CN-121980334-A - Early damage identification and intelligent early warning method based on real-time networking strain

CN121980334ACN 121980334 ACN121980334 ACN 121980334ACN-121980334-A

Abstract

The early damage identification and intelligent early warning method based on real-time networking strain comprises the steps of arranging a plurality of sensors to collect strain information, carrying out normalization, temperature drift compensation and standardization treatment on the strain information to obtain a fusion data time sequence corresponding to a sensitive area, carrying out statistics on characteristics to obtain damage index results of the characteristics, obtaining corresponding damage labels, constructing a neural network model for damage state classification, training to obtain a neural network model for completing training, carrying out online damage identification, obtaining damage grade probability and carrying out early warning interpretation. The invention can realize real-time identification, intelligent early warning and state evaluation of the damage of the complex structure, thereby improving the safety and reliability of the test.

Inventors

HUI TIANLI
SONG WENLI
WANG JIAZE
REN HONGTAO
LI QIANG
LI DASONG
TANG XIAOJUN
JIA MINTAO
YANG YAODONG
Lang Jinchi
XIN LIANG
LI WEIYU
HUO SHULIN
GUAN DONGFANG

Assignees

北京卫星制造厂有限公司

Dates

Publication Date: 20260505
Application Date: 20251231

Claims (8)

1. The early damage identification and intelligent early warning method based on real-time networking strain is characterized by comprising the following steps: 1) A plurality of sensors are distributed in the sensitive area and used for collecting strain information; 2) Carrying out normalization, temperature drift compensation and standardization processing on the strain information to obtain a fusion data time sequence corresponding to the sensitive area ; 3) Fused data time series corresponding to sensitive area Statistical characteristics including fusion of time sequence of data Strain mean, strain standard deviation, and overall rate of change of (a) Dynamic feature extraction Hurst index ; 4) Performing damage grade division on the damage index result to obtain a corresponding damage label ; 5) Constructing a neural network model for classifying the damage states by adopting a three-layer bidirectional recurrent neural network; 6) The damage index result obtained in the step 3) and the damage label obtained in the step 4) are processed Dividing the neural network model into a training set and a verification set, and training the neural network model constructed in the step 5) to obtain a trained neural network model ; 7) Utilizing the trained neural network model of step 6) Carrying out online damage identification on the structural product to obtain the probability of damage grade; 8) And D), carrying out early warning interpretation on the damage condition of the structural product by the probability of the damage level output in the step 7).
2. The method for early damage identification and intelligent early warning based on real-time networking strain according to claim 1, wherein step 2) obtains a fused data time sequence corresponding to the sensitive area The method of (1) comprises the following steps: 21 Zero normalization processing is carried out on the acquired strain information to obtain strain information after zero normalization processing ; Wherein, the Numbering the sensors; represent the first The individual sensors are at the moment Is provided for the strain reading of (a), Is the first Nominal values of the individual sensors; 22 Strain information after zero bit normalization processing Performing temperature drift compensation processing to obtain strain information of the temperature drift compensation processing ; Wherein, the As a function of the temperature-sensitive coefficient, The change amount between the temperature of the sensor mounting position and the reference temperature value is used; 23 The strain information of the temperature drift compensation process is standardized to obtain a fusion data time sequence corresponding to the sensitive area ; Wherein, the In order to normalize the data it is, In order to weight the fusion coefficients, Is the total number of sensors.
3. The early damage identification and intelligent early warning method based on real-time networking strain according to claim 2, wherein the data time sequence is fused in step 3) Strain mean, strain standard deviation, and overall rate of change of (a) Dynamic feature extraction Hurst index The method comprises the following steps: Wherein, the For the ith sensor to be extremely poor in strain in the window, 。
4. The method for early damage identification and intelligent pre-warning based on real-time networking strain of claim 3, wherein in step 4) Wherein, the method comprises the steps of, =1 Indicates a lossless state, The expression =2 indicates a mild injury, =3 Represents severe injury; Determining damage tags The method of (1) comprises the following steps: the slight damage corresponds to the peak strain increment of less than 5 percent and the strain standard deviation The change rate is less than 30%, the integral change rate of the power spectral density is less than 3%, and the Hurst index Less than 0.7, the high-frequency area spectrum energy ratio of the waveform signal is less than 0.15; the moderate damage corresponds to the peak strain increment of 5-20% and the strain standard deviation 30% -100% Rise over initial state, hurst index Increasing to 0.7-0.85, and the spectrum energy ratio of the waveform signal in the high frequency region is between 0.15 and 0.3; the serious damage corresponds to the peak strain increment exceeding 20 percent and the strain standard deviation A rate of change of greater than 100%, a Hurst index The frequency spectrum energy ratio of the waveform signal in the high frequency region is more than 0.85 and more than 0.3; If the damage index result with the characteristics belongs to different damage grades, defining a damage label by using the most serious damage grade 。
5. The method for early damage identification and intelligent pre-warning based on real-time networking strain according to any one of claims 1 to 4, wherein in step 5), the input of the neural network model is: Wherein, the Representing j-th dimension input feature at time t, d being input feature dimension, the corresponding features being strain mean, strain standard deviation, overall rate of change And dynamic feature extraction Hurst index When d is equal to 4.
6. The early damage identification and intelligent early warning method based on real-time networking strain of claim 5, wherein the output of the neural network model is: Wherein, the To predict Is a probability of (2).
7. The method for early damage identification and intelligent pre-warning based on real-time networking strain of claim 6, wherein the loss function of the neural network model is cross entropy.
8. The early damage identification and intelligent early warning method based on real-time networking strain according to claim 7, wherein the method for early warning interpretation in step 8) specifically comprises the following steps: first-level early warning of 0.7 0.85, Slight damage; Secondary early warning 0.85 0.95, Moderate injury; three-stage early warning: 0.95, severe injury. Wherein, the Is that Is a probability of (2).

Description

Early damage identification and intelligent early warning method based on real-time networking strain Technical Field The invention belongs to the technical field of structural health monitoring and mechanical test intellectualization, and particularly relates to an early damage identification and intelligent early warning method based on a real-time networking strain signal, which is particularly suitable for application scenes of real-time monitoring and damage early warning on structural state change in a mechanical test process of a large complex structure under a dynamic load condition. Background In the structural mechanical test process, especially in the structure bearing multiple source complex loads (including static pressure, impact load, alternating fatigue and the like), the strain distribution state of a local area of the material can be microscopically changed along with the load action, so that initial damage such as microcracks, interface separation and the like can be formed. If the initial damage is not identified and controlled in time, the initial damage is easily expanded into structural failure events such as weld cracking, shell perforation or stress overrun yield, and the test safety and the accuracy of structural service assessment are seriously affected. Currently, structural damage detection mainly relies on nondestructive detection methods such as visual inspection, ultrasonic detection, radiation detection, and acoustic emission techniques. These methods generally require interrupting the test for detection or analyzing the results after the test, lack of real-time and dynamic continuity, and cannot effectively track the response change of the structure in the loading process, and often miss the window period of early damage early warning. Although partial technology introduces strain sensor arrays to monitor strain response of the structure, most of the existing deployment modes adopt fixed point location and sparse distribution strategies, so that full-coverage monitoring of key stress areas in multi-area and multi-scale complex structures (such as multi-cavity shells and large-scale welding cylinders) is difficult to realize, and typical damage evolution characteristics such as crack initiation paths, sudden change responses caused by stress concentration and the like cannot be captured. On the other hand, the traditional strain data analysis method is mostly based on static data or an offline statistical model, and cannot process the time sequence, non-stationarity and mutation characteristics of the structural strain response. These limitations make it difficult for lesions to be effectively identified at an early stage, resulting in a large model early warning lag or identification error. In addition, most of current damage identification models adopt rule-based setting or linear regression strategies, lack deep pattern mining and modeling capabilities for multi-dimensional dynamic strain signals, and cannot meet the modeling requirements of nonlinear evolution processes under complex damage mechanisms. In particular in stress coupling and multi-scale strain responses, systematic intelligent recognition and response have not yet formed a closed loop. Thus, the prior art has key shortcomings in: 1. the space monitoring resolution ratio is insufficient, the number of strain sensors is limited, the layout density is low, a high risk area cannot be covered, and global state sensing is difficult to form; 2. The dynamic modeling capability is insufficient, key time sequence characteristics such as abrupt change of strain rate, spectrum drift and the like in the structure loading process cannot be effectively extracted; 3. Most models rely on manual threshold setting, lack of prediction capability based on data driving, and early warning is insensitive; 4. The feedback and self-adaptation capability of the system is weak, and the recognition strategy cannot be automatically adjusted according to the new damage mode due to the lack of a real-time model correction and feedback mechanism. In view of the foregoing, there is an urgent need to develop a novel structural health monitoring method with high-density networking strain monitoring capability, a fused time sequence intelligent modeling algorithm, and real-time identification and hierarchical response to structural damage states, so as to meet the requirements of evaluating the safety and stability of a complex structure in a test process. The existing structural damage detection and early warning technology has a plurality of defects in the mechanical test process of coping with complex structures, and is mainly characterized in the following aspects: the existing data analysis method is mainly based on static strain values for evaluation, lacks modeling capability of dynamic change of a strain signal along with loading time, and cannot effectively identify early abnormal behaviors in the structure response process; The exist