ChE 436: Temperature Control Lab

Featured in CACHE News, December 2014.

MATLAB / Arduino Temperature Control Lab

This lab is an application of feedback control for a temperature control device. Complete instructions are available below in the temperature control lab description. The lab can be run on personal computers or on workstations in the UO Lab. Kits are available for temporary use and should remain in the UO Lab. The lab files are intended for MATLAB 2013B and earlier. For the latest versions, see the new Temperature Control Lab.

Students have an opportunity to use this lab throughout the semester to learn principles of system dynamics and control. In particular, this lab reinforces:

The difference between manual and automatic control
Step tests to generate dynamic data
Fitting dynamic data to build a simple First Order Plus Deadtime (FOPDT) model
Obtaining parameters for a PID control from standard tuning rules
Tuning the PID controller to improve performance
Stability analysis
Dynamic modeling with first principles
Dynamic data reconciliation
Model Predictive Control

The three important elements for a control loop are the measurement device (thermistor temperature sensor), an actuator (voltage to the transistor), and capability to perform computerized control (USB interface). At maximum output the transistor dissipates 3.1 W of power with a voltage of 8.9 V and current of 0.35 A. The heat generated by the transistor transfers by radiation, convection, and conduction to the temperature sensor.

Stability Analysis and PID Tuning in Simulink

Root Locus Stability Analysis

Solve and Linearize an Energy Balance Model with MATLAB

MIMO (Multiple Input / Multiple Output) Version

Generative AI Learning

Use these prompts around your lab sessions. Direct the AI - the step tests, fits, and tuning decisions are yours; the AI checks, quizzes, and debugs with you.

"Quiz me with 5 questions, one at a time, on the Arduino temperature control lab: how to run a step test and read Kp, tau-p, and theta-p graphically from the response, why the heater-to-sensor path shows dead time, how an FOPDT fit turns into PID tuning constants through a correlation (IMC or Ziegler-Nichols), what actuator saturation at 100% heater does to integral action, and why a controller tuned on the model can behave differently on the hardware. Grade my answers and list my misconceptions."

"Here are the FOPDT parameters I fit from my lab step test: Kp={value} degC/%, tau-p={value} s, theta-p={value} s, and the tuning I computed from them: {Kc, tau-I, tau-D and the correlation used}. Audit the chain: are the parameters physically plausible for a small resistively-heated thermistor system, does the tuning follow from the correlation, and what closed-loop behavior should I predict before I run it? Ask me to commit to a predicted overshoot and settling time, then tell me what to watch for on the hardware."

Debugging MATLAB/Arduino connection or code problems with AI is encouraged - paste the exact error message and your script. Make the AI explain the fix, and re-run the step test yourself: never let it invent data for a run that did not happen.

App: The browser-based TCLab Control Studio and TCLab Simulation Studio are digital twins of this lab's heater/thermistor hardware. Practice the whole workflow before your scheduled lab time: run a step test, fit Kp, tau-p, theta-p with sliders against a live objective function, and try your PID tuning - then confirm on the real device, where the fan of a neighboring laptop is a genuine disturbance.

What to Turn In

The deliverable is unchanged: a 2-page executive summary per group (see the grading rubric above). You may use Generative AI to help draft and format it, but you must guide it to the correct data, fits, justifications, and assumptions. A strong summary answers these questions with your own plots and numbers:

Show the step-test data with the FOPDT fit overlaid, and report Kp, tau-p, theta-p with units. How well should this model extrapolate to a different heater level, and why?
Report your PID tuning, the correlation it came from, and the closed-loop setpoint response (plot). Where did predicted and observed behavior differ, and what explains the gap?
Show one disturbance or saturation test and explain what the integral term did during it.
Where did AI assist (debugging, drafting, checking), and what did you verify before trusting it? Include one prediction you committed to before a run and whether the hardware agreed.

Dynamics and Control