About

Dr. Burcu Akinci is the William D. and Nancy W. Strecker Dean of the College of Engineering at Carnegie Mellon University and a full professor in the Department of Civil and Environmental Engineering. Her research focuses on climate-resilient environmental systems and technologies, sustainable energy and transportation systems, and intelligent engineered systems and society. She is involved in advancing engineering solutions that address environmental challenges and improve societal resilience through innovative research and leadership.

Research topics

• Computer Science
• Mathematics
• Artificial Intelligence
• Transport engineering
• Environmental health
• Machine Learning
• Econometrics
• Engineering
• Statistics
• Algorithm
• Medicine
• Environmental science

Selected publications

• Reducing changeover waste in customized manufacturing: A causal-informed approach to dynamic sequence optimization

  Advanced Engineering Informatics · 2026-04-11

  articleOpen access
  PublisherDOI

• Human-Machine Collaboration for Coordinating Design Customization and Production Systems

  2025-08-17

  article

  Abstract The modern manufacturing industry faces increasing challenges in rapidly machine setups for adapting customized product designs, with frequent adjustments consuming up to 30% of total production time. Current approaches rely on human operators’ trial-and-error setups, which, while incorporating valuable experiential knowledge, are unstable, labor-intensive, error-prone, and difficult to transfer to new tasks. This limitation restricts the integration of human expertise with intelligent machine algorithm resulting in unstable and inefficient setup processes that struggle to meet the demands of customized manufacturing. To address these limitations, we propose a Human-knowledge-Embed Bayesian Optimization (HK-Embed BO) method that leverages Shapley Additive Explanations (SHAP) values to capture human implicit knowledge in a transferable format. This extracted knowledge guides Bayesian Optimization (BO) by dynamically shaping the search space, effectively steering the algorithm toward regions identified by human insight as promising. By integrating interpretable human-domain knowledge with BO, our approach is expected to improve efficiency by reducing unnecessary trial-and-error iteration, enhance adaptability to new design variations through knowledge transfer, and bolster the robustness of the optimization process under customized manufacturing’s dynamic environment. This method has been validated in a roll-forming process for manufacturing customized HVAC components. These results show that the proposed method reduces wasted time compared to both human and machine approaches, requiring 14.6% fewer trial-and-errors than human experts and 76.7% fewer than vanilla BO baseline. Additionally, it decreases the standard deviation of trial-and-error iterations by 63.5% and 83.3% compared to human operators and vanilla BO, respectively.

  PublisherDOI

• Explainable Production Time and Material Waste Prediction of Customized Building Mechanical System Components

  2025-12-11

  articleSenior authorCorresponding

  Explainable production time and material waste for customized building mechanical system components are critical for resolving production bottlenecks related to diverse building needs. Customized building mechanical system elements often lead to unpredictable production and material waste in modular building practices. As a result, production managers could not implement reliable planning for customized orders, failing to prevent delays and cost escalations. One challenge is that existing production time and material waste prediction methods cannot explain the impacts of some customized features of components that cause time and material waste. Such explanations are needed for proactive production system management in reducing time waste in customized building manufacturing. A challenge to implementing process waste prediction and explanation algorithms is the lack of comprehensive and balanced data sets that capture diverse products’ production histories for reliable prediction with accountable explanations. This study examines an explainable waste prediction method that overcomes those challenges. Causal inference is utilized to uncover causal relationships, addressing the scarcity of production history data for customized products. The results show that the proposed methods achieve more reliable production time and material waste prediction for customized ventilation ducts while identifying critical features responsible for delays and cost escalation.

  PublisherDOI

• A Markov Decision Process-Based Method to Generate Strategies for Automated Updating of Building Energy Models

  2025-12-11

  articleSenior authorCorresponding

  Building energy models (BEMs), which are used to evaluate building energy performance, need to be updated when there is a change in the design of a building. Currently, the BEM updating process relies on domain experts to create and implement model updating strategies, which is time-consuming. Existing studies related to automated updating of BEMs mainly focus on calibrating parameters in the existing model. However, for design changes resulting in more than just parameter recalibration, such as room reconfiguration, there is a lack of methods to automatically modify existing BEMs to reflect the change other than regenerating them, but updating BEMs through regeneration can also result in losing prior expert knowledge embedded in the original BEMs. The study described in this paper draws inspiration from software engineering and draws a parallel between the problem of BEM updating at the structure level and model repair tasks, considering the similarity shared by the two. More specifically, the process of generating BEM updating strategies, the sequence of actions required to modify a BEM towards the final model state reflecting the design changes, is cast as a sequential decision-making problem formalized using a Markov Decision Process (MDP). This MDP-based formalism is implemented on a campus building reconfiguration case that involves updating the corresponding Modelica model. The implementation results demonstrated the feasibility of the proposed method and at the same time highlighted two critical points for increasing the robustness of utilizing MDP-based methods for automated BEM updating. These are (1) the choice of representation for model states and (2) the choice of an end state for a given model updating objective. Future work will explore ways to address these challenges.

  PublisherDOI

• Pruning Bayesian networks for computationally tractable multi-model calibration

  Frontiers in Aerospace Engineering · 2025-05-30 · 1 citations

  articleOpen accessSenior author

  Anomaly response in aerospace systems increasingly relies on multi-model analysis in digital twins to replicate the system’s behaviors and inform decisions. However, computer model calibration methods are typically deployed on individual models and are limited in their ability to capture dependencies across models. In addition, model heterogeneity has been a significant issue in integration efforts. Bayesian Networks are well suited for multi-model calibration tasks as they can be used to formulate a mathematical abstraction of model components and encode their relationship in a probabilistic and interpretable manner. The computational cost of this method however increases exponentially with the graph complexity. In this work, we propose a graph pruning algorithm to reduce computational cost while minimizing the loss in calibration ability by incorporating domain-driven metrics for selection purposes. We implement this method using a Python wrapper for BayesFusion software and show that the resulting prediction accuracy outperforms existing pruning approaches which rely primarily on statistics.

  PublisherOA PDFDOI

• Ontologies at Work: Analyzing Information Requirements for Model Predictive Control in Buildings

  2024-10-29 · 2 citations

  articleOpen accessSenior author

  Model Predictive Control (MPC) has shown significant potential for improving energy efficiency, indoor air quality and occupant comfort of buildings. MPC-based control algorithms have also shown the ability to shift loads and optimize for multiple objectives, including but not limited to reducing the green-house gas emissions, energy costs and peak demand. However, one of the main implementation challenges of these control algorithms is the integration and configuration effort needed to deploy a supervisory MPC controller in a building. By assigning standardized references to information sources and control points in buildings, existing studies have shown that semantic ontologies and corresponding queries have the potential to ease the deployment of such controllers. Yet, the use of semantic information to ease the deployment processes of MPC controllers is still limited. In this paper, we review three MPC experiments and synthesize the information requirements of these optimization problems. We then turn to existing and upcoming semantic ontologies such as Brick, SAREF and ASHRAE Standard 223 to represent these requirements, evaluating their potential to support the implementation of an MPC controller. This investigation concludes with a discussion of existing opportunities and open questions that the community should explore to support more streamlined MPC implementations.

  PublisherDOI

• Integrated Calibration of Simulation Models for Autonomous Space Habitat Operations

  2024-03-02 · 3 citations

  articleSenior author

  Space habitats for exploration beyond low earth orbit need to provide the crew with enhanced capabilities for earth-independent operations. Mission control has traditionally been the main decision maker in anomaly response procedures, but this role will be limited in deep space due to increased communication delays. Digital simulation models are used by ground control for troubleshooting tasks and are likely to be essential assets for the crew to test "what-if" scenarios when important faults are detected onboard. Migrating models from mission control to a space habitat is however challenging as these models are typically heterogeneous and rely on the knowledge of sub-system specialists to be operated. Efforts have been made to automate the integration of multiple simulation models, but their calibration, i.e., the assignment of model parameters that best represent the system behavior, typically remains expert-driven and focused on individual models. To alleviate this reliance on experts and facilitate integration without human intervention, we propose leveraging the interpretable representation ability of probabilistic graphical models to encode dependencies between simulation models at the time of calibration. In this mathematical abstraction, nodes represent random variables, and edges embed causal relationships as conditional probability distributions. We build a graphical model hierarchically with a first layer of nodes representing subsystem states, and a second layer for the simulation model parameters, e.g., a set of possible slopes and intercepts of a regression model. The two layers are mapped probabilistically using domain knowledge from sub-system specialists thereby enabling the migration of reasoning capabilities from mission control to a space habitat. The created network is used to infer the most likely set of simulation parameters given the believed system state which is derived from a diagnosis module. We study the proposed mechanism by implementing it in a docking scenario. In this scenario, the crew of an incoming vehicle is performing a system readiness check before docking to a space station. An algorithm detects a CO2 removal fault, and we perform calibration accordingly using a graphical model. Three types of simulation models are being integrated via calibration, namely: (i) machine learning models trained on empirical data from a testbed of the ISS Carbon Dioxide Removal Assembly, (ii) physics-based models that were designed for this same testbed and (iii) knowledge-based models derived from NASA’s flight rules on admissible CO2 concentrations in a space habitat. We explore a method to implement such a graphical model that consists of (i) selecting a subset of system states as degrees of faulty behaviors, (ii) identifying their dependencies, i.e., defining the likelihood of cascading fault symptoms across subsystems, and (iii) selecting the simulation models that are expected to provide the most insight onboard and which can be parameterized given the network of system states established in the previous two steps. Our study reveals challenges that are to be solved for implementing this graphical model-based calibration. Specifically, it identifies a need to formalize the creation of the network for the subsystem states and to assess how to leverage existing standards for simulation model interfaces.

  PublisherDOI

• <i>FSBrick</i>: an information model for representing fault-symptom relationships in heating, ventilation, and air conditioning systems

  Data-Centric Engineering · 2024-01-01 · 1 citations

  articleOpen access

  Abstract Current fault diagnosis (FD) methods for heating, ventilation, and air conditioning (HVAC) systems do not accommodate for system reconfigurations throughout the systems’ lifetime. However, system reconfiguration can change the causal relationship between faults and symptoms, which leads to a drop in FD accuracy. In this paper, we present Fault-Symptom Brick ( FSBrick ), an extension to the Brick metadata schema intended to represent information necessary to propagate system configuration changes onto FD algorithms, and ultimately revise FSRs. We motivate the need to represent FSRs by illustrating their changes when the system reconfigures. Then, we survey FD methods’ representation needs and compare them against existing information modeling efforts within and outside of the HVAC sector. We introduce the FSBrick architecture and discuss which extensions are added to represent FSRs. To evaluate the coverage of FSBrick , we implement FSBrick on (i) the motivational case study scenario, (ii) Building Automation Systems’ representation of FSRs from 3 HVACs, and (iii) FSRs from 12 FD method papers, and find that FSBrick can represent 88.2% of fault behaviors, 92.8% of fault severities, 67.9% of symptoms, and 100% of grouped symptoms, FSRs, and probabilities associated with FSRs. The analyses show that both Brick and FSBrick should be expanded further to cover HVAC component information and mathematical and logical statements to formulate FSRs in real life. As there is currently no generic and extensible information model to represent FSRs in commercial buildings, FSBrick paves the way to future extensions that would aid the automated revision of FSRs upon system reconfiguration.

  PublisherOA PDFDOI

• Pre-trained language model based method for building information model to building energy model transformation at metamodel level

  Proceedings of the ... ISARC · 2024-05-27 · 3 citations

  articleSenior author

  Pre-trained language model based method for building information model to building energy model transformation at metamodel level Zhichen Wang, Mario Bergés, Burcu Akinci Pages 17-25 (2024 Proceedings of the 41st ISARC, Lille, France, ISBN 978-0-6458322-1-1, ISSN 2413-5844) Abstract: Building energy model (BEM) creation based on building information models (BIM) can save model re-creation time during building design phase. However, current BIM-to-BEM transformation is at the model level so the BEM is re-generated every time when the change happens to the BIM. Since design changes happen frequently and the generated BEM needs fine-tuning, such model regeneration is still time-consuming. Mapping rules between BIM and BEM are needed to achieve component level transformation, so that only the corresponding part in BEM instead of whole BEM are updated when changes happen to the BIM. These mapping rules can be defined explicitly using model transformation languages. However, these rule-based transformation methods have limitations in scalability. To solve this issue, this study proposes a pre-trained language model (PLM) based method to construct the mapping relationships between BIM and different types of BEM at the metamodel level. In summary, we formulate the BIM-BEM mapping as a machine translation task and solve it using PLM. For evaluation we collected and generated 35 pairs of BIM and BEM metamodels, and these metamodels are preprocessed into formatted texts that are readable by PLM. The 82% matching accuracy is achieved by proposed method which is higher than the 61% accuracy achieved by a baseline model in previous work. This paper shows the potential to utilize PLMs to facilitate the BIM to BEM transformation from BIM to a varying type of BEMs at the metamodel level. Future work will be focus on the realizing instance-level mapping and transformation. Keywords: Model transformation, Building information model, Building energy model, Interoperability, Natural language processing, Pre-trained language model DOI: https://doi.org/10.22260/ISARC2024/0004 Download fulltext Download BibTex Download Endnote (RIS) TeX Import to Mendeley

  PublisherDOI

• Digital Twin Technologies for Autonomous Environmental Control and Life Support Systems

  Journal of Aerospace Information Systems · 2024-01-08 · 14 citations

  articleSenior author

  Environmental control and life support systems will require enhanced self-awareness and self-sufficiency as human spaceflights are designed to reach further destinations. These requirements have led to the development of autonomous technologies and systems to enable more Earth independence, while at the same time relying more heavily on the knowledge contained in their computational models (as opposed to the knowledge of ground control experts). For environmental control and life support systems, these consist of disparate models often tailored to specific subsystems and use cases, such as temperature control and [Formula: see text] removal. Therefore, there is a need for technologies supporting the integration of existing models. We propose to extend existing digital twin frameworks to integrate models and serve as a tool for answering onboard queries during operation. Toward this vision, we identify research directions for three types of technologies: i) streamlining the process of merging information models without redundancies by leveraging model-based systems engineering languages; ii) the calibration of simulation models that requires a better understanding of how to encode domain knowledge into a probabilistic representation of subsystem states and model parameters; and iii) automating the query understanding process and constructing a mapping between information and simulation models.

  PublisherDOI

Recent grants

• ITR/IM+AP Early Defect Detection and Management at Construction Sites Using Integrated Project Models, Laser Scanners and Embedded Sensor Systems

  NSF · $2.1M · 2001–2008

• CAREER: Development of an Integrated Framework for Capturing, Modeling and Visualizing the History of Construction Projects

  NSF · $407k · 2005–2011

• MRI: Acquisition of a High-Performance Laser Scanner for Research and Education in Civil Engineering, Architecture, Robotics and Computer Graphics

  NSF · $207k · 2004–2008

Frequent coauthors

• James H. Garrett

  Shell (United States)

  89 shared

• Mario Bergés

  Carnegie Mellon University

  55 shared

• Semiha Kiziltas

  45 shared

• Lúcio Soibelman

  University of Southern California

  43 shared

• Esin Ergen

  Istanbul Technical University

  41 shared

• Semiha Ergan

  San Diego State University

  28 shared

• Pingbo Tang

  Carnegie Mellon University

  27 shared

• Carl T. Haas

  26 shared

Education

• B.S., Civil Engineering

  Middle East Technical University

  1991

• Other

  Bilkent University

  1993

• M.S., Civil and Environmental Engineering

  Stanford University

  1995

• Ph.D., Civil and Environmental Engineering

  Stanford University

  2000

Awards & honors

• ASCE Distinguished Member