Machine learning based systems and methods for real time, model based diagnosis

Patent metadata

Field	Value
Publication number	US-11544422-B2
Application number	US-201916572280-A
Country	US
Kind code	B2
Filing date	Sep 16, 2019
Priority date	Sep 16, 2019
Publication date	Jan 3, 2023
Grant date	Jan 3, 2023

Abstract

The disclosure following relates generally to complex simulations, and fault diagnosis. In some embodiments, a component that is causing a delayed simulation time of a system is determined. A component of reduced complexity is designed, and the component of reduced complexity is used to replace the original component in the system. Fault diagnosis may then be conducted using the updated system with the reduced complexity component, thus decreasing the time taken to diagnose the fault.

First claim

What is claimed is:

1. A device comprising:
at least one processor; and
at least one memory including computer program code;
the at least one memory and the computer program code configured to, with the at least one processor, cause the device to:
determine that a component of a first model of a physical system is more complex than other components of the first model;
generate training data from the first model of the physical system;
using the generated training data, design a reduced complexity component for the said component of the first model;
generate a second model of the physical system by replacing the component in the first model with the reduced complexity component;

simulate the physical system using the second model over a predefined time horizon, and update simulation outputs;
use the simulation outputs and compute mean square errors between simulated values of variables of the physical system and observed values of the variables; and
diagnose a fault in the system using computed values of mean square errors.

2. A device comprising:
at least one processor; and
at least one memory including computer program code;
the at least one memory and the computer program code configured to, with the at least one processor, cause the device to:
determine a set of constitutive equations of a physical system, the constitutive equations based on generalized mass spring dampers (gMSD);
solve a constrained optimization problem that is based on the determined set of constitutive equations;
learn a representation of a component of the physical system using a solution of the constrained optimization problem;
evaluate a loss function for the optimization problem using model based simulations using the learned representation of the component;
estimate values of a number of parameters of the physical system using the optimization and simulations; and
determine a fault condition of the physical system if the estimated values of the parameters deviate from their nominal values as indicated by mean square errors (MSE) between those values.

3. The device of claim 2 , wherein the at least one memory and the computer program code are further configured to, with the at least one processor, cause the device at least to:
impose a dissipativity condition on the component, wherein the dissipativity condition:
(i) ensures that the system can be simulated; and
(ii) is based on an energy of the component.

4. The device of claim 2 , wherein the constitutive equations emulate physical laws.

5. The device of claim 2 , wherein the constitutive equations are of the form f(x; w)=0, where x includes port variables and internal variables, and w is a vector of parameters of the component.

6. The device of claim 2 , wherein the constrained optimization problem is a minimization problem.

7. The device of claim 2 , wherein the fault condition is determined based on a mean square error (MSE) of a given parameter.

8. The device of claim 2 , wherein the fault condition is determined based on ∥θ i −θ i *∥≥ε i , where:
ε i is a fault specific threshold;
{θ 1 , . . . , θ L } is a set of fault parameters; and
θ i * are nominal values for i∈{1, . . . , L}.

9. The device of claim 2 , wherein the system is a rail switch system and the component is a rail.

10. The device of claim 2 , wherein:
the system is a rail switch system and the component is a rail; and
the at least one memory and the computer program code are further configured to, with the at least one processor, cause the device to determine the set of constitutive equations further by using force, position, velocity and acceleration as training data.

11. A device comprising:
at least one processor; and
at least one memory including computer program code;
the at least one memory and the computer program code configured to, with the at least one processor, cause the device to:
use a causal map to assign causality relations between at least one input variable of a component and at least one output variable of the component in a physical system by:
first, training parameters of the component from training data; and
second, fine tuning the parameters of the component by using a model of the physical system, wherein the physical system includes the component;

perform simulations of the physical system using the model and the fine tuned parameters and update simulation outputs;
use the simulation outputs and compute mean square errors between simulated values of variables of the physical system and observed values of the variables; and
determine a fault condition of the system using computed values of mean square errors.

12. The device of claim 11 , wherein the causal map uses a neural network (NN).

13. The device of claim 11 , wherein the fault condition is determined based on a mean square error (MSE) of a given parameter.

14. The device of claim 11 , wherein the system is a rail switch system and the component is a rail.

15. The device of claim 11 , wherein the system is an electrical system, and the fault condition is one of a short connection and an open connection.

16. The device of claim 11 , wherein the system is a mechanical system, and the fault condition is one of a broken flange or a stuck flange.

17. The device of claim 11 , wherein the system is a fluid system, and the fault condition is one of a blocked pipe or a leaking pipe.

Assignees

• Palo Alto Res Ct Inc

Inventors

• Matei Ion
• Zhenirovskyy Maksym
• De Kleer Johan
• Feldman Alexander

Classifications

• G06F17/13 Primary

  Differential equations (using digital differential analysers G06F7/64) · CPC title

• G06F30/27

  using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model · CPC title

• G06F18/214

  Generating training patterns; Bootstrap methods, e.g. bagging or boosting · CPC title

• G06N5/01

  Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound · CPC title

• G06N5/022

  Knowledge engineering; Knowledge acquisition · CPC title