Short answer

When presenting complex simulation data, especially in immersive environments like MR, consider using AI tools that have been specifically trained on the relevant domain to ensure accurate and understandable interpretation for all users.

Field
Modelling
Source
Technologies (2026)
Method
Quantitative evaluation and domain adaptation of AI models.
Evidence
Strong effect

Fine-tuning general-purpose Vision-Language Models (VLMs) on domain-specific datasets significantly enhances their ability to interpret complex Computational Fluid Dynamics (CFD) visualizations within Mixed Reality (MR) environments, enabling non-expert stakeholders to quantitatively understand indoor environmental performance. This modelling research insight is drawn from a 2026 study published in Technologies. Using Quantitative evaluation and domain adaptation of ai models., researchers explored how this design variable affects real-world outcomes. The key design takeaway: When presenting complex simulation data, especially in immersive environments like MR, consider using AI tools that have been specifically trained on the relevant domain to ensure accurate and understandable interpretation for all users.

Study
ModellingNew This WeekStrong effect

Domain-adapted VLMs improve CFD visualization interpretation in MR by 30% for non-experts

Fine-tuning general-purpose Vision-Language Models (VLMs) on domain-specific datasets significantly enhances their ability to interpret complex Computational Fluid Dynamics (CFD) visualizations within Mixed Reality (MR) environments, enabling non-expert stakeholders to quantitatively understand indoor environmental performance.

Technologies · 2026

01

Key Findings

  • 01Baseline general-purpose VLM achieved less than 30% accuracy in interpreting MR CFD visualizations.
  • 02Fine-tuning the VLM on a domain-specific dataset improved accuracy to over 60% across all interpretation categories.
  • 03Domain adaptation enabled VLMs to quantitatively interpret physical information from CFD visualizations, moving beyond simple color associations.
02

Application

Design takeaway

When presenting complex simulation data, especially in immersive environments like MR, consider using AI tools that have been specifically trained on the relevant domain to ensure accurate and understandable interpretation for all users.

How to apply

Incorporate AI-driven interpretation layers into MR visualizations of design simulations to provide real-time, understandable feedback to stakeholders.

Project actions

  • 01Explore how AI could help interpret your design models (e.g., CAD, simulations).
  • 02Consider if your target users would benefit from AI-assisted interpretation of technical data.
03

Method & Evidence

AimTo investigate the effectiveness of domain adaptation in improving the quantitative interpretation of mixed-reality CFD visualizations by Vision-Language Models (VLMs) for non-expert users.
MethodQuantitative evaluation and domain adaptation of AI models.
ProcedureA novel dataset of MR images with superimposed CFD results (indoor temperature and airflow) was created, paired with domain-specific Q&A annotations requiring legend-based reasoning. A general-purpose VLM (Qwen2.5-VL) was fine-tuned using this dataset, and its performance was compared to the baseline model in interpreting the visualizations.
ContextBuilt environmental design, Mixed Reality (MR) simulations, Computational Fluid Dynamics (CFD) visualization interpretation.

Variables

IVDomain adaptation of Vision-Language Models.
DVAccuracy of CFD visualization interpretation.
CVType of CFD visualization (MR, indoor temperature/airflow), general-purpose VLM architecture (Qwen2.5-VL), dataset characteristics, evaluation metrics.
04

Strengths & Limitations

Strengths

  • +Addresses a critical need for interpreting complex simulation data in design.
  • +Introduces a novel dataset and methodology for domain adaptation of VLMs in MR.
  • +Provides quantitative evidence of performance improvement.

Limitations

Replicating the domain adaptation process requires significant computational resources and expertise in AI training. Access to specialized datasets might also be a challenge.

Reliability & validity

Reliability is supported by the quantitative evaluation of model accuracy. Validity is enhanced by the creation of a novel, domain-specific dataset designed to test legend-based reasoning, directly addressing the research aim. However, generalizability might be limited by the specific VLM and CFD parameters tested.

Think critically

To what extent can VLMs truly replace human expertise in interpreting complex design simulations, and what are the ethical considerations of relying on AI for critical design decisions?

05

Design Principles

"Technical simulation data can be made accessible to non-experts through AI-powered interpretation tools, provided these tools are adapted to the specific domain and visualization context."

This research highlights how advanced AI, specifically VLMs, can bridge the gap between technical simulations (like CFD) and user comprehension. For design, it demonstrates the potential of AI to make complex modelling outputs accessible, facilitating better design decisions and user-centred design by allowing a wider range of stakeholders to engage with and understand simulation results.

06

What This Means for Your Design

Computers that can 'see' and 'understand' text can be taught to explain complex 3D visualizations of building simulations in mixed reality, making it easier for anyone to grasp how a building will perform.

How to use in your project

  • 1.Use this research to justify the need for user-friendly interpretation of your design models, especially if they involve complex simulations.
  • 2.Consider how AI could be a 'tool' in your design process to enhance user understanding or stakeholder communication.
07

Add to My Project

08

Quick Cite

Paragraph starter

This study demonstrates that Vision-Language Models (VLMs) can be effectively adapted to interpret complex mixed-reality (MR) visualizations of Computational Fluid Dynamics (CFD) results, significantly improving accuracy for non-expert users. By fine-tuning a general VLM on a domain-specific dataset, interpretation accuracy increased from under 30% to over 60%, enabling quantitative reasoning based on legend information rather than relying on learned natural image associations. This highlights the potential for AI to bridge the gap between technical simulations and user comprehension in built environmental design, facilitating more informed design decisions and user engagement.

09

Source

Technologies

Quantitative Evaluation and Domain Adaptation of Vision–Language Models for Mixed-Reality Interpretation of Indoor Environmental Computational Fluid Dynamics Visualizations

journal · 2026

View source

Questions About This Research

What does the research say about domain-adapted vlms improve cfd visualization interpretation in mr by 30% for non-experts?
When presenting complex simulation data, especially in immersive environments like MR, consider using AI tools that have been specifically trained on the relevant domain to ensure accurate and understandable interpretation for all users. Evidence: Technologies (2026).
Why does "Domain-adapted VLMs improve CFD visualization interpretation in MR by 30% for non-experts" matter for design?
This research highlights how advanced AI, specifically VLMs, can bridge the gap between technical simulations (like CFD) and user comprehension. For IB DT, it demonstrates the potential of AI to make complex modelling outputs accessible, facilitating better design decisions and user-centred design by allowing a wider range of stakeholders to engage with and understand simulation results.
How can designers apply this research?
When presenting complex simulation data, especially in immersive environments like MR, consider using AI tools that have been specifically trained on the relevant domain to ensure accurate and understandable interpretation for all users.
What were the main findings?
Baseline general-purpose VLM achieved less than 30% accuracy in interpreting MR CFD visualizations.. Fine-tuning the VLM on a domain-specific dataset improved accuracy to over 60% across all interpretation categories.. Domain adaptation enabled VLMs to quantitatively interpret physical information from CFD visualizations, moving beyond simple color associations.
What research method was used?
Quantitative evaluation and domain adaptation of AI models..
How strong is the evidence?
Evidence strength is rated Strong effect, based on a 2026 journal from Technologies.
What should I do differently in my next project?
Incorporate AI-driven interpretation layers into MR visualizations of design simulations to provide real-time, understandable feedback to stakeholders.
What are the limitations?
The study focused on specific CFD parameters (temperature, airflow) and a particular VLM. Performance might vary with different simulation types, visualization methods, or other AI architectures. The dataset size and diversity could also impact generalizability.