Method for classification based diagnosis with partial system model information

Patent metadata

Field	Value
Publication number	US-11915112-B2
Application number	US-201916559069-A
Country	US
Kind code	B2
Filing date	Sep 3, 2019
Priority date	Sep 3, 2019
Publication date	Feb 27, 2024
Grant date	Feb 27, 2024

Abstract

A classification-based diagnosis for detecting and predicting faults in physical system (e.g. an electronic circuit or rail switch) is disclosed. Some embodiments make use of partial system model information (e.g., system topology, components behavior) to simplify the classifier complexity (e.g., reduce the number of parameters). Some embodiments of the method use a Bayesian approach to derive a classifier structure.

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:
perform a classification-based diagnosis for at least one of detecting and predicting faults in a physical system using partial system model information, said partial system model information including at least one of a system topology and component behavior, wherein said classification-based diagnosis includes:
during an offline learning phase, learning parameters of an unknown component of the physical system, 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:
learn the parameters of the unknown component by solving:

β

j
*

=

arg
⁢

min

β
j

⁢

∑
i

⁢



y

0
:
T

(
i
)

-

y
^

0
:
T

(
i
)



2

where β is the parameters, j is a possible mode, and y 0:T is an output of a system that includes the unknown component; and
during an online learning phase, predicting a current mode based on the learned parameters of the unknown component.

2. The device of claim 1 , 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:
learn the parameters of the unknown component by solving an optimization problem.

3. The device of claim 1 , 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:
during the offline learning phase, estimate a switching parameter; and
during the online learning phase, predict the current mode further based on the switching parameter.

4. The device of claim 1 , 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:
during the offline learning phase:
estimate an error covariance matrix; and
estimate a switching parameter based on an argmin function and the estimated covariance matrix; and

during the online learning phase, predict the current mode further based on the switching parameter.

5. The device of claim 1 , 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:
during the online learning phase, predict the current mode according to an argmax function, wherein an input of the argmax function is the learned parameters.

6. The device of claim 1 , 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:
during the online learning phase, predict the current mode by solving:

j
*

=

arg
⁢

max
j

⁢

p
⁡

(

θ
=

j
|

y

0
:
T

;

β
j
*

,

η
*

)

where β is the parameters, j is a possible mode, θ is a random variable, η is a switching parameter, and y 0:T is an output of a system that includes the unknown component.

7. A method comprising:
performing a classification-based diagnosis for at least one of detecting and predicting faults in a physical system using partial system model information, said partial system model information including at least one of a system topology and component behavior, wherein said classification-based diagnosis includes:
during an offline learning phase, learning parameters of an unknown component, wherein the learning of the parameters of the unknown component occurs by solving:

β

j
*

=

arg
⁢

min

β
j

⁢

∑
i

⁢



y

0
:

T

Assignees

• Palo Alto Res Ct Inc
• Xerox Corp

Inventors

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

Classifications

• G06N3/09

  Supervised learning · CPC title

• G06N3/0499

  Feedforward networks · CPC title

• G06N20/20 Primary

  Ensemble learning · CPC title

• G06F17/10

  Complex mathematical operations {(function generation by table look-up G06F1/03; evaluation of elementary functions by calculation G06F7/544)} · CPC title

• G06F17/11 Primary

  for solving equations {, e.g. nonlinear equations, general mathematical optimization problems (optimization specially adapted for a specific administrative, business or logistic context G06Q10/04)} · CPC title