutional neural network (CNN) architectures and report modest performance over random among the 56 test conditions, highlighting certain shortcomings in effectively leveraging CGR in conjunction with deep CNN architectures. Many of the insights from this work will serve to orient subsequent protein-related studies involving CGR-based encoding and be generally applicable to disparate domains seeking to leverage CGR for sequence-type data.