Agentic AI represents a further evolution of AI-assisted engineering.
While a conventional AI model generally receives an input and produces a prediction, an AI agent can be designed to plan and execute a sequence of actions toward an engineering objective.
In future or user-developed agentic workflow, PyRocky could serve as the programmable execution layer between an AI agent and Ansys Rocky.
For example, an engineer could define an objective such as:
- Evaluate the transfer chute performance between 800 and 1,200 tons per hour and identify operating conditions that maintain stable discharge while minimizing maximum particle impact velocity.
An engineering AI agent could then be configured to:
- Interpret the requested engineering objective.
- Select the Ansys Rocky parameters that should be varied.
- Generate the required simulation cases.
- Use PyRocky to launch and control the simulations.
- Monitor simulation execution and identify failed cases.
- Extract mass flow, velocity, collision, and wear-related results.
- Compare the results against predefined engineering constraints.
- Select the next simulation points using an optimization strategy.
- Produce a technical summary of the evaluated configurations.
- Present recommendations for review by the responsible engineer.
In this architecture, the AI agent would coordinate the workflow, while PyRocky would provide controlled access to Ansys Rocky’s simulation capabilities.
This combination could reduce the time required to investigate large design spaces and enable more interactive engineering workflows based on natural-language objectives.
However, such systems must be developed with appropriate engineering controls.
AI-generated actions should be restricted by validated parameter ranges, predefined unit systems, approved simulation templates, and explicit acceptance criteria. All automatically generated models should preserve a complete record of input parameters, software versions, solver settings, warnings, and results.
Most importantly, engineering validation must remain a human responsibility. An AI agent may assist with case generation, execution, data analysis, and recommendation, but it cannot independently guarantee that:
- The selected contact model is physically appropriate.
- Material parameters have been properly calibrated.
- Particle shapes and size distributions represent the real material.
- Boundary conditions accurately reproduce the equipment's actual operation.
- The numerical time step and solver settings are adequate.
- The simulation has reached the required physical regime.
- The resulting design is safe or suitable for industrial implementation.
For this reason, the most reliable architecture is a human-in-the-loop workflow in which the AI agent assists the simulation engineer, while the engineer remains responsible for model validation, physical interpretation, and final design approval.
