Search

The escalating demand for free-form buildings presents a formidable technological hurdle in the realm of prefabrication manufacturing and assembly. This research draws inspiration from recent advancements in large language models (LLMs) and seeks to harness their transformative potential in tandem with Building Information Modeling (BIM) to advance the Design for Manufacture and Assembly (DfMA) method. The ultimate aim is to catapult the prefabrication industry into an era characterized by unparalleled design freedom and heightened efficiency. The research includes contributions from the following four areas: (1) A review of the literature on DfMA theories, specifically fabrication and construction. (2) The development of a BIM-based DfMA method framework, including the methodological basis, Design for Manufacturing (DfM), and Design for Assembly (DfA). (3) A pipeline for LLMs driven BIM-based DfMA method framework and a validation using the Decathlon Solar Team design-build practice at the University of Washington’s Lotus House. (4) An evaluation of 692 conversation scenarios to verify the feasibility of our proposal. We get a 91.1% user agreement on average based on GPT4. The research results could guide the workflow in building a lifecycle for DfMA practices in the future smart manufacturing era.

Free-form surfaces are commonly present in several engineering components, from molds to more sophisticated components of aeronautical engines. These parts are usually finished by machining, more specifically, ball-end milling. In this process, the contact between the tool and the part constantly changes along the toolpath, resulting in several manufacturing problems that damage the surface and compromise the performance of the part. Besides the ordinary cutting parameters, these machined surfaces are deeply influenced by the tool tip center and the elastoplastic deformation of the material through the shearing and plowing pair. Knowing the exact orientation of the tool in relation to the surface determines all aspects of the milling process and is necessary for process modeling. The current CAD/CAM software platforms is not able to predict such surface topography. In this direction, the current work presents a software routine developed and implemented on an open interface CAD/CAM software (Siemens® NX). The normal vectors of the surface to be machined, the cutter contact (CC), and cutting location (CL) data from the toolpath calculated by the CAD/CAM were used to obtain a complete discretization of the cutter-workpiece position along the toolpath. This information was used to predict the surface topography and identify the cutting-edge elements. Then, the developed model was used to evaluate the free-form surface of a blade milled on 5-axis together with confocal imaging analysis. The results show that the methodology developed can predict the topography aspects of a free-form machined surface and support the analysis of milling problems such as run-out.

We design and integrate free‐form metagratings with irregular nanoscale patterns into diffractive augmented reality (AR) waveguides to achieve simultaneous improvements in eye‐glow attenuation and outcoupling efficiency. The freeform metagrating demonstrates over 40% reduction in eye‐glow while enhancing outcoupling efficiency by 8% compared to the conventional binary grating. [ABSTRACT FROM AUTHOR]

Interpretability fosters trust when humans and artificial intelligence (AI) systems interact, and its value for neural networks in particular cannot be overstated. In this paper, we analyse the emergence and value of neural modularity as it relates to interpretability. We compare the modularity evolved through connectivity constraints in terms of network Q-scores and examine the interpretable qualities of the resultant networks with functional subset regression. The connectivity constraints compared here include those proposed by previous research as well as several novel variations formulated to express neuron input competition and connections per neuron variance. Networks were evolved using HyperNEAT on a free-form substrate. The results indicate that the connection costs successfully promote the evolution of neural modularity across a variety of tasks and show that the novel connection cost variations are competitive with previously explored connection costs. The interpretability assessment shows that while the evolved networks' interpretable qualities are task dependent, two of the compared connection costs deliver statistically different functional module overlap distributions. However, recovered subnetwork module accuracies remain low, highlighting the key points for future research. [ABSTRACT FROM AUTHOR]