What Dyad 3.0 Adds to Physics-Based Simulation Software
Dyad 3.0 is an AI-based physics simulation platform that combines autonomous agents, scientific machine learning and digital twin technology to build, validate and deploy models of complex physical systems from natural-language requirements and engineering data. JuliaHub positions Dyad as a bridge between general-purpose AI and verification-grade engineering tools, targeting teams who build aircraft, EVs, semiconductors, utilities, HVAC systems and medical devices. Instead of manually assembling every model element, engineers can feed requirements, prior designs and test data into Dyad’s agents, which propose model structures, run physics-based simulations and refine designs while respecting safety and operational limits. The aim is to shorten design cycles without sacrificing the validation discipline that physical products demand. According to JuliaHub, Dyad is already in production use with Fortune 100 customers, underlining that its agentic workflows are more than a laboratory experiment.
Agentic Model Generation and the Rise of ‘Agentic Simulation’
The core change in Dyad 3.0 is agentic model generation, where autonomous software agents interpret engineering intent and carry out much of the simulation workflow. Engineers can submit a requirements document plus historical test data and a plain-language request, and Dyad’s agents assemble candidate models, explore design variations, apply physical and safety constraints, and produce validated control code ready for hardware deployment. The engineer still sets direction and signs off on trade-offs, but repetitive tasks such as model construction, controller tuning, simulation execution and toolchain integration become automated. JuliaHub describes this as a new category of “agentic simulation”, arguing that conventional physics-based simulation software was not built around autonomous, natural-language workflows, while pure AI agents lack the physical validation substrate. By pairing agents with scientific machine learning and high-fidelity solvers, Dyad 3.0 aims to let teams explore larger, multi-physics design spaces with the same headcount.
FMU Interoperability and the Broader Engineering Toolchain
For many engineering organizations, a new platform must coexist with decades of models and processes, which is where Dyad 3.0’s Functional Mock-up Unit (FMU) interoperability matters. The release includes major FMU advances so Dyad models can integrate with existing tools that support the FMI standard, turning the platform into a peer rather than a replacement in established workflows. This helps teams wrap Dyad’s agent-driven models into larger system simulations or reuse validated components from legacy tools inside Dyad. FMU interoperability strengthens Dyad’s position as physics-based simulation software that can sit in the middle of a multi-vendor stack, enabling co-simulation, controller-in-the-loop tests and shared digital twin assets. For engineering leaders, this reduces migration risk and supports gradual adoption: teams can keep current toolchains while adding AI-driven exploration and automated model generation around them, instead of rebuilding every model from scratch.
Digital Twin Technology for Predictive Maintenance
Dyad 3.0 extends beyond design-phase analysis into digital twin technology and predictive maintenance tools. JuliaHub has expanded the platform’s workflows to help teams design and optimize industrial predictive maintenance applications, using simulation models tied to operational data. In this setup, a Dyad model becomes a live counterpart to an asset in the field, running physics-based simulation against streaming telemetry to estimate hidden states, predict failures and test maintenance strategies virtually before they are applied. Because physical and safety constraints are enforced throughout, engineers can encode operating envelopes, regulatory limits and fault scenarios directly in the twin. The same agentic capabilities that speed model creation can then be used to explore what-if maintenance schedules or control changes. Enterprise deployment improvements around installation, configuration, security, compliance and lifecycle management are meant to make these digital twins practical in regulated, distributed engineering organizations.
HVAC Workflows as a Template for Industry-Focused Simulations
JuliaHub highlights HVAC as a flagship example of how agentic simulation can target specific industries. Dyad 3.0 introduces agent-driven HVAC system design with fast modeling tools, accurate refrigerant splines, an expanded component library and templates for common system architectures. A concrete example is data center cooling circuits, where engineers must size chillers, study performance under typical load profiles and tune control systems. In Dyad, an agent can construct the cooling circuit model, generate and run varied load profiles, and evaluate controller performance under each scenario. This creates a repeatable pattern for other sectors: mix domain-specific libraries with agentic model generation, validate behavior through physics-based simulation, and then export control code or FMUs for deployment. As Dyad’s multibody dynamics preview develops toward robotics and vehicle dynamics through 2026, similar targeted workflows could emerge across more motion-intensive applications.