Department of Architecture

There are three imperial residences in Kyoto: Gosho (京都御所), rebuilt in 1855 and used for formal affairs even today; Shūgakuin (修学院離宮), a summer retreat on mountain slopes built in the mid-seventeenth century; and Katsura Imperial Retreat (桂離宮), slightly older than Shūgakuin. Upon the death of the Hachijō imperial line in 1881, Katsura came into the hands of the reigning household; shortly afterward, the Imperial Household Ministry was formed and took responsibility for the care of such sites. Sometimes grouped with the other residences, Nijō Palace was originally built not for the imperial household but for the warriors who effectively ruled Japan from the seventeenth to the middle of the nineteenth century; today, it too is managed by the Imperial Household Agency (the scope and name of the Imperial Household Ministry having changed at the end of World War II). Of these four, Katsura, with its extensive grounds and esteemed teahouses in addition to a large, shoin-style residence, is best known of all, used both at home and abroad to illustrate arguments about architecture and national tradition. Yet even so, much remains to be said about the complex, as demonstrated by this brief descriptive bibliography. Download High-Resolution PDF

Structural member sizing directly influences material consumption, embodied carbon, and overall structural efficiency. Recent advances in graph representations have opened new possibilities for structural optimization, with applications such as topology optimization and surrogate modeling. We propose a graph-based representation learning and clustering pipeline that uses structural simulation results to support member sizing and cross-section optimization. In this representation, connection points and linear members in the structural system are modeled as two node types in a heterogeneous graph. Mechanical responses, geometric properties, and topological information are encoded as features, and structural context is propagated through message passing. We then train a Heterogeneous Graph Attention Network (HeteroGAT) encoder with a contrastive objective constructed from mechanically similar and topologically adjacent member pairs. Finally, we cluster the learned embeddings with Gaussian Mixture Models (GMM). The resulting clusters provide a practical basis for structural optimization. Specifically, we use the clustering results as grouping rules for cross-section optimization in Karamba3D, such that members within the same cluster share a common profile. Using steel member sizing as a case study, we evaluate the cross-section assignment strategies on a whole-building structural system and demonstrate the method through multiple application examples. The results show that the proposed method achieves a rationalized section system with significantly fewer cross-section types and stable convergence, while maintaining structural performance comparable to the original design, making it a practical tool for early-stage steel structural design.

Structural design is deeply expertise-driven: engineers rely on intuition, experience, and both explicit and tacit knowledge of load paths, structural typologies, first principles and in-depth calculations to arrive at a design solution. Current computational workflows in structural design largely focus on optimization, form-finding and dimensioning of members, where the design space is predefined and algorithms iteratively refine a solution. Machine learning, however, opens the possibility to explore broader design spaces in which geometry and topology are not fixed in advance. Imitation learning (IL), a paradigm closely related to reinforcement learning, offers a pathway to integrate skills from design examples, rather than explicit rulesets. Recently, flow matching, a generative framework that learns a vector field transporting noise toward data over time, has emerged as a prominent method for modeling complex, multimodal action distributions that standard one-shot predictors often fail to capture. By learning from engineers’ demonstrations with a flow-based imitation policy, we transfer structural intuition into a design agent without dense reward engineering or computationally expensive finite element analysis at every decision step. To study this idea, we create a 2D structural testing environment. Coupled with a finite element analysis solver to track and rank design outcomes, we create a training ground for the IL algorithm. Using the environment we tested building of pin-jointed steel truss bridges as graphs and trained an imitation policy to predict chunks of continuous placement actions and capture longer-term design intent. Our results show that the flow-based policies can learn structural intuition, generating diverse feasible bridge designs that reflect established engineering principles. The work introduces a reproducible benchmark for assembly-constrained structural design and examines the limitation and the potential of generative imitation learning as an alternative framework for structural design exploration beyond top-down optimization.

Residential demand response (DR) is a valuable resource for grid reliability, but remains challenging because the highly heterogeneous residential building stock leads to widely varying and hard-to-predict load and comfort responses during DR events. Although prior research has estimated the technical potential of DR-capable technologies for achieving energy demand savings, little is known about how they affect thermal comfort. In particular, it remains unclear how indoor thermal conditions due to DR depend on the thermal envelope characteristics of the housing stock. To address this gap, this study provides a systematic, location-specific assessment of indoor thermal performance during DR-events across the US housing stock using both typical DR weather data and detailed building metadata. We evaluate how envelope characteristics influence indoor temperatures during realistic simulated summer and winter DR events across 37 US locations, applying both temperature threshold and rate of temperature change criteria to estimate region-level probabilities of discomfort. Additionally, we show the impact of distinct weather patterns that intensify or abate thermal stress on comfort outcomes. Results show a near-universal overheating risk in summer DR events, where comfort outcomes are strongly influenced by rapid risk of comfort violations. In contrast, overall winter DR discomfort risk is lower, risk escalation is more gradual and shows greater sensitivity to event duration. These findings offer a data-driven quantification of comfort risk across diverse climates and building envelopes, demonstrating the need for region-specific DR scheduling and discomfort mitigation strategies tailored to local weather patterns and the performance of existing residential buildings.

Conventional measurement and verification (M&V) methods for estimating energy savings rely on comparing pre- and post-retrofit performance. They are often time-consuming and unreliable, especially when non-routine events, such as step changes or more gradual changes in building operation, occur during the M&V process. When those events are unrelated to the retrofit intervention and significantly affect building energy consumption, the results will be confounded when the analyst applies the conventional M&V method. In this study, we demonstrated that switchable interventions, such as most HVAC control retrofits, can benefit from a new M&V method that randomly samples whether to implement the baseline or the intervention strategy at a fixed interval (e.g., daily). We tested this novel randomized M&V method on a large public dataset (hourly energy data over 2 years for 639 buildings) covering various climate zones and commercial building types, using a virtual chilled water supply temperature reset based on outdoor weather as the intervention. The results show that, compared to the conventional method, the randomized method provides more accurate savings estimations with a median of 74% accuracy improvement and is faster (typically 36 weeks instead of 104 weeks, ∼65% reduction in duration). Additionally, we found that when non-routine events are present (e.g., occupancy pattern change), the randomized method estimates savings that are much closer to the ground-truth values than the conventional method, demonstrating significantly improved reliability. We also assessed the impact of normalizing for different weather, starting the M&V at different dates of the year, continuing randomization with a different sampling ratio after satisfying all stopping criteria, and dropping samples affected by carryover effects when switching between strategies. For each scenario, we identified the optimal sampling interval using the large dataset.

Coarse mode respiratory aerosols can carry viral loads over long distances and have very different dynamics than submicron particles, but experimental studies under realistic conditions remain limited. To study the differential impacts on exposure under different mixing conditions, we co-released 7–10 µm particles and carbon dioxide (CO2)—which served as an indicator of gas and submicron particle dynamics—in a 158 m3 room at LBNL’s FLEXLAB facility with an overhead heating, ventilation, and air conditioning (HVAC) system. The room was arranged as a distanced meeting then a classroom with eight heated manikins and a researcher. Spatial variability was measured using 16 particle counters and 22–26 CO2 sensors throughout the space. Conditions included: HVAC off or supply air at 1000-1060 m3 h-1 at neutral, cooling, or heating temperatures; with and without 20% outdoor air; and added HVAC filtration, portable air cleaners (PACs), or a physical barrier between the speaker and occupants. We found that good mixing via neutral or cooling supply air or use of PACs under heating lowered coarse particle exposure at some locations, but increased exposure for one-quarter to two-thirds of manikins compared to poor mixing under heating. A physical barrier reduced direct transfer of coarse particles during heating, but less during cooling. High spatial variability shows that a single measurement cannot represent occupant exposure. Instantaneous air mixing assumptions overstate the effectiveness of ventilation, HVAC filtration, and upper-room germicidal ultraviolet disinfection for coarse particles, as relatively few particles reach the return grille or upper room under most conditions.

Human-factor analysis typically employs correlation analysis and significance testing to identify relationships between variables. However, these descriptive (‘what-is’) methods, while effective for identifying associations, are often insufficient for answering causal (‘what-if’) questions. Their application in such contexts often overlooks confounding and colliding variables, potentially leading to bias and suboptimal or incorrect decisions. We advocate for explicitly distinguishing descriptive from interventional questions in human-factor analysis, and applying causal inference frameworks specifically to these problems to prevent methodological mismatches. This approach disentangles complex variable relationships and enables counterfactual reasoning. Using post-occupancy evaluation (POE) data from the Center for the Built Environment’s (CBE) Occupant Survey as a demonstration case, we show how causal discovery generates testable hypotheses about intervention hierarchies and directional relationships that traditional associational analysis cannot explore. The systematic distinction between causally associated and independent variables, combined with intervention prioritization capabilities, offers broad applicability to complex human-centric systems, for example, in building science or ergonomics, where understanding intervention effects is critical for optimization and decision-making.