Systems and methods for machine learning based flexural wave absorber
A flexural wave absorption system detects, with a sensor attached to a beam, an incident wave propagating in the beam. The system determines, based on a signal from the sensor generated in response to the incident wave, an amplitude and phase of the incident wave propagating in the beam and controls an actuator connected to the beam to generate a suppression wave, based on the amplitude and the phase of the incident wave, that reduces a coefficient of reflection of the incident wave across a broadband frequency range.
1. A flexural wave absorption system, comprising:
a sensor configured to be connected to a beam;
an actuator configured to be connected to the beam;
a processor; and
a memory communicably coupled to the processor and storing:
a wave module including instructions that when executed by the processor cause the processor to:
determine an amplitude and a phase of an incident wave propagating in the beam based on a signal from the sensor; and
control the actuator to generate a suppression wave, based on the amplitude and the phase of the incident wave; that reduces a coefficient of reflection of the incident wave resulting in a reflected wave; and
a neural network module including instructions that when executed by the processor cause the processor to:
determine that a sampling rate of the sensor is less than a threshold;
cause in response to a determination that the sampling rate is less than the threshold, a time delay to be imposed on a feedback signal input into a neural network to predict a voltage Va to apply to the actuator to generate a cancellation wave, wherein the feedback signal is received from the sensor; and
control the actuator to generate the cancellation wave based on the time-delayed feedback signal that indicates multiple wave components present in the beam, the multiple wave components including the incident wave, the suppression wave, and the reflected wave.
2. The flexural wave absorption system of claim 1 , wherein the cancellation wave is configured to cancel the multiple wave components indicated by the feedback signal.
3. The flexural wave absorption system of claim 1 , wherein:
the neural network module further includes instructions to train the neural network to determine a transfer function between the voltage V applied to the actuator and a voltage is detected by the sensor, and the neural network is configured to predict the voltage V to generate the cancellation wave based on the transfer function.
4. The flexural wave absorption system of claim 3 , wherein the instructions to train the neural network include instructions to obtain time-series data to train the neural network by applying sine-modulated signals to multiple filters that approximate frequency responses across a range of multiple frequencies.
5. The flexural wave absorption system of claim 4 , wherein the neural network is implemented as a recurrent neural network, the multiple filters are implemented as infinite impulse response filters, and both the sensor and the actuator are implemented as piezoelectric patches.
6. A method of absorbing a flexural wave, comprising:
detecting, with a sensor attached to a beam, an incident wave propagating in the beam;
determining, based on a signal from the sensor generated in response to the incident wave, an amplitude and a phase of the incident wave;
controlling an actuator, connected to the beam, to generate a suppression wave, based on the amplitude and the phase of the incident wave, that reduces a coefficient of reflection of the incident wave resulting in a reflected wave;
determining that a sampling rate of the sensor is less than a threshold:
causing, in response to a determination that the sampling rate is less than the threshold, a time delay to be imposed on a feedback signal input into a neural network to predict a voltage Va to apply to the actuator to generate a cancellation wave, wherein the feedback signal is received from the sensor; and
controlling the actuator to generate the cancellation wave based on the time-delayed feedback signal that indicates multiple wave components present in the beam, the multiple wave components including the incident wave, the suppression wave, and the reflected wave.
7. The method of claim 6 , wherein the cancellation wave cancels the multiple wave components indicated by the feedback signal.
8. The method of claim 6 , further comprising:
training the neural network to determine a transfer function between the voltage applied to the actuator and a voltage V s detected by the sensor; and
predicting, via the neural network, the voltage Va to generate the cancellation wave based on the transfer function.
9. The method of claim 8 , wherein the training the neural network comprises obtaining time-series data to train the neural network by applying sine-modulated signals to multiple filters that approximate frequency responses across a range of multiple frequencies.
10. The method of claim 9 , wherein the neural network is implemented as a recurrent neural network, the multiple filters are implemented as infinite impulse response filters, and both the sensor and the actuator are implemented as piezoelectric patches.
11. A non-transitory computer-readable medium for absorbing a flexural wave, including instructions that, when executed by one or more processors, cause the one or more processors to:
receive, from a sensor attached to a beam, a signal generated in response to an incident wave propagating in the beam;
determine, based on the signal, an amplitude and a phase of the incident wave;
control an actuator, connected to the beam, to generate a suppression wave, based on the amplitude and the phase of the incident wave, that reduces a coefficient of reflection of the incident wave resulting in a reflected wave;
determine that a sampling rate of the sensor is less than a threshold;
cause, in response to a determination that the sampling rate is less than the threshold, a time delay to be imposed on a feedback signal input into a neural network to predict a voltage Va to apply to the actuator to generate a cancellation wave, wherein the feedback signal is received from the sensor; and
control the actuator to generate the cancellation wave based on the time-delayed feedback signal that indicates multiple wave components present in the beam, the multiple wave components including the incident wave, the suppression wave, and the reflected wave.
12. The non-transitory computer-readable medium of claim 11 , wherein the cancellation wave is configured to cancel the multiple wave components indicated by, the feedback signal.
13. The non-transitory computer-readable medium of claim 11 , further comprising instructions to:
train the neural network to determine a transfer function between the voltage V a applied to the actuator and a voltage V s detected by the sensor; and
predict, via the neural network, the voltage V a to generate the cancellation wave based on the transfer function.
14. The non-transitory computer-readable medium of claim 13 , wherein the instructions to train the neural network include instructions to obtain time-series data to train the neural network by applying sine-modulated signals to multiple filters that approximate frequency responses across a range of multiple frequencies.
15. A system, comprising:
a processor; and
a memory storing a neural network and a module, wherein the module includes instructions that cause the processor to:
determine that a sampling rate, of a sensor configured to detect an incident wave propagating in a beam, is less than a threshold; and
cause, in response to a determination that the sampling rate is less than the threshold, a time delay to be imposed on a feedback signal, from the sensor, input into the neural network configured to predict a voltage Va to apply to an actuator configured to generate a cancellation wave.
16. The system of claim 15 , wherein the module further includes instructions to:
determine an amplitude and a phase of the incident wave propagating in the beam based on a signal from the sensor;
control the actuator to generate a suppression wave, based on the amplitude and the phase of the incident wave, that reduces a coefficient of reflection of the incident wave, but fails to completely absorb the incident wave resulting in a reflected wave; and
control the actuator to generate the cancellation wave based on feedback signal that indicates multiple wave components present in the beam, the multiple wave components including the incident wave, the suppression wave, and the reflected wave.
17. The system of claim 16 , wherein the cancellation wave is configured to cancel the multiple wave components indicated by the feedback signal.
18. The system of claim 15 , wherein:
the module further includes instructions to train the neural network to determine a transfer function between the voltage V a applied to the actuator and a voltage V s detected by the sensor, and
the neural network is configured to predict the voltage Via to generate the cancellation wave based on the transfer function.
19. The system of claim 18 , wherein the instructions to train the neural network include instructions to obtain time-series data to train the neural network by applying sine-modulated signals to multiple filters that approximate frequency responses across a range of multiple frequencies.
20. The system of claim 19 , wherein the neural network is implemented as a recurrent neural network, the multiple filters are implemented as infinite impulse response filters, and both the sensor and the actuator are implemented as piezoelectric patches.