The interest of timber structures has in recent years increased, primarily due tothe environmental benets of timber. This has created an increased demand forstructural engineers with timber expertise. At the same time the concept of structuralparametric design have become more popular. This new way of working withdesigns enables for architects and engineers to explore dierent geometries in earlystages of a project. However, the combination of a parametric workow and timberdesign have so far been limited due to the complexity of the material.This thesis aims to create an parametric workow within the visual programmingenvironment Grasshopper. This enables analysis of structural design simultaneouslywith a cross sectional and topological optimization of timber structures. The structuralanalysis is performed with Karamba which is a plug-in tool to the Grasshopperenvironment. The design verication based on Eurocode EN-1995 have been manuallyscripted in python components. The parametric model have been applied to acase where the main bearing bearing of a glass roof is to be designed. Three dierentgeometries have been evaluated with regard to cross sectional dimensions andgeometrical shape.The framework with a truss turned out to be a preferable design if only consideringweight, deection and utilization. The truss frame provides the lowest weight and thesecond smallest displacement. Furthermore, a comparison of the structural analysisand design have been performed with the FEM-program Robot. The compassionshow similar results, increasing the reliability of the Grasshopper model and theresults from this tool. It conrms it is possible to perform generative design oftimber structures within the same interface.The Grasshopper model is limited and can not handle all variations of 2D timberstructures. The complexity and variation of such calculations in conjunction with theEurocode have not been implemented during the time-span of this thesis. However,it is general within the limitations of the case study meaning a variety of framegeometries can be evaluated.

In collaboration with Kongsberg Automotive, the thesis focuses on material selection and redesigning the shift fork for additive manufacturing using topology optimization. The shift fork is a component in the gear shifting mechanism in the automotive industry. The current shift fork at Kongsberg is manufactured from aluminum using die-casting. This design and material do not withstand huge dynamic loads in commercial vehicles. The material to withstand the loading conditions and is widely available across powder manufacturers is selected using the weighted properties method. The topology optimization of the design resulted in a 50 % reduction in mass. The shift fork's two legs undergo uneven load distribution due to eccentricity. The optimized models are simulated using Finite Element Analysis to validate the design. The optimized design is obtained such that the difference in displacement between both legs is within 50 %. Numerous metal powder manufacturers and 3D printing service providers were contacted to understand the current additive manufacturing market.

<p>In this dissertation, we study three important problems in modern deep learning: adversarial robustness, visualization, and partially monotonic function modeling. In the first part, we study the trade-off between robustness and standard accuracy in deep neural network (DNN) classifiers. We introduce sensible adversarial learning and demonstrate the synergistic effect between pursuits of standard natural accuracy and robustness. Specifically, we define a sensible adversary which is useful for learning a robust model while keeping high natural accuracy. We theoretically establish that the Bayes classifier is the most robust multi-class classifier with the 0-1 loss under sensible adversarial learning. We propose a novel and efficient algorithm that trains a robust model using implicit loss truncation. Our  experiments demonstrate that our method is effective in promoting robustness against various attacks and keeping high natural accuracy. </p> <p>In the second part, we study nonlinear dimensional reduction with the manifold assumption, often called manifold learning. Despite the recent advances in manifold learning, current state-of-the-art techniques focus on preserving only local or global structure information of the data. Moreover, they are transductive; the dimensional reduction results cannot be generalized to unseen data. We propose iGLoMAP, a novel inductive manifold learning method for dimensional reduction and high-dimensional data visualization. iGLoMAP preserves both local and global structure information in the same algorithm by preserving geodesic distance between data points. We establish the consistency property of our geodesic distance estimators. iGLoMAP can provide the lower-dimensional embedding for an unseen, novel point without any additional optimization. We  successfully apply iGLoMAP to the simulated and real-data settings with competitive experiments against state-of-the-art methods.</p> <p>In the third part, we study partially monotonic DNNs. We model such a function by using the fundamental theorem for line integrals, where the gradient is parametrized by DNNs. For the validity of the model formulation, we develop a symmetric penalty for gradient modeling. Unlike existing methods, our method allows partially monotonic modeling for general DNN architectures and monotonic constraints on multiple variables. We empirically show the necessity of the symmetric penalty on a simulated dataset.</p>

The workflow and process of graphic design is today streamlined by the usage of our tools such as softwares. However, this doesn’t mean that designers must restrict themselves from working outside an established practice largely determined by such software. Implementing programming and generative design as a method in the development of graphic design can lead to new insights and different perspectives. Proven to be an asset in the creation of visual material the method falls short as generative design is co-dependent on both the designer and user. Though complicated and expensive, generative design proves useful when working with realtime data and as a tool for generating new ideas. Working experimentally with under-utilized tools can help the development of new trends and methods within the field of graphic design.

The AEC-industry need to respond to multiple requirements from regulations and clients, leading to that building projects are becoming increasingly complex to handle for designers. CAD or computer aided design is a way to handle these challenges, and within CAD a new method is emerging: Computational design enables users to generate and explore design solutions automatically. The purpose of this study was to investigate how computational design can be used and what limitations architects and engineers experience with it.   A qualitative research approach was chosen to get in depth understanding. To get variation 16 semi-structured interviews were conducted as primary data collection coupled with a literature review as theoretical framework. The thesis found that computational design applies both design thinking and computational thinking, it is an iterative process that generates design by altering parameters or algorithms and affects the intended design. In this thesis it also needs to be part of the AEC design process. Computational design was found to have most potential in early stage but can be useful for engineers in later stage as well.  Computational design can be used to increase workflow efficiency through automation and rapid feedback which can improve communication and collaboration. It can also increase solution performance by generating design based on multiple objectives. Furthermore, it enables users to expand their solution space and solve complex problems too difficult to solve otherwise. Computational design can be used to analyze early building concepts, analyze floorplans, to optimize material consumption, material choices, structural elements, energy efficiency, daylight, and acoustic requirements. Challenges found were on an individual level a steep learning curve, increased complexity, collection of trustworthy data and interpretation of data. Challenges on an organizational level were fear of automation, low support from leaders, low understanding of the subject from clients and colleges, unsuitable business models, and traditional processes. Furthermore, performance was found to be limited by computers and software capabilities.  Future research should focus on investigating solutions for the many challenges identified in this thesis. Additionally, further applications should be investigated in a narrower scope; a specific type of building or a general element, preferably avoiding repetition of applications in this study. It would also be of interest to investigate challenges of participants on an international scale, experienced with generative design and textual programming languages since these were found to be difficult to learn and apply. / Byggprojekt har blivit mer komplexa för projektörer och arkitekter att hantera då AEC branschen behöver uppfylla flera lagar och regler samt krav från beställare. En teknik för att möta detta behov är med CAD eller computer-aided-design, och inom CAD har en ny metod växt fram: Computational design möjliggör att automatiskt generera och utforska design lösningar. Syftet med den här studien är att undersöka hur computational design kan användas och vilka begränsningar arkitekter och ingenjörer upplever med det.  En kvalitativ metod valdes för att få en djupare förståelse. För att öka variationen av studien genomfördes 16 semi-strukturerade intervjuer som primär data kombinerat med en litteraturstudie som teoretiskt ramverk. Det framkom att computational design applicerar både designtänkande och computational thinking, är en iterativ process som genererar design genom att ändra parametrar eller algoritmer och påverkar den avsedda designen. I den här studien är computational design också en del av branschens design process. Det framkom också att Computational design har störst potential i tidigt skede men kan också vara användbar för ingenjörer i senare skede. Computational design kan användas till att öka arbetsflödets effektivitet genom automation och snabb feedback, vilket kan förbättra kommunikation och samarbete. Det kan också öka prestandan för lösningar genom att generera design baserat på flera kriterier. Vidare så möjliggör det för användare att öka antalet möjliga designlösningar och lösa komplexa problem som är för svåra att hantera traditionellt. Computational design kan användas till att analysera byggnadskoncept, analysera planlösningar, samt till att optimera materialanvändning, materialval, konstruktioner, energieffektivitet, dagsljus, och akustik. De utmaningar som hittades var på en individuell nivå: brant inlärningskurva, ökad komplexitet, hitta tillförlitliga data och tolka data. På organisationsnivå var utmaningarna: rädsla för automatisering, lågt stöd från ledningen, låg förståelse för ämnet av kunder och kollegor, olämpliga affärsmodeller, och traditionella processer. Prestanda begränsades även av förmågan hos datorer och mjukvaror. Framtida forskning bör fokusera på att undersöka lösningar på de utmaningarna som identifierades i den här studien. Ytterligare applikationer bör också undersökas i högre detalj till exempel som en kategori av byggnader eller ett specifikt byggnadselement som förekommer generellt, företrädesvis utan repetition av denna studie. Det vore också av intresse att undersöka utmaningar för deltagare internationellt som har erfarenhet av generativ design och textuell programmering, då dessa visades ha större svårighet.

There is a gap in scientific knowledge regarding designing functional parts that may not fail, and this project came to define these parts as critical structures. The proposed design process is called the Additive Design Process for Critical Structure, which synthesizes required activities found in the literature review necessary to produce theoretically safe design structures. Although this proposed design process does not meet the requirements of a safe design as intended and must be further studied before the proposed design process can be adapted. The project’s ambition was to integrate the design’s safety with value components, referred to as elements/activities/tools/processes that could contribute to innovation and value creation, to exploit the advantages of additive manufacturing in the design process. The research conducted in this project adapted and applied Design Research Methodology (DRM), written by Blessing &amp; Chakrabarti (2009). Two main research questions were studied that lay a foundation for this thesis, presented below. The project combined quantitative and qualitative research methods to generate the necessary knowledge and then apply/test the derived knowledge to answer these research questions. RQ1: What activities can this project identify to synthesize an additive design process in constructing critical structures for Additive Manufacturing (AM)? RQ2: What are the possible value components to include in the additive design process that would contribute to innovation concerning lead time, weight, and mass customization? The results show that the proposed design process, Additive Design Process for Critical Structured, did not meet the theoretical safe design. However, the findings still suggest that the required activities to achieve a safe design are by introducing defined and explicit protective measurements in the design process. The protective measurement parameters identified in this project were safety factors and Finite Element Analysis (FEA); the question of why the design process does not meet the requirements of producing a theoretical safe design is unknown today and needs further study. Concerning the second RQ, the results showed that Generative Design (GD) was this project's most innovative value component. Adapting GD contributed to shortening the product development time, liberating the design engineer to explore a bolder concept, reducing weight, and allowing the design engineers to generate mass customization. Keywords: Design for Additive Manufacturing, Design Process, Generative Design, Method, Critical Structures, Safety factor.

<p dir="ltr">The United States government and Environmental Protection Agency have mandated that vehicles must meet 54.5 miles per gallon(<i>07/29/2011: President Obama Announces Historic 54.5 Mpg Fuel Efficiency Standard/Consumers Will Save $1.7 Trillion at the Pump, $8K per Vehicle by 2025</i>, n.d.). Faced with increased governmental regulations, manufacturers must find new ways to improve their internal combustion engines. Examining the various components of internal combustion engines, there is potential to optimize individual components for mass reduction and thus improving vehicle performance. Engine Components, specifically connecting rods, are essential to the operation of an engine. While connecting rod designs are highly refined, breakthroughs in additive manufacturing technology have given way to novel approaches in the optimization process. Autodesk Inc. has provided an innovative, generative design space to reduce design time and explore complex optimization. The generative design process provides engineers with unique designs while considering many parameters including material, load cases, and manufacturing processes. The study applied generative design structural load casing to a connecting rod of a single cylinder Ryobi engine to optimize for metal additive manufacturing. The generated outcomes were subjected to finite element analysis to determine their feasibility against a traditional drop forged or die cast design. The results compared three generated geometries against three common additive manufacturing materials in ANSYS Mechanical. The generated geometries were tested for equivalent (Von Mises) stress, equivalent strain, and total deformation. The study, using preliminary forces in a static situation, found that mass reductions of up to 19% were achievable while maintaining performance capabilities of the original design.</p>

"Inverse design" formulates the design process as an inverse problem; optimal values of a parameterized design space are sought so to best reproduce quantitative outcomes from the forwards dynamics of the design's intended environment. Arguably, two subtasks are necessary to iteratively solve such a design problem; the generation and evaluation of designs. This thesis work documents two experiments leveraging machine learning (ML) to facilitate each subtask. Included first is a review of relevant physics and machine learning theory. Then, analysis on the theoretical foundations of ensemble methods realizes a novel equation describing the effect of Bagging and Random Forests on the expected mean squared error of a base model. Complex models of design evaluation may capture environmental dynamics beyond those that are useful for a design optimization. These constitute unnecessary time and computational costs. The first experiment attempts to avoid these by replacing EGSnrc, a Monte Carlo simulation of coupled electron-photon transport, with an efficient ML "surrogate model". To investigate the benefits of surrogate models, a simulated annealing design optimization is twice conducted to reproduce an arbitrary target design, once using EGSnrc and once using a random forest regressor as a surrogate model. It is found that using the surrogate model produced approximately an 100x speed-up, and converged upon an effective design in fewer iterations. In conclusion, using a surrogate model is faster and (in this case) also more effective per-iteration. The second experiment of this thesis work leveraged machine learning for design generation. As a proof-of-concept design objective, the work seeks to efficiently sample 2D Ising spin model configurations from an optimized design space with a uniform distribution of internal energies. Randomly sampling configurations yields a narrow Gaussian distribution of internal energies. Convolutional neural networks (CNN) trained with NeuroEvolution, a mutation-only genetic algorithm, were used to statistically shape the design space. Networks contribute to sampling by processing random inputs, their outputs are then regularized into acceptable configurations. Samples produced with CNNs had more uniform distribution of internal energies, and ranged across the entire space of possible values. In combination with conventional sampling methods, these CNNs can facilitate the sampling of configurations with uniformly distributed internal energies.

No description available.

The commercial exoskeletons are often heavy and bulky, thus reducing the weight and simplifying the form factor becomes a critical task. This thesis details the process of designing and making a low-profile, cable-driven arm exoskeleton. Many advanced methods are explored: 3D scanning, generative design, soft material, compliant joint, additive manufacturing, and 3D latticing. The experiments on TPU kerf cut found that the stress-strain curve of the sample can be modified by changing the cut pattern, it is even possible to control the linear region. The TPU TPMS test showed that given the same volume, changing the lattice parameters can result in different bending stress-strain curves. This thesis also provides many prototypes, test data, and samples for future reference. / Master of Science / Wearing an exoskeleton should be easy and stress-free, but many of the available models are not ergonomic nor user-friendly. To make an exoskeleton that is inviting and comfortable to wear, various nontraditional methods are used. The arm exoskeleton prototype has a lightweight and ergonomic frame, the joints are soft and compact, the cable-driven system is safe and low-profile. This design also brings aesthetics to the exoskeleton which closes the gap between engineering and design.