Evaluating the Potential of Design for Additive Manufacturing Heuristic Cards to Stimulate Novel Product Redesigns

Additive manufacturing (AM) affords those who wield it correctly the benefits of shape, material, hierarchical, and functional complexity. However, many engineers and designers lack the training and experience necessary to take full advantage of these benefits. They require training, tools, and methods to assist them in gaining the enhanced design freedom made possible by additive manufacturing. This work, which is an extension of the authors’ previous work, explores if design heuristics for AM, presented in a card-based format, are an effective mechanism for helping designers achieve the design freedoms enabled by AM. The effectiveness of these design heuristic cards is demonstrated in an experiment with 27 product design students, by showing that there is an increase in the number of unique capabilities of AM being utilized, an increase in the AM novelty, and an increase in the AM flexibility of the generated concepts, when given access to the cards. Additionally, similar to the previous work, an increase in the number of interpreted heuristics and AM modifications present in the participants’ designs when they are provided with the heuristic cards is shown. Comparisons are also made between 8-heuristic and 29-heuristic experiments, but no conclusive statements regarding these comparisons can be drawn. Further user studies are planned to confirm the efficacy of this format at enhancing the design freedoms achieved in group and team design scenarios.

Topology Design With Conditional Generative Adversarial Networks

Deep convolutional neural networks have gained significant traction as effective approaches for developing detailed but compact representations of complex structured data. Generative networks in particular have become popular for their ability to mimic data distributions and allow further exploration of them. This attribute can be utilized in engineering design domains, in which the data structures of finite element meshes for analyzing potential designs are well suited to the deep convolutional network approaches that are being developed at a rapid pace in the field of image processing. This paper explores the use of conditional generative adversarial networks (cGANs) as a means of generating a compact latent representation of structures resulting from classical topology optimization techniques. The constraints and contextual factors of a design problem, such as mass fraction, material type, and load location, can then be specified as input conditions to generate potential topologies in a directed fashion. The trained network can be used to aid concept generation, such that engineers can explore a variety of designs relevant to the problem at hand with ease. The latent variables of the generator can also be used as design parameters, and the low dimensionality enables tractable computational design without analytical sensitivities. This paper demonstrates these capabilities and discusses avenues for further developments that would enable the engineering design community to further leverage generative machine learning techniques to their full potential.

Gaussian Process Based Crack Initiation Modeling for Design of Battery Anode Materials

Silicon-based anode is one of the promising candidates for the next generation lithium ion batteries (LIBs) to achieve high power/energy density. However, the major drawback limiting the practical application of Si anode is that Si experiences significant volume change during its lithiation/de-lithiation cycles, which induces high stress and causes degradation and pulverization of the anode. This study focuses on the crack initiation performances of Si anode during the de-lithiation process. A multi-physics based finite element (FE) model is built to simulate the electrochemical process and crack generation during de-lithiation. On top of that, a Gaussian Processes (GP) based surrogate model is developed to assist the exploration of the crack initiation performances within the anode design space. It is found that, the thickness of the Si coating layer TSi, the yield strength σFc of Si material, the cohesive strength between Si and substrate σFs, and the curvature of the substrate ρ have large impacts on the cracking behavior of Si. This coupled FE simulation-GP surrogate model framework is also applicable to other types of LIB electrodes.

Multimaterial Topology Optimization of Thermoelectric Generators

Over 50% of the energy from power plants, vehicles, oil refining, and steel or glass making process is released to the atmosphere as waste heat. As an attempt to deal with the growing energy crisis, the solid-state thermoelectric generator (TEG), which converts the waste heat into electricity using Seebeck phenomenon, has gained increasing popularity. Since the figures of merit of the thermoelectric materials are temperature dependent, it is not feasible to achieve high efficiency of the thermoelectric conversion using only one single thermoelectric material in a wide temperature range. To address this challenge, this paper proposes a method based on topology optimization to optimize the layouts of functional graded TEGs consisting of multiple materials. The objective of the optimization problem is to maximize the output power and conversion efficiency as well. The proposed method is implemented using the Solid Isotropic Material with Penalization (SIMP) method. The proposed method can make the most of the potential of different thermoelectric materials by distributing each material into its optimal working temperature interval. Instead of dummy materials, both the P and N-type electric conductors are optimally distributed with two different practical thermoelectric materials, namely Bi2Te3 & PbTe for P-type, and Bi2Te3 & CoSb3 for N-type respectively, with the yielding conversion efficiency around 12.5% in the temperature range Tc = 25°C and Th = 400°C. In the 2.5D computational simulation, however, the conversion efficiency shows a significant drop. This could be attributed to the mismatch of the external load and internal TEG resistance as well as the grey region from the topology optimization results as discussed in this paper.

Stress-Constrained Design of Functionally Graded Lattice Structures With Spline-Based Dimensionality Reduction

This paper presents a computationally tractable approach for designing lattice structures for stiffness and strength. Yielding in the mesostructure is determined by a worst-case stress analysis of the homogenization simulation data. This provides a physically meaningful, generalizable, and conservative way to estimate structural failure in three-dimensional functionally graded lattice structures composed of any unit cell architectures. Computational efficiency of the design framework is ensured by developing surrogate models for the unit cell stiffness and strength as a function of density. The surrogate models are then used in the coarse-scale analysis and synthesis. The proposed methodology further uses a compact representation of the material distribution via B-splines, which reduces the size of the design parameter space while ensuring a smooth density variation that is desirable for manufacturing. The proposed method is demonstrated in compliance minimization studies using two types of unit cells with distinct mechanical properties. The effects of B-spline mesh refinement and the presence of a stress constraint on the optimization results are also investigated.

Using Bayesian Optimization With Knowledge Transfer for High Computational Cost Design: A Case Study in Photovoltaics

Engineering design is usually an iterative procedure where many different configurations are tested to yield a desirable end performance. When the design objective can only be measured by costly operations such as experiments or cumbersome computer simulations, a thorough design procedure can be limited. The design problem in these cases is a high cost optimization problem. Meta model-based approaches (e.g. Bayesian optimization) and transfer optimization are methods that can be used to facilitate more efficient designs. Transfer optimization is a technique that enables using previous design knowledge instead of starting from scratch in a new task. In this work, we study a transfer optimization framework based on Bayesian optimization using Gaussian Processes. The similarity among the tasks is determined via a similarity metric. The framework is applied to a particular design problem of thin film solar cells. Planar multilayer solar cells with different sets of materials are optimized to obtain the best opto-electrical efficiency. Solar cells with amorphous silicon and organic absorber layers are studied and the results are presented.

Simulation-Based Process Optimization of Metallic Additive Manufacturing Under Uncertainty

The presence of various uncertainty sources in metal-based additive manufacturing (AM) process prevents producing AM products with consistently high quality. Using electron beam melting (EBM) of Ti-6A1-4V as an example, this paper presents a data-driven framework for process parameters optimization using physics-informed computer simulation models. The goal is to identify a robust manufacturing condition that allows us to constantly obtain equiaxed materials microstructures under uncertainty. To overcome the computational challenge in the robust design optimization under uncertainty, a two-level data-driven surrogate model is constructed based on the simulation data of a validated high-fidelity multi-physics AM simulation model. The robust design result, indicating a combination of low preheating temperature, low beam power and intermediate scanning speed, was acquired enabling the repetitive production of equiaxed-structure products as demonstrated by physics-based simulations. Global sensitivity analysis at the optimal design point indicates that among the studied six noise factors, specific heat capacity and grain growth activation energy have largest impact on the microstructure variation.

But Will it Print?: Assessing Student Use of Design for Additive Manufacturing and Exploring its Effect on Design Performance and Manufacturability

Additive manufacturing (AM) enables engineers to improve the functionality and performance of their designs by adding complexity at little to no additional cost. However, AM processes also exhibit certain unique limitations, such as the presence of support material, which must be accounted for to ensure that designs can be manufactured feasibly and cost-effectively. Given these unique process characteristics, it is important for an AM-trained workforce to be able to incorporate both opportunistic and restrictive design for AM (DfAM) considerations into the design process. While AM/DfAM educational interventions have been discussed in the literature, limited research has investigated the effect of these interventions on students’ use of DfAM. Furthermore, limited research has explored how DfAM use affects the performance of students’ AM designs. This research explores this gap through an experimental study with 123 undergraduate students. Specifically, participants were exposed to either restrictive DfAM or dual DfAM (both opportunistic and restrictive) and then asked to participate in an AM design challenge. The students’ final designs were evaluated for (1) performance with respect the design objectives and constraints, and (2) the use of the various aspects of DfAM. The results showed that the use of certain DfAM considerations, such as minimum feature size and support material mass, successfully predicted the performance of the AM designs. Further, while the variations in DfAM education did not influence the performance of the AM designs, it did have an effect on the students’ use of certain DfAM concepts in their final designs. These results highlight the influence of DfAM education in bringing about an increase in students’ use of DfAM. Moreover, the results demonstrate the potential influence of DfAM in reducing build time and build material of the students’ AM designs, thus improving design performance and manufacturability.

Gaussian Process Emulation for Big Data in Data-Driven Metamaterials Design

Our main contribution is to introduce a novel method for Gaussian process (GP) modeling of massive datasets. The key idea is to build an ensemble of independent GPs that use the same hyperparameters but distribute the entire training dataset among themselves. This is motivated by our observation that estimates of the GP hyperparameters change negligibly as the size of the training data exceeds a certain level, which can be found in a systematic way. For inference, the predictions from all GPs in the ensemble are pooled to efficiently exploit the entire training dataset for prediction. We name our modeling approach globally approximate Gaussian process (GAGP), which, unlike most largescale supervised learners such as neural networks and trees, is easy to fit and can interpret the model behavior. These features make it particularly useful in engineering design with big data. We use analytical examples to demonstrate that GAGP achieves very high predictive power that matches or exceeds that of state-of-the-art machine learning methods. We illustrate the application of GAGP in engineering design with a problem on data-driven metamaterials design where it is used to link reduced-dimension geometrical descriptors of unit cells and their properties. Searching for new unit cell designs with desired properties is then accomplished by employing GAGP in inverse optimization.

Checking the Automated Construction of Finite Element Simulations From Dirichlet Boundary Conditions

From engineering analysis and topology optimization to generative design and machine learning, many modern computational design approaches require either large amounts of data or a method to generate that data. This paper addresses key issues with automatically generating such data through automating the construction of Finite Element Method (FEM) simulations from Dirichlet boundary conditions. Most past work on automating FEM assumes prior knowledge of the physics to be run or is limited to a small number of governing equations. In contrast, we propose three improvements to current methods of automating the FEM: (1) completeness labels that guarantee viability of a simulation under specific conditions, (2) type-based labels for solution fields that robustly generate and identify solution fields, and (3) type-based labels for variational forms of governing equations that map the three components of a simulation set — specifically, boundary conditions, solution fields, and a variational form — to each other to form a viable FEM simulation. We implement these improvements using the FEniCS library as an example case. We show that our improvements increase the percent of viable simulations that are run automatically from a given list of boundary conditions. This paper’s procedures ultimately allow for the automatic — i.e., fully computer-controlled — construction of FEM multi-physics simulations and data collection required to run data-driven models of physics phenomena or automate the exploration of topology optimization under many physics.