CN116030464B - Welding arcing state real-time detection method based on semantic segmentation - Google Patents
Welding arcing state real-time detection method based on semantic segmentation Download PDFInfo
- Publication number
- CN116030464B CN116030464B CN202310311021.4A CN202310311021A CN116030464B CN 116030464 B CN116030464 B CN 116030464B CN 202310311021 A CN202310311021 A CN 202310311021A CN 116030464 B CN116030464 B CN 116030464B
- Authority
- CN
- China
- Prior art keywords
- arcing
- current
- welding
- semantic segmentation
- point
- 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
Links
- 238000003466 welding Methods 0.000 title claims abstract description 66
- 238000000034 method Methods 0.000 title claims abstract description 49
- 230000011218 segmentation Effects 0.000 title claims abstract description 44
- 238000011897 real-time detection Methods 0.000 title claims abstract description 11
- 238000012549 training Methods 0.000 claims abstract description 11
- 238000012937 correction Methods 0.000 claims abstract description 8
- 238000005070 sampling Methods 0.000 claims description 22
- 238000001514 detection method Methods 0.000 claims description 11
- 238000011176 pooling Methods 0.000 claims description 8
- 230000008033 biological extinction Effects 0.000 claims description 5
- 238000007781 pre-processing Methods 0.000 claims description 5
- 238000012360 testing method Methods 0.000 claims description 5
- 238000012795 verification Methods 0.000 claims description 5
- 238000010791 quenching Methods 0.000 claims description 2
- 230000000171 quenching effect Effects 0.000 claims description 2
- 238000013461 design Methods 0.000 abstract description 4
- 230000007547 defect Effects 0.000 description 2
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 206010063385 Intellectualisation Diseases 0.000 description 1
- 230000006978 adaptation Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000011179 visual inspection Methods 0.000 description 1
Classifications
-
- 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
Abstract
The invention discloses a welding arcing state real-time detection method based on semantic segmentation. The method comprises the steps of building a semantic segmentation model based on an FCN network structure, training, classifying input data, outputting data labels of all current points in each sample, determining an arcing section in an actual welding process according to the labels, and detecting the arcing continuity of the welding process in real time. In addition, aiming at the conditions of continuous arcing, instantaneous arcing and the like possibly existing in the actual welding process, the invention also designs an auxiliary correction method, solves the possible redundancy problem of the prediction result of the semantic segmentation model, and can more accurately judge the duration of the arcing state.
Description
Technical Field
The invention belongs to the technical field of welding arcing state detection, and particularly relates to a real-time welding arcing state detection method based on semantic segmentation.
Background
Intelligent welding is one of the most important research subjects in the field of intelligent manufacturing. The sensing technology and the information processing thereof are key elements for realizing the intellectualization and automation of the welding process. Welding arcing condition detection refers to detecting the arcing time and end time of each weld, which plays a very important role in welding technology, and welding arcing continuity is closely related to welding quality and welder's ability. On the one hand, frequent arcing, namely, too short arcing continuity, can cause welding consistency problems, so that too many welding joints are caused, and defect hidden dangers are more likely to exist; on the other hand, the arcing continuity can reflect the operation capability of a welder, and long-time welding and short-time welding of the welder each day can be counted through arcing state detection, so that the method has very important significance for process and personnel performance evaluation.
Considering that the welding process is full of intense light, heat and noise, the current welding arcing continuity detection mainly depends on visual inspection and spot check after welding is finished, and has the defects of inaccurate detection, large delay, difficult closed loop control and the like.
Disclosure of Invention
The invention aims to: aiming at the problems in the background art, the invention provides a real-time detection method for the welding arcing state based on semantic segmentation, which is characterized in that high-frequency current data are collected and input into a semantic segmentation model based on an FCN network structure, the welding state of each current data point in each sample is predicted, and an arcing section is judged. Aiming at the counting redundancy existing in the arcing process, the invention designs an auxiliary correction judging method, adjusts the model prediction result and enables the output arcing state interval to be more accurate.
The technical scheme is as follows: a welding arcing state real-time detection method based on semantic segmentation comprises the following steps:
s1, collecting high-frequency current data in a welding process and preprocessing, wherein the preprocessing comprises sliding window sampling of the current data; carrying out data annotation on the sample sampled by the sliding window, and carrying out dimension expansion; finally, dividing a training set, a verification set and a test set according to a preset proportion;
s2, building a semantic segmentation model based on an FCN network structure; inputting the samples preprocessed in the step S1 into a semantic segmentation model, and outputting the welding state of each current point in each sample;
s3, training a semantic segmentation model, determining an arcing current point and an arc extinction current point, and outputting a welding arcing continuity detection result;
and S4, performing welding arcing continuity real-time detection based on the trained semantic segmentation model, designing an arcing continuity correction method to perform auxiliary adjustment, and finally outputting corrected arcing continuity results.
Further, the preprocessing of the current data in step S1 mainly includes:
s1.1, sliding window sampling is carried out on the collected high-frequency current data according to a preset window length; wherein the sampling time is T, and the sampling frequency is F;
s1.2, carrying out data marking on a sliding window sample, and endowing each current point with a label; searching an arcing current point start_point and an arcing current point end_point in a sample, marking all current points between the arcing current point and the arcing current point as 1, representing that welding is performed, marking other current points as 0, representing that welding is not performed; when the sample has no arcing current, all the samples are marked as 0; when the sample maintains normal welding current, all the samples are marked as 1;
s1.3, performing dimension expansion on each sample and the corresponding label;
and S1.4, carrying out random disorder on all the preprocessed samples with labels, and dividing the samples into a training set train_data, a verification set dev_data and a test set test_data according to a preset proportion.
Further, the semantic segmentation model based on the FCN network structure in the step S2 includes a plurality of TCN network blocks and convolutional network blocks that are cascaded in sequence; each stage of TCN network block is connected with a layer of maximum pooling downsampling layer; the output of the largest pooling downsampling layer after inputting the current sample and each stage of TCN network block is x1, x2, … and xi in sequence; sequentially inputting xi into a convolution network block, and outputting x as the final-stage convolution network block;
after the output x passes through the first layer up-sampling layer, adding the output x with the output of the maximum pooling down-sampling layer after at least one stage of TCN network block, and inputting the output x into the second layer up-sampling layer; obtaining a final output X; the final output X and the input current sample have the same length and respectively represent the prediction result of each current point in the input current sample.
Further, the method for correcting the arcing continuity in the step S4 specifically includes:
s4.1, acquiring new high-frequency welding current data in real time, carrying out window drawing acquisition based on a preset time step, inputting the current data of each window into a semantic segmentation model, and outputting the welding state of each data point in each window in real time; taking a section from a current point of which the current first label is 1 to a current point of which the current first label is 0 as an arcing section, and taking a continuous current section of which the current label is 0 as an arcing section;
step S4.2, when the point proportion of the current point label with the residual label of 0 is detected to be more than 99%, the arc quenching is considered not to occur at the moment, and the current point label with the residual label of 0 is set to be 1;
step S4.3, when the point proportion of the tag 1 is detected to be lower than 1%, the current point tag with the tag 1 is considered to be not started at the moment, and the current point tag with the tag 1 is set to 0;
step S4.4, when a plurality of continuous current point intervals with the labels of 0 exist in the window, if the lengths of the current point intervals with the labels of 0 are lower than 40% of the total window length, the conditions of arc starting failure and arc restarting exist at the moment, and the current point labels with the labels of 0 are set to be 1;
and S4.5, when a continuous current interval with the tag of 0 exists at the junction of two adjacent windows and the interval accounts for less than 40% of the window length, the new arcing is considered not to occur, the current point tags with all the tags of 0 are set to be 1, and otherwise, the new arcing process is judged.
Compared with the prior art, the technical scheme adopted by the invention has the following beneficial effects:
(1) According to the invention, the traditional semantic segmentation model based on the FCN network structure is applied to the classification process of high-frequency time sequence data, and by setting a plurality of layers of TCN network blocks and convolution network blocks and fusing the output of each level of TCN network blocks, a high-level semantic feature map and a low-level feature map containing rich position information can be fused, so that the feature map with more abundant information can be obtained, and the accurate feature learning of the model is realized.
(2) The method designed by the invention adopts a detection mode of 'sequence to sequence', and outputs the prediction results of different current points in each sample through a semantic segmentation model, so that the current arcing time and the ending time can be accurately judged, and an accurate arcing continuity detection result can be obtained.
(3) The invention designs an auxiliary correction judging method based on the semantic segmentation model, corrects the problems of instantaneous stop, continuous arcing and the like by adjusting the label, solves the redundancy problem of the model statistical result, and can acquire more accurate arcing and arcing judging results.
Drawings
FIG. 1 is a flow chart of a welding arcing state real-time detection method based on semantic segmentation;
fig. 2 is a schematic structural diagram of a semantic segmentation model according to an embodiment of the present invention.
Description of the embodiments
The invention provides a welding arcing state real-time detection method based on semantic segmentation. The method comprises the steps of building a semantic segmentation model based on an FCN network structure, training, classifying input data, outputting data labels of all current points in each sample, determining an arcing section in an actual welding process according to the labels, and detecting the arcing continuity of the welding process in real time. In addition, aiming at the conditions of continuous arcing, instantaneous arcing and the like possibly existing in the actual welding process, the invention also designs an auxiliary correction method, solves the possible redundancy problem of the prediction result of the semantic segmentation model, and can more accurately judge the duration of the arcing state. The invention is further explained below with reference to the drawings. As shown in figure 1 of the drawings,
step S1, firstly, collecting high-frequency current data in the welding process, and preprocessing the current data. In particular, the method comprises the steps of,
s1.1, sliding window sampling is carried out on the collected high-frequency current data according to a preset window length; wherein the sampling time is T, and the sampling frequency is F;
s1.2, carrying out data marking on a sliding window sample, and endowing each current point with a label; searching an arcing current point start_point and an arcing current point end_point in a sample, marking all current points between the arcing current point and the arcing current point as 1, representing that welding is performed, marking other current points as 0, representing that welding is not performed. When the sample has no arcing current, all are marked as 0, and when the sample maintains normal welding current, all are marked as 1.
Step S1.3, performing dimension expansion on each sample and the corresponding label.
And S1.4, carrying out random disorder on all the preprocessed samples with labels, and dividing the samples into a training set train_data, a verification set dev_data and a test set test_data according to a preset proportion.
And S2, building a semantic segmentation model based on the FCN network structure.
The invention aims to judge whether each current point in a sample belongs to a welding state or not by constructing a semantic segmentation model based on an FCN network structure, and further judges the state of each current point in the sample by the semantic segmentation model.
The semantic segmentation model structure comprises a plurality of TCN network blocks and convolution network blocks which are sequentially cascaded; as shown in fig. 2, in the embodiment of the present invention, 4-level TCN network blocks are used, where each level TCN network block is connected to a maximum pooled downsampling layer; and (3) inputting the current sample into a first-stage TCN network block TCNBlock1, and connecting the output to a layer of maximum pooling downsampling layer to obtain the first-stage output as x1. And inputting x1 into a second-stage TCN network block TCNBlock2, and connecting the output to a layer of maximum pooling downsampling layer to obtain second-stage output x2. Similarly, a third stage output x3 and a fourth stage output x4 are obtained, respectively. And sequentially inputting the 4 th-stage output to a 2-stage convolution network block ConvBlock1 and a ConvBlock2 to obtain an output x.
Here, for the TCN network blocks of level 1-4, the convolution kernel size is the same, and kernel_size=3 is taken in this embodiment. The output dimension out_channel increases sequentially, and in this embodiment, output dimensions of TCN network blocks at each stage are 16, 32, 64, and 128 respectively. The expansion ratio of the expansion_rate increases sequentially, and 2, 2 and 3 are respectively taken as 2, 0, 2, 1 and 2.
The largest pooled downsampling layer connected by each stage of TCN network block adopts the same structural parameters, and in this embodiment, the pooled layer size pool_size=2, and step size stride=2.
Here for the 2-level convolutional network blocks ConvBlock1 and ConvBlock2, wherein the output dimension out_channel of ConvBlock 1=160, the convolution kernel size kernel_size=7, the step size stride=1, the output dimension out_channel of ConvBlock 2=1, the convolution kernel size kernel_size=1.
In the practical application process, the number of stages of the TCN network block and the convolution network block is not lower than 2, and the output x can be obtained by ensuring that the output dimension and the convolution kernel size of the convolution network block at the last stage are both 1.
And then, after the output X passes through the first up-sampling layer, adding the output X with the output X2 of the 2 nd-stage TCN network block to obtain a first up-sampling output X1, inputting the X1 into the second up-sampling layer, adding the output X1 with the output X1 of the 1 st-stage TCN network block to obtain a second up-sampling output X2, and inputting the X2 into a third up-sampling layer to obtain a final output X.
The semantic segmentation network structure built in the embodiment further learns relevant characteristics of high-frequency current data by fusing the outputs of the first-stage TCN network block and the second-stage TCN network block on the basis of the output x of the convolution network. When the network is actually built, the method can be used for fusing the output of the multistage TCN network blocks according to the need, and fusing the high-level semantic feature map with the low-level feature map containing rich position information, so that the feature map with more abundant information can be obtained.
The output X of the semantic segmentation model is a vector with the same length as the input current sample, wherein each element represents whether the current at the point is in a welding state or not, if the current is in the welding state, the corresponding result of the current point is marked 1, and if no arcing occurs or the arc extinguishing is finished, the corresponding result of the current point is marked 0.
S3, training a semantic segmentation model;
and setting a learning rate of learning_rate=0.001, training a semantic segmentation model according to the number of samples batch_size=16, setting a loss function as a cross entropy loss function, and fitting network parameters. The finally trained model can judge whether the current points in each sample are in a welding state or not, and the arcing current points and the arc extinction current points can be accurately judged.
And S4, performing welding arcing continuity real-time detection based on the trained semantic segmentation model, designing an arcing continuity correction method, and judging the arcing continuity in real time. In particular, the method comprises the steps of,
and S4.1, acquiring new high-frequency welding current data in real time, carrying out window drawing acquisition based on a preset time step, inputting the current data of each window into a semantic segmentation model, and outputting the welding state of each data point in each window in real time.
And S4.2, because the arc starting failure of the welding machine exists in the actual welding process, the arc starting and extinguishing processes are repeated, meanwhile, the condition that the current is 0 in a certain shorter period exists in the welding process, and the trained semantic segmentation model can detect whether the arc starting and the welding occur in real time, but cannot intelligently identify the intermittent special condition. Based on the requirement, a correction judging method is designed, the semantic segmentation model is further assisted to judge, and more accurate arcing and extinction judging results can be obtained. Specifically, the current point detection results in each window are counted,
(1) When the point proportion of the tag 1 is detected to be more than 99%, the arc extinction is not considered to occur at the moment, and the current point tag with the rest tag of 0 is set to be 1.
(2) When the point proportion of the current point with the label of 1 is detected to be lower than 1%, the current point with the label of 1 is regarded as not being started at the moment, and the current point with the label of 1 is set to 0.
(3) When a plurality of continuous current point intervals with the labels of 0 exist in the window, if the lengths of the current point intervals with the labels of 0 are lower than 40% of the total window length, the conditions of arc starting failure and arc restarting exist at the moment, and the current point labels with the labels of 0 are set to be 1. Since arc striking failure re-striking is common in the actual welding process, the existence of continuous arc striking can lead to the fact that the model statistical result contains continuous arc striking intervals, so that statistical redundancy is caused, and the continuous arc striking intervals are considered as one continuous arc striking interval in the process of statistical arc striking intervals.
(4) When a continuous current interval with the tag of 0 exists at the junction of two adjacent windows and the interval accounts for less than 40% of the window length, the new arcing is considered not to occur, the current point tags with all the tags of 0 are set to be 1, and otherwise, the new arcing is judged.
The foregoing is merely a preferred embodiment of the present invention and it should be noted that modifications and adaptations to those skilled in the art may be made without departing from the principles of the present invention, which are intended to be comprehended within the scope of the present invention.
Claims (3)
1. A welding arcing state real-time detection method based on semantic segmentation is characterized by comprising the following steps:
s1, collecting high-frequency current data in a welding process, and carrying out sliding window sampling on the current data; carrying out data annotation on the sample sampled by the sliding window, and carrying out dimension expansion; finally, dividing a training set, a verification set and a test set according to a preset proportion;
s2, building a semantic segmentation model based on an FCN network structure; inputting the samples preprocessed in the step S1 into a semantic segmentation model, and outputting the welding state of each current point in each sample; the semantic segmentation model comprises a plurality of TCN network blocks and convolution network blocks which are sequentially cascaded; each stage of TCN network block is connected with a layer of maximum pooling downsampling layer; the output of the largest pooling downsampling layer after inputting the current sample and each stage of TCN network block is x1, x2, … and xi in sequence; sequentially inputting xi into a convolution network block, and outputting x as the final-stage convolution network block;
after the output x passes through the first layer up-sampling layer, adding the output x with the output of the maximum pooling down-sampling layer after at least one stage of TCN network block, and inputting the output x into the second layer up-sampling layer; obtaining a final output X; the final output X has the same length as the input current sample and respectively represents the prediction result of each current point in the input current sample;
s3, training a semantic segmentation model, determining an arcing current point and an arc extinction current point, and outputting a welding arcing continuity detection result;
and S4, performing welding arcing continuity real-time detection based on the trained semantic segmentation model, designing an arcing continuity correction method to perform auxiliary adjustment, and finally outputting corrected arcing continuity results.
2. The method for detecting the welding arcing state in real time based on semantic segmentation according to claim 1, wherein the preprocessing of the current data in step S1 mainly comprises:
s1.1, sliding window sampling is carried out on the collected high-frequency current data according to a preset window length; wherein the sampling time is 1S, and the sampling frequency is F;
s1.2, carrying out data marking on a sliding window sample, and endowing each current point with a label; searching an arcing current point start_point and an arcing current point end_point in a sample, marking all current points between the arcing current point and the arcing current point as 1, representing that welding is performed, marking other current points as 0, representing that welding is not performed; when the sample has no arcing current, all the samples are marked as 0; when the sample maintains normal welding current, all the samples are marked as 1;
s1.3, expanding each sample and the corresponding label in the last dimension;
and S1.4, carrying out random disorder on all the preprocessed samples with labels, and dividing the samples into a training set train_data, a verification set dev_data and a test set test_data according to a preset proportion.
3. The method for detecting the welding arcing state in real time based on semantic segmentation according to claim 1, wherein the method for correcting the arcing continuity in step S4 specifically comprises the following steps:
s4.1, acquiring new high-frequency welding current data in real time, carrying out window drawing acquisition based on a preset time step, inputting the current data of each window into a semantic segmentation model, and outputting the welding state of each data point in each window in real time; taking a section from a current point of which the current first label is 1 to a current point of which the current first label is 0 as an arcing section, and taking a continuous current section of which the current label is 0 as an arcing section;
step S4.2, when the point proportion of the current point label with the residual label of 0 is detected to be more than 99%, the arc quenching is considered not to occur at the moment, and the current point label with the residual label of 0 is set to be 1;
step S4.3, when the point proportion of the tag 1 is detected to be lower than 1%, the current point tag with the tag 1 is considered to be not started at the moment, and the current point tag with the tag 1 is set to 0;
step S4.4, when a plurality of continuous current point intervals with the labels of 0 exist in the window, if the lengths of the current point intervals with the labels of 0 are lower than 40% of the total window length, the conditions of arc starting failure and arc restarting exist at the moment, and the current point labels with the labels of 0 are set to be 1;
and S4.5, when a continuous current interval with the tag of 0 exists at the junction of two adjacent windows and the interval accounts for less than 40% of the window length, the new arcing is considered not to occur, the current point tags with all the tags of 0 are set to be 1, and otherwise, the new arcing process is judged.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310311021.4A CN116030464B (en) | 2023-03-28 | 2023-03-28 | Welding arcing state real-time detection method based on semantic segmentation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310311021.4A CN116030464B (en) | 2023-03-28 | 2023-03-28 | Welding arcing state real-time detection method based on semantic segmentation |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116030464A CN116030464A (en) | 2023-04-28 |
| CN116030464B true CN116030464B (en) | 2023-10-24 |
Family
ID=86074354
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310311021.4A Active CN116030464B (en) | 2023-03-28 | 2023-03-28 | Welding arcing state real-time detection method based on semantic segmentation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116030464B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113076691A (en) * | 2019-09-23 | 2021-07-06 | 华为技术有限公司 | Direct current arc detection method, device, equipment, system and storage medium |
| CN114200258A (en) * | 2021-11-23 | 2022-03-18 | 昆明理工大学 | Electric arc arcing detection method based on electric signals |
| CN114282689A (en) * | 2022-03-04 | 2022-04-05 | 苏芯物联技术(南京)有限公司 | Universal welding arcing real-time intelligent detection method and system |
| CN114559133A (en) * | 2022-04-27 | 2022-05-31 | 苏芯物联技术(南京)有限公司 | Universal welding arc starting continuity real-time detection method and system |
| CN115457077A (en) * | 2022-08-08 | 2022-12-09 | 南京理工大学 | Passive visual weld joint tracking method based on deep learning semantic segmentation |
-
2023
- 2023-03-28 CN CN202310311021.4A patent/CN116030464B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113076691A (en) * | 2019-09-23 | 2021-07-06 | 华为技术有限公司 | Direct current arc detection method, device, equipment, system and storage medium |
| CN114200258A (en) * | 2021-11-23 | 2022-03-18 | 昆明理工大学 | Electric arc arcing detection method based on electric signals |
| CN114282689A (en) * | 2022-03-04 | 2022-04-05 | 苏芯物联技术(南京)有限公司 | Universal welding arcing real-time intelligent detection method and system |
| CN114559133A (en) * | 2022-04-27 | 2022-05-31 | 苏芯物联技术(南京)有限公司 | Universal welding arc starting continuity real-time detection method and system |
| CN115457077A (en) * | 2022-08-08 | 2022-12-09 | 南京理工大学 | Passive visual weld joint tracking method based on deep learning semantic segmentation |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116030464A (en) | 2023-04-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113870260B (en) | Welding defect real-time detection method and system based on high-frequency time sequence data | |
| CN113067754B (en) | Semi-supervised time series anomaly detection method and system | |
| CN111160628A (en) | Air pollutant concentration prediction method based on CNN and double-attention seq2seq | |
| CN114559133B (en) | Real-time detection method and system for arc striking continuity of universal welding | |
| CN114282689B (en) | Universal welding arcing real-time intelligent detection method and system | |
| CN112884747A (en) | Automatic bridge crack detection system integrating cyclic residual convolution and context extractor network | |
| CN107341508B (en) | Fast food picture identification method and system | |
| CN111260156B (en) | Cash flow prediction model construction method and cash flow prediction method and device | |
| CN110276385B (en) | Similarity-based mechanical part residual service life prediction method | |
| CN112270335A (en) | Method and system for predicting welding quality defects of lap joint and computer readable storage medium | |
| CN115471487A (en) | Insulator defect detection model construction and insulator defect detection method and device | |
| CN116030464B (en) | Welding arcing state real-time detection method based on semantic segmentation | |
| CN113506269A (en) | Turnout and non-turnout rail fastener positioning method based on deep learning | |
| CN111178156B (en) | Time sequence feature effective window extraction method applied to machine learning | |
| CN114820499A (en) | Pipeline welding seam detection method | |
| CN114565149A (en) | CGA fusion model-based time series data prediction method and device and computer equipment | |
| CN117372377A (en) | Broken line detection method and device for monocrystalline silicon ridge line and electronic equipment | |
| CN116204830B (en) | Welding abnormality real-time detection method based on path aggregation network | |
| CN110879820A (en) | Industrial data processing method and device | |
| CN110866172B (en) | Data analysis method for block chain system | |
| CN111143934A (en) | Structural deformation prediction method based on time convolution network | |
| CN115147432A (en) | First arrival picking method based on depth residual semantic segmentation network | |
| CN102521805B (en) | Video word processing method based on interframe information | |
| CN113239727A (en) | Person detection and identification method | |
| CN117057942B (en) | Intelligent financial decision big data analysis system |
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 |