# US 2025/0156718 A1 · Training device, training method, and device and method of diagnosing anomaly of equipment using signal reconstruction model

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

- Number: US 2025/0156718 A1
- Document ID: US20250156718A1
- Status: Published application
- Kind: A1
- Application number: 18/946,191
- Filed: 2024-11-13
- Published: 2025-05-15
- Earliest claimed: 2023-11-09
- Assignees: TECH UNIVERSITY OF KOREA Industry Academic Cooperation Foundation; Reshenie Corp.
- Applicants: Tech University of Korea Industry Academic Cooperation Foundation; Reshenie Corp.
- Inventors: Yong Kwan LEE; Ki Chang PARK
- Web page: https://patentgrep.com/patent/US20250156718A1

## CPC

- G06N 3/088: Non-supervised learning, e.g. competitive learning
- G01N 29/12: by measuring frequency or resonance of acoustic waves {(measuring frequency or resonant frequency of mechanical vibrations or acoustic waves in general G01H1/06, G01H3/04, G01H13/00; acoustic resonators G10K11/04; vibration or shock testing of structures G01M7/00)}
- G01N 29/4427: {with stored values, e.g. threshold values}
- G01N 29/4454: {Signal recognition, e.g. specific values or portions, signal events, signatures}
- G05B 23/02: Electric testing or monitoring
- G06N 3/0442: characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
- G06N 3/0455: Auto-encoder networks; Encoder-decoder networks
- G06N 3/0464: Convolutional networks [CNN, ConvNet]
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G01N: INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES (measuring or testing apparatus or processes other than immunoassay, involving enzymes or microorganisms C12M, C12Q)
- 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)

## Abstract

A training device, a training method, a failure diagnosis device, and a failure diagnosis method for diagnosing an anomaly of equipment using a signal reconstruction model are disclosed. The training device includes a memory configured to store instructions executable by a processor, and a processor, wherein the processor is configured to obtain a vibration frequency signal that measures vibration occurring when equipment is normally operated through a vibration sensor, train a signal reconstruction model configured to generate a reconstructed signal corresponding to the vibration frequency signal based on the obtained vibration frequency signal, obtain a reconstructed signal corresponding to the vibration frequency signal through the trained signal reconstruction model, determine a reconstruction error value representing a difference between the vibration frequency signal and the reconstructed signal, and determine a threshold value for the reconstruction error value for each of a plurality of predefined frequency bands.

## Claims

### Claim 1

A training device comprising:
a memory configured to store instructions executable by a processor; and
a processor,
wherein the processor is configured to:
obtain a vibration frequency signal that measures vibration occurring when equipment is normally operated through a vibration sensor,
train a signal reconstruction model configured to generate a reconstructed signal corresponding to the vibration frequency signal based on the obtained vibration frequency signal,
obtain a reconstructed signal corresponding to the vibration frequency signal through the trained signal reconstruction model,
determine a reconstruction error value representing a difference between the vibration frequency signal and the reconstructed signal, and
determine a threshold value for the reconstruction error value for each of a plurality of predefined frequency bands.

### Claim 2 (depends on 1)

The training device of claim 1, wherein the signal reconstruction model comprises at least one of an autoencoder, a stacked autoencoder, a long short-term memory (LSTM) autoencoder, and a convolutional autoencoder model.

### Claim 3 (depends on 1)

The training device of claim 1, wherein the processor is further configured to determine the reconstruction error value based on at least one of an average of differences between the vibration frequency signal and the reconstructed signal, an average of a square of the differences, and a square root of the average of the square of the differences.

### Claim 4 (depends on 1)

The training device of claim 1, wherein the processor is further configured to determine a threshold value for the reconstruction error value to be a maximum value of the reconstruction error value or a three-sigma rule.

### Claim 5

A failure diagnosis device for diagnosing an anomaly of equipment using a signal reconstruction model, the failure diagnosis device comprising:
a memory configured to store instructions executable by a processor; and
a processor,
wherein the processor is configured to:
obtain a vibration frequency signal that measures vibration occurring in equipment through a vibration sensor,
obtain a reconstructed signal corresponding to the vibration frequency signal from a trained signal reconstruction model by inputting the vibration frequency signal to the signal reconstruction model,
determine a reconstruction error value representing a signal difference between the vibration frequency signal and the reconstructed signal for each predefined frequency band, and
determine whether the equipment is abnormal based on the reconstruction error value determined for each frequency band and a threshold value determined for each frequency band.

### Claim 6 (depends on 5)

The failure diagnosis device of claim 5, wherein the processor is further configured to, when at least one of the reconstruction error values determined for each frequency band is greater than the threshold value determined for each frequency band, determine that the equipment is abnormal.

### Claim 7 (depends on 5)

The failure diagnosis device of claim 5, wherein the processor is further configured to, when all reconstruction error values determined for each frequency band are less than or equal to the threshold value determined for each frequency band, determine that the equipment is in a normal state.

### Claim 8 (depends on 5)

The failure diagnosis device of claim 5, wherein the signal reconstruction model comprises at least one of an autoencoder, a stacked autoencoder, a long short-term memory (LSTM) autoencoder, and a convolutional autoencoder model.

### Claim 9 (depends on 5)

The failure diagnosis device of claim 5, wherein the processor is further configured to determine the reconstruction error value based on at least one of an average of differences between an input signal and the reconstructed signal, an average of a square of the differences, and a square root of the average of the square of the differences.

### Claim 10 (depends on 5)

The failure diagnosis device of claim 5, wherein the processor is further configured to, as a result of comparing the reconstruction error value determined for each frequency band with the threshold value determined for each frequency band, when a frequency band in which the reconstruction error value is greater than the threshold value is detected, estimate an anomaly type based on the detected frequency band.

### Claim 11

A failure diagnosis method performed by a failure diagnosis device for diagnosing an anomaly of equipment using a signal reconstruction model, the failure diagnosis method comprising:
obtaining a vibration frequency signal that measures vibration occurring in equipment through a vibration sensor;
obtaining a reconstructed signal corresponding to the vibration frequency signal from a trained signal reconstruction model by inputting the vibration frequency signal to the signal reconstruction model;
determining a reconstruction error value representing a signal difference between the vibration frequency signal and the reconstructed signal for each predefined frequency band; and
determining whether the equipment is abnormal based on reconstruction error values determined for each frequency band and a threshold value determined for each frequency band.

### Claim 12 (depends on 11)

The failure diagnosis method of claim 11, wherein the determining of the anomaly comprises:
when at least one of the reconstruction error values determined for each frequency band is greater than the threshold value determined for each frequency band, estimating that the equipment is abnormal.

### Claim 13 (depends on 11)

The failure diagnosis method of claim 11, further comprising:
when all reconstruction error values determined for each frequency band are less than or equal to the threshold value determined for each frequency band, determining that the equipment is in a normal state.

### Claim 14 (depends on 11)

The failure diagnosis method of claim 11, wherein the signal reconstruction model comprises at least one of an autoencoder, a stacked autoencoder, a long short-term memory (LSTM) autoencoder, and a convolutional autoencoder model.

### Claim 15 (depends on 11)

The failure diagnosis method of claim 11, wherein the determining of the reconstruction error value comprises:
determining the reconstruction error value based on an average of differences between an input signal and the reconstructed signal, an average of a square of the differences, and a square root of the average of the square of the differences.

### Claim 16 (depends on 11)

The failure diagnosis method of claim 11, further comprising:
as a result of comparing the reconstruction error value determined for each frequency band with the threshold value determined for each frequency band, when a frequency band in which the reconstruction error value is greater than the threshold value is detected, estimate an anomaly type based on the detected frequency band.

### Claim 17 (depends on 11)

The failure diagnosis method of claim 11, wherein the signal reconstruction model is trained by:
obtaining a vibration frequency signal that measures vibration occurring when equipment is normally operated through a vibration sensor;
training the signal reconstruction model configured to generate a reconstructed signal corresponding to the vibration frequency signal based on the obtained vibration frequency signal;
obtaining a reconstructed signal corresponding to the vibration frequency signal through the trained signal reconstruction model;
determining a reconstruction error value representing a difference between the vibration frequency signal and the reconstructed signal; and
determining a threshold value for the reconstruction error value for each of a plurality of predefined frequency bands.

### Claim 18 (depends on 17)

The failure diagnosis method of claim 17, wherein the determining of the reconstruction error value comprises:
determining the reconstruction error value based on one of an average of differences between the vibration frequency signal and the reconstructed signal, an average of a square of the differences, and a square root of the average of the square of the differences.

### Claim 19 (depends on 17)

The failure diagnosis method of claim 17, wherein the determining of the threshold value comprises:
determining a threshold value for the reconstruction error value to be a maximum value of the reconstruction error value or a three-sigma rule.