Reduced-order modeling (ROM) of time-dependent and parameterized differential equations aims to accelerate the simulation of complex high-dimensional systems by learning a compact latent manifold representation that captures the characteristics of the solution fields and their time-dependent dynamics. Although high-fidelity numerical solvers generate the training datasets, they have thus far been excluded from the training process, causing the learned latent dynamics to drift away from the discretized governing physics. This mismatch often limits generalization and forecasting capabilities. In this work, we propose **Ph**ysics-**i**nformed **ROM** ($\Phi$-ROM) by incorporating differentiable PDE solvers into the training procedure. Specifically, the latent space dynamics and its dependence on PDE parameters are shaped directly by the governing physics encoded in the solver, ensuring a strong correspondence between the full and reduced systems. Our model outperforms state-of-the-art data-driven ROMs and other physics-informed strategies by accurately generalizing to new dynamics arising from unseen parameters, enabling long-term forecasting beyond the training horizon, maintaining continuity in both time and space, and reducing the data cost. Furthermore, $\Phi$-ROM learns to recover and forecast the solution fields even when trained or evaluated with sparse and irregular observations of the fields, providing a flexible framework for field reconstruction and data assimilation. We demonstrate the framework’s robustness across various PDE solvers and highlight its broad applicability by providing an open-source JAX implementation that is readily extensible to other PDE systems and differentiable solvers, available at https://phi-rom.github.io.

Requirements elicitation, a critical yet time-consuming and challenging step in product development, often fails to capture the full spectrum of user needs. This paper introduces a novel framework that leverages Large Language Models (LLMs) to automate and enhance the requirements elicitation process. LLMs are used to generate a vast array of simulated user agents, enabling the exploration of a much broader range of user needs and unforeseen use cases. These agents then engage in product experience scenarios, based on the Action, Observation, and Challenge framework introduced in this paper, inspired by Chain-of-Thought reasoning to produce structured responses. Subsequent agent interviews uncover valuable user needs. We validate our framework with two experiments. First, we explore different methodologies for diverse agent generation, discussing their advantages and shortcomings. We measure the diversity of identified user needs and demonstrate that serial agent generation leads to greater diversity. Second, we show how our framework effectively mimics empathic lead user interviews, identifying a greater number of latent needs than conventional human interviews. Our work highlights the potential of using LLMs to accelerate early-stage product development, reduce costs, and increase innovation.

The Boundary representation (B-rep) format is the de-facto shape representation in computer-aided design (CAD) to model solid and sheet objects. Recent approaches to generating CAD models have focused on learning sketch-and-extrude modeling sequences that are executed by a solid modeling kernel in postprocess to recover a B-rep. In this paper we present a new approach that enables learning from and synthesizing B-reps without the need for supervision through CAD modeling sequence data. Our method SolidGen, is an autoregressive neural network that models the B-rep directly by predicting the vertices, edges, and faces using Transformer-based and pointer neural networks. Key to achieving this is our Indexed Boundary Representation that references B-rep vertices, edges and faces in a well-defined hierarchy to capture the geometric and topological relations suitable for use with machine learning. SolidGen can be easily conditioned on contexts e.g., class labels, images, and voxels thanks to its probabilistic modeling of the B-rep distribution. We demonstrate qualitatively, quantitatively, and through perceptual evaluation by human subjects that SolidGen can produce high quality, realistic CAD models.

Requirements elicitation, a critical, yet time-consuming and challenging step in product development, often fails to capture the full spectrum of user needs. This may lead to products that fall short of expectations. This paper introduces a novel framework that leverages Large Language Models (LLMs) to automate and enhance the requirements elicitation process. LLMs are used to generate a vast array of simulated users (LLM agents), enabling the exploration of a much broader range of user needs and unforeseen use cases. These agents engage in product experience scenarios, through explaining their actions, observations, and challenges. Subsequent agent interviews and analysis uncover valuable user needs, including latent ones. We validate our framework with three experiments. First, we explore different methodologies for diverse agent generation, discussing their advantages and shortcomings. We measure the diversity of identified user needs and demonstrate that context-aware agent generation leads to greater diversity. Second, we show how our framework effectively mimics empathic lead user interviews, identifying a greater number of latent needs than conventional human interviews. Third, we showcase that LLMs can be used to analyze interviews, capture needs, and classify them as latent or not. Our work highlights the potential of using LLM agents to accelerate early-stage product development, reduce costs, and increase innovation.

The analysis of building geometry is an essential capability for generative design, multi-agent simulation, and other computational paradigms being explored in architectural research and practice. We propose a set of voxel-based algorithms that operate on a common data structure to support pathfinding, visibility calculations, direct sunlight calculations, and custom workflows that combine these basic spatial analyses. Our data structure is a binary 3D voxel model represented using a memory-efficient encoding that allows large buildings to be analyzed at fine resolutions. The algorithms are based on underutilized methods from the level set mathematics community that allow short travel paths and sightlines to be approximated without ray casting. The proposed approach has been made available in a visual programming package called VASA (Voxel-based Architectural Space Analysis). We present an example that demonstrates how voxel-based operations and analyses can be chained together to produce human-centric metrics.

We present a nonintrusive approach to robust structural topology optimization. Specifically, we consider optimization of mean- and variance-based robustness metrics of a linear functional output associated with the linear elasticity equation in the presence of probabilistic uncertainties in the loading and material properties. To provide an efficient approximation of higher-dimensional problems, we approximate the solution to the governing stochastic partial differential equations using the anchored ANOVA Petrov-Galerkin projection scheme. We then develop a nonintrusive quadrature-based formulation to evaluate the robustness metric and the associated shape derivative. The formulation is nonintrusive in the sense that it works with any level-set-based topology optimization code that can provide deterministic displacements, outputs, and shape derivatives for selected stochastic parameter values. We demonstrate the effectiveness of the proposed approach on various problems under loading and material uncertainties.

We present an optimization framework that allows designers and engineers to develop conceptual designs for structures that can be manufactured using a network of connected beams. We leverage level set-based topology optimization along with a graph representation extracted from the skeleton of the level set. This graph is used to create an idealized beam network of the structure and the level set and the beam networks co-evolve during the course of the optimization. We propose a constraint that moves the level set geometry closer to this beam network and ultimately drives the two representations to converge. The graph representation also enables us to enforce manufacturability in the beam network, including minimum and maximum cross-section size, angle snapping and cross-section continuity. By applying the beam constraint during topology optimization we are able to generate designs that satisfy the structural requirements and can also be fabricated as beam networks. We present several example sets that illustrate the beam networks that can be generated, how they compare to results without the beam constraint as well as the impact of the manufacturability methods.

We present Neon, a new programming model for grid-based computation with an intuitive, easy-to-use interface that allows domain experts to take full advantage of single-node multi-GPU systems. Neon decouples data structure from computation and back end configurations, allowing the same user code to operate on a variety of data structures and devices. Neon relies on a set of hierarchical abstractions that allow the user to write their applications as if they were sequential applications, while the runtime handles distribution across multiple GPUs and performs optimizations such as overlapping computation and communication without user intervention. We evaluate our programming model on several applications: a Lattice Boltzmann fluid solver, a finite-difference Poisson solver and a finite-element linear elastic solver. We show that these applications can be implemented concisely and scale well with the number of GPUsachieving more than 99% of ideal efficiency.

We introduce UV-Net, a novel neural network architecture and representation designed to operate directly on Boundary representation (B-rep) data from 3D CAD models. The B-rep format is widely used in the design, simulation and manufacturing industries to enable sophisticated and precise CAD modeling operations. However, B-rep data presents some unique challenges when used with modern machine learning due to the complexity of the data structure and its support for both continuous non-Euclidean geometric entities and discrete topological entities. In this paper, we propose a unified representation for B-rep data that exploits the U and V parameter domain of curves and surfaces to model geometry, and an adjacency graph to explicitly model topology. This leads to a unique and efficient network architecture, UV-Net, that couples image and graph convolutional neural networks in a compute and memory-efficient manner. To aid in future research we present a synthetic labelled B-rep dataset, SolidLetters, derived from human designed fonts with variations in both geometry and topology. Finally we demonstrate that UV-Net can generalize to supervised and unsupervised tasks on five datasets, while outperforming alternate 3D shape representations such as point clouds, voxels, and meshes.

We present a method for enforcing manufacturability constraints in generated parts such that they will be automatically ready for fabrication using a subtractive approach. We primarily target multi-axis CNC milling approaches but the method should generalize to other subtractive methods as well. To this end we take as user input: the radius of curvature of the tool bit, a coarse model of the tool head and optionally a set of milling directions. This allows us to enforce the following manufacturability conditions: (1) surface smoothness such that the radius of curvature of the part does not exceed the milling bit radius, (2) orientation such that every part of the surface to be milled is visible from at least one milling direction, (3) accessibility such that every surface patch can be reached by the tool bit without interference with the tool or head mount. We will show how to efficiently enforce the constraint during level set–based topology optimization modifying the advection velocity such that at each iteration the topology optimization maintains a descent optimization direction and does not violate any of the manufacturability conditions. This approach models the actual subtractive process by carving away material accessible to the machine at each iteration until a local optimum is achieved.

SPICE is an imaging spectrometer operating at vacuum ultraviolet (VUV) wavelengths, 70.4 - 79.0 nm and 97.3 - 104.9 nm. It is a facility instrument on the Solar Orbiter mission, which carries 10 science instruments in all, to make observations of the Sun's atmosphere and heliosphere, at close proximity to the Sun, i.e to 0.28 A.U. at perihelion. SPICE's role is to make VUV measurements of plasma in the solar atmosphere. SPICE is designed to achieve spectral imaging at spectral resolution >1500, spatial resolution of several arcsec, and two-dimensional FOV of 11 x16arcmins. The many strong constraints on the instrument design imposed by the mission requirements prevent the imaging performance from exceeding those of previous instruments, but by being closer to the sun there is a gain in spatial resolution. The price which is paid is the harsher environment, particularly thermal. This leads to some novel features in the design, which needed to be proven by ground test programs. These include a dichroic solar-transmitting primary mirror to dump the solar heat, a high in-flight temperature (60deg. C) and gradients in the optics box, and a bespoke variable-line-spacing grating to minimise the number of reflective components used. The tests culminate in the system-level test of VUV imaging performance and pointing stability. We will describe how our dedicated facility with heritage from previous solar instruments, is used to make these tests, and show the results, firstly on the Engineering Model of the optics unit, and more recently on the Flight Model. For the keywords, select up to 8 key terms for a search on your manuscript's subject.

The TopSat camera is a low cost remote sensing imager capable of producing 2.5 metre resolution panchromatic imagery, funded by the British National Space Centre’s Mosaic programme. The instrument was designed and assembled at the Space Science & Technology Department of the CCLRC’s Rutherford Appleton Laboratory (RAL) in the UK, and was launched on the 27th October 2005 from Plesetsk Cosmodrome in Northern Russia on a Kosmos-3M. The camera utilises an off-axis three mirror system, which has the advantages of excellent image quality over a wide field of view, combined with a compactness that makes its overall dimensions smaller than its focal length. Keeping the costs to a minimum has been a major design driver in the development of this camera. The camera is part of the TopSat mission, which is a collaboration between four UK organisations; QinetiQ, Surrey Satellite Technology Ltd (SSTL), RAL and Infoterra. Its objective is to demonstrate provision of rapid response high resolution imagery to fixed and mobile ground stations using a low cost minisatellite. The paper “Development of the TopSat Camera” presented by RAL at the 5th ICSO in 2004 described the opto-mechanical design, assembly, alignment and environmental test methods implemented. Now that the spacecraft is in orbit and successfully acquiring images, this paper presents the first results from the camera and makes an initial assessment of the camera’s in-orbit performance.

Design problems are complex and not well-defined in the early stages of projects. To gain an insight into these problems, designers envision a space of various alternative solutions and explore various performance trade-offs, often manually. To assist designers with rapidly generating and exploring a design space, researchers introduced the concept of design synthesis methods. These methods promote innovative thinking and provide solutions that can augment a designer's abilities to solve problems. Recent advances in technology push the boundaries of design synthesis methods in various ways: a vast number of novel solutions can be generated using high-performance computing in a timely manner, complex geometries can be fabricated using additive manufacturing, and integrated sensors can provide feedback for the next design generation using the Internet of things (IoT). Therefore, new synthesis methods should be able to provide designs that improve over time based on the feedback they receive from the use of the products. To this end, the objective of this study is to demonstrate a design synthesis approach that, based on high-level design requirements gathered from sensor data, generates numerous alternative solutions targeted for additive manufacturing. To demonstrate this method, we present a case study of design iteration on a car chassis. First, we installed various sensors on the chassis and measured forces applied during various maneuvers. Second, we used these data to define a high-level engineering problem as a collection of design requirements and constraints. Third, using an ensemble of topology and beam-based optimization techniques, we created a number of novel solutions. Finally, we selected one of the design solutions and because of some manufacturability constraints we, 3D-printed a prototype for the next generation of design at one third scale. The results show that designs generated from the proposed method were up to 28% lighter than the existing design. This paper also presents various lessons learned to help engineers and designers with a better understanding of challenges applying new technologies in this research.

Sensitivity analysis is the use of computational and statistical methods to investigate and analyze the sensitivity of a model. Within the context of product development and manufacturing it is a tool used to quantify the relative importance of known process variation within the space of design tolerances and observed manufacturing defects. In this work sensitivity analysis is shown to provide numerous benefits to the development of a CPV system, including; ranking the relative importance of manufacturing tolerances, understanding how defects interact to influence device efficiency, predicting distributions of product efficiency and manufacturing yield, comparing product designs, and providing feedback for product tolerance requirements.