Tuning a PID controller is more of an art than a science, especially when you're doing it by hand. The goal is to find the right balance between a fast response and a stable system. We'll walk through the most common manual method: the trial-and-error approach. It’s an iterative process that helps you develop an intuition for how each gain affects the system's behavior.

Start by setting your integral ($K_i$) and derivative ($K_d$) gains to zero. This turns your controller into a simple P-only controller. Now, begin slowly increasing the proportional gain ($K_p$) and observing the system's response to a setpoint change. With a low $K_p$, the system will be sluggish and may not reach the setpoint. As you increase $K_p$, the response will get faster, but it will also start to overshoot the target.

Your goal is to find the ultimate gain, known as $K_u$. This is the value of $K_p$ where the system's output starts to oscillate with a constant amplitude, never settling down. This is the edge of instability. Once you find this point, note down the value of $K_u$ and also measure the time period of one full oscillation, called the ultimate period ($T_u$). These two values are your baseline.

A common practice is to set the final $K_p$ to about half of the ultimate gain ($K_p = 0.5 * K_u$). This gives you a good margin of safety away from instability. At this point, you'll likely notice the system settles below the setpoint. This persistent difference is the steady-state error.

With $K_p$ set, the next step is to manage the overshoot. This is the job of the derivative term. Keep $K_i$ at zero and begin to slowly increase the $K_d$ gain. The D term acts like a brake, resisting rapid changes in the system's output. As you increase $K_d$, you'll see the overshoot from the P term decrease, and the oscillations will be dampened more quickly.

Adjust $K_d$ until you have a response with minimal overshoot that you're comfortable with. The system should now settle quickly, but it will still have that steady-state error.

Finally, it's time to eliminate the steady-state error using the integral term. Begin adding a very small amount of $K_i$. The I term looks at the accumulated error over time. As long as there is an error, the integral term will continue to grow, pushing the system's output further until the error reaches zero.

Increase $K_i$ slowly. Too much integral gain can reintroduce overshoot and oscillations, undoing the work of your D term. It can also lead to a problem called integrator windup. Find a small value for $K_i$ that eliminates the steady-state error in a reasonable amount of time without compromising stability. You may need to go back and slightly readjust $K_p$ and $K_d$ to get the perfect response.

This iterative process gives you direct, hands-on experience with the trade-offs involved in PID tuning. You learn to balance a quick response (high $K_p$) with stability (appropriate $K_d$) and precision (minimal $K_i$).

Ready to test your knowledge on manual tuning?

While manual tuning is effective, there are also more structured methods that use the $K_u$ and $T_u$ values you found to calculate starting gains. But mastering this hands-on approach builds a strong foundation for any tuning task.