HVAC
Design smarter HVAC systems.
Simulate faster.
Optimize for performance.
HVAC systems are becoming more complex, while engineering teams are still relying on fragmented tools for modeling, simulation, optimization, and control. Dyad from JuilaHub brings these workflows together in a unified environment built for modern, AI-driven engineering.
With Dyad HVAC, engineers can design and simulate thermal-fluid systems, explore performance, calibrate models against real-world data, optimize designs, and develop intelligent control strategies—all from a single model and workflow.
From vapor-compression cycles to complete thermal management systems, JuliaHub helps engineering teams move faster from concept to simulation to validated digital twin.
HVAC challenges
Where HVAC Control Software and Hardware Design Gets Hard
Continuous variable-speed control, multi-refrigerant support, growing connectivity, and UL 60730 safety classification are stacking real software-engineering complexity onto HVAC teams. Dyad addresses all four from one calibrated model instead of a chain of disconnected tools.
Continuous Control
One calibrated model validates PID, MPC, and gain-scheduled control across the full modulation range.
Expanding Connectivity
One source of truth keeps firmware, control logic, and physics model from drifting apart.
Equipment Variation
Extensive built-in refrigerant models — swap properties and hardware models without redesigning the entire system.
UL 60730 Traceability
Compiles straight to embedded code — physics, control logic, and code as one traceable artifact.
Agentic Engineering
Agentic engineering tools revolutionize how software-driven machines are developed. Instead of an engineer hand-building and hand-validating every model, an AI agent can research the physics, assemble and calibrate the model, and generate embedded code alongside the engineer — turning weeks of manual modeling work into hours of directed iteration. Read more about it in our Agentic Engineering white paper.
Download
proven applications
Accelerate Every Stage of HVAC Development
See how Dyad delivers next-generation model-based design and simulation capabilities offers a game-changing alternative to Simulink and Modelica tools by fusing native agentic AI with scientific machine learning (SciML).
Build High-Fidelity HVAC Models
Create detailed thermal-fluid models using a library of production-ready HVAC components, including: heat-exchangers, compressors, valves, fans, conditioned spaces, pipes, controller blocks.
Accelerate with Agentic Engineering
Dyad's agentic capabilities can create and modify models, generate tests, run simulations, analyze results, and validate changes - while keeping engineers informed and incontrol.
Extensive Refrigerant Models
Design and optimize thermodynamic systems up to 1000x faster with best in class spline-based refrigerant property models rather than standard equation of state approaches.
Simulate from One Model
Run steady-state and transient simulations from the same model definition, eliminating the need to rebuild systems for different types of analysis.
How we compare
Built for the AI Era of Model-Based Design
Legacy tools like Simulink1 and those that are Modelica1-based have earned three decades of engineering trust — but neither was architected for a world where an AI agent researches physics, assembles models, and iterates alongside the engineer. Dyad was built for that world from day one. Here is how Dyad compares against the leading legacy tools2.
Capability
Dyad
Legacy Tools
Modeling paradigm
Unifies acausal, causal & state-machine in one language
Mixed. Usually acausal or causal
AI-native model development
Dyad Agent builds, tests & iterates models from plain language
Not a native capability
Model calibration
Automatic-differentiation calibration — exact gradients, not guesswork
Usually manual trial-and-error
Missing-physics discovery
SciML/UDE recovers unmodeled physics as a readable equation
Not supported
Thermofluid simulation speed
Solutions leverage Juila-based solvers that are optimized for high performance
General-purpose solver
Control design
Native PID autotuning, linear & nonlinear MPC
Separate toolbox
Embedded code & safety traceability
Single model compiles to embedded C with UL 60730 Class B/C traceability
Manual traceability
Model optimization & calibration
Purpose-built for optimizing & calibrating against noisy, sparse, or chaotic data
Add-on license
End-to-end workflow
Model, calibrate, control & code-gen in one environment
Separate tools per stage
Model deployment
Compiles directly to native binaries, shared libraries, or embedded targets
Requires exporting to FMU
case studies
Engineering Breakthroughs in HVAC
Discover how engineering teams are delivering unmatched innovation and accelerating time-to-market
common questions
What HVAC Engineers Ask Before Switching
Adopting a new modeling tool means real questions about your existing models, your safety-certification path, and what "AI-augmented" actually means day to day. Here are straight answers to the ones we hear most.
Do we have to throw away our existing Simulink and Modelica models to adopt Dyad?
No. Dyad delivers a pre-built library of the most widely used HVAC components. For less common or proprietary components, Dyad supports importing Functional Mock-up Units (FMUs), the open standard both Simulink and Modelica can export to.
Is Dyad mature enough for safety-relevant control software, like UL 60730 Class B/C logic?
Dyad compiles the calibrated model directly to embedded C, so the physics, control logic, and generated code exist as one traceable artifact — giving you a much shorter path to the requirements-to-code traceability a Class B/C safety case requires. That said, classification and certification itself remains your team's and your certification body's responsibility; Dyad supports the case, it doesn't grant the certification.
Does Dyad handle our specific refrigerants, including the shift to A2L blends like R-32 and R-454B?
Dyad's HVAC library ships with built-in thermodynamic property models for 20+ refrigerants — R1234YF, R1234ZEE, R125A, R134A, R143A, R290, R32, R407C, R410A, R448A, R449A, R452A, R452B, R454B, R454C, R455A, R457A, R459B, R507A, R717 — plus moist and dry air. Dyad can also support proprietary refrigerants by training splines to accurately represent those materials.
How is the Dyad Agent different from just asking an AI coding assistant to write Simulink or Modelica code?
The Dyad Agent doesn't generate disconnected code from a prompt — it operates on the same underlying model as the graphical editor and the textual language, so it researches the relevant physics, assembles matching components from a physically-typed library, compiles the result, and validates it, with the engineer reviewing and directing each step. It's iterating on an actual engineering model, not producing a guess at syntax.
What does it cost to try Dyad?
Dyad Studio is a free, so evaluating it against a real model doesn't require a procurement cycle. A JuliaHub account unlocks the cloud-native compute behind it, including multi-GPU training for the AI components, when you're ready to scale beyond a local evaluation.
LEGAL DISCLAIMERS & NOTICES
1 Trademark Notice: MATLAB, Simulink, and Stateflow are registered trademarks of The MathWorks, Inc. All other product names, trademarks, and registered trademarks mentioned on this website—including those related to Modelica-based tools, libraries, or environments (such as Dymola, OpenModelica, or Wolfram System Modeler)—are the property of their respective owners. Reference to these third-party trademarks is for nominative, descriptive, and informational purposes only to identify the compatibility, interoperability, or alternative nature of our products. Use of these names does not imply any affiliation, sponsorship, endorsement, or approval by The MathWorks, Inc. or any Modelica Association member or tool vendor.
2 Data Accuracy & Comparison Notice: The product comparison data presented on this page is based on publicly available information, official documentation, and feature specifications as of September 2026. While we make every reasonable effort to ensure this information is accurate and up-to-date, product features, licensing terms, and technical capabilities change frequently. Product performance metrics may vary based on specific simulation configurations, hardware environments, and user implementations. This information is provided "as-is" for general comparative purposes only. We recommend that users verify current specifications directly with the respective vendors before making a purchasing decision.