Semi-Supervised Deep Learning for High-Dimensional Uncertainty Quantification

Conventional uncertainty quantification methods usually lacks the capability of dealing with high-dimensional problems due to the curse of dimensionality. This paper presents a semi-supervised learning framework for dimension reduction and reliability analysis. An autoencoder is first adopted for mapping the high-dimensional space into a low-dimensional latent space, which contains a distinguishable failure surface. Then a deep feedforward neural network (DFN) is utilized to learn the mapping relationship and reconstruct the latent space, while the Gaussian process (GP) modeling technique is used to build the surrogate model of the transformed limit state function. During the training process of the DFN, the discrepancy between the actual and reconstructed latent space is minimized through semi-supervised learning for ensuring the accuracy. Both labeled and unlabeled samples are utilized for defining the loss function of the DFN. Evolutionary algorithm is adopted to train the DFN, then the Monte Carlo simulation method is used for uncertainty quantification and reliability analysis based on the proposed framework. The effectiveness is demonstrated through a mathematical example.

TopologyGAN: Topology Optimization Using Generative Adversarial Networks Based on Physical Fields Over the Initial Domain

In topology optimization using deep learning, load and boundary conditions represented as vectors or sparse matrices often miss the opportunity to encode a rich view of the design problem, leading to less than ideal generalization results. We propose a new data-driven topology optimization model called TopologyGAN that takes advantage of various physical fields computed on the original, unoptimized material domain, as inputs to the generator of a conditional generative adversarial network (cGAN). Compared to a baseline cGAN, TopologyGAN achieves a nearly 3× reduction in the mean squared error and a 2.5× reduction in the mean absolute error on test problems involving previously unseen boundary conditions. Built on several existing network models, we also introduce a hybrid network called U-SE(Squeeze-and-Excitation)-ResNet for the generator that further increases the overall accuracy. We publicly share our full implementation and trained network.

Network Redundancy: A Key Design Factor for Cooling Networks

Microvascular composite offers a variety of multi-functionality based on the choice of fluid flowing through the embedded microchannels. The design of the microchannel network in microvascular composites is quite challenging. Indeed, the design is often expected to have high cooling efficiency, satisfy the manufacturing and operating constraints, and also have redundancy to increase the temperature uniformity and alleviate the destructive effects of potential microchannel blockage. In this study, we present a design optimization framework to satisfy these requirements. We use the Hybrid Topology/Shape (HyTopS) optimization scheme to design a redundant blockage-tolerant cooling network. In this method, the optimizer can change the topology of the design during the shape optimization process. Being able to modify the topology of the network enables the optimizer to provide network redundancy to effectively optimize the design for blockage tolerance. We also solve several numerical examples to show the unique features of the proposed method.

Structure, Process, and Material Influences for 3D Printed Lattices Designed With Mixed Unit Cells

Emerging 3D printing technologies are enabling the design and fabrication of novel architected structures with advantageous mechanical responses. Designing complex structures, such as lattices, with a targeted response is challenging because build materials, fabrication process, and topological design have unique influences on the structure’s mechanical response. Changing any factor may have unanticipated consequences, even for simpler lattice structures. Here, we conduct mechanical compression experiments to investigate varied lattice design, fabrication, and material combinations using stereolithography printing with a biocompatible polymer. Mechanical testing demonstrates that a higher ultraviolet curing time increases elastic modulus. Material testing demonstrated that anisotropy does not strongly influence lattice mechanics. Designs were altered by comparing homogenous lattices of single unit cell types and heterogeneous lattices that combine two types of unit cells. Unit cells for heterogeneous structures include a Cube design for a high elastic modulus and Cross design for improved shear response. Mechanical testing of three heterogeneous layouts demonstrated how unit cell organization influences mechanical outcomes, therefore enabling the tuning of an elastic modulus that surpasses the law of averages designed for application-dependent mechanical needs. These findings provide a foundation for linking design, process, and material for engineering 3D printed structures with preferred properties, while also facilitating new directions in design automation and optimization.

Generative Adversarial Networks With Synthetic Training Data for Enforcing Manufacturing Constraints on Topology Optimization

Topology optimization is a powerful tool to generate mechanical designs that use minimal mass to achieve their function. However, the designs obtained using topology optimization are often not manufacturable using a given manufacturing process. There exist some modifications to the traditional topology optimization algorithm that are able to impose manufacturing constraints for a limited set of manufacturing methods. These approaches have the drawback that they are often based on heuristics to obtain the manufacturability constraint and thus cannot be applied generally to multiple manufacturing methods. In order to create a general approach to imposing manufacturing constraints on topology optimization, generative adversarial networks (GANs) are used. GANs have the capability to produce samples from a distribution defined by training data. In this work, the GAN is trained by generating synthetic 3D voxel training data that represent the distribution of designs that can be created by a particular manufacturing method. Once trained, the GAN forms a mapping from a latent vector space to the space of manufacturable designs. The topology optimization is then performed on the latent vector space ensuring that the design obtained is manufacturable. The effectiveness of this approach is demonstrated by training a GAN on designs intended to be manufacturable on a 3-axis computer numerically controlled (CNC) milling machine.

Lattice Structure Optimization With Orientation-Dependent Material Properties

Advances in additive manufacturing techniques have enabled the production of parts with complex internal geometries. However, the layer-based nature of additive processes often results in mechanical properties that vary based on the orientation of the feature relative to the build plane. Lattice structures have been a popular design application for additive manufacturing due to their potential uses in lightweight structural applications. Many recent works have explored the modeling, design, and fabrication challenges that arise in the multiscale setting of lattice structures. However, there remains a significant challenge in bridging the simplified computational models used in the design process and the more complex properties actually realized in fabrication. This work develops a design approach that captures orientation-dependent material properties that have been observed in metal AM processes while remaining suitable for use in an iterative design process. Exemplar problems are utilized to investigate the potential design changes and performance improvements that can be attained by taking the directional dependence of the manufacturing process into account in the design of lattice structures.

Structural Design Exploration for Product Architecture Design

This paper proposed a design support method based on structuralization and analysis of various design candidates of product architecture design. The product architecture is a basic scheme that assigns the function of the product to physical components. In the conventional modular design method, a concise model, i.e., a graph or a matrix, is used to express the interactions of the system’s components and aims to support the designer grasping the system behavior. The Design Structure Matrix (DSM) is a representative model of system architecture and enables quantitative evaluation of design candidates. While various design candidates are generated through mathematical operations, it is difficult to understand their relationships from simple comparisons because of discrete behavior and the size of the problem. It must be a critical issue at the stage of selecting and interpreting the design candidates. In the proposed method, the design candidates are classified and structuralized as a dendrogram by the hierarchical clustering method. The comparison of clusters of each branch of dendrogram clarifies the system leverage points. The information of the system is summarized into the hierarchical tree-shaped graph that corresponds to the dendrogram. The designer can explore the design candidates with such a graph-based based interpretation of underlying structures effectively.

Scalable Set-Based Design Optimization and Remanufacturing for Meeting Changing Requirements

Engineering design problems often have open-ended requirements, especially in the early stages of development. Set-based design is a paradigm for exploring, and keeping under consideration, several alternatives so that commitment to a single design can be delayed until requirements are settled. In addition, requirements may change over the lifetime of a component or a system. Novel manufacturing technologies enable designs to be remanufactured to meet changed requirements. By considering this capability during the set-based design optimization process, solutions can be scaled to meet evolving requirements and customer specifications even after commitment. Such an ability can also support a circular economy paradigm based on the return of used or discarded components and systems to working condition. We propose a set-based design methodology to obtain scalable optimal solutions that can satisfy changing requirements through remanufacturing. We first use design optimization and surrogate modeling to obtain parametric optimal designs. This set of parametric optimal designs is then reduced to scalable optimal designs by observing a set of transition rules for the manufacturing process used (additive or subtractive). The methodology is demonstrated by means of a structural aeroengine component that is remanufactured by direct energy deposition of a stiffener to meet higher loading requirements.

A Weighted Network Modeling Approach for Analyzing Product Competition

Statistical network models allow us to study the co-evolution between the products and the social aspects of a market system, by modeling these components and their interactions as graphs. In this paper, we study competition between different car models using network theory, with a focus on how product attributes (like fuel economy and price) affect which cars are considered together and which cars are finally bought by customers. Unlike past work, where most systems have been studied with the assumption that relationships between competitors are binary (i.e., whether a relationship exists or not), we allow relationships to take strengths (i.e., how strong a relationship is). Specifically, we use valued Exponential Random Graph Models and show that our approach provides a significant improvement over the baselines in predicting product co-considerations as well as in the validation of market share. This is also the first attempt to study aggregated purchase preference and car competition using valued directed networks.

A Model Predictive Control-Based Energy Management Strategy Considering Electric Vehicle Battery Thermal and Cabin Climate Control

The energy management strategy plays a critical role in scheduling the operations and enhancing the overall efficiency for electric vehicles. This paper proposes an effective model predictive control-based (MPC) energy management strategy to simultaneously control the battery thermal management system (BTMS) and the cabin air conditioning (AC) system for electric vehicles (EVs). We aim to improve the overall energy efficiency, while retaining soft constraints from both BTMS and AC systems. It is implemented by optimizing the operation and discharging schedule to avoid peak load and by directly utilizing the regenerative power instead of recharging. Compared to the systematic performance without any control coordination between BTMS and AC, results reveal that there are a 4.3% reduction for the recharging energy, and a 6.5% improvement for the overall energy consumption that gained from the MPC-based energy management strategy. Overall the MPC-based energy management is a promising solution to enhance the efficiency for electric vehicles.