# US 2026/0202382 A1 · Method and system for detecting anomalies of mechanical components, in particular aircraft components, by classifying spectrograms of acoustic signals

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

- Number: US 2026/0202382 A1
- Document ID: US20260202382A1
- Status: Published application
- Kind: A1
- Application number: 19/131,365
- Filed: 2023-11-23
- Published: 2026-07-16
- Earliest claimed: 2022-11-28
- Assignees: Leonardo S.P.A.
- Applicants: Leonardo S.P.A.
- Inventors: Francesco CALABRÒ; Federico COLANGELO; Roberto SANGUINI
- Web page: https://patentgrep.com/patent/US20260202382A1

## CPC

- G01N 29/045: {by imparting shocks to the workpiece and detecting the vibrations or the acoustic waves caused by the shocks (measuring resonant frequency G01H13/00; measuring strength properties by application of mechanical stress G01N3/00)}
- G01N 29/4445: {Classification of defects}
- G01N 29/4481: {Neural networks}
- G01N 2291/0289: Internal structure, e.g. defects, grain size, texture
- G01N 2291/2694: Wings or other aircraft parts
- 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)

## Abstract

A method implemented by means of computer so as to detect anomalies of an unknown component, including determining a plurality of zones of the unknown component and performing at least once the steps of: generating a spectrogram relative to an acoustic signal generated by striking a portion of a zone of the unknown component; between a plurality of binary classifiers each one relative to a corresponding zone of the unknown component, selecting the binary classifier relative to the struck zone, each one of said binary classifiers classifying spectrograms relative to acoustic signals generated by striking the corresponding zone on respective two classes indicative of a spectrogram relative to an acoustic signal generated by striking an undamaged version or a damaged version of the corresponding zone respectively; by means of the selected binary classifier, performing a classification of the spectrogram in one of the respective two classes; and detecting the presence of an anomaly in the struck zone of the unknown component, on the basis of the classification performed by the selected binary classifier.

## Claims

### Claim 1

A method implemented by a computer for detecting anomalies of an unknown component, comprising determining a plurality of zones of the unknown component and carrying out at least once the steps of:
generating a spectrogram relative to an acoustic signal generated by striking a portion of a zone of the unknown component;
among a plurality of binary classifiers each one relative to a corresponding zone of the unknown component, selecting the binary classifier relative to the struck zone, each one of said binary classifiers being configured to classify spectrograms relative to acoustic signals generated by striking the corresponding zone on two respective classes indicative of a spectrogram relative to an acoustic signal generated by striking an undamaged version or a damaged version of the corresponding zone respectively;
through the selected binary classifier, performing a classification of said spectrogram in one of the respective two classes; and
detecting the presence of an anomaly in said struck zone of the unknown component, on the basis of the classification performed by the selected binary classifier.

### Claim 2 (depends on 1)

The method according to claim 1, wherein each binary classifier has been trained in a supervised manner on the basis of:
respective first training spectrograms, relative to acoustic signals generated by striking portions of the corresponding zones of training components identical to the unknown component and without any damage, said first training spectrograms being associated with the corresponding first class; and
respective second training spectrograms, relative to acoustic signals generated by striking portions of the corresponding zones of training components identical to the unknown component and with damages in said corresponding zones and/or by striking portions of zones different from the corresponding zone of training components identical to the unknown component and without any damage and/or by striking portions of zones different from the corresponding zone of training components identical to the unknown component and with damages in said zones different from the corresponding zone, said second training spectrograms being associated with the corresponding second class.

### Claim 3 (depends on 1)

The method according to claim 1, wherein selecting the binary classifier relative to the struck zone comprises:
classifying, by means of a multiclass classifier, the spectrogram in a corresponding class among a plurality of zone classes equal to the number of zones of the unknown component, each one of said zone classes being indicative of a spectrogram relative to an acoustic signal generated by striking a corresponding zone; and
selecting the binary classifier on the basis of the classification performed by the multiclass classifier.

### Claim 4 (depends on 3)

The method according to claim 3, wherein the multiclass classifier has been trained by performing the steps of:
determining a plurality of subregions of the unknown component; and subsequently
training the multiclass classifier on the basis of a set of training spectrograms relative to acoustic signals generated by striking portions of the subregions of training components identical to the unknown component and without any damage, each training spectrogram of said set being associated with a corresponding subregion class indicative of the subregion to which the training spectrogram refers, so that the multiclass classifier is configured to perform classifications on a number of subregion classes equal to the number of subregions, each one of said subregion classes being indicative of a spectrogram relative to an acoustic signal generated by striking the corresponding subregion; and subsequently
defining the zone classes so that each zone class is identical to a corresponding subregion class or is indicative of a corresponding set of subregion classes, and subsequently configuring the zone classifier so that it performs classifications on said plurality of zone classes.

### Claim 5 (depends on 4)

The method according to claim 4, wherein said step of defining the zone classes comprises:
classifying through the zone classifier a plurality of test spectrograms relative to acoustic signals generated by striking portions of the subregions of training components identical to the unknown component and without any damage, so that each test spectrogram is classified in a corresponding subregion class;
calculating a confusion matrix of the classifications of the test spectrograms; and
on the basis of the confusion matrix, detecting the presence of sets of two or more subregions such that the test spectrograms relative to said two or more subregions have been classified in a confused manner between one another in a manner that respects a threshold condition; and
for each detected set of subregions, aggregating the corresponding subregion classes so as to form a corresponding zone class; and
for each subregion which does not belong to any detected set of subregions, setting a corresponding zone class equal to the subregion class.

### Claim 6 (depends on 4)

The method according to claim 4, wherein, in the respective training, each binary classifier has been initialized on the basis of the zone classifier.

### Claim 7 (depends on 6)

The method according to claim 6, wherein the zone classifier and the binary classifiers are convolutional neural networks, each one of which comprises a respective feature extraction stage; and wherein the binary classifiers have been initialized so that the respective feature extraction stages are identical to the feature extraction stage of the zone classifier.

### Claim 8 (depends on 1)

A method for detecting anomalies comprising the steps of:
performing the method implemented by a computer according to claim 1;
performing said strike of a portion of a zone of the unknown component;
and wherein generating a spectrogram depending on the acoustic signal comprises:
acquiring the acoustic signal; and
calculating the spectrogram on the basis of the acquired acoustic signal.

### Claim 9 (depends on 8)

The method for detecting anomalies according to claim 8,
wherein said strike is performed periodically.

### Claim 10 (depends on 1)

A processing system comprising means configured to perform the method according to claim 1.

### Claim 11 (depends on 10)

A system comprising:
the processing system according to claim 10;
a striking device configured to mechanically strike single portions of zones of the unknown component, so as to generate corresponding acoustic signals; and
a microphone, coupled to the processing system and configured to acquire the acoustic signals.

### Claim 12 (depends on 1)

A computer program comprising instructions that, when the program is performed by a computer, cause the computer to perform the method according to claim 1.

### Claim 13 (depends on 12)

A computer medium readable by a computer, on which the computer program is stored according to claim 12.
