Virtual sensor for estimating online unmeasurable variables via successive time derivatives

Patent metadata

Field	Value
Publication number	US-11526724-B2
Application number	US-201916546887-A
Country	US
Kind code	B2
Filing date	Aug 21, 2019
Priority date	Aug 21, 2019
Publication date	Dec 13, 2022
Grant date	Dec 13, 2022

Abstract

Provided is a processor-implemented method and a processor in a vehicle for estimating the value of a quantity for which a physical sensor is not available for measurement. The method includes: receiving a plurality of measured signals representing values of measurable variables; computing, in real-time, time derivatives of the measured signals; and applying a trained feedforward neural network, in real-time, to estimate values for a plurality of unmeasurable variables, the unmeasurable variables being variables that are unmeasurable in real-time, the feedforward neural network having been trained using test data containing time derivatives of values for the measurable variables and values for the unmeasurable variables; wherein the vehicle uses the estimated values for the unmeasurable variables for vehicle operation.

First claim

What is claimed is:

1. A processor-implemented method in a vehicle for estimating a quantity for which a physical sensor is not available to measure the quantity, the method comprising:
receiving a plurality of measured signals representing values of measurable variables;
computing, in real-time, time derivatives of the measured signals by applying a model-free derivatives estimator to compute the time derivatives; and
applying a trained feedforward neural network, in real-time, to estimate values for a plurality of unmeasurable variables, the unmeasurable variables being variables that are unmeasurable in real-time, the feedforward neural network having been trained using test data containing time derivatives of values for the measurable variables and values for the unmeasurable variables;
wherein the vehicle uses the estimated values for the unmeasurable variables for vehicle operation.

2. The method of claim 1 , wherein the trained feedforward neural network was trained using a process comprising:
collecting test data that includes values for the unmeasurable variables and corresponding values for the measurable variables;
estimating successive time derivatives of the values for the measured variables; and
training the feedforward neural network to find a mathematical model that correlates the time derivatives for the measured values to the values attained by the unmeasurable variables.

3. The method of claim 2 , wherein the estimating successive time derivatives of the values for the measured variables is implemented using a model-free derivatives estimator.

4. The method of claim 1 , wherein the model-free derivatives estimator comprises a high-gain observer, a sliding mode observer or a super-twisting algorithm.

5. The method of claim 1 , wherein the model-free derivatives estimator is represented by the following discrete time system: ξ(k+1)=Eξ(k)+Fv(k), ψ(k)=Gξ(k)+Hv(k), wherein ξ is a state vector, v is a vector of the measured variables, and ψ is the vector of time derivatives of v.

6. The method of claim 5 , wherein E, F, G, and H are matrices computed using a Forward Euler method, a Backward Euler method, or a Tustin method.

7. A processor configured as a virtual sensor in a vehicle to estimate a quantity for which a physical sensor is not available for measurement, the processor configured to:
receive a plurality of measured signals representing values of measurable variables;
compute, in real-time, time derivatives of the measured signals by applying a model-free derivatives estimator to compute the time derivatives; and
apply a trained feedforward neural network, in real-time, to estimate values for a plurality of unmeasurable variables, the unmeasurable variables being variables that are unmeasurable in real-time, the feedforward neural network having been trained using test data containing time derivatives of values for the measurable variables and values for the unmeasurable variables;
wherein the vehicle uses the estimated values for the unmeasurable variables for vehicle operation.

8. The processor of claim 7 , wherein the trained feedforward neural network was trained using a process comprising:
collecting test data that includes values for the unmeasurable variables and corresponding values for the measurable variables;
estimating successive time derivatives of the values for the measured variables; and
training the feedforward neural network to find a mathematical model that correlates the time derivatives for the measured values to the values attained by the unmeasurable variables.

9. The processor of claim 8 , wherein the estimating successive time derivatives of the values for the measured variables is implemented using a model-free derivatives estimator.

10. The processor of claim 7 , wherein the model-free derivatives estimator comprises a high-gain observer, a sliding mode observer or a super-twisting algorithm.

11. The processor of claim 7 , wherein the model-free derivatives estimator is represented by the following discrete time system: ξ(k+1)=Eξ(k)+Fv(k), ψ(k)=Gξ(k)+Hv(k), wherein ξ is a state vector, v is a vector of the measured variables, and ψ is the vector of time derivatives of v.

12. The processor of claim 11 , wherein E, F, G, and H are matrices computed using a Forward Euler method, a Backward Euler method, or a Tustin method.

13. A non-transitory computer readable storage medium embodying programming instruction for performing a method in a vehicle, the method comprising:
receiving a plurality of measured signals representing values of measurable variables;
computing, in real-time, time derivatives of the measured signals by applying a model-free derivatives estimator to compute the time derivatives; and
applying a trained feedforward neural network, in real-time, to estimate values for a plurality of unmeasurable variables, the unmeasurable variables being variables that are unmeasurable in real-time, the feedforward neural network having been trained using test data containing time derivatives of values for the measurable variables and values for the unmeasurable variables;
wherein the vehicle uses the estimated values for the unmeasurable variables for vehicle operation.

14. The non-transitory computer readable storage medium of claim 13 , wherein the trained feedforward neural network was trained using a process comprising:
collecting test data that includes values for the unmeasurable variables and corresponding values for the measurable variables;
estimating successive time derivatives of the values for the measured variables; and
training the feedforward neural network to find a mathematical model that correlates the time derivatives for the measured values to the values attained by the unmeasurable variables.

15. The non-transitory computer readable storage medium of claim 14 , wherein the estimating successive time derivatives of the measured variables is implemented using a model-free derivatives estimator.

16. The non-transitory computer readable storage medium of claim 13 , wherein the model-free derivatives estimator comprises a High-Gain observer, a sliding mode observer or a super-twisting algorithm.

17. The non-transitory computer readable storage medium of claim 13 , wherein the model-free derivatives estimator is represented by the following discrete time system: ξ(k+1)=Eξ(k)+Fv(k), ψ(k)=Gξ(k)+Hv(k), wherein:
ξ is a state vector, v is a vector of the measured variables, and ψ is the vector of time derivatives of v; and
E, F, G, and H are matrices computed using a Forward Euler method, a Backward Euler method, or a Tustin method.

Assignees

• Gm Global Tech Operations Llc

Inventors

• Pedicini Carmen
• Possieri Corrado
• Alfieri Vincenzo
• Conte Giuseppe

Classifications

• G06N3/045

  Combinations of networks · CPC title

• G06N3/08

  Learning methods · CPC title

• B60W2050/0012

  Feedforward or open loop systems · CPC title

• G05B13/027

  using neural networks only · CPC title

• G06N3/04 Primary

  Architecture, e.g. interconnection topology · CPC title