Control mapping for hydraulic machines

1. A machine control system, comprising:

memory storing a set of model weights trained using a machine learning system based on a training dataset, wherein:

the training dataset correlates preset displacements for a hydraulic valve of a machine to movement parameters of a component of the machine measured when the hydraulic valve is set according to the preset displacements and the component is moved at least in part by the hydraulic valve;

an input data receiver configured to receive:

a velocity command indicating a requested velocity of the component of the machine; and

machine state data from one or more sensors positioned on the machine; and

a control mapping model configured to use the set of model weights to map a combination of the velocity command and the machine state data into a predicted displacement of the hydraulic valve that causes movement of the component at the requested velocity in response to the velocity command; and

the control mapping model causes the hydraulic valve to move to the predicted displacement.

2. The machine control system of claim 1 , wherein the velocity command is an operator command input by an operator of the machine via a control element of the machine.

3. The machine control system of claim 1 , wherein the movement parameters include at least one of positional coordinates, velocity, or acceleration of the component as measured during testing when the hydraulic valve is set according to the preset displacements and the component is moved at least in part by the hydraulic valve.

4. The machine control system of claim 1 , wherein the training dataset correlates preset displacements for the hydraulic valve along with other preset parameter values to the movement parameters.

5. The machine control system of claim 1 , wherein the machine learning system comprises a radial basis function network with a fully-connected neural network layer.

6. The machine control system of claim 5 , wherein the radial basis function network is configured to train the set of model weights at least partially using backpropagation to adjust the set of model weights over a plurality of iterations.

7. The machine control system of claim 1 , wherein the control mapping model causes an electric current to be provided to the hydraulic valve that causes the hydraulic valve to move to the predicted displacement.

8. The machine control system of claim 7 , wherein the hydraulic valve comprises a valve solenoid, and the electric current is a valve solenoid current.

9. The machine control system of claim 1 , wherein the memory stores a plurality of sets of model weights corresponding to different types of events, and wherein the control mapping model selects the set of model weights from the plurality of sets of model weights based on an event type associated with the velocity command.

10. The machine control system of claim 1 , wherein the machine state data includes one or more of: pump pressure, hydraulic cylinder pressure, and positional data associated with the component or other components of the machine.

11. A machine, comprising:

a substantially rigid frame;

a hydraulic valve configured to open to different displacements;

a component connected to the substantially rigid frame, wherein the hydraulic valve is configured to move the component relative to the substantially rigid frame;

one or more sensors carried by the machine that are configured to measure machine state data;

a machine control system configured to control displacement of the hydraulic valve, wherein the machine control system comprises:

memory storing a set of model weights trained using a machine learning system based on a training dataset, wherein:

the training dataset correlates preset displacements for the hydraulic valve to movement parameters of the component measured when the hydraulic valve is set according to the preset displacements and the component is moved at least in part by the hydraulic valve;

an input data receiver configured to receive:

a velocity command indicating a requested velocity of the component; and

the machine state data from the one or more sensors; and

a control mapping model configured to use the set of model weights to map a combination of the velocity command and the machine state data into a predicted displacement of the hydraulic valve that causes movement of the component at the requested velocity in response to the velocity command; and

the control mapping model causes the hydraulic valve to move to the predicted displacement.

12. The machine of claim 11 , wherein:

the machine learning system comprises a radial basis function network with a fully-connected neural network layer, and

the machine learning system is configured to train the set of model weights at least partially using backpropagation to adjust the set of model weights over a plurality of iterations.

13. The machine of claim 11 , wherein:

the hydraulic valve comprises a valve solenoid, and

the control mapping model causes a valve solenoid current to be provided to the valve solenoid to cause the hydraulic valve to move to the predicted displacement.

14. The machine of claim 11 , wherein:

the memory of the machine control system stores a plurality of sets of model weights corresponding to different types of events, and

the control mapping model selects the set of model weights from the plurality of sets of model weights based on an event type associated with the velocity command.

15. The machine of claim 11 , wherein the machine state data includes one or more of: pump pressure, hydraulic cylinder pressure, and positional data associated with the component or other components of the machine.

16. A method, comprising:

obtaining test sequence data by performing a plurality of test sequences on a machine according to a parameter grid that defines a plurality of predefined valve displacements, wherein performing a first test sequence of the plurality of test sequences includes:

setting a hydraulic valve to a first valve displacement of the plurality of predefined valve displacements;

actuating the hydraulic valve, wherein actuating the hydraulic valve causes movement of a component of the machine; and

measuring first test sequence data indicative of movement of the component caused by actuation of the hydraulic valve in association with the first valve displacement;

training a set of model weights using a machine learning system by:

using the test sequence data in training data;

using the plurality of predefined valve displacements in training labels; and

iteratively adjusting the set of model weights until the set of model weights predicts the training labels from the training data such that a loss metric is below a predefined threshold;

receiving a velocity command indicating a requested velocity of the component;

receiving machine state data from one or more sensors on the machine; and

generating a predicted displacement of the hydraulic valve by using the set of model weights to map a combination of the velocity command and the machine state data to the predicted displacement,

wherein controlling displacement of the hydraulic valve, based at least in part on the predicted displacement, causes movement of the component at the requested velocity in response to the velocity command.

17. The method of claim 16 , wherein:

the machine learning system comprises a radial basis function network with a fully-connected neural network layer, and

the machine learning system is configured to train the set of model weights at least partially using backpropagation to iteratively adjust the set of model weights over a plurality of iterations.

18. The method of claim 16 , wherein:

the hydraulic valve comprises a valve solenoid, and

controlling displacement of the hydraulic valve is based on a valve solenoid current provided to the valve solenoid that causes the hydraulic valve to move to the predicted displacement.

19. The method of claim 16 , further comprising:

determining an event type associated with the velocity command; and

selecting, from a plurality of sets of model weights that correspond to different types of events, the set of model weights associated with the event type.

20. The method of claim 16 , wherein the machine state data includes one or more of pump pressure, hydraulic cylinder pressure, and positional data associated with the component or other components of the machine.