CN114529548B - Mechanical part stress corrosion detection method - Google Patents

Mechanical part stress corrosion detection method Download PDF

Info

Publication number
CN114529548B
CN114529548B CN202210434064.7A CN202210434064A CN114529548B CN 114529548 B CN114529548 B CN 114529548B CN 202210434064 A CN202210434064 A CN 202210434064A CN 114529548 B CN114529548 B CN 114529548B
Authority
CN
China
Prior art keywords
sample
scc
samples
time
hydrogen evolution
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202210434064.7A
Other languages
Chinese (zh)
Other versions
CN114529548A (en
Inventor
王耀琼
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nantong Zhongkuang Metal New Material Co ltd
Original Assignee
Nantong Zhongkuang Metal New Material Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Nantong Zhongkuang Metal New Material Co ltd filed Critical Nantong Zhongkuang Metal New Material Co ltd
Priority to CN202210434064.7A priority Critical patent/CN114529548B/en
Publication of CN114529548A publication Critical patent/CN114529548A/en
Application granted granted Critical
Publication of CN114529548B publication Critical patent/CN114529548B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00Image analysis
    • G06T7/0002Inspection of images, e.g. flaw detection
    • G06T7/0004Industrial image inspection
    • G06T7/0008Industrial image inspection checking presence/absence
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/22Matching criteria, e.g. proximity measures
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00Image enhancement or restoration
    • G06T5/20Image enhancement or restoration using local operators
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T5/00Image enhancement or restoration
    • G06T5/70Denoising; Smoothing
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00Image analysis
    • G06T7/10Segmentation; Edge detection
    • G06T7/13Edge detection
    • GPHYSICS
    • G16INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16CCOMPUTATIONAL CHEMISTRY; CHEMOINFORMATICS; COMPUTATIONAL MATERIALS SCIENCE
    • G16C20/00Chemoinformatics, i.e. ICT specially adapted for the handling of physicochemical or structural data of chemical particles, elements, compounds or mixtures
    • G16C20/10Analysis or design of chemical reactions, syntheses or processes
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/20Special algorithmic details
    • G06T2207/20024Filtering details
    • G06T2207/20032Median filtering
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/30Subject of image; Context of image processing
    • G06T2207/30108Industrial image inspection
    • G06T2207/30136Metal

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Chemical & Material Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Evolutionary Biology (AREA)
  • Artificial Intelligence (AREA)
  • Analytical Chemistry (AREA)
  • Chemical Kinetics & Catalysis (AREA)
  • Crystallography & Structural Chemistry (AREA)
  • Computing Systems (AREA)
  • Quality & Reliability (AREA)
  • Testing Resistance To Weather, Investigating Materials By Mechanical Methods (AREA)

Abstract

The invention relates to a method for detecting stress corrosion of a mechanical part, which comprises the following steps: in the primary batch test of one material, measuring a plurality of materials with different stress intensities by using a constant load method, and constructing SCC (SCC) occurrence time and hydrogen evolution characteristic sequence under constant load; analyzing the confidence degrees of the hydrogen evolution sequences of different test groups, and determining the sequence similarity among the samples so as to train the TCN; the characteristics of the corrosion process are reflected by analyzing the hydrogen evolution characteristics of the image, the future corrosion rate is predicted, and the manual judgment is assisted to judge whether to start key observation.

Description

Method for detecting stress corrosion of mechanical part
Technical Field
The invention relates to the field of artificial intelligence, in particular to a method for detecting stress corrosion of a mechanical part.
Background
The constant load method is the same as a smooth sample in an evaluation method that the sample loading load is kept unchanged in the test process, also belongs to a pass/fail evaluation method, and can be used for acceptance of material qualification. One end of the specimen was fixed and a constant tensile static load was applied to the other end, and then the specimen was immersed in a corrosive medium and the time of SCC occurrence was recorded. Since the stress level continuously increases after the crack is initiated, and the crack accelerates the propagation, information such as the crack propagation rate cannot be obtained, the time cost for observing the constant load SCC is high, and the crack propagation information cannot be observed in advance. The constant load method is usually used for carrying out a plurality of materials or researching stress corrosion sensitivity tests of a plurality of variables in a high-pressure environment, and the cost is higher.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention adopts the following technical scheme:
a method for detecting stress corrosion of a mechanical part comprises the following steps:
the method comprises the following steps: in the primary batch test of a material, measuring a plurality of materials with different stress intensities by using a constant load method, and constructing SCC occurrence time and a hydrogen evolution characteristic sequence under constant load;
step two: analyzing the confidence degrees of the hydrogen evolution sequences of different test groups, and determining the sequence similarity among the samples so as to train the TCN;
step three: the TCN is used for predicting the occurrence time of SCC in subsequent material tests, thereby reminding a user to observe the corrosion condition of the metal sample in advance.
Further, the step one is specifically as follows: the method comprises the steps of using a constant load measuring device to measure the constant load SCC time, obtaining readings of a stressometer and an image sequence of a hydrogen evolution process at the same time, applying prestress to the SCC constant load measuring device which consists of a material to be measured and fixed ends at two ends, immersing the SCC constant load measuring device into a corrosive solution, simulating the corrosion aging process of metal, observing the occurrence time of a fracture phenomenon, processing the image sequence of defoaming every time to obtain hydrogen evolution characteristic vectors, and constructing the image sequence of each test
Figure 27780DEST_PATH_IMAGE001
As characteristic data for this test.
Further, the second step is specifically as follows: calculating the confidence of the hydrogen evolution bubble sequence, determining the sequence similarity among samples,
using the Kmeans algorithm of K =2, two classification results were obtained based on the above sample distances, with one group having a small intra-cluster evaluation SCC time as a short-term prediction group and the other group as a long-term prediction group.
Further, the third step is specifically: using long and short TCNs for all samples, comparing the prediction residuals of TCNs based on the aforementioned hydrogen evolution characteristics S, and if the S prediction result residual of long TCNs is smaller than that of short TCNs, the sample is considered to be a long-time corroded sample, and the sample may not be observed before SCC occurs in the rest of samples, and vice versa, because the sample is likely to be a sample with a shorter SCC.
The beneficial effects of the invention are:
the characteristics of the corrosion process are reflected by analyzing the hydrogen evolution characteristics of the image, the future corrosion rate is predicted, and the manual judgment is assisted to judge whether to start key observation.
Drawings
FIG. 1 is a schematic of the present invention.
Detailed Description
The invention is described in detail below with reference to the figures and examples.
The method comprises the following steps: in a primary batch test of one material, a plurality of materials are measured with different stress intensities using a constant load method, and SCC occurrence time and hydrogen evolution characteristic sequences under constant load are constructed.
Constant load SCC time measurements were made using a constant load measurement apparatus as in fig. 1, with simultaneous strain gauge readings and hydrogen evolution process image sequences. The SCC constant load measuring device is composed of a material to be measured and two end fixing ends, prestress is applied to the SCC constant load measuring device, the SCC constant load measuring device is immersed in corrosive solution, the corrosion aging process of metal is simulated, and the occurrence time of a fracture phenomenon is observed. Firstly, an implementer uses a testing device similar to that shown in the figure, wherein the testing device comprises a stress meter, a force application mechanism and the like required by the conventional constant load method, and a vibration device is additionally added and used for shaking the measuring device to eliminate bubbles caused by hydrogen evolution reaction. Therefore, the practitioner can operate using the same flow as the conventional constant load test method. The testing device is different from the conventional testing device in that a vibration device is added, and an implementer gives a time interval t based on experience, so that the device can be started when a certain amount of bubbles are attached to the metal surface and do not float upwards to a large extent when a hydrogen evolution reaction occurs on the metal surface, and the attached bubbles are separated from the metal surface. The function is to analyze the rate characteristics of hydrogen evolution through image characteristics. It is therefore necessary to periodically activate the vibration unit during the process in order to defoam the surface-evolving hydrogen bubbles. The vibration unit can be clung to the experimental container, applies vibration in all directions to the container, can load a plurality of experimental containers onto a test bed, and regularly vibrates the mobile test bed to achieve the purpose of defoaming. Before the vibration is started, acquiring a frame of image as a hydrogen evolution characteristic image of the sample
Figure 671251DEST_PATH_IMAGE002
Where n is the nth acquisition. The specific mechanisms and mechanisms for defoaming are common in the art and are not discussed further herein. Thus, an image sequence before each defoaming process is obtained
Figure 6417DEST_PATH_IMAGE003
After shaking of the image
Figure 922289DEST_PATH_IMAGE004
Image at the beginning of test
Figure 1104DEST_PATH_IMAGE005
And SCC time T, wherein N is the collection times before the SCC occurs. And processing the image sequence defoamed each time to obtain a hydrogen evolution characteristic vector. The hydrogen evolution characteristics were extracted for each image sample as follows. This feature is an intensity feature that reflects the rate of hydrogen evolution over a period of time. If it is the first time, based on the initial image obtained after a metal part
Figure 499081DEST_PATH_IMAGE005
On the contrary, the image after the last vibration is based on
Figure 208411DEST_PATH_IMAGE006
For the current nth acquired image
Figure 159050DEST_PATH_IMAGE002
The operation under the figure is performed. Because the observation process time is long, the observable image indexes such as the solution color, the turbidity and the like of the image in the acquisition environment can be changed greatly, and therefore, the frame difference image is constructed by using the following characteristics, so that the hydrogen evolution characteristics are described: firstly, median filtering processing is carried out on an initial image and an image to be processed, and the median filtering processing is used for carrying out simple noise reduction on collected image noise and eliminating the collected image noise obviouslySalt and pepper noise, and the influence of characteristics such as corrosion texture of the metal surface on subsequent evaluation is reduced. The median filter kernel size is preferably 5, and the implementer may fine tune or skip this step depending on the particular image quality and resolution.
The three-channel RGB image is processed as follows:
Figure 41555DEST_PATH_IMAGE007
the edge feature extraction method comprises the following steps of (because the darkest value in three channels is used as the pixel value of a final gray image when the rusty color of metal is possible, the saturation degree of the edge of the bubble is low due to the reflection characteristic of the bubble, so that the influence of the color after the metal corrosion on the judgment of the subsequent bubble is reduced by reducing the response intensity to the color and reducing the influence of the brighter pixel value of the color on the judgment of the subsequent bubble, and the local contrast of the edge pixel of the bubble is improved.) carrying out edge feature extraction on a preprocessed image: and constructing convolution kernels Gx and Gy of the Sobel operator in two directions, performing convolution operation on the original image to obtain edge responses in the x direction and the y direction, and summing the edge responses of the two images to obtain edge responses in all directions. To this end, processed images are obtained separately
Figure 394039DEST_PATH_IMAGE008
,
Figure 336587DEST_PATH_IMAGE009
Since the relationship between the edge strength of the bubbles on the metal surface of the original and the corroded metal surface is that the edge strength of the metal surface of the original is lower than the edge strength of the bubble portion, the hydrogen evolution characteristics of the metal hydrogen evolution are evaluated as follows: a determination is made as to whether the pixel belongs to a bubble,
Figure 457078DEST_PATH_IMAGE010
calculating an evaluation index
Figure 877695DEST_PATH_IMAGE011
Wherein larger S means more bubbles are attached and the hydrogen evolution rate in the current period is larger. Repeating the above process to obtain the final productAfter that, the following processing is carried out on the S:
Figure 350264DEST_PATH_IMAGE012
the ratio of S to the first hydrogen evolution evaluation index was obtained as follows: thus obtaining the hydrogen evolution characteristic S. To this end, each test is constructed
Figure 401397DEST_PATH_IMAGE001
As characteristic data for this test.
Step two: and analyzing the confidence degrees of the hydrogen evolution sequences of different test groups, and determining the sequence similarity among the samples so as to train the TCN. Due to the different SCC characteristics of some metal samples under different stresses during hydrogen evolution:
and calculating the confidence coefficient of the hydrogen evolution bubble sequence, because the hydrogen evolution characteristic S can only reflect the hydrogen evolution rate and the reaction strength to a certain degree, and the SCC time cannot be predicted on the basis of the reaction rate for some samples completely linearly, the TCN needs to be trained to obtain the prediction result of the SCC, and the mode of the sequence of each sample has a certain rule of smooth change, so the type and the importance of the sample can be adjusted on the basis of the change characteristic S. The confidence of the bubble sequence is firstly calculated, and the unreliable observation results in the sequence are eliminated, so that the correlation correctness between the sequence and other sequences is ensured. The confidence coefficient is calculated in the following way: because the corrosion rate in the SCC test process cannot be guaranteed to be increased, the evaluation index S has the following rule: if a compact oxide layer is formed in the metal in the corrosion process, S may become larger first and then smaller finally, and if the compact oxide layer is not formed, S is changed from small to large, so that the rule necessarily accords with three times of trinomial fitting:
Figure 326628DEST_PATH_IMAGE013
substituting S into M, and performing least square fitting to finally obtain a residual R of each S relative to the final M. In this example, samples with large Top-25% residual were discarded and M was refitted. The implementer adjusts the number of samples n in the experiment, and n of the sequence with the minimum n in the embodiment is 20. Finally obtaining the sum R of the residual errors and the model M, because
Figure 550936DEST_PATH_IMAGE014
Has absolute accuracy for bubble characteristics at the initial stage of reaction, so
Figure 612433DEST_PATH_IMAGE015
Conf in the range of [0,1]. Wherein
Figure 83734DEST_PATH_IMAGE016
As a result of the fitting and the fact
Figure 496261DEST_PATH_IMAGE014
R is the sum of the residuals of S relative to M in M fitted after removing the samples with large residuals. Determining sequence similarity among samples, wherein each sample is made of the same metal, and hydrogen evolution rate can be greatly changed at a certain moment due to different stresses, because the essence of the SCC is that stress stretching aggravates oxidation reaction of the metal in a corrosive environment, so that a contact surface of a micro-crack area is continuously enlarged, and therefore a similarity model needs to be established for a corrosion sequence of the metal under each stress to find a group of samples which are obviously accelerated to corrode due to the micro-cracks and further cause the SCC. Here, a similarity model is constructed:
Figure 524260DEST_PATH_IMAGE017
wherein, the PPMCC is the sequence similarity of two S characteristics, if the two S characteristics are in a corresponding relationship, the accelerated corrosion periods of the two S characteristics are similar, the speeds are similar, otherwise, the accelerated corrosion time points caused by the micro-crack area and the specific accelerated phenomenon are different. Trunc is a truncation function, defined herein as truncating to 0 less than 0, and thus the value range of Trucn is [0,1 ]]. The partial expression of the two confidence degrees represents the difference ratio of the two confidence degrees, if the difference ratio is larger, the fitting effect of the two confidence degrees is worse, and the confidence degree value range [0,1 ]]. Thus, a similarity comparison between two samples is completed, and the range of the similarity is converted into the sample distance:
Figure 705842DEST_PATH_IMAGE018
using a Kmeans algorithm of K =2, two classification results are obtained based on the above sample distance, where one group with a small intra-cluster evaluation SCC time is used as a short-term prediction group, and the other group is used as a long-term prediction group. Training samples for TCNs were constructed for both sets of SCC large and small, respectively. The size of the cause restriction window of the TCN is 5, and the implementer should properly adjust the window size according to the actual defoaming frequency. Wherein the temporally large group uses the future 5 th S value in the current sequence as a predicted value, and the temporally small group uses the future 2 nd S value as a predicted value. The actual future nth value is selected by the implementer based on the SCC time of the material to be tested. Wherein the definition of the future predicted value is: for example, when the window of TCN slides to a group with a large time
Figure 833198DEST_PATH_IMAGE019
When, the tag of the TCN at the current time is
Figure 733021DEST_PATH_IMAGE020
The reason for using different future predictors is: if the SCC time is long, it means that the following phenomenon occurs during the corrosion process: the microcrack appears later, and the main factor is that the applied stress is smaller, so the prediction advance of the TCN can be increased. Otherwise, the prediction lead of the TCN is reduced, and a researcher is helped to observe the corrosion condition in advance. The invention uses TCN to predict future value, the mode of training TCN is the former part of input sequence, and the next n values of sequence are used as label, so that TCN can learn the next n predicted values in current sequence mode. Because of the way TCNs are trained, the scale of the network is well known and the present invention is not described in detail. The Loss function of the TCN is mean square error Loss, and because the sample size of the TCN is small, samples need to be amplified, and meanwhile, the distribution of the small samples is ensured to be in accordance with the characteristics of SCC in actual test, the invention adopts a mode that two samples are mixed pairwise and weight is monitored for the Loss: the specific method comprises the following steps: determining a weight Q for each sample in a batch of samples: firstly, picking out samples of the same cluster, and selecting two of the samples to calculate
Figure 299132DEST_PATH_IMAGE021
And mixing based on the ratio of the confidence degrees of the two: the confidence ratio of a and b is calculated in the way
Figure 289216DEST_PATH_IMAGE022
So as to obtain the mixing ratio of the two samples, and because the lengths of the ab and the two samples are different, the length from the longest sample to the shortest sample is cut off to obtain a model
Figure 649790DEST_PATH_IMAGE023
Obtaining a mixed sample
Figure 36909DEST_PATH_IMAGE024
The corresponding sample weight Q is calculated in the same manner as the confidence Conf given above, and when the self is mixed, the weight is 1. In this embodiment, the batch size is 32, and Q corresponding to the entire batch needs to be normalized for the weight Q, so that the sum of all Q is 1, and a scaling coefficient is obtained
Figure 406710DEST_PATH_IMAGE025
Multiply each sample Loss in a batch by E, so that the weight of each sample in the total Loss is not the same:
Figure 234989DEST_PATH_IMAGE026
the purpose of adjusting the Loss weight here is that for the augmented samples, because the similarity of some samples is low, the mixed samples are reasonable to a certain extent, but all samples cannot be guaranteed to be reasonable, so that the mixed correlation of two samples is superposed on the basis of the correlation between clusters to obtain a final weight Q, and normalization is performed on the basis of Q in one batch, so that the under-fitting phenomenon caused by the noise of the samples cannot occur in the training process, and meanwhile, the over-fitting phenomenon caused by too few samples is overcome on the basis of the amplification result of S in each erosion process.
Step three: the TCN is used for predicting the generation time of the SCC in subsequent material tests, so that a user is reminded of observing the corrosion condition of the metal sample in advance. Using long and short TCNs for all samples, comparing the prediction residuals of TCNs based on the aforementioned hydrogen evolution characteristics S, and if the S prediction result residual of long TCNs is smaller than that of short TCNs, the sample is considered to be a long-time corroded sample, and the sample may not be observed before SCC occurs in the rest of samples, and vice versa, because the sample is likely to be a sample with a shorter SCC.
The above embodiments are merely illustrative and should not be construed as limiting the scope of the invention, which is intended to be covered by the claims.

Claims (5)

1. A method for detecting stress corrosion of a mechanical part is characterized by comprising the following steps:
the method comprises the following steps: in the primary batch test of one material, measuring a plurality of materials with different stress intensities by using a constant load method, and constructing SCC (SCC) occurrence time and hydrogen evolution characteristic sequence under constant load;
step two: analyzing the confidence degrees of the hydrogen evolution characteristic sequences of different test groups, and determining the sequence similarity among samples so as to train the TCN;
step three: in a subsequent material test, the occurrence time of the SCC is predicted by using the TCN, so that a user is reminded of observing the corrosion condition of the metal sample in advance;
the first step is specifically as follows: the method comprises the steps of using a constant load measuring device to measure the constant load SCC time, obtaining readings of a stressometer and an image sequence of a hydrogen evolution process at the same time, applying prestress to the SCC constant load measuring device which consists of a material to be measured and fixed ends at two ends, immersing the SCC constant load measuring device into a corrosive solution, simulating the corrosion aging process of metal, observing the occurrence time of a fracture phenomenon, processing the image sequence of defoaming every time to obtain hydrogen evolution characteristic vectors, and constructing the image sequence of each test
Figure DEST_PATH_IMAGE002
As characteristic data of the test; wherein S is a hydrogen evolution characteristic, and T is SCC occurrence time;
the second step is specifically as follows: calculating the confidence coefficient of the hydrogen evolution bubble sequence, determining the sequence similarity among samples to obtain a sample distance, and obtaining two classification results based on the sample distance by using a Kmeans algorithm with K =2, wherein one group with small SCC (continuous content distribution) evaluation time in a cluster is used as a short-term prediction group, and the other group is used as a long-term prediction group;
the third step is specifically as follows: using long and short time TCNs for all samples, comparing the prediction residuals of TCNs based on the aforementioned hydrogen evolution characteristics S, and considering that a sample is a long time corroded sample if the S prediction residual of a long time TCN is smaller than that of a short time, the sample may not be observed before SCC occurs in the rest of samples, and the sample may be observed otherwise because the sample is likely to be a sample with a shorter time SCC.
2. The method for detecting stress corrosion of a mechanical part according to claim 1, wherein the second step further comprises: if a compact oxide layer is formed in the metal corrosion process, the hydrogen evolution characteristic S may become larger firstly and then smaller, and if the compact oxide layer is not formed, S is fitted from small to large by a trinomial form:
Figure DEST_PATH_IMAGE004
(ii) a Substituting S into M, and performing least square fitting to finally obtain the sum R of the residual error of each S relative to the final M; discarding the samples with big Top-25% residual error, and refitting M; n is the number of samples, n of the minimum sequence is 20, and x is an independent variable of a fitting formula; finally obtaining the sum R of the residual errors and a model M,
Figure DEST_PATH_IMAGE006
conf in the [0,1 ] range](ii) a Wherein
Figure DEST_PATH_IMAGE008
As a result of the fitting and the fact
Figure DEST_PATH_IMAGE010
The size of the difference in (a) to (b),
Figure 663464DEST_PATH_IMAGE010
r is the bubble characteristics at the initial stage of reaction, and is the sum of residual errors of S relative to M in the M fitted after the sample with large residual errors is removed; determining sequence similarity among samples, and constructing a similarity model:
Figure DEST_PATH_IMAGE012
wherein PPMCC is the sequence similarity of two S characteristics, if the two S characteristics are in a corresponding relationship, the accelerated corrosion periods of the two S characteristics are similar and the speeds are similar, otherwise, the accelerated corrosion time points caused by the microcrack region are different from the specific accelerated phenomena; trunc is a truncation function defined herein as truncating to 0 less than 0, and thus lying in a [0,1 ] range](ii) a Convert similarity to sample distance:
Figure DEST_PATH_IMAGE014
using a Kmeans algorithm of K =2, two classification results are obtained based on the above sample distances, where one group with a small intra-cluster evaluation SCC time is used as a short-term prediction group, and the other group is used as a long-term prediction group.
3. The method for detecting the stress corrosion of the mechanical part according to claim 2, wherein training samples of the TCN are constructed for two groups of large SCC and small SCC respectively.
4. The method of claim 3, wherein the window size of the CAUSER CONVOLUTION window of the TCN is 5, wherein the group with the larger time uses the future 5 th S value in the current sequence as the predicted value, and the group with the smaller time uses the future 2 nd S value as the predicted value, for example, the group with the larger time, when the window of the TCN slides to the position where the window of the TCN is located
Figure DEST_PATH_IMAGE016
When, the tag of the TCN at the current time is
Figure DEST_PATH_IMAGE018
Determining a lot sizeWeight Q of each of the samples: firstly, picking out samples of the same cluster, and selecting two of the samples to calculate
Figure DEST_PATH_IMAGE020
And mixing based on the proportion of the confidence degrees of the two: the confidence ratio of a and b is calculated in the way of
Figure DEST_PATH_IMAGE022
So as to obtain the mixing ratio of two samples, and since the lengths of the two samples ab are different, the length from the longest sample to the shortest sample is cut off to obtain the model
Figure DEST_PATH_IMAGE024
Obtaining a mixed sample
Figure DEST_PATH_IMAGE026
The corresponding sample weight Q is calculated in the same manner as the confidence Conf given above, and when the self is mixed, the weight is 1.
5. The method of claim 4, wherein the batch size is 32, and the Q corresponding to the entire batch needs to be normalized for the weight Q, so that all Q sums up to 1, resulting in a scaling factor
Figure DEST_PATH_IMAGE028
Multiply each sample Loss in a batch by E, so that the weight of each sample in the total Loss is not the same:
Figure DEST_PATH_IMAGE030
wherein the Loss function is the function of the Loss,
Figure DEST_PATH_IMAGE032
as a function of the loss for the ith sample,
Figure DEST_PATH_IMAGE034
is the scaling factor of the ith sample.
CN202210434064.7A 2022-04-24 2022-04-24 Mechanical part stress corrosion detection method Active CN114529548B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210434064.7A CN114529548B (en) 2022-04-24 2022-04-24 Mechanical part stress corrosion detection method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210434064.7A CN114529548B (en) 2022-04-24 2022-04-24 Mechanical part stress corrosion detection method

Publications (2)

Publication Number Publication Date
CN114529548A CN114529548A (en) 2022-05-24
CN114529548B true CN114529548B (en) 2022-07-15

Family

ID=81627947

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210434064.7A Active CN114529548B (en) 2022-04-24 2022-04-24 Mechanical part stress corrosion detection method

Country Status (1)

Country Link
CN (1) CN114529548B (en)

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1502424A (en) * 2002-11-20 2004-06-09 Posco��ʽ���� Fault diagnosis apparatus and method for hot fine rolling band steel
US7013249B1 (en) * 2001-07-16 2006-03-14 Kinder Morgan, Inc. Method for detecting near neutral/low pH stress corrosion cracking in steel gas pipeline systems
CN109357994A (en) * 2018-10-30 2019-02-19 中国长江电力股份有限公司 The corrosion accelerated test evaluation method of electrical equipment copper product under a kind of function of current
CN111781129A (en) * 2019-04-04 2020-10-16 应急管理部化学品登记中心 High-efficiency metal material corrosion rate measuring device and method
CN214041024U (en) * 2020-11-27 2021-08-24 山东科技大学 Metal corrosion rate measuring device based on hydrogen evolution method
CN113466427A (en) * 2021-07-07 2021-10-01 中国石油大学(华东) Experimental system for testing hydrogen brittleness resistance and service life of material in high-pressure hydrogen environment

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7414395B2 (en) * 2006-03-27 2008-08-19 General Electric Company Method and apparatus inspecting pipelines using magnetic flux sensors

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7013249B1 (en) * 2001-07-16 2006-03-14 Kinder Morgan, Inc. Method for detecting near neutral/low pH stress corrosion cracking in steel gas pipeline systems
CN1502424A (en) * 2002-11-20 2004-06-09 Posco��ʽ���� Fault diagnosis apparatus and method for hot fine rolling band steel
CN109357994A (en) * 2018-10-30 2019-02-19 中国长江电力股份有限公司 The corrosion accelerated test evaluation method of electrical equipment copper product under a kind of function of current
CN111781129A (en) * 2019-04-04 2020-10-16 应急管理部化学品登记中心 High-efficiency metal material corrosion rate measuring device and method
CN214041024U (en) * 2020-11-27 2021-08-24 山东科技大学 Metal corrosion rate measuring device based on hydrogen evolution method
CN113466427A (en) * 2021-07-07 2021-10-01 中国石油大学(华东) Experimental system for testing hydrogen brittleness resistance and service life of material in high-pressure hydrogen environment

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
不同温度加热后镀铁层组织结构和机械性能的研究;黄水兴等;《福州大学学报(自然科学版)》;19860801(第04期);全文 *
利用极化曲线推测中性水溶液中钛合金表面的氧化还原反应机理;姜应律等;《北京科技大学学报》;20040825(第04期);全文 *

Also Published As

Publication number Publication date
CN114529548A (en) 2022-05-24

Similar Documents

Publication Publication Date Title
CN114627111B (en) Textile defect detection and identification device
CN109142514B (en) Defect detection device and method based on pulse eddy current array
CN111879709B (en) Lake water body spectral reflectivity inspection method and device
CN109584225B (en) Unsupervised defect detection method based on self-encoder
CN116665057A (en) River channel water quality monitoring method and system based on image processing
CN115420806A (en) Nondestructive ultrasonic detection method based on neural network and image fusion
CN116840135A (en) Steel gate accelerated degradation test bed with simultaneous effects of corrosion and fatigue and post-degradation running state evaluation method
CN114529548B (en) Mechanical part stress corrosion detection method
CN115272322A (en) Alloy defect quantitative detection method based on data processing
CN109030132B (en) Preparation method of creep damage reference block, damage detection method and system
CN116990229B (en) Defect detection method and system for copper plating layer surface of product
CN116433661B (en) Method, device, equipment and medium for detecting semiconductor wafer by multitasking
CN111366600B (en) Method, device and system for detecting sweetness of fruit and storage medium
RU2406637C1 (en) Method of estimating available index of ship hull
CN111898314B (en) Lake water parameter inspection method and device, electronic equipment and storage medium
CN108229727B (en) Method and system for predicting service life distribution of complete machine of medical instrument
CN116165213A (en) Stainless steel chain product quality inspection method and system
CN113607068B (en) Method for establishing and extracting recognition model of photoacoustic measurement signal characteristics
CN115311238A (en) Photovoltaic module defect damage identification method based on image analysis
CN113218998B (en) Eddy current thermal imaging defect identification method based on global Moran index
CN115496706A (en) Surface corrosion ratio measuring method based on digital image processing
CN112069621B (en) Method for predicting residual service life of rolling bearing based on linear reliability index
JPH1019826A (en) Apparatus for measuring corrosion of metallic material
CN111189906B (en) On-line intelligent judging and classifying identification method for alternating current magnetic field defects
RU2730541C1 (en) Method for prediction of pipeline failure rate

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant