Granted Patent B2

US 12,203,890 · App. 17/830,859 · Granted Jan 21, 2025

Inventors: Hung-Chih Chang (Hsinchu, TW); Yao-Long Tsai (Kaohsiung, TW); Li-Hua Wang (Hsinchu, TW)

Assignee: INDUSTRIAL TECHNOLOGY RESEARCH INSTITUTE

G01N29/12G01N29/14G01N29/348G06N3/04G06N3/08G06N20/10G01N2291/0289

Abstract

Claims (61)

1. An acoustic resonance diagnostic method for detecting structural degradation of an under-test section of an under-test structure, the method comprising:

building a database and storing reference acoustic signals in the database as training acoustic signals which are derived from mathematic calculations and/or based on historic acoustic data of structures similar to the under-test section;

classifying the training acoustic signals into transient training acoustic signals and steady training acoustic signals according to states of the reference acoustic signals;

capturing a plurality of under-test sound wave signals as captured under-test acoustic signals from the under-test section and storing the captured under-test acoustic signals also in the database;

using an artificial intelligence (AI) to filter out acoustic signals in the database that are invalid for detecting structural degradation of the under-test section;

building a training model using a deep neural network (DNN);

inputting at least two filtered transient training acoustic signals as training acoustic signals to the training model to carry out a training of the training model;

building a diagnostic model according to a result of the training using a convolutional neural network (CNN);

inputting a remaining portion of the filtered training acoustic signals, being the filtered steady training acoustic signals, as verifying acoustic signals to the diagnostic model, to verify that the diagnostic model has been trained thus can successfully detect transient acoustic signals, the diagnostic model then outputs detection signals and labels the detection signals as leakage frequency, metal frequency, ambient frequency, or noise frequency;

inputting the captured and filtered under-test acoustic signals to the trained diagnostic model;

determining a structural degradation state of the under-test section by comparing the inputted captured and filtered under-test acoustic signals to the labeled detection signals in the trained diagnostic model.

2. The acoustic resonance diagnostic method for detecting structural degradation according to claim 1 , wherein the step of using the artificial intelligence (AI) to filter out the acoustic signals in the database that are invalid for detecting structural degradation of the under-test section comprises:

performing an 2D graphic comparison process to compare a frequency-domain waveform of the captured under-test signals with characteristic curves of historical frequency-domain waveforms with different prototypes of structural degradation provided by historical data so as to select a plurality of frequency segments from the captured under-test acoustic signals, wherein characteristic curves of the selected frequency segments match portions of the characteristic curves in the historical frequency-domain waveforms;

applying a support vector machine (SVM) learning using a classification and/or a regression analysis to remove portions of the captured under-test acoustic signals from the selected frequency segments according to a predetermine threshold value;

reunion remaining portions of the selected frequency segments.

3. The acoustic resonance diagnostic method for detecting structural degradation according to claim 1 , further comprising providing at least one sound vibration on the under-test structure.

4. The acoustic resonance diagnostic method for detecting structural degradation according to claim 3 , wherein the at least one sound vibration has an amplitude value greater than that of the plurality of under-test sound wave signals previously captured.

5. The acoustic resonance diagnostic method for detecting structural degradation according to claim 3 , wherein the at least one sound vibration has a frequency to form a standing wave in the under-test structure.

6. The acoustic resonance diagnostic method for detecting structural degradation according to claim 5 , wherein the frequency of the standing wave is estimated using the following equation:

f

n

=

2

⁢

nV

L

;

wherein f n is the frequency of the standing wave; L is a distance between a sensing unit used to capture the plurality of under-test sound wave signals and a vibration generator for providing the at least one sound vibration; V is speed of sound; and n is a mode number of the standing wave.

7. The acoustic resonance diagnostic method for detecting structural degradation according to claim 5 , further comprising identifying a leakage position by referring to a relationship between a sound pressure change rate of the at least one sound vibration and an unit wavelength of the standing wave.

8. An acoustic resonance diagnostic system for detecting structural degradation of a plurality of under-test sections of an under-test structure, the system comprising:

a database to store reference acoustic signals as training acoustic signals which are derived from mathematic calculations and/or based on historic acoustic data of structures similar to the under-test sections, the training acoustic signals are classified into transient training acoustic signals and steady training acoustic signals according to states of the reference acoustic signals;

a plurality of sound wave sensing units, to capture a plurality of under-test sound waves as captured under-test acoustic signals from the under-test sections, the captured under-test acoustic signals are then stored also in the database;

a communication module to signal-connect the sound wave sensing units to the database;

a data pre-processing module, using an AI to filter out the acoustic signals in the database that are invalid for detecting structural degradation of the under-test sections;

an acoustic resonance diagnostic module to perform the following steps:

building a training model using a deep neural network;

inputting at least two filtered transient training acoustic signals as training acoustic signals to the training model to carry out a training of the training model;

building a diagnostic model according to a result of the training using a neural network;

inputting a remaining portion of the filtered training acoustic signals, being the

filtered steady training acoustic signals, as verifying acoustic signals to the diagnostic model, to verify that the diagnostic model has been trained thus can successfully detect transient acoustic signals, the diagnostic model then outputs detection signals and labels the detection signals as leakage frequency, metal frequency, ambient frequency, or noise frequency;

inputting the captured and filtered under-test acoustic signals to the trained diagnostic model; and

determining at least one structural degradation state of the under-test sections by comparing the inputted captured and filtered under-test acoustic signals to the labeled detection signals in the diagnostic model.

9. The acoustic resonance diagnostic system for detecting structural degradation according to claim 8 , wherein the plurality of sound wave sensing units are fixed on and/or adjacent to the under-test section.

10. The acoustic resonance diagnostic system for detecting structural degradation according to claim 8 , wherein the step of using the AI to filter out the acoustic signals in the database that are invalid for detecting structural degradation of the under-test sections comprises:

performing an 2D graphic comparison process to compare a frequency-domain waveform of the captured under-test acoustic signals with characteristic curves of historical frequency-domain waveforms with different prototypes of structural degradation provided by historical data to select a plurality of frequency segments from the captured under-test acoustic signals, wherein characteristic curves of the selected frequency segments match portions of the characteristic curves in the historical frequency-domain waveforms;

applying a support vector machine (SVM) learning using a classification and/or a regression analysis to remove portions of the captured under-test acoustic signals from the selected frequency segments according to a predetermine threshold value;

reunion remaining portions of the selected frequency segments.

11. The acoustic resonance diagnostic system for detecting structural degradation according to claim 8 , further comprising a vibration generator used to provide at least one sound vibration on the under-test structure.

12. The acoustic resonance diagnostic system for detecting structural degradation according to claim 11 , wherein the at least one sound vibration has an amplitude value greater than that of the plurality of under-test sound wave signals previously captured.

13. The acoustic resonance diagnostic system for detecting structural degradation according to claim 11 , wherein the at least one sound vibration has a frequency to form a standing wave in the under-test structure.

14. The acoustic resonance diagnostic system for detecting structural degradation according to claim 13 , wherein the frequency of the standing wave is estimated using the following equation:

f

n

=

2

⁢

nV

L

;

wherein f n is the frequency of the standing wave; L is a distance between a sensing unit used to capture the plurality of under-test sound wave signals and a vibration generator for providing the at least one sound vibration; V is speed of sound; and n is a mode number of the standing wave.

15. The acoustic resonance diagnostic system for detecting structural degradation according to claim 13 , further comprising a human-machine interface for identifying a leakage position by referring to a relationship between a sound pressure change rate of the at least one sound vibration and an unit wavelength of the standing wave.

Assignments (1)

Priority Claims (1)

Continuity (3)

References Cited (32)