CN111476311B - Anchor chain flash welding quality online detection method based on increment learning - Google Patents

Anchor chain flash welding quality online detection method based on increment learning Download PDF

Info

Publication number
CN111476311B
CN111476311B CN202010309872.1A CN202010309872A CN111476311B CN 111476311 B CN111476311 B CN 111476311B CN 202010309872 A CN202010309872 A CN 202010309872A CN 111476311 B CN111476311 B CN 111476311B
Authority
CN
China
Prior art keywords
samples
sample
unqualified
electrode position
new
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
CN202010309872.1A
Other languages
Chinese (zh)
Other versions
CN111476311A (en
Inventor
苏世杰
王真心
潘纬鸣
陈赟
唐文献
付灵懿
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.)
Jiangsu University of Science and Technology
Original Assignee
Jiangsu University of Science and Technology
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 Jiangsu University of Science and Technology filed Critical Jiangsu University of Science and Technology
Priority to CN202010309872.1A priority Critical patent/CN111476311B/en
Publication of CN111476311A publication Critical patent/CN111476311A/en
Application granted granted Critical
Publication of CN111476311B publication Critical patent/CN111476311B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
    • G06F18/24133—Distances to prototypes
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
    • Y02P90/30—Computing systems specially adapted for manufacturing

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Software Systems (AREA)
  • Molecular Biology (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Image Analysis (AREA)
  • Analysing Materials By The Use Of Radiation (AREA)

Abstract

The invention discloses an online detection method for the flash welding quality of an anchor chain based on incremental learning, which comprises the following steps: collecting M unqualified samples and N qualified samples; the signal length of the sample is consistent through piecewise linear interpolation; normalizing the signals; calculating the distance between unqualified samples, and establishing a distance matrix; synthesizing a new sample after randomly extracting the unqualified samples, and calculating the distance between the new sample and the existing unqualified samples; judging whether the number of unqualified samples is the same as the number of qualified samples; constructing a convolutional neural network and training a model; outputting a prediction result of the model on the test set; training the model by using the new added sample; and outputting the prediction result of the model after incremental learning on the test set. The invention effectively increases the number of unqualified samples, solves the problem of unbalance of the samples, and improves the generalization capability of the model; the problem that the number of samples for flash welding of the anchor chain is increased is solved.

Description

Anchor chain flash welding quality online detection method based on increment learning
Technical Field
The invention belongs to a welding quality detection method, and particularly relates to an online detection method for flash welding quality of an anchor chain based on incremental learning.
Background
With the development of ocean engineering and shipping industry, the anchor chain is developed from a plant rope, a forging anchor chain and a casting anchor chain at first to a welding anchor chain at present. The anchor chain is formed by connecting a plurality of chain rings, is a special chain for buffering external force applied to ships and ocean engineering equipment, needs to keep good mechanical properties for a long time in a severe seawater environment, and the quality of the chain can directly influence the life and property safety of related personnel. At present, the quality detection of the anchor chain mainly passes through a pull test before delivery, and for the whole anchor chain, only one chain ring is broken due to the quality problem, so that the whole anchor chain can fail, and the loss caused by the failure is huge.
In the production process of the anchor chain, flash welding is a core link of the production of the anchor chain, and key mechanical performance indexes such as tensile load, breaking load, impact load and the like of the anchor chain are determined to a great extent. Because a series of complex physical and chemical reactions can occur during the flash welding of the anchor chain, a simple and effective mathematical model is difficult to establish, and meanwhile, the surface characteristics after welding are not obvious, so that the timely and effective detection of the welding quality is a long-felt technical problem in the industry, and the patent of China patent CN109242023A discloses an online assessment method for the flash welding quality of the anchor chain based on DTW and MDS. The disadvantage is that the invention does not take into account that in the actual production of the anchor chain, the number of acceptable samples is far greater than the number of unacceptable samples, a typical phenomenon of data imbalance. The patent of Chinese patent CN109886298A discloses a welding seam quality detection method based on a convolutional neural network, and the welding seam quality detection method based on the convolutional neural network carries out welding seam quality analysis on a welding seam image, avoids complicated detection steps, can automatically position and capture a welding seam area, and realizes effective detection of the welding seam quality. The invention has the defects that the anchor chain flash welding joint is sealed inside the base material, an effective welding line image cannot be obtained, the number of samples which are not considered to be welded is increased, new samples can bring new welding characteristics, and a fixed model is difficult to adapt to the changes, so that the long-term detection of welding quality is not facilitated.
Disclosure of Invention
The invention aims to: the invention aims to provide an online detection method for the flash welding quality of an anchor chain based on incremental learning, which solves the problems that a sample is unbalanced, a newly added sample cannot be adapted, and the detection accuracy is insufficient.
The technical scheme is as follows: the invention discloses an online detection method for the flash welding quality of an anchor chain based on incremental learning, which comprises the following steps:
(1) Collecting M unqualified samples S= { S 1 ,S 2 ,…,S i ,…,S M And N qualified samples t= { T 1 ,T 2 ,…,T i ,…,T N }；
(2) The signal length of the unqualified sample is consistent with that of the qualified sample through piecewise linear interpolation;
(3) Normalizing signals of the unqualified samples and the qualified samples;
(4) Calculating the distance between unqualified samples, and establishing a distance matrix;
(5) Randomly extracting samples from the unqualified samples, finding out samples closest to the extracted samples according to the distance matrix, synthesizing new unqualified samples by the extracted samples and the samples closest to the extracted samples, and calculating the distance between the new samples and the existing unqualified samples;
(6) Judging whether the number of the new unqualified samples is the same as the number of the qualified samples, if so, executing the step (7); if not, executing the step (5);
(7) Constructing a convolutional neural network, selecting half of new unqualified samples and half of qualified samples as training sets to train the model, substituting the rest of new unqualified samples and qualified samples as test sets into the model, and outputting a prediction result of the model on the test sets;
(8) Judging whether a new sample is added, if yes, executing the step (9); if not, completing model training;
(9) Training the model by using the new added sample, and updating parameters of the model after training is completed;
(10) Substituting the unqualified samples and the qualified samples of the test set into the model after incremental learning to obtain the prediction probability of the samples belonging to each category, and outputting the prediction result of the model after incremental learning.
Wherein, the step (2) specifically comprises the following steps:
will L i(1) A number of integers successively increasing
Figure GDA0004161636740000021
As a reject sampleS i Interpolation node in warm-up phase, +.>
Figure GDA0004161636740000022
Is the electrode position signal A of the interpolation node corresponding to the preheating stage i(1) The piecewise linear interpolation function R (x) is expressed as:
Figure GDA0004161636740000023
wherein the method comprises the steps of
Figure GDA0004161636740000024
For preheating stage electrode position signal A i(1) Is specifically expressed as:
Figure GDA0004161636740000031
Figure GDA0004161636740000032
Figure GDA0004161636740000033
if x is E [ x ] j ,x j+1 ]Electrode position signal A in preheating stage i(1) The piecewise linear interpolation function R (x) of (c) is expressed as:
Figure GDA0004161636740000034
wherein the failed sample s= { S 1 ,S 2 ,…,S i ,…,S M And pass samples t= { T 1 ,T 2 ,…,T i ,…,T N The signals of the electrode are composed of the electrode position signal and the current signal, and the ith disqualified sample is represented as S i ＝[A i ,B i ]In the preheating stageThe lengths of the continuous flashing stage and the upsetting stage are L respectively i(1) 、L i(2) And L i(3) Electrode position signal A of failed sample i Represented by A i ＝[A i(1) ,A i(2) ,A i(3) ]The electrode position signal of the preheating stage is expressed as
Figure GDA0004161636740000035
The electrode position signal of the successive flashing phases is indicated as +.>
Figure GDA0004161636740000036
The electrode position signal of the upsetting phase is denoted +.>
Figure GDA0004161636740000037
Current signal B i Denoted as B i ＝[B i(1) ,B i(2) ,B i(3) ]The current signal of the preheating phase is denoted +.>
Figure GDA0004161636740000038
The current signal of the continuous flash phase is expressed as
Figure GDA0004161636740000039
The current signal of the upsetting phase is denoted +.>
Figure GDA00041616367400000310
The ith qualifying sample is denoted as T i ＝[C i ,D i ]The lengths of the preheating stage, the continuous light stage and the upsetting stage are respectively Q i(1) 、Q i(2) And Q i(3) Electrode position signal C of qualified sample i Denoted as C i ＝[C i(1) ,C i(2) ,C i(3) ]The electrode position signal of the preheating phase is indicated as +.>
Figure GDA00041616367400000311
The electrode position signal of the continuous flash is expressed as
Figure GDA00041616367400000312
The upset electrode position signal is denoted +.>
Figure GDA0004161636740000041
Current signal D i Denoted as D i ＝[D i(1) ,D i(2) ,D i(3) ]The current signal of the preheating stage is expressed as
Figure GDA0004161636740000042
The current signal of the successive flash phases is denoted +.>
Figure GDA0004161636740000043
The current signal of the upsetting phase is denoted +.>
Figure GDA0004161636740000044
The length of the unqualified sample after piecewise linear interpolation is G, and the lengths of the preheating stage, the continuous flashing stage and the upsetting stage are G respectively 1 、G 2 And G 3 The electrode position signal of the preheating stage after interpolation is
Figure GDA0004161636740000045
The electrode position signal of the same continuous flashing phase is +.>
Figure GDA0004161636740000046
The electrode position signal in the upsetting phase is +.>
Figure GDA0004161636740000047
As a failed sample S i Electrode position signal A of (2) i Piecewise linear interpolation results, similarly B' i ＝[B′ i(1) ,B′ i(2) ,B′ i(3) ]As a failed sample S i Current signal B of (2) i Results after piecewise linear interpolation, S' i ＝[A′ i ,B′ i ]As a failed sample S i Performing piecewise linear interpolation on the qualified sample according to the result after piecewise linear interpolation, and performing the same process on the qualified sampleThe length of the preheating stage, the continuous flashing stage and the upsetting stage is G respectively 1 、G 2 And G 3 ，C′ i ＝[C′ i(1) ,C′ i(2) ,C′ i(3) ]For a qualified sample T i Electrode position signal C of (2) i Piecewise linear interpolation, similarly D' i ＝[D′ i(1) ,D′ i(2) ,D′ i(3) ]For qualified sample S i Current signal D of (2) i Results after piecewise linear interpolation, T' i ＝[C′ i ,D′ i ]For a qualified sample T i And (5) piecewise linear interpolation.
The step (3) specifically comprises the following steps:
for each failed sample S' i Normalization is performed using the following formula:
Figure GDA0004161636740000048
Figure GDA0004161636740000049
wherein a' i,max For electrode position signal A' i Maximum value of a' i,min For electrode position signal A' i Is the minimum value of A i ＝[a″ i,1 ,a″ i,2 ,…,a″ i,j, …,a″ i,G ]For electrode position signal A' i Normalized results; b' i,max Is a current signal B' i Maximum value of b' i,min Is a current signal B' i Minimum value of B i ＝[b″ i,1 ,b″ i,2 ,…,b″ i,j ,…,b″ i,″G ]Is a current signal B' i Normalized result, S i ＝[A″ i ,B″ i ]For unacceptable samples S' i Normalizing the signal of the qualified sample according to the above formula, and similarly normalizing the signal of the electrode position to C' i ＝[c″ i,1 ,c″ i,2 ,…,c″ i,j ,…,c″ i,G ]The current signal is D i ＝[d″ i,1 ,d″ i,2 ,…,d″ i,j ,…,d″ i,G ]，T″ i ＝[C″ i ,D″ i ]For a qualified sample T' i Normalized results.
The step (4) specifically comprises the following steps:
disqualified sample S i And S j The calculation formula of the distance between the two is as follows:
Figure GDA0004161636740000051
according to the calculated distance, a distance matrix H of M x M is established:
Figure GDA0004161636740000052
wherein h is i,j Representing a failed sample S i And S j A distance therebetween; a' i,k Representing a failed sample S i Electrode position signal A "") i The kth number of (a); a' j,k Representing a failed sample S j Electrode position signal A "") j The kth number of (a); b' i,k Representing a failed sample S i Current signal B "") i The kth number of (a); b' j,k Representing a failed sample S j Current signal B "") j The k number of (a).
The specific steps of calculating the distance between the new sample and the existing unqualified sample in the step (5) are as follows:
for unqualified sample S i And S j Electrode position signals of (a) are synthesized:
r＝rand(0.1,0.9)
A new ＝r×A″ i +(1-r)×A″ j
wherein r represents a number between 0.1 and 0.9 randomly generated, A new Representing the electrode position signal after synthesis;
for unqualified sample S i And S j Is synthesized by the current signals of:
B new ＝r×B″ i +(1-r)×B″ j
B new representing the current signal after synthesis;
S new ＝[A new ,B new ]to add a row-by-column distance matrix H for storing new samples S new Distance from the existing failed sample.
The step (7) specifically comprises the following steps:
constructing a convolutional neural network model, wherein the convolutional neural network model comprises 1 input layer, 2 convolutional layers, 2 pooling layers, 1 full-connection layer and 1 output layer; wherein, the 2 convolution layers respectively adopt 8 channels and 16 channels, and the convolution kernel sizes are 1*3; the 2 pooling layers are all in maximum pooling, and the sizes of pooling cores are 1*2; the number of neurons of the full-connection layer is 3200; the 2 categories of the output layer are qualified and unqualified;
updating parameters of the model by adopting back propagation and gradient descent, enabling a drop rate dropout=0.2 during training, and storing the model after iterative training to obtain a model for detecting the flash welding quality of the anchor chain;
substituting the unqualified samples and the qualified samples of the test set into the model to obtain the prediction probability of each category of unqualified samples and qualified samples, wherein the category with higher probability is the prediction result of the model.
The step (9) specifically comprises the following steps:
extracting M 'unqualified samples and N' qualified samples from the newly added samples, and performing piecewise linear interpolation processing on signals of the samples according to the step (2);
normalizing the signal of the sample according to the step (3);
calculating the distance between M ' newly added unqualified samples according to the step (4), calculating the distance between the newly added unqualified samples and the existing unqualified samples, and adding M ' rows and M ' columns to a distance matrix H for storing the calculated distance;
extracting one sample from M' newly added unqualified samples, finding the sample closest to the new sample to synthesize a new sample, and adding a row and a column to a distance matrix H for storing the distance between the synthesized new sample and the existing unqualified samples;
judging whether the number of the unqualified samples is the same as the number of the qualified samples, if so, executing the next step, if not, executing the previous step;
and inputting the newly added sample into the model for training, and updating parameters of the model after training is completed.
The beneficial effects are that: the invention uses the piecewise linear interpolation method to make the data length of the samples consistent, normalizes the samples to eliminate the dimension influence between the electrode position signal and the current signal, calculates the distance between the unqualified samples by using Euclidean distance, synthesizes new samples by randomly extracting the unqualified samples, effectively increases the number of the unqualified samples, solves the imbalance problem of the samples, and improves the generalization capability of the model; a convolutional neural network is established, and real-time detection of the flash welding quality of the anchor chain is realized; the model is trained by adopting an incremental learning method, so that the problem that the number of samples for flash welding of an anchor chain is increased gradually is solved, and the problem that the model is forgotten catastrophically is avoided; the invention can timely and effectively detect the flash welding quality of the anchor chain with higher accuracy, improves the production efficiency, reduces the production cost and ensures the safety of ship navigation and ocean engineering platforms.
Drawings
FIG. 1 is a schematic flow chart of the present invention;
FIG. 2 is a graph of electrode position signals for a failed sample and a failed sample;
FIG. 3 is a graph of current signals for a failed sample and a failed sample;
FIG. 4 is a graph of two piecewise linear interpolated and normalized electrode position signals and a synthesized electrode position signal;
FIG. 5 is a graph of two piecewise linear interpolated and normalized current signals and a synthesized current signal;
FIG. 6 is a block diagram of a convolutional neural network;
FIG. 7 is a graph of predicted results of incremental learning on a test set.
Detailed Description
The invention is further described below with reference to the accompanying drawings.
The invention discloses an online detection method for the flash welding quality of an anchor chain based on incremental learning, which has the flow shown in figure 1 and specifically comprises the following steps:
(1) Collecting 40 unqualified samples S= { S 1 ,S 2 ,…,S i ,…,S 40 And 400 qualified samples t= { T 1 ,T 2 ,…,T i ,…,T 400 }。
(2) The signal lengths of the samples are made uniform by piecewise linear interpolation.
The signals of the 40 failed samples and the 400 failed samples are each composed of an electrode position signal and a current signal. The i-th failed sample is denoted as S i ＝[A i ,B i ]The lengths of the preheating stage, the continuous flashing stage and the upsetting stage are L respectively (as shown in fig. 2 and 3) i(1) 、L i(2) And L i(3) . Electrode position signal A i Represented by A i ＝[A i(1) ,A i(2) ,A i(3) ]The electrode position signal of the preheating stage is expressed as
Figure GDA0004161636740000071
The electrode position signal for successive flash phases is expressed as
Figure GDA0004161636740000072
The electrode position signal of the upsetting phase is denoted +.>
Figure GDA0004161636740000073
Current signal B i Denoted as B i ＝[B i(1) ,B i(2) ,B i(3) ]The current signal of the preheating phase is denoted +.>
Figure GDA0004161636740000074
The current signal of the successive flash phases is denoted +.>
Figure GDA0004161636740000075
The current signal of the upsetting phase is denoted +.>
Figure GDA0004161636740000081
The ith qualifying sample is denoted as T i ＝[C i ,D i ]The lengths of the preheating stage, the continuous flashing stage and the upsetting stage are respectively Q (as shown in fig. 2 and 3) i(1) 、Q i(2) And Q i(3) . Electrode position signal C of qualified sample i Denoted as C i ＝[C i(1) ,C i(2) ,C i(3) ]The electrode position signal of the preheating phase is indicated as +.>
Figure GDA0004161636740000082
The electrode position signal for successive flash phases is expressed as
Figure GDA0004161636740000083
The electrode position signal of the upsetting phase is denoted +.>
Figure GDA0004161636740000084
Current signal D i Denoted as D i ＝[D i(1) ,D i(2) ,D i(3) ]The current signal of the preheating stage is expressed as
Figure GDA0004161636740000085
The current signal of the continuous flash is denoted +.>
Figure GDA0004161636740000086
The current signal of the upsetting phase is denoted +.>
Figure GDA0004161636740000087
Will L i(1) A number of integers successively increasing
Figure GDA0004161636740000088
As a reject sample S i Interpolation node in warm-up phase, +.>
Figure GDA0004161636740000089
Is the electrode position signal A of the interpolation node corresponding to the preheating stage i(1) The piecewise linear interpolation function R (x) is expressed as:
Figure GDA00041616367400000810
wherein the method comprises the steps of
Figure GDA00041616367400000811
For preheating stage electrode position signal A i(1) Is specifically expressed as:
Figure GDA00041616367400000812
Figure GDA00041616367400000813
Figure GDA0004161636740000091
if x is E [ x ] j ,x j+1 ]Electrode position signal A in preheating stage i(1) The piecewise linear interpolation function R (x) of (c) is expressed as:
Figure GDA0004161636740000092
the length of the unqualified sample after piecewise linear interpolation is G, and the lengths of the preheating stage, the continuous flashing stage and the upsetting stage are G respectively 1 、G 2 And G 3 The electrode position signal in the preheating stage is
Figure GDA0004161636740000093
The electrode position signal of the successive flashing phases is +.>
Figure GDA0004161636740000094
The electrode position signal of the upsetting stage is
Figure GDA0004161636740000095
As a failed sample S i Electrode position signal A of (2) i Piecewise linear interpolation results, similarly B' i ＝[B′ i(1) ,B′ i(2) ,B′ i(3) ]As a failed sample S i Current signal B of (2) i Results after piecewise linear interpolation, S' i ＝[A′ i ,B′ i ]As a failed sample S i And (5) piecewise linear interpolation. Performing piecewise linear interpolation on the qualified sample, wherein the length of the qualified sample is G, and the lengths of the preheating stage, the continuous flashing stage and the upsetting stage are G respectively 1 、G 2 And G 3 ，C′ i ＝[C′ i(1) ,C′ i(2) ,C′ i(3) ]For a qualified sample T i Electrode position signal C of (2) i Piecewise linear interpolation, similarly D' i ＝[D′ i(1) ,D′ i(2) ,D′ i(3) ]For qualified sample S i Current signal D of (2) i Results after piecewise linear interpolation, T' i ＝[C′ i ,D′ i ]For a qualified sample T i And (5) piecewise linear interpolation.
(3) For each failed sample S' i Normalization is performed by adopting the following formula:
Figure GDA0004161636740000096
Figure GDA0004161636740000097
wherein a' i,max For electrode position signal A' i Maximum value of a' i,min For electrode position signal A' i Is the minimum value of A i ＝[a″ i,1 ,a″ i,2 ,…,a″ i,j ,…,a″ i,G ]For electrode position signal A' i Normalized results; b' i,max Is a current signal B' i Maximum value of b' i,min Is a current signal B' i Minimum value of B i ＝[b″ i,1 ,b″ i,2 ,…,b″ i,j ,…,b″ i , G] Is a current signal B' i Normalized results; s' i ＝[A″ i ,B″ i ]For unacceptable samples S' i The normalized results are shown in fig. 4 and 5. Normalizing the signal of the qualified sample according to the formula, wherein the normalized electrode position signal is C i ＝[c″ i,1 ,c″ i,2 ,…,c″ i,j ,…,c″ i,G ]The current signal is D i ＝[d″ i,1 ,d″ i,2 ,…,d″ i,j ,…,d″ i,G ]，T″ i ＝[C″ i ,D″ i ]For a qualified sample T' i Normalized results.
(4) Calculate the unqualified sample S i And S j Distance between:
Figure GDA0004161636740000101
the distance between 40 unqualified samples is calculated according to the formula, and a distance matrix H of 40 x 40 is established:
Figure GDA0004161636740000102
wherein h is i,j Representing a failed sample S i And S j A distance therebetween; a' i,k Representing a failed sample S i Electrode position signal A "") i The kth number of (a); a' j,k Representing a failed sample S j Electrode position signal A "") j The kth number of (a); b' i,k Representing a failed sample S i Current signal B "") i The kth number of (a); b' j,k Representing a failed sample S j Current signal B "") j The k number of (a).
(5) Randomly extracting a failed sample S' from the failed samples i Finding out and disqualifying sample S' according to distance matrix H i A nearest disqualified sample S j 。
(6) Synthesis of a New sample S Using an unacceptable sample new ＝[A new ,B new ]As shown in fig. 4 and 5, a row-by-column distance matrix H is added for storing new samples S new The distance between the sample and the existing unqualified sample is specifically as follows:
(6.1) for failed samples S i And S j Electrode position signals of (a) are synthesized:
r＝rand(0.1,0.9)
A new ＝r×A″ i +(1-r)×A″ j
wherein r represents a number between 0.1 and 0.9 randomly generated, A new Representing the electrode position signal after synthesis.
(6.2) for failed samples S i And S j Is synthesized by the current signals of:
B new ＝r×B″ i +(1-r)×B″ j
B new representing the current signal after synthesis.
(6.3)S new ＝[A new ,B new ]To add a row-by-column distance matrix H for storing new samples S new Distance from the existing failed sample.
(7) Judging whether the number of the unqualified samples is the same as the number of the qualified samples, if so, executing the step 8; and if not, executing the step 5 and the step 6.
(8) Construction of convolutional neural networks the model was trained as shown in fig. 6.
Constructing a convolutional neural network model, wherein the convolutional neural network model comprises 1 input layer, 2 convolutional layers, 2 pooling layers, 1 full-connection layer and 1 output layer; wherein, the 2 convolution layers respectively adopt 8 channels and 16 channels, and the convolution kernel sizes are 1*3; the 2 pooling layers are all in maximum pooling, and the sizes of pooling cores are 1*2; the number of neurons of the full-connection layer is 3200; the 2 categories of the output layer are pass and fail.
Half of the failed samples and half of the qualified samples are randomly selected from the sample set to be used as a training set, and the rest of the failed samples and the qualified samples are used as a test set.
And updating parameters of the model by adopting back propagation and gradient descent, so as to prevent network overfitting, enabling the drop rate dropout=0.2 during training, and storing the model after iterative training to obtain the model for detecting the flash welding quality of the anchor chain.
(9) Substituting the unqualified samples and the qualified samples of the test set into the model to obtain the prediction probability of each category of unqualified samples and qualified samples, wherein the category with higher probability is the prediction result of the model.
(10) Judging whether a new sample is added, if yes, executing the step 11; and if not, finishing model training.
(11) Training the model by using the newly added sample.
And (11.1) extracting 20 unqualified samples and 200 qualified samples from the newly added samples, and performing piecewise linear interpolation processing on signals of the samples.
(11.2) normalizing the signal of the sample.
And (11.3) calculating the distance between 20 newly added unqualified samples, calculating the distance between the newly added unqualified samples and the existing unqualified samples, and adding 20 rows and 20 columns to a distance matrix H for storing the calculated distance.
(11.4) extracting one sample from the 20 newly added unqualified samples, finding the sample closest to the newly added sample to synthesize a new sample, adding one row by one row to a distance matrix H, and storing the distance between the synthesized new sample and the existing unqualified sample.
(11.5) judging whether the number of unqualified samples is the same as the number of qualified samples, if so, executing the step 11.6; if not, executing step 11.4.
And (11.6) inputting the newly added sample into the current model for training, and updating parameters of the model after training is completed.
(12) Substituting the unqualified samples and the qualified samples of the test set into the model after incremental learning to obtain the prediction probability that the samples belong to each category, and outputting the prediction result of the model after incremental learning, wherein the result is shown in fig. 7.

Claims (6)

1. The online detection method for the flash welding quality of the anchor chain based on incremental learning is characterized by comprising the following steps of:
(1) Collecting M unqualified samples S= { S 1 ,S 2 ,…,S i ,…,S M And N qualified samples t= { T 1 ,T 2 ,…,T i ,…,T N }；
(2) Unifying signal lengths of the unacceptable samples and acceptable samples by piecewise linear interpolation, comprising:
will L i(1) A number of integers successively increasing
Figure FDA0004094707100000011
As a reject sample S i The interpolation node in the warm-up phase,
Figure FDA0004094707100000012
is the electrode position signal A of the interpolation node corresponding to the preheating stage i(1) The piecewise linear interpolation function R (x) is expressed as:
Figure FDA0004094707100000013
wherein the method comprises the steps of
Figure FDA0004094707100000014
For preheating stage electrode position signal A i(1) Is specifically expressed as:
Figure FDA0004094707100000015
Figure FDA0004094707100000016
Figure FDA0004094707100000017
if x is E [ x ] j ,x j+1 ]Electrode position signal A in preheating stage i(1) The piecewise linear interpolation function R (x) of (c) is expressed as:
Figure FDA0004094707100000018
wherein the failed sample s= { S 1 ,S 2 ,…,S i ,…,S M And pass samples t= { T 1 ,T 2 ,…,T i ,…,T N The signals of the electrode are composed of the electrode position signal and the current signal, and the ith disqualified sample is represented as S i ＝[A i ,B i ]The lengths of the preheating stage, the continuous flashing stage and the upsetting stage are L respectively i(1) 、L i(2) And L i(3) Electrode position signal A of failed sample i Represented by A i ＝[A i(1) ,A i(2) ,A i(3) ]The electrode position signal of the preheating stage is expressed as
Figure FDA0004094707100000021
The electrode position signal of the successive flashing phases is indicated as +.>
Figure FDA0004094707100000022
The electrode position signal of the upsetting phase is denoted +.>
Figure FDA0004094707100000023
Current signal B i Denoted as B i ＝[B i(1) ,B i(2) ,B i(3) ]The current signal of the preheating phase is denoted +.>
Figure FDA0004094707100000024
The current signal of the continuous flash phase is expressed as
Figure FDA0004094707100000025
The current signal of the upsetting phase is denoted +.>
Figure FDA0004094707100000026
The ith qualifying sample is denoted as T i ＝[C i ,D i ]The lengths of the preheating stage, the continuous light stage and the upsetting stage are respectively Q i(1) 、Q i(2) And Q i(3) Electrode position signal C of qualified sample i Denoted as C i ＝[C i(1) ,C i(2) ,C i(3) ]The electrode position signal of the preheating phase is indicated as +.>
Figure FDA0004094707100000027
The electrode position signal of the continuous flash is expressed as
Figure FDA0004094707100000028
The upset electrode position signal is denoted +.>
Figure FDA0004094707100000029
Current signal D i Denoted as D i ＝[D i(1) ,D i(2) ,D i(3) ]The stream signal of the preheating stage is expressed as
Figure FDA00040947071000000210
The current signal of the successive flash phases is denoted +.>
Figure FDA00040947071000000211
Current signal representation of the upsetting phase +.>
Figure FDA00040947071000000212
The length of the unqualified sample after piecewise linear interpolation is G, and the lengths of the preheating stage, the continuous flashing stage and the upsetting stage are G respectively 1 、G 2 And G 3 The electrode position signal of the preheating stage after interpolation is
Figure FDA00040947071000000213
The electrode position signal of the same continuous flashing phase is +.>
Figure FDA00040947071000000214
The electrode position signal in the upsetting phase is +.>
Figure FDA00040947071000000215
A′ i ＝[A′ i(1) ,A′ i(2) ,A′ i(3) ]As a failed sample S i Electrode position signal A of (2) i Piecewise linear interpolation results, similarly B' i ＝[B′ i(1) ,B′ i(2) ,B′ i(3) ]As a failed sample S i Current signal B of (2) i Results after piecewise linear interpolation, S' i ＝[A′ i ,B′ i ]As a failed sample S i The result after piecewise linear interpolation carries out piecewise linear interpolation on the qualified sample according to the formula, and the length of the qualified sample is G, and the lengths of the preheating stage, the continuous flashing stage and the upsetting stage are G respectively 1 、G 2 And G 3 ，C′ i ＝[C′ i(1) ,C′ i(2) ,C′ i(3) ]For a qualified sample T i Electrode position signal C of (2) i Piecewise linear interpolated junctionFruit, emblica officinalis D' i ＝[D′ i(1) ,D′ i(2) ,D′ i(3) ]For qualified sample S i Current signal D of (2) i Results after piecewise linear interpolation, T i ′＝[C′ i ,D′ i ]For a qualified sample T i A result after piecewise linear interpolation;
(3) Normalizing signals of the unqualified samples and the qualified samples;
(4) Calculating the distance between unqualified samples, and establishing a distance matrix;
(5) Randomly extracting samples from the unqualified samples, finding out samples closest to the extracted samples according to the distance matrix, synthesizing new unqualified samples by the extracted samples and the samples closest to the extracted samples, and calculating the distance between the new samples and the existing unqualified samples;
(6) Judging whether the number of the new unqualified samples is the same as the number of the qualified samples, if so, executing the step (7); if not, executing the step (5);
(7) Constructing a convolutional neural network, selecting half of new unqualified samples and half of qualified samples as training sets to train the model, substituting the rest of new unqualified samples and qualified samples as test sets into the model, and outputting a prediction result of the model on the test sets;
(8) Judging whether a new sample is added, if yes, executing the step (9); if not, completing model training;
(9) Training the model by using the new added sample, and updating parameters of the model after training is completed;
(10) Substituting the unqualified samples and the qualified samples of the test set into the model after incremental learning to obtain the prediction probability of the samples belonging to each category, and outputting the prediction result of the model after incremental learning.
2. The online detection method of the flash welding quality of the anchor chain based on the incremental learning of claim 1, wherein the step (3) is specifically:
for each failed sample S' i Normalization is performed using the following formula:
Figure FDA0004094707100000031
Figure FDA0004094707100000032
/>
wherein a' i,max For electrode position signal A' i Maximum value of a' i,min For electrode position signal A' i Is the minimum value of A i ＝[a″ i,1 ,a″ i,2 ,…,a″ i,j ,…,a″ i,G ]For electrode position signal A' i Normalized results; b' i,max Is a current signal B' i Maximum value of b' i,min Is a current signal B' i Minimum value of B i ＝[b″ i,1 ,b″ i,2 ,…,b″ i,j ,…,b″ i,G ]Is a current signal B' i Normalized result, S i ＝[A″ i ,B″ i ]For unacceptable samples S' i Normalizing the signal of the qualified sample according to the above formula, and similarly normalizing the signal of the electrode position to C' i ＝[c″ i，1 ,c″ i,2 ,…,c″ i,j ,…,c″ i,G ]The current signal is D i ＝[d″ i,1 ,d″ i,2 ,…,d″ i,j ,…,d″ i,G ]，T i ″＝[C″ i ,D″ i ]For a qualified sample T i ' normalized results.
3. The online detection method of the flash welding quality of the anchor chain based on the incremental learning of claim 1, wherein the step (4) is specifically:
disqualified sample S i And S j The calculation formula of the distance between the two is as follows:
Figure FDA0004094707100000041
according to the calculated distance, a distance matrix H of M x M is established:
Figure FDA0004094707100000042
wherein h is i,j Representing a failed sample S i And S j A distance therebetween; a' i,k Representing a failed sample S i Electrode position signal A "") i The kth number of (a); a' j,k Representing a failed sample S j Electrode position signal A "") j The kth number of (a); b' i,k Representing a failed sample S i Current signal B of (2) i "the kth number in"; b' j,k Representing a failed sample S j Current signal B "") j The k number of (a).
4. The online detection method for the flash welding quality of the anchor chain based on incremental learning according to claim 1, wherein the specific step of calculating the distance between the new sample and the existing failed sample in the step (5) is as follows:
for unqualified sample S i And S j Electrode position signals of (a) are synthesized:
r＝rand(0.1,0.9)
A new ＝r×A″ i +(1-r)×A″ j
wherein r represents a number between 0.1 and 0.9 randomly generated, A new Representing the electrode position signal after synthesis;
for unqualified sample S i And S j Is synthesized by the current signals of:
B new ＝r×B″ i +(1-r)×B″ j
B new representing the current signal after synthesis; s is S new ＝[A new ,B new ]Is synthesized byAdding a row and a column to a distance matrix H for storing new samples S new Distance from the existing failed sample.
5. The online detection method of the flash welding quality of the anchor chain based on the incremental learning of claim 1, wherein the step (7) is specifically:
constructing a convolutional neural network model, wherein the convolutional neural network model comprises 1 input layer, 2 convolutional layers, 2 pooling layers, 1 full-connection layer and 1 output layer; wherein, the 2 convolution layers respectively adopt 8 channels and 16 channels, and the convolution kernel sizes are 1*3; the 2 pooling layers are all in maximum pooling, and the sizes of pooling cores are 1*2; the number of neurons of the full-connection layer is 3200; the 2 categories of the output layer are qualified and unqualified;
updating parameters of the model by adopting back propagation and gradient descent, enabling a drop rate dropout=0.2 during training, and storing the model after iterative training to obtain a model for detecting the flash welding quality of the anchor chain;
substituting the unqualified samples and the qualified samples of the test set into the model to obtain the prediction probability of each category of unqualified samples and qualified samples, wherein the category with higher probability is the prediction result of the model.
6. The online detection method for the flash welding quality of the anchor chain based on incremental learning of claim 1, wherein the step (9) is specifically:
extracting M 'unqualified samples and N' qualified samples from the newly added samples, and performing piecewise linear interpolation processing on signals of the samples according to the step (2);
normalizing the signal of the sample according to the step (3);
calculating the distance between M ' newly added unqualified samples according to the step (4), calculating the distance between the newly added unqualified samples and the existing unqualified samples, and adding M ' rows and M ' columns to a distance matrix H for storing the calculated distance;
extracting one sample from M' newly added unqualified samples, finding the sample closest to the new sample to synthesize a new sample, and adding a row and a column to a distance matrix H for storing the distance between the synthesized new sample and the existing unqualified samples;
judging whether the number of the unqualified samples is the same as the number of the qualified samples, if so, executing the next step, if not, executing the previous step;
and inputting the newly added sample into the model for training, and updating parameters of the model after training is completed.
CN202010309872.1A 2020-04-20 2020-04-20 Anchor chain flash welding quality online detection method based on increment learning Active CN111476311B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010309872.1A CN111476311B (en) 2020-04-20 2020-04-20 Anchor chain flash welding quality online detection method based on increment learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010309872.1A CN111476311B (en) 2020-04-20 2020-04-20 Anchor chain flash welding quality online detection method based on increment learning

Publications (2)

Publication Number Publication Date
CN111476311A CN111476311A (en) 2020-07-31
CN111476311B true CN111476311B (en) 2023-05-26

Family

ID=71755948

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010309872.1A Active CN111476311B (en) 2020-04-20 2020-04-20 Anchor chain flash welding quality online detection method based on increment learning

Country Status (1)

Country Link
CN (1) CN111476311B (en)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108038471A (en) * 2017-12-27 2018-05-15 哈尔滨工程大学 A kind of underwater sound communication signal type Identification method based on depth learning technology
CN109242023A (en) * 2018-09-12 2019-01-18 江苏科技大学 A kind of anchor chain flash welding quality online evaluation method based on DTW and MDS
CN109239301A (en) * 2018-09-12 2019-01-18 江苏科技大学 A kind of anchor chain flash welding quality online evaluation method

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108038471A (en) * 2017-12-27 2018-05-15 哈尔滨工程大学 A kind of underwater sound communication signal type Identification method based on depth learning technology
CN109242023A (en) * 2018-09-12 2019-01-18 江苏科技大学 A kind of anchor chain flash welding quality online evaluation method based on DTW and MDS
CN109239301A (en) * 2018-09-12 2019-01-18 江苏科技大学 A kind of anchor chain flash welding quality online evaluation method

Also Published As

Publication number Publication date
CN111476311A (en) 2020-07-31

Similar Documents

Publication Publication Date Title
WO2022077901A1 (en) Bearing failure mode diagnosis method using small sample data sets, and system
CN109272123B (en) Sucker-rod pump working condition early warning method based on convolution-circulation neural network
CN107092596A (en) Text emotion analysis method based on attention CNNs and CCR
CN110108456A (en) A kind of rotating machinery health evaluating method of depth convolutional neural networks
CN111337768A (en) Deep parallel fault diagnosis method and system for dissolved gas in transformer oil
CN104793606B (en) Industrial method for diagnosing faults based on improved KPCA and HMM
CN111651931A (en) Blast furnace fault diagnosis rule derivation method based on deep neural network
CN114913150A (en) Intelligent identification method for concrete dam defect time sequence image
CN110276385B (en) Similarity-based mechanical part residual service life prediction method
CN106897774A (en) Multiple soft measurement algorithm cluster modeling methods based on Monte Carlo cross validation
CN115673596B (en) Welding abnormity real-time diagnosis method based on Actor-Critic reinforcement learning model
CN112101445B (en) Continuous casting billet subcutaneous slag inclusion defect real-time forecasting method based on supervision neural network
CN108345284A (en) A kind of quality dependent failure detection method becoming gauge block based on two
CN114519311A (en) Prediction method, system, storage medium and application of total harbor basin wave effective wave height
CN117312816B (en) Special steel smelting effect evaluation method and system
CN113378967A (en) Structural health monitoring multivariate data anomaly diagnosis method based on convolutional neural network and transfer learning
CN112149750A (en) Water supply network pipe burst identification data driving method
CN115841004A (en) Strip steel hot rolling process mechanical property soft measurement method and device based on multidimensional data
CN111476311B (en) Anchor chain flash welding quality online detection method based on increment learning
CN113592939A (en) Deep learning method for judging size of narrow blood vessel based on coronary angiography image
CN112231971A (en) Blast furnace fault diagnosis method based on relative integral trend diffusion fault sample generation
CN115795380A (en) Flue gas acid making data cleaning and optimizing method based on isolated forest and weighted random forest
CN116677569A (en) Fan tower stress monitoring and early warning method and system
CN114239397A (en) Soft measurement modeling method based on dynamic feature extraction and local weighted deep learning
CN114462717A (en) Small sample gas concentration prediction method based on improved GAN and LSTM

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