PID Tuning Assistant

• Fits the process model for you. Load a step test as a CSV file. The app fits a first-order-plus-dead-time model, grades the fit, and warns you when the data is too poor to tune from.
• Compares the tuning rules side by side. Ziegler-Nichols, Cohen-Coon and IMC/Lambda run on the same simulated setpoint step, with overshoot, settling time, IAE, ISE and controller travel for each.
• Lets you tune by hand, live. Drag Kp, Ti and Td and the response updates as you move. Add measurement noise or output limits to see how the loop copes.
• Checks the AI before you trust it. Ask for a change, such as less overshoot, and the AI must answer with specific gains. The app simulates them and shows current against suggested before you can apply them. Unstable suggestions are flagged.
• Keeps your data on your computer. There is no CSI server. The AI receives the fitted model, gains and simulated results, never your raw data, and you use your own Anthropic API key.
• Documents the result. Save a report of the model, the gains and the metrics, so the next engineer knows why the loop is tuned the way it is.

Version 0.1 tunes one loop at a time, from one open-loop step test, against a first-order model. Results come from a simulation of a fitted model: validate any new gains cautiously on the real process.