I want to prepare data sets to train neural network to estimate 3 phase induction motor speed. I am using Denfoss make FC-302 AC drive to run 3 HP IM at different speed. I can export rms value of AC drive terminal voltage and phase current ,power,torque. etc to excel sheet.
From excel data (RMS voltage and current) i can implement abc-Alpha beta transformation in simulink. This alpha beta componentes of voltage and current can be used to obtain rotor flux . But i am feeling i am missing the phase relation when recording RMS voltages and currents.This may result in incorrect rotor flux value.
And also i am unable find Induction motor resistance ,mutual inductance and leakage inductance value in technical specification provided by the manufacturers.
Savitha Pr Follow these procedures to obtain data sets for training a neural network for sensorless speed estimate of an induction motor:
1. Create a test rig: Connect the induction motor and the AC drive as directed by the manufacturer.
2. Run the motor at various speeds: Record the RMS values of the AC drive terminal voltage and phase current, power, and torque while you run the motor at different speeds.
3. Carry out the ABC-Alpha Beta transformation: To conduct the ABC-Alpha Beta transformation in Simulink, use the exported RMS values of the AC drive terminal voltage and phase current.
4. Determine the rotor flux: Calculate the rotor flux using the alpha-beta components of voltage and current.
5. Keep track of the motor speed: Measure and record the motor speed on each run.
6. Repeat the above steps many times: Repeat the preceding procedures to expand the size of your data collection.
7. Preparing the data: Remove any noise, outliers, and discrepancies from the data.
8. Divide the data into two sets: training and validation: Divide the data into training and validation sets for use in training and assessing the neural network's performance.
If you are unable to determine the induction motor resistance, mutual inductance, and leakage inductance values, you may need to utilize additional approaches such as finite element analysis or physical testing.
Just two, three small comments:
if you record the RMS values, then these are (very likely) averaged steady-state values and not instantaneous values w.r.t. to single sampling points in the time domain. Of course, you can use these values trying to build a data-driven speed estimator for the IM drive, but as the base data is covering steady state only, the resulting estimator will very likely be not feasible during transient drive operation.
Parameter identification of unknown IM is also a classical drive control problem and a multitude of identification procedures on different complexity levels exist, cf. e.g. https://www.mdpi.com/1996-1073/11/9/2194
A small google scholar search will provide you many further papers both on IM parameter identification and machine learning-based sensorless control.
- Badges
- Science topic
- Similar topics
- Engineering
- Control Theory
- Control Systems
- Control Systems Engineering