Neutral point oscillations in a 4-leg inverter under unbalanced loads using PI c

[VadimDCL]

Hello everyone,

I’m working with a 4-leg inverter topology (artificial neutral) supplying unbalanced / asymmetric phase currents.
The objective is to stabilize the neutral point even when phase currents are intentionally asymmetric.

System overview

1.4-leg inverter with active neutral leg
2.Neutral controlled via feedback (neutral current / voltage)
3.Classical PI controller
4.Digital control implementation

Observed problem

Under unbalanced phase currents, the neutral point exhibits periodic oscillations at 150 Hz.
Despite extensive PI tuning:
1.the controller either reacts too slowly
2.or becomes overly aggressive
3.but the 150 Hz oscillation at the neutral point remains

Adjusting gains improves performance only marginally.
The behavior appears structurally limited, rather than being a simple tuning issue.

What was tried
1.different PI gains
2.integrator limiting
3.increased sampling rate

The oscillatory behavior persists under asymmetric load conditions.

Question

Is this a fundamental limitation of PI control when stabilizing an artificial neutral in a 4-leg inverter with unbalanced phase currents?

Have you encountered similar behavior, or used alternative control structures for this type of problem?

Any insights would be appreciated.

[jbb]

Sorry for quick format but I’m a bit squeezed for time…

It’s clearly possible to somehow control a 4-wire inverter. I’ve never done it myself but I’m sure methods exist.

You mention PI controllers, but these could be deployed in various ways for the system in question.

Have you got one per phase (+1 for neutral)? That can be troublesome, because the phases all interact with each other and the PI controller is trying to keep up with a 50 Hz sine wave.

Have you done a Clarke transform to convert the A, B, C phase currents into alpha, beta, nought components? This will decouple the phases from each other, which is a win.  You would still need to deploy 3 PI controllers, which would operate on alpha, beta and nought components and (again) keep up with the 50 Hz sone wave.

https://en.wikipedia.org/wiki/Alpha–beta_transformation

A possible next step is to then do a Park Transformation. This concerts the AC alpha, beta, nought signals into DC d, q, nought (same nought) signals. These are well suited to PI controllers.

https://en.wikipedia.org/wiki/Direct-quadrature-zero_transformation

Hope this provides some ideas

[VadimDCL]

Yes, from the perspective of controlling the three-phase inverter itself, I fully agree with you. However, the 4th leg (artificial neutral) has a separate control loop. In that loop there is no phase information as such — there is a single feedback signal (current or voltage).

[jonpaul]

See the many papers , app notes and seminars on SMPS/PFC control systems "sub harmonic oscillation"

 (current mode instability)

j

[uer166]

There's no fundamental issue here I believe.

Have you tried plotting the control loop internal current setpoint for the N leg? Perhaps it's railing due to high imbalance.

One simple addition here without going to a rotational reference frame is adding a feed forward component to the Neutral leg control. Simply add up the L1/L2/L3 currents (what neutral leg needs to source or sink), multiply by a constant, and give to your current controller inner loop or directly to the PWM duty cycle, depending on details. That'll make the PI loop work a lot less hard and be essentially the equivalent of a rotational frame but a lot simpler.

[Phoenix]

Hello everyone,
Under unbalanced phase currents, the neutral point exhibits periodic oscillations at 150 Hz.
Despite extensive PI tuning:
1.the controller either reacts too slowly
2.or becomes overly aggressive
3.but the 150 Hz oscillation at the neutral point remains

If the frequency of the reference or disturbance into a system is above DC/0Hz a PI controller can not eliminate all the error - some error will remain.

While I've never developed a 4 phase leg inverter some control techniques that I can think of that might help
- Proportional plus resonant controller (the resonant replaces the integral and drives the error to 0 at its frequency)
- Feedforward control (feedforward a 150Hz correction factor if this can be estimated somehow)
- Other predictive controllers

[uer166]

150Hz correction factor if this can be estimated somehow

Presumably there is knowledge of the currents in the other 3 legs. If so, it's known more or less ~perfectly which would simplify the problem massively.