Deep learning models for electric motor winding temperature estimation and control
A motor control system includes a motor including a plurality of windings, a first sensor configured to sense a first operating parameter of the motor, a second sensor configured to sense a second operating parameter of the motor, and memory hardware configured to store a machine learning model and computer-executable instructions. The machine learning model is trained to generate a winding temperature estimation output based on motor operating parameter inputs. The motor control system includes processor hardware configured to execute the instructions and use the machine learning model to cause the motor control system to generate a winding temperature estimation output using the machine learning model based on the first operating parameter and the second operating parameter, the temperature estimation output indicative of a predicted temperature of the plurality of windings, and control the motor based on the winding temperature estimation output.
1. A motor control system comprising:
a motor including a plurality of windings;
a first sensor configured to sense a first operating parameter of the motor;
a second sensor configured to sense a second operating parameter of the motor;
memory hardware configured to store a machine learning model and computer-executable instructions, the machine learning model trained to generate a winding temperature estimation output based on motor operating parameter inputs; and
processor hardware configured to execute the instructions and use the machine learning model to cause the motor control system to,
generate the winding temperature estimation output using the machine learning model based on the motor operating parameter inputs, the using the machine learning model including inputting an input feature vector to the machine learning model, the input feature vector derived from a combination of current parameter values of the first operating parameter and the second operating parameter and summarized historical parameter values, of the first operating parameter and the second operating parameter, over a specified time period, and the temperature estimation output indicative of a predicted temperature of the plurality of windings; and
control the motor based on the winding temperature estimation output.
2. The motor control system of claim 1 , wherein the processor hardware is further configured to execute the instructions and use the machine learning model to cause the motor control system to
generate the winding temperature estimation output based on a plurality of motor operating parameters, the plurality of motor operating parameters including at least two of an oil inlet temperature, a rotor oil flow, a stator oil flow, a speed of the motor, a torque of the motor, a direct current (DC) bus voltage of the motor, a quadrature (q)-axis voltage of the motor, a direct (d)-axis voltage of the motor, a q-axis current of the motor or a d-axis current of the motor, the first operating parameter and the second operating parameter being part of the plurality of operating parameters.
3. The motor control system of claim 2 , wherein the plurality of motor operating parameters does not include a sensed temperature of the motor or a sensed temperature of the plurality of windings.
4. The motor control system of claim 1 , further comprising a temperature sensor configured to sense a temperature of the plurality of windings, wherein the instructions further include:
using the sensed winding temperature from the temperature sensor to control operation of the motor when the temperature sensor has a normal operating condition; and
using the winding temperature estimation output to control operation of the motor in response to a failure of the temperature sensor.
5. The motor control system of claim 1 , wherein the processor hardware is configured to execute the instructions to cause the motor control system to
compare the winding temperature estimation output of the machine learning model to a winding temperature threshold; and
transmit an alarm signal in response to the winding temperature estimation output exceeding the winding temperature threshold.
6. The motor control system of claim 1 , wherein the first sensor includes at least one of an oil temperature sensor, an oil flow sensor, a current sensor or a voltage sensor.
7. The motor control system of claim 1 , wherein the machine learning model includes at least one of a feed forward network (FFN), a temporal convolution neural network (TCN), a convolution neural network, a long short-term memory (LSTM) network, an XGBoost model, a transformer model, a linear regression model, a decision tree model, a random forest model, a gradient boosting machine (GBM), a recurrent neural network (RNN) or a multilayer perceptron (MLP) model.
8. The motor control system of claim 7 , wherein the machine learning model is the multilayer perceptron (MLP) model.
9. The motor control system of claim 1 , the processor hardware is configured to execute the instructions to cause the motor control system to
determine a rolling mean of the first operating parameter over the specified time period; and
supply the determined rolling mean of the first operating parameter as an input to the machine learning model to generate the winding temperature estimation output.
10. The motor control system of claim 1 , wherein the processor hardware is configured to execute the instructions to cause the motor control system to generate at least one of:
a shaft torque estimation output indicative of a predicted torque of at least one of a shaft or rotor of the motor; and
an insulated-gate bipolar transistor (IGBT) temperature estimation output indicative of a predicted temperature of an IGBT of the motor.
11. A computer system comprising:
memory hardware configured to store a machine learning model, motor operating parameter vector inputs, and computer-executable instructions, each motor operating parameter vector input including at least one operating parameter of a motor having a plurality of windings; and
processor hardware configured to execute the instructions, the instructions including:
training the machine learning model with the motor operating parameter vector inputs to generate a winding temperature estimation output, the winding temperature estimation output indicative of a predicted temperature of the plurality of windings based on a plurality of operating parameters of the motor;
obtaining a first operating parameter of the motor via a first sensor;
obtaining a second operating parameter of the motor via a second sensor;
supplying an input feature vector, based on the first operating parameter and the second operating parameter, as an input to the machine learning model to generate the winding temperature estimation output indicative of the predicted temperature of the plurality of windings, the input feature vector derived from a combination of current parameter values of the first operating parameter and the second operating parameter and summarized historical parameter values of the first operating parameter and the second operating parameter; and
executing at least one control instruction based on the winding temperature estimation output to control operation of the motor.
12. The computer system of claim 11 , wherein the instructions further include generating synthetic data to define the motor operating parameter vector inputs for training the machine learning model, wherein generating the synthetic data includes:
generating timestamp data values for multiple motor operating parameters at multiple points of time within a specified time period;
scaling the timestamp data values to normalized input vector values; and
for each of the multiple motor operating parameters, determining at least one of a mean operating parameter, a standard deviation and a rolling average of the normalized input vector values corresponding to the motor operating parameter, wherein
the at least one of the mean, the standard deviation and the rolling average is determined within a specified time window size, and
a same specified time window size is applied to each of the multiple motor operating parameters.
13. The computer system of claim 11 , wherein the machine learning model includes at least one of a feed forward network (FFN), a temporal convolution neural network (TCN), a convolution neural network, a long short-term memory (LSTM) network, an XGBoost model, a transformer model, a linear regression model, a decision tree model, a random forest model, a gradient boosting machine (GBM), a recurrent neural network (RNN) or a multilayer perceptron (MLP) model.
14. The computer system of claim 13 , wherein the machine learning model is the multilayer perceptron (MLP) model.
15. The computer system of claim 11 , wherein:
supplying inputs to the machine learning model includes supplying a plurality of motor operating parameters to the machine learning model including the first operating parameter and the second operating parameter; and
the plurality of operating parameters includes at least one of an oil inlet temperature, a rotor oil flow, a stator oil flow, a speed of the motor, a torque of the motor, a direct current (DC) bus voltage of the motor, a q-axis voltage of the motor, a d-axis voltage of the motor, a q-axis current of the motor or a d-axis current of the motor.
16. The computer system of claim 15 , wherein the plurality of motor operating parameters does not include a sensed temperature of the motor or a sensed temperature of the plurality of windings.
17. The computer system of claim 11 , wherein:
the instructions further include determining a rolling mean of the first operating parameter over a specified time period; and
supplying the inputs to the machine learning model includes supplying the determined rolling mean of the first operating parameter as an input to the machine learning model to generate the winding temperature estimation output.
18. The computer system of claim 11 , wherein executing the control instruction includes:
comparing the winding temperature estimation output of the machine learning model to a winding temperature threshold; and
transmitting an alarm signal in response to the winding temperature estimation output exceeding the winding temperature threshold.
19. The computer system of claim 11 , wherein the supplying the first operating parameter and the second operating parameter as inputs to the machine learning model generates at least one of:
a shaft torque estimation output indicative of a predicted torque of at least one of a shaft or rotor of the motor; and
an insulated-gate bipolar transistor (IGBT) temperature estimation output indicative of a predicted temperature of an IGBT of the motor.
20. A computerized method comprising:
obtaining motor operating parameter vector inputs, each motor operating parameter vector input including at least one operating parameter of a motor having a stator, a rotor, and a plurality of windings, the rotor configured to rotate relative to the stator, and the plurality of windings wound about a portion of at least one of the stator and the rotor;
training a machine learning model with the motor operating parameter vector inputs to generate a winding temperature estimation output, the winding temperature estimation output indicative of a predicted temperature of the plurality of windings based on a plurality of operating parameters of the motor;
obtaining a first operating parameter of the motor via a first sensor;
obtaining a second operating parameter of the motor via a second sensor;
supplying an input feature vector to the machine learning model to generate the winding temperature estimation output indicative of the predicted temperature of the plurality of windings, the input feature vector derived from a combination of current parameter values of the first operating parameter and the second operating parameter and summarized historical parameter values of the first operating parameter and the second operating parameter; and
executing at least one control instruction based on the winding temperature estimation output to control operation of the motor.