# US 2026/0243830 A1 · Systems and methods for motor fault detection

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2026/0243830 A1
- Document ID: US20260243830A1
- Status: Published application
- Kind: A1
- Application number: 19/053,786
- Filed: 2025-02-14
- Published: 2026-08-20
- Earliest claimed: 2025-02-14
- Assignees: Mitsubishi Electric Research Laboratories, Inc.
- Applicants: Mitsubishi Electric Research Laboratories, Inc.
- Inventors: Dehong Liu; Peikun Guo; Bingnan Wang; Yebin Wang
- Web page: https://patentgrep.com/patent/US20260243830A1

## CPC

- G01R 31/343: {in operation}
- G05B 23/0221: {Preprocessing measurements, e.g. data collection rate adjustment; Standardization of measurements; Time series or signal analysis, e.g. frequency analysis or wavelets; Trustworthiness of measurements; Indexes therefor; Measurements using easily measured parameters to estimate parameters difficult to measure; Virtual sensor creation; De-noising; Sensor fusion; Unconventional preprocessing inherently present in specific fault detection methods like PCA-based methods}
- G01M 15/02: Details or accessories of testing apparatus
- G01R 31/346: {Testing of armature or field windings}
- G05B 23/024: {Quantitative history assessment, e.g. mathematical relationships between available data; Functions therefor; Principal component analysis [PCA]; Partial least square [PLS]; Statistical classifiers, e.g. Bayesian networks, linear regression or correlation analysis; Neural networks}
- G01R: MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES (indicating correct tuning of resonant circuits H03J3/12)
- G05B: CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS (systems for controlling or regulating non-electric variables G05D; systems for regulating electric or magnetic variables G05F; control devices or systems insofar as characterised by mechanical features only G05G)
- G01M: TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR

## Abstract

A method for motor fault detection comprises acquiring sensor data from each sensor of a plurality of sensors associated with a motor. Each sensor of the plurality of sensors provides data corresponding to a unique modality. The method further comprises extracting features from the sensor data of each sensor and mapping the extracted features to a shared latent space using a deep generalized canonical correlation analysis model that aligns the features by learning nonlinear transformations that maximize correlations among the sensor data of different sensors. An attention-based fusion mechanism is applied to the aligned features to dynamically compute attention weights for each sensor modality based on a relevance of a respective sensor modality to a fault detection task. The method further comprises generating a fused feature representation by aggregating the aligned features weighted by the attention-based fusion mechanism and generating motor fault data based on the fused feature representation.

## Claims

### Claim 1

A method for motor fault detection, comprising:
acquiring sensor data from each sensor of a plurality of sensors associated with a motor, wherein each sensor of the plurality of sensors provides data corresponding to a unique modality of a plurality of modalities;
extracting features from the sensor data of each sensor of the plurality of sensors using one or more feature extraction models;
mapping the extracted features to a shared latent space using at least one Deep Generalized Canonical Correlation Analysis (DGCCA) model, wherein the at least one DGCCA model aligns the features by learning nonlinear transformations that maximize correlations among the sensor data of different sensors of the plurality of sensors;
applying an attention-based fusion mechanism to the DGCCA-aligned features to dynamically compute attention weights for each sensor modality of the plurality of modalities based on a relevance of a respective sensor modality to a fault detection task;
generating a fused feature representation by aggregating the DGCCA-aligned features weighted by the attention-based fusion mechanism; and
generating motor fault data based on the fused feature representation.

### Claim 2 (depends on 1)

The method of claim 1, further comprising:
generating one or more control signals, based on the motor fault data, wherein the one or more control signals specify values of one or more operational parameters of the motor; and
controlling the motor, based on the one or more control signals.

### Claim 3 (depends on 1)

The method of claim 1, wherein extracting the features from the sensor data of each sensor of the plurality of sensors further comprises:
extracting time-domain features for non-periodic faults; and
extracting frequency-domain features using Discrete Fourier Transform (DFT) for periodic faults.

### Claim 4 (depends on 1)

The method of claim 1, wherein the at least one DGCCA model aligns features from two or more sensor modalities of the plurality of modalities, wherein the plurality of modalities include at least vibration, current, and velocity.

### Claim 5 (depends on 1)

The method of claim 1, wherein the attention-based fusion mechanism assigns weights to sensor modalities according to their relative importance to the fault detection task.

### Claim 6 (depends on 1)

The method of claim 1, further comprising:
preprocessing the sensor data to remove noise and normalizing the preprocessed data to have zero mean and unit variance.

### Claim 7 (depends on 1)

The method of claim 1, wherein generating the motor fault data comprises at least one of:
classifying, using a neural network, motor faults associated with the motor; or
predicting, using the neural network, fault severity levels associated with the motor faults associated with the motor.

### Claim 8 (depends on 1)

The method of claim 1, wherein the plurality of sensors comprises at least two of vibration sensors, current sensors, velocity sensors, or air gap sensors.

### Claim 9 (depends on 1)

The method of claim 1, wherein mapping the extracted features to a shared latent space further comprises aligning each feature of the extracted features into a plurality of fault-specific latent spaces using a plurality of DGCCA models, wherein each DGCCA model of the plurality of DGCCA models aligns the extracted feature into a unique fault-specific latent space of the plurality of fault-specific latent spaces.

### Claim 10 (depends on 9)

The method of claim 9, wherein the attention-based fusion mechanism comprises a plurality of attention networks, each attention network of the plurality of attention networks corresponding to a specific DGCCA model of the plurality of DGCCA models and a specific fault type, and each attention network of the plurality of attention networks dynamically computing attention weights based on the aligned features of a corresponding fault type.

### Claim 11 (depends on 1)

The method of claim 1, wherein mapping the extracted features to a shared latent space further comprises:
using one DGCCA model to align features from all sensor modalities into a shared latent space, wherein the aligned features are used for detecting multiple fault types.

### Claim 12 (depends on 11)

The method of claim 11, wherein the attention-based fusion mechanism comprises:
a multi-head attention network, wherein each attention head of the multi-head attention network corresponds to a specific fault type, and wherein each attention head of the multi-head attention network dynamically assigns weights to the aligned features based on the fault type being analyzed.

### Claim 13

A system for motor fault detection, comprising:
memory configured to store a feature extraction model, a Deep Generalized Canonical Correlation Analysis (DGCCA) model, an attention-based fusion model, and computer-executable instructions;
at least one processor configured to execute the instructions to:
acquire sensor data from each sensor of a plurality of sensors configured to collect multi-modal data from a motor, wherein each sensor of the plurality of sensors provides data corresponding to a unique modality of a plurality of modalities;
extract, using the feature extraction model, features from the sensor data of each sensor of the plurality of sensors;
align, using the DGCCA model, the extracted features into a shared latent space by learning nonlinear transformations that maximize correlations among the sensor data of different sensors of the plurality of sensors;
dynamically assign, using the attention-based fusion model, attention weights to the DGCCA-aligned features based on their relevance to a fault detection task;
generate a fused feature representation by aggregating the DGCCA-aligned features weighted by the attention-based fusion model; and
generate motor fault data based on the fused feature representation.

### Claim 14 (depends on 13)

The system of claim 13, wherein the at least one processor is further configured to:
generate one or more control signals, based on the motor fault data, wherein the one or more control signals specify values of one or more operational parameters of the motor; and
control the motor, based on the one or more control signals.

### Claim 15 (depends on 13)

The system of claim 13, wherein to extract the features from the sensor data of each sensor of the plurality of sensors, the at least one processor is further configured to:
extract time-domain features for non-periodic faults; and
extract frequency-domain features using Discrete Fourier Transform (DFT) for periodic faults.

### Claim 16 (depends on 13)

The system of claim 13, wherein the DGCCA model aligns features from two or more sensor modalities of the plurality of modalities, wherein the plurality of modalities include at least vibration, current, and velocity.

### Claim 17 (depends on 13)

The system of claim 13, wherein to generate the motor fault data the at least one processor is configured to execute a neural network to perform at least one of:
classification of motor faults associated with the motor; or
prediction of fault severity levels associated with the motor faults associated with the motor.

### Claim 18 (depends on 13)

The system of claim 13, wherein to align the extracted features to a shared latent space the at least one processor is configured to align each feature of the extracted features into a plurality of fault-specific latent spaces using a plurality of DGCCA models, wherein each DGCCA model of the plurality of DGCCA models aligns the extracted feature into a unique fault-specific latent space of the plurality of fault-specific latent spaces.

### Claim 19 (depends on 18)

The system of claim 18, wherein the attention-based fusion model comprises a plurality of attention networks, each attention network of the plurality of attention networks corresponding to a specific DGCCA model of the plurality of DGCCA models and a specific fault type, and each attention network of the plurality of attention networks dynamically computing attention weights based on the aligned features of a corresponding fault type.

### Claim 20

A non-transitory computer-readable medium storing instructions that, when executed by a processor, cause the processor to perform a method for motor fault detection, the method comprising:
acquiring sensor data from each sensor of a plurality of sensors associated with a motor, wherein each sensor of the plurality of sensors provides data corresponding to a unique modality of a plurality of modalities;
extracting features from the sensor data of each sensor of the plurality of sensors using one or more feature extraction models;
mapping the extracted features to a shared latent space using at least one Deep Generalized Canonical Correlation Analysis (DGCCA) model, wherein the at least one DGCCA model aligns the features by learning nonlinear transformations that maximize correlations among the sensor data of different sensors of the plurality of sensors;
applying an attention-based fusion mechanism to the DGCCA-aligned features to dynamically compute attention weights for each sensor modality of the plurality of modalities based on a relevance of a respective sensor modality to a fault detection task;
generating a fused feature representation by aggregating the DGCCA-aligned features weighted by the attention-based fusion mechanism; and
generating motor fault data based on the fused feature representation.
