CN-122017016-A - Construction method of acoustic time sequence image and application of acoustic time sequence image in metal plate crack detection

CN122017016ACN 122017016 ACN122017016 ACN 122017016ACN-122017016-A

Abstract

The invention discloses a construction method of an acoustic time sequence image and application thereof in crack detection of a metal plate, wherein excitation is applied to the metal plate, and acoustic array non-contact measurement surface sound field information is utilized to obtain sound pressure sampling sequences at a plurality of spatial positions; the method comprises the steps of segmenting a sound pressure sequence according to a preset time interval, combining a sensor space arrangement relation, mapping sound pressure data of each time interval into a two-dimensional sound pressure amplitude distribution diagram, arranging the two-dimensional sound pressure amplitude distribution diagram according to time sequence, constructing and forming an acoustic time sequence image for reflecting the change of the sound pressure distribution along with time and space, and realizing metal plate crack detection based on the acoustic time sequence image. According to the invention, through time sequence recombination and structural expression of sound field data, the data scale and the calculation complexity are obviously reduced while key information is maintained, and the engineering realizability of the crack detection method is effectively improved.

Inventors

ZHU LU
ZHANG XIAOZHENG
LI JIA
FAN YUFEI
BI CHUANXING
LIU KAI

Assignees

合肥工业大学

Dates

Publication Date: 20260512
Application Date: 20260127

Claims (9)

1. A construction method of an acoustic time sequence image is characterized by comprising the steps of applying excitation on a metal plate, utilizing acoustic array non-contact measurement surface sound field information to obtain sound pressure sampling sequences at a plurality of space positions, segmenting the sound pressure sequences according to a preset time interval, mapping sound pressure data of each time period into a two-dimensional sound pressure amplitude distribution diagram by combining a sensor space arrangement relation, and constructing the two-dimensional sound pressure amplitude distribution diagram to form the acoustic time sequence image in a time sequence mode, wherein the sound pressure amplitude distribution diagram is used for reflecting the change of sound pressure distribution along with time and space.
2. The method for constructing the acoustic time sequence image according to claim 1, wherein the method comprises the following steps: step 1, acoustic signal acquisition, namely exciting a metal plate, and carrying out non-contact measurement on a sound field on the surface of the metal plate by utilizing an acoustic array to acquire sound pressure sampling sequences at a plurality of spatial positions in the same measuring plane; Step 2, time segmentation processing is carried out on the sound pressure sampling sequence according to a preset time interval to obtain sound pressure sampling signals in a plurality of time segments; step 3, spatial mapping processing, namely mapping the corresponding sound pressure sampling data in each time section into a two-dimensional sound pressure amplitude distribution diagram by combining the spatial arrangement relation of the sensors in the acoustic array; And 4, image mapping and time sequence arrangement, namely arranging the two-dimensional sound pressure amplitude distribution graphs according to time sequence, and constructing and forming an acoustic time sequence image, wherein the acoustic time sequence image consists of a plurality of two-dimensional sound pressure amplitude distribution graphs arranged in time sequence.
3. The method for constructing the acoustic time sequence image according to claim 2 is characterized in that the time segmentation processing is to segment a sound pressure sampling sequence with high time resolution according to a preset time interval, reconstruct original high sampling rate acoustic data into an acoustic time sequence image formed by two-dimensional sound pressure amplitude distribution graphs corresponding to a plurality of time segments, so that the data scale for crack detection is reduced from time sequence level data to image sequence level data, the data storage requirement is obviously reduced while key space-time characteristic information of a sound field is reserved, the engineering realizability and real-time processing capacity of the crack detection method are improved, and the acoustic time sequence image can be constructed in the form of a two-dimensional gray level image, a multi-channel color image or a multi-characteristic channel image.
4. The method for constructing an acoustic time series image according to claim 2, wherein in step 2, the time-division processing is performed as follows: S (T) is used for representing sound pressure sampling signals, the sound pressure sampling signals are time functions corresponding to sound pressure sampling sequences, the total duration is T, and the total duration T is divided into N equal-length time periods so as to Characterizing the start time of the nth time period to The end time of the nth time period is characterized, For the preset time interval(s), The following steps are: ; ; Sound pressure data sequence in nth time period The method comprises the following steps: , ; Wherein, the 。
5. The method for constructing an acoustic time series image according to claim 2, wherein in step 3, the spatial mapping process is performed as follows: To be used for The representation being located at spatial coordinates Sound pressure data acquired by sound pressure sensor at the position, namely the sound pressure data Corresponding sound pressure amplitude characteristics Obtained from formula (1) or formula (2): (1); (2)。
6. The method of constructing an acoustic time series image according to claim 5, wherein the image mapping and time series arrangement are performed in step 4 by a method of characterizing the sound pressure amplitude Normalization processing is carried out, and mapping is carried out to pixel points in the two-dimensional image Image channel intensity values of (2) : Wherein the mapping function Is a linear normalized form represented by formula (3): (3); Wherein, the And Respectively representing a reference minimum value and a reference maximum value of the corresponding sound pressure amplitude characteristic, c is an image channel index, The gray scale intensity is a proportionality coefficient and is used for mapping the normalized gray scale intensity to a preset image display range; The mapping function f (), or is in a nonlinear mapping, piecewise mapping or adaptive mapping form; the reference minimum and the reference maximum are determined from a single two-dimensional sound pressure amplitude profile, a plurality of time segments, or a preset statistical range.
7. A crack detection model for crack detection of a metal plate, which is characterized in that the crack detection model is constructed based on the acoustic time sequence image in claim 1 and used for crack detection of the metal plate, and comprises the following components: The local space-time characteristic extraction module is used for carrying out joint characteristic extraction on the acoustic time sequence image in the space dimension and the time dimension to obtain characteristic representation representing the change characteristic of the local sound field; the time sequence feature modeling module is used for performing global dependency modeling on the local space-time features in the time dimension so as to obtain global time sequence feature representation for representing the evolution rule of the sound field; the feature fusion and discrimination module is used for carrying out fusion processing on the local space-time features and the global time sequence features and outputting a detection result of the metal plate crack based on the fused features; the crack detection model is trained through an acoustic time sequence image data set formed by a plurality of acoustic time sequence images, the acoustic time sequence images are used as training samples to be input into the crack detection model, and model parameters are learned and optimized to obtain a parameterized crack detection model for detecting the metal plate cracks; The crack detection model performs feature mapping on the acoustic time sequence image to obtain feature representation for crack discrimination, wherein the feature mapping relation is represented as follows: (4); In the formula (4) Representing an acoustic timing image consisting of a T-frame two-dimensional sound pressure amplitude profile to Representing a nonlinear mapping function for global spatiotemporal feature modeling and temporal feature extraction, And representing the time-space characteristics of the fused sound field.
8. The crack detection model for metal sheet crack detection as set forth in claim 7, wherein the training process of the crack detection model includes: Acquiring an acoustic time sequence image sample containing known crack state and crack position information, and constructing an acoustic time sequence image training data set with crack marking information; Inputting the training data set of the acoustic time sequence image into a crack detection model, and obtaining a crack detection prediction result through forward feature mapping; Constructing a model training loss function based on the crack detection prediction result and the corresponding crack marking information, and carrying out iterative updating on parameters of the crack detection model by adopting a parameter optimization algorithm; and when the model training loss function meets a preset convergence condition, obtaining a parameterized crack detection model for detecting the metal plate cracks.
9. A metal plate crack detection method is characterized in that: Constructing an acoustic time sequence image of the metal plate to be detected according to the method of claim 1, wherein the acoustic time sequence image is the acoustic time sequence image of the metal plate to be detected; inputting the acoustic time sequence image of the metal plate to be detected into a trained parameterized crack detection model of the metal plate to be detected; Performing feature mapping and distinguishing processing on the acoustic time sequence image of the metal plate to be detected by using the parameterized crack detection model of the metal plate to be detected, and outputting a detection result of whether the metal plate to be detected has cracks or not and corresponding crack position information; To be used for And (3) representing a metal plate crack detection result output by the model: ; To be used for Representing an acoustic time sequence image input consisting of a T-frame two-dimensional sound pressure amplitude distribution diagram arranged in time sequence; To be used for Representing the parameters as The parameterized crack detection model comprises a nonlinear mapping structure of local space-time feature extraction and global time sequence dependency modeling.

Description

Construction method of acoustic time sequence image and application of acoustic time sequence image in metal plate crack detection Technical Field The invention relates to the technical field of nondestructive testing and intelligent recognition, in particular to a method for constructing an acoustic time sequence image based on acoustic array acquisition signals and recognizing and positioning metal plate cracks by combining a deep learning model. Background The metal plate has good specific strength, workability and formability, and is widely applied to the manufacture and assembly of various components such as a vehicle body, a ship body, a wing, a shell, a tank body and the like in the field of engineering and machinery manufacturing. In the service process of the metal plate, the metal plate is influenced by factors such as manufacturing defects, external load, impact and vibration, environmental corrosion, fatigue accumulation, structural aging and the like, is easy to generate micro cracks locally, gradually expands under the action of cyclic load, and finally can cause fracture and failure accidents. Therefore, the crack position of the metal plate is timely and reliably identified, and the method provides basis for preventing safety accidents, reducing maintenance cost and optimizing material type selection and process, and has important engineering significance. Among the existing crack detection methods, the manual detection method relies on experience, has low efficiency and is difficult to find early micro defects, the contact detection method is sensitive to surface states and coupling conditions, has limited detection efficiency and is difficult to rapidly cover a large-area structure, and part of the methods have insufficient adaptability to curved surfaces/thin plates or coating members. The acoustic detection method has the advantages of non-contact, high sensitivity and the like, and can identify early cracks under the condition of not damaging the structure. With the development of a data driving method, deep learning is gradually introduced into a nondestructive testing scene, wherein a Convolutional Neural Network (CNN) can replace the traditional manual characteristic and classifier process through end-to-end learning, and the deep learning has better characteristic extraction capability and robustness in a two-dimensional image recognition task. However, in the sheet metal crack detection task, the sound field information contains not only spatial distribution but also changes continuously with time and excitation frequency. The two-dimensional convolution only extracts the features on the plane, so that the dynamic change rule of the sound field in the time dimension cannot be effectively captured, and the time sequence response features under different frequency excitation are difficult to characterize. Therefore, when the sound pressure cloud image is simply processed by using the 2D CNN, key time sequence related information is lost, and the recognition effect is remarkably reduced under the condition that cracks are small or the sound field difference is weak. In order to consider the space and time dimension characteristics at the same time, the 3D CNN is utilized to realize joint modeling of the dynamic characteristics through convolution operation in a space-time domain, so that the problem that the 2D CNN ignores time information is solved well. But 3D CNNs still rely on fixed receptive field local convolution operations in nature, which have shortcomings in handling long timing signals or cross-regional global dependencies. Particularly, 3D CNN has low sensitivity to global changes in sound field at different excitation frequencies, and it is difficult to effectively capture remote related information in the acoustic wave propagation process. Disclosure of Invention The invention aims to avoid the defects of the prior art, provides a construction method of an acoustic time sequence image and application of the acoustic time sequence image in metal plate crack detection, aims to realize high-precision positioning and classification of metal plate cracks under a non-contact condition, and is particularly suitable for micro crack identification under a multi-frequency excitation condition. The invention adopts the following technical scheme for realizing the purpose: The method for constructing the acoustic time sequence image is characterized by comprising the steps of applying excitation on a metal plate, utilizing acoustic array non-contact measurement surface sound field information to obtain sound pressure sampling sequences at a plurality of spatial positions, segmenting the sound pressure sequences according to a preset time interval, combining a sensor space arrangement relation, mapping sound pressure data of each time period into a two-dimensional sound pressure amplitude distribution diagram, and constructing the two-dimensional sound pressure amplitude distribution