CN-121980335-A - Model-based multisource integrated detection data characterization and evaluation method

CN121980335ACN 121980335 ACN121980335 ACN 121980335ACN-121980335-A

Abstract

The invention relates to a model-based multisource integrated detection data characterization and evaluation method, which comprises the steps of monitoring positions of damage on structural products by adopting a plurality of multisource detection systems, wherein the multisource detection systems comprise a plurality of sound field detection systems, a multisensory remote sensing system, a strain measurement system, an array eddy current detection system and a laser ultrasonic detection system, carrying out standardization processing on original signals to obtain a standardized data set, carrying out feature selection to obtain effective data of each multisource detection system to form an optimal feature subset, carrying out nonlinear mapping processing on the optimal feature subset, and mapping an original high-order feature space to a low-dimensional embedded space to obtain a low-dimensional feature matrix. The invention is beneficial to realizing the high-efficiency integration of multi-source data and the improvement of the performance interpretation capability.

Inventors

TANG XIAOJUN
WANG WEN
SONG WENLI
ZHAO ZILE
LI XU
Yan Xushuai
XIN LIANG
XU LIXIA
YANG YAODONG
TIAN XIN
ZHU XIAOXI
LIU YANG
Lang Jinchi
REN HONGTAO

Assignees

北京卫星制造厂有限公司

Dates

Publication Date: 20260505
Application Date: 20251231

Claims (7)

1. The model-based multi-source integrated detection data characterization and evaluation method is characterized by comprising the following steps of: 1) Each type of the original signals output by the multi-source detection system is respectively marked as a characteristic dimension, and the characteristic dimensions of all the multi-source detection systems are numbered according to any sequence; The multi-source detection system comprises a plurality of sound field detection systems, a multi-vision remote sensing system, a strain measurement system, an array eddy current detection system and a laser ultrasonic detection system; 2) Normalizing the original signal to obtain a normalized data set ; 3) Feature selection is carried out to obtain an optimal feature subset composed of effective data of each multi-source detection system ; 4) Adopting a nonlinear dimension reduction method to obtain the optimal feature subset in the step 3) Performing nonlinear mapping processing, and mapping from an original high-order feature space to a low-dimensional embedded space to obtain a low-dimensional feature matrix 。
2. The method for characterizing and evaluating model-based multi-source integrated test data as defined in claim 1, wherein, The sound field detection system obtains the position coordinates of damage through sensing the acoustic emission data of stress release or structural damage, and the output original signal type includes vector sound pressure waveform Time domain amplitude, frequency of (a); The multi-vision remote sensing system is used for acquiring remote sensing image data, performing three-dimensional deformation monitoring and surface defect identification to obtain position coordinates of damage occurrence, wherein the output original signal type comprises an image frame sequence Frame rate and image matrix of (a); the strain measurement system is used for measuring the surface strain response by collecting voltage data of the strain response of the key part to obtain the position coordinate of the damage occurrence, and the output original signal type comprises voltage signals Time domain amplitude and frequency of (a); The array eddy current detection system comprises detecting impedance data of conductive material, identifying surface defect to obtain position coordinates of damage, and outputting original signal type including impedance change curve Amplitude change rate, impedance modulus and phase angle characteristics; The laser ultrasonic detection system is used for identifying the defects in the detected structure by detecting echo data in the structure to obtain the position coordinates of the damage occurrence, and the output original signal type comprises echo signals Echo peak amplitude, time of flight and envelope curve energy distribution characteristics.
3. The method for characterizing and evaluating model-based multi-source integrated test data according to claim 2, wherein the step 3) is a method for obtaining valid data of each multi-source test system, specifically: 31 Obtaining a feature subset P in a feature selection model of the shoal optimization algorithm; 32 Solving the order of optimization objective function When the value is minimum, the corresponding feature subset P is combined into a final output solution to form an optimal feature subset 。
4. The method for characterizing and evaluating model-based multi-source integrated detection data according to claim 3, wherein the feature subset P in the feature selection model of the shoal optimization algorithm is defined as follows: Wherein, the ; At the position of Representing step 2) of the normalized dataset All normalized data of the jth feature dimension in the list are selected into the current feature subset; At the position of When the data of all sampling points representing the j-th feature dimension are not selected into the current feature subset; j is the number of the feature dimension, j belongs to [1, d ]; the total feature dimension of a single sampling point; Wherein, the Represent the first Normalized data corresponding to the sample points, The total number of sampling points for each multi-source detection system.
5. The model-based multi-source integrated test data characterization and evaluation method according to claim 4, wherein the objective function is optimized The method comprises the following steps: Wherein, the Calculating the deviation between the predicted output value of the model and the real standardized data for the model deviation by using the loss function R; For model deviations The feature dimensions of the corresponding feature subset, ≤d。
6. The model-based multi-source integrated test data characterization and evaluation method according to any one of claims 2-5, wherein step 4) is a low-dimensional feature matrix The method specifically comprises the following steps: wherein H is a single hidden layer feedforward structure mapping function; the optimal feature subset obtained in the step 3); representing the preserved effective feature dimension; L is the number of hidden layer nodes, is a positive integer greater than 1 and less than the total number n of sampling points of the multi-source monitoring system; Is a bias term, which is a random number in [ -1, 1]; is an activation function; γ is the output layer weight, which is obtained by minimizing the loss function R.
7. The model-based multi-source integrated detection data characterization and evaluation method according to claim 6, wherein the loss function R is specifically: Wherein T is an output target label; Lambda is the regularization coefficient; 。

Description

Model-based multisource integrated detection data characterization and evaluation method Technical Field The invention belongs to the technical field of structural performance detection and intelligent data analysis, and particularly relates to a model-based multi-source integrated detection data characterization and evaluation method. Background With the rapid development of sensor networks, nondestructive testing, structural Health Monitoring (SHM) and other technologies, the state monitoring means of structural products in the manufacturing and service stages are gradually evolving from traditional single-mode detection to multi-mode fusion detection. Typical technical means include strain gauge, eddy current array sensor, laser ultrasound, acoustic detection, multi-vision remote sensing, etc., which can realize synchronous acquisition of multi-parameter data such as stress release, deformation measurement, crack identification, abnormal sound acquisition, etc. The signal types generated by the detection systems are obviously different, and the detection systems cover various data forms such as vector fields, images, waveforms, sound spectrums and the like, and present typical multi-source heterogeneous data structures. However, the existing multi-source detection system still has the following key problems in terms of data fusion, analysis and evaluation: The multi-source data fusion capability is insufficient. At present, methods such as feature level fusion and decision level fusion are often adopted, but the time sequence relevance and the space mapping structure of data are often ignored in the processing process, and a physical relevance mechanism based on a structural model is lacked, so that the fusion result has serious information loss and insufficient precision. The evaluation method is single and lacks a quantification mechanism. The traditional structure detection data evaluation relies on expert experience, univariate statistics or two-dimensional chart display, and is difficult to support dynamic joint evaluation of high-dimensional and multi-parameter variables, and reproducibility and engineering applicability are lacking. In summary, the current structure product has the parallel acquisition capability of multiple sensors in the test and detection process, but the problems of weak fusion mechanism, lag of evaluation algorithm and the like still exist in a data processing link. The defects severely restrict the precision, efficiency and practicability of test results, and a new method for data processing and evaluation of fusion modeling driving and real-time interactive response is needed to be provided so as to realize efficient integration of multi-source data and improvement of performance interpretation capability. Disclosure of Invention Aiming at the problems of data fracture, incomparable information and discontinuous state evaluation in the manufacturing, assembling and service stages of the existing structure health monitoring technology, the invention provides a model-based multisource integrated detection data characterization and evaluation method, and the state monitoring of the whole process from manufacturing to service of the structure is realized by constructing a cross-stage multisource perception network, a standardized processing mechanism and a unified fusion algorithm model. The technical scheme of the invention is as follows: the model-based multisource integrated detection data characterization and evaluation method comprises the following steps: 1) Each type of the original signals output by the multi-source detection system is respectively marked as a characteristic dimension, and the characteristic dimensions of all the multi-source detection systems are numbered according to any sequence; The multi-source detection system comprises a plurality of sound field detection systems, a multi-vision remote sensing system, a strain measurement system, an array eddy current detection system and a laser ultrasonic detection system; 2) Normalizing the original signal to obtain a normalized data set ; 3) Feature selection is carried out to obtain an optimal feature subset composed of effective data of each multi-source detection system; 4) Adopting a nonlinear dimension reduction method to obtain the optimal feature subset in the step 3)Performing nonlinear mapping processing, and mapping from an original high-order feature space to a low-dimensional embedded space to obtain a low-dimensional feature matrix。 Preferably, the sound field detection system obtains the position coordinates of damage occurrence by sensing acoustic emission data when stress is released or the structure is damaged, and the output original signal type comprises vector sound pressure waveformTime domain amplitude, frequency of (a); The multi-vision remote sensing system is used for acquiring remote sensing image data, performing three-dimensional deformation monitoring and surface defect identifica