IP Library Granted Patent US 12688402
Granted Patent B2
US 12688402 · App. 18/793,331 · Granted Jul 21, 2026

Method and system of machine fault classification using label-consistent convolutional dictionary learning

Inventors: Saurabh Sahu (Bangalore, IN); Kriti Kumar (Bangalore, IN); Achanna Anil Kumar (Bangalore, IN); Mariswamy Girish Chandra (Bangalore, IN); Angshul Majumdar (New Delhi, IN)
Assignee: Tata Consultancy Services Limited
G06N3/0464G06F11/2252G06F11/2257G06N5/04
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12688402
App. No.
18/793,331
Granted
Jul 21, 2026
Kind
B2
Abstract

Existing Convolutional Dictionary Learning (CDL) based machine fault classification do not utilize label information while learning the dictionary, hence the representation learned are not class-discriminative. Method and system disclosed herein provide a label-consistent convolutional dictionary learning approach for machine fault classification. The approach involves generating a training data for a classifier, wherein coefficients forming a plurality of class-discriminative features form the training data. The training data is then used to train a classifier, which is then used to perform machine fault classification for a given test data.

Claims (314)

1 . A processor implemented method, comprising:

receiving, via one or more hardware processors, a training sensor data (S) and a set of classification labels (Q) associated with the training sensor data, as input data;

performing, via the one or more hardware processors, a joint optimization on the input data to learn a plurality of parameters comprising a) one or more dictionary atoms {D m }, one or more coefficients {X m } associated with each of the one or more dictionary atoms, and weights {W m } associated with a label consistency term for each of the one or more coefficients for m th filter in a sequence of filters, wherein performing the joint optimization comprises iteratively updating the parameters {D m }, {X m }, and {W m }, till an objective function represented in the joint optimization is converged, wherein a plurality of class-discriminative dictionaries are learnt for each of a plurality of filters, by performing the joint optimization, wherein the joint optimization is represented as

arg

min

{

D

m

}

{

X

m

}

{

W

m

}

1

2

m

=

1

M

D

m

X

m

-

S

F

2

+

λ

m

=

1

M

X

m

1

+

η

2

Q

-

m

=

1

M

W

m

X

m

F

2

s

.

t

.

D

m

2

=

1

m

,

where, Q∈R C×K represents one hot encoded classification labels for C-classes, and K is the total number of training samples;

generating, via the one or more hardware processors, a plurality of coefficients forming a plurality of class-discriminative features, from the plurality of learnt class-discriminative dictionaries and the training sensor data (S); concatenating, via the one or more hardware processors, the plurality of coefficients of the plurality of filters to obtain a concatenated matrix of coefficients wherein the concatenated matrix of coefficients form a training data for a classifier, wherein the classifier is trained using the concatenated matrix of coefficients forming the plurality of class-discriminative features and the set of classification labels; and

using, via the one or more hardware processors, the trained classifier for performing machine fault classification.

2 . The method of claim 1 , wherein performing the joint optimization comprises:

learning one or more coefficients of each of the plurality of filters based on the training sensor data, the set of classification labels and a dictionary comprising the plurality of filters, and weights associated with a label consistency term comprising the plurality of filters;

learning one or more weights associated with a label consistency term of each of the plurality of filters based on the set of classification labels and the coefficients ({X m }) comprising the plurality of filters; and

learning the one or more dictionary atoms of each of the plurality of filters, based on the training sensor data and the coefficients comprising the plurality of filters.

3 . The method of claim 1 , wherein the classifier is used for classification of test data, comprising:

obtaining a test input comprising a test sensor data;

computing a plurality of test coefficients for each of the plurality of filters, using the test sensor data and learnt dictionary associated with each of the plurality of filters;

concatenating the plurality of test coefficients of the plurality of filters to obtain a concatenated matrix of test coefficients; and

estimating a set of output labels for the test sensor data using the concatenated matrix of test coefficients and the trained classifier.

4 . The method of claim 1 , wherein the training sensor data comprises one of a time series data and an image data.

5 . A system, comprising:

one or more hardware processors;

a communication interface; and

a memory storing a plurality of instructions, wherein the plurality of instructions cause the one or more hardware processors to:

receive a training sensor data (S) and a set of classification labels (Q) associated with the training sensor data, as input data;

perform a joint optimization on the input data to learn a plurality of parameters comprising a) one or more dictionary atoms {D m }, one or more coefficients {X m } associated with each of the one or more dictionary atoms, and weights {W m } associated with a label consistency term for each of the one or more coefficients for m th filter in a sequence of filters, wherein performing the joint optimization comprises iteratively updating the parameters {D m }, {X m }, and {W m }, till an objective function represented in the joint optimization is converged, wherein a plurality of class-discriminative dictionaries are learnt for each of a plurality of filters, by performing the joint optimization, wherein the joint optimization is represented as

arg

min

{

D

m

}

{

X

m

}

{

W

m

}

1

2

m

=

1

M

D

m

X

m

-

S

F

2

+

λ

m

=

1

M

X

m

1

+

η

2

Q

-

m

=

1

M

W

m

X

m

F

2

s

.

t

.

D

m

2

=

1

m

,

where, Q∈R C×K represents one hot encoded classification labels for C-classes, and K is the total number of training samples;

generate a plurality of coefficients forming a plurality of class-discriminative features, from the plurality of learnt class-discriminative dictionaries and the training sensor data (S);

concatenate the plurality of coefficients of the plurality of filters to obtain a concatenated matrix of coefficients wherein the concatenated matrix of coefficients form a training data for a classifier, wherein the classifier is trained using the concatenated matrix of coefficients forming the plurality of class-discriminative features and the set of classification labels; and

use the trained classifier for performing machine fault classification.

6 . The system of claim 5 , wherein the one or more hardware processors are configured to perform the joint optimization by:

learning one or more coefficients of each of the plurality of filters based on the training sensor data, the set of classification labels and a dictionary comprising the plurality of filters, and weights associated with a label consistency term comprising the plurality of filters;

learning one or more weights associated with a label consistency term of each of the plurality of filters based on the set of classification labels and the coefficients ({X m }) comprising the plurality of filters; and

learning the one or more dictionary atoms of each of the plurality of filters, based on the training sensor data and the coefficients comprising the plurality of filters.

7 . The system of claim 5 , wherein the one or more hardware processors are configured to use the classifier for classification of test data, by:

obtaining a test input comprising a test sensor data;

computing a plurality of test coefficients for each of the plurality of filters, using the test sensor data and learnt dictionary associated with each of the plurality of filters;

concatenating the plurality of test coefficients of the plurality of filters to obtain a concatenated matrix of test coefficients; and

estimating a set of output labels for the test sensor data using the concatenated matrix of test coefficients and the trained classifier.

8 . The system of claim 5 , wherein the training sensor data comprises one of a time series data and an image data.

9 . One or more non-transitory machine-readable information storage mediums comprising one or more instructions which when executed by one or more hardware processors cause:

receiving a training sensor data (S) and a set of classification labels (Q) associated with the training sensor data, as input data;

performing, via the one or more hardware processors, a joint optimization on the input data to learn a plurality of parameters comprising a) one or more dictionary atoms {D m }, one or more coefficients {X m } associated with each of the one or more dictionary atoms, and weights {W m } associated with a label consistency term for each of the one or more coefficients for m th filter in a sequence of filters, wherein performing the joint optimization comprises iteratively updating the parameters {D m }, {X m }, and {W m }, till an objective function represented in the joint optimization is converged, wherein a plurality of class-discriminative dictionaries are learnt for each of a plurality of filters, by performing the joint optimization, wherein the joint optimization is represented as

arg

min

{

D

m

}

{

X

m

}

{

W

m

}

1

2

m

=

1

M

D

m

X

m

-

S

F

2

+

λ

m

=

1

M

X

m

1

+

η

2

Q

-

m

=

1

M

W

m

X

m

F

2

s

.

t

.

D

m

2

=

1

m

,

where, Q∈R C×K represents one hot encoded classification labels for C-classes, and K is the total number of training samples;

generating a plurality of coefficients forming a plurality of class-discriminative features, from the plurality of learnt class-discriminative dictionaries and the training sensor data (S);

concatenating the plurality of coefficients of the plurality of filters to obtain a concatenated matrix of coefficients wherein the concatenated matrix of coefficients form a training data for a classifier, wherein the classifier is trained using the concatenated matrix of coefficients forming the plurality of class-discriminative features and the set of classification labels; and

using the trained classifier for performing machine fault classification.

10 . The one or more non-transitory machine-readable information storage mediums of claim 9 , wherein performing the joint optimization comprises:

learning one or more coefficients of each of the plurality of filters based on the training sensor data, the set of classification labels and a dictionary comprising the plurality of filters, and weights associated with a label consistency term comprising the plurality of filters;

learning one or more weights associated with a label consistency term of each of the plurality of filters based on the set of classification labels and the coefficients ({X m }) comprising the plurality of filters; and

learning the one or more dictionary atoms of each of the plurality of filters, based on the training sensor data and the coefficients comprising the plurality of filters.

11 . The one or more non-transitory machine-readable information storage mediums of claim 9 , wherein the one or more instructions which when executed by the one or more hardware processors cause the classifier to perform classification of test data, by:

obtaining a test input comprising a test sensor data;

computing a plurality of test coefficients for each of the plurality of filters, using the test sensor data and learnt dictionary associated with each of the plurality of filters;

concatenating the plurality of test coefficients of the plurality of filters to obtain a concatenated matrix of test coefficients; and

estimating a set of output labels for the test sensor data using the concatenated matrix of test coefficients and the trained classifier.

12 . The one or more non-transitory machine-readable information storage mediums of claim 9 , wherein the training sensor data comprises one of a time series data and an image data.