I am working on a sensorless control scheme of PMSM based on Adaptive Luenberger (state) Obsever. It is relatively easy to set the observer gain to reach stable and successful observer tracking in simulation. When it comes to real-time implementation (I use a DSP hardware with program cycle time of 150 micro sec. which is more than sufficient for a high quality sensor-based control which is also easily achievable in my case).
I could only obtain an intermittent or unstable tracking in real time when I use the same gain I used for simulation (although off-line simulation is made very close to reflect real time dynamics by making current and speed controllers discrete ones at the same sampling time used in real time where speed control loop is executing once each 30 executions of current control loop)
I know that there are still many differences between off-line simulation and real time system (specially those attributed to noise and parameter uncertainty) that can result in different response but I wonder if somebody can help advising a systematic tuning approach like for example those procedures we usually use for tuning PID controllers.
Question: In your simulation, was your 150us incorporated in the design? Or is your simulation continuous?
Well Elmer, I tried to make my simulation as close as possible to the real time experiment, it is a discrete simulation where multitasking is enabled (i.e. different block execution rates) meaning that the basic simulation time is small enough to capture the actual motor dynamics with acceptable accuracy ( I've chosen it to be 1.5 micro sec.), the PWM period is 150 micro second (SVPWM Freq is 6.6667 KHz). The current control is done in the synchronously rotating dq reference frame once every 150 micro sec. exactly the same as the real time. The speed control loop in simulation is also executed once every 30 executions of the current control loop; again it is the same as its counterpart in the real time application. Low pass speed filtering for feedback loop is enabled in both simulation and real time with same cutoff characteristics.
Here you are Elmer, the script is to be run before you run the model since all parameters are just substituted as symbols in SIMULINK model and numerated in the script
hello sir,
can you post some of the mathematical dynamic equation that u implemented in this model especially for dq current estimation and svpwm?
- Badges
- Science topic
- Similar topics
- Filing