CN110174409A - A kind of cut deal periodicity defect control method based on real-time detection result - Google Patents
A kind of cut deal periodicity defect control method based on real-time detection result Download PDFInfo
- Publication number
- CN110174409A CN110174409A CN201910514268.XA CN201910514268A CN110174409A CN 110174409 A CN110174409 A CN 110174409A CN 201910514268 A CN201910514268 A CN 201910514268A CN 110174409 A CN110174409 A CN 110174409A
- Authority
- CN
- China
- Prior art keywords
- cut deal
- defect
- periodicity defect
- periodicity
- real
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
- G01N2021/8854—Grading and classifying of flaws
- G01N2021/8877—Proximity analysis, local statistics
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
- G01N2021/8883—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges involving the calculation of gauges, generating models
Abstract
The present invention provides a kind of cut deal periodicity defect control method based on real-time detection result, belongs to plate quality control technology field.Cut deal surface defects detection system, the surface defects detection model based on convolutional neural networks and the periodicity defect detection model based on shot and long term memory network are built, when detecting periodicity defect, periodicity defect control program is determined by master station.The present invention can control cut deal surface quality based on the periodicity defect data of real-time detection, avoid because of quality of lot accident caused by there is periodicity defect, effectively improve cut deal product quality, production efficiency and economic benefit.
Description
Technical field
The present invention relates to steel plate defect technical field of control method, particularly relate to that there is periodic cut deal surface defect
Control method.
Background technique
Cut deal product mainly has structural steel, auto truck beam, bridge plate, Steels for High Rise Buildings, boiler vessel plate, low conjunction
Jin Gang, die steel etc. are widely used in the industries such as metallurgy, machinery, automobile, bridge, ship, electric power, building.Its surface quality
It is the important link for influencing cut deal product quality, cut deal surface defect especially periodicity defect is carried out effectively in real time
Control is the key technology emphasis and difficult point of high-efficiency high-quality production cut deal, and prevents the decisive of quality of lot accident generation
Link receives the attention of domestic and foreign scholars.
Due to the influence of high temperature and production technology, traditional cut deal surface defects detection is based on human eye ocular estimate.?
The plate surface quality situation before and after straightener can be monitored online in good means not yet during current Rolling Production,
General reached after cold bed cools down by steel plate is inspected by random samples again.When finding defect problem after the cooling of cold bed inspection desk, if
It is the rolling defect that milling train or straightener generate, then at least 25 steel plates of continuous production, the production of this continuous defect
It is raw that great economic loss can also be generated if defect is serious to subsequent handling generation extreme influence.Lower surface of steel plate is difficult to directly
Inspection is connect, turnover panel or setting lower surface reflective mirror need to be passed through and is detected by way of arranging professional to use sampling observation.It will
Steel plate hangs shear line to panel turnover machine turnover panel inspection, and increases production process task difficulty and production efficiency, and turnover panel seat in the plane
After cold bed, once the defects of secondary scuffing problem, which occurs, in lower surface will cause high-volume mass defect plate.In summary, it uses
Manual Visual Inspection and the subjective factor of the examined personnel of the testing result of sampling observation influence, and lack accuracy, reliability, continuity, complete
Whole property can not be effectively detected and be controlled to periodicity defect, not can avoid the generation of quality of lot accident.
With the development of artificial intelligence technology and CCD imaging technique, the surface detection technique based on machine vision is obtained
It is widely applied, patent " a kind of online test method of small defect on metal plate strip surface " (201410137363.X), " steel plate
Surface defect inspection apparatus and surface defect inspection method " (201680014977.6), " multistage special based on convolutional neural networks
Surface Defects in Steel Plate detection method of sign " (201810338076.3) etc. proposes Surface Defects in Steel Plate online test method, can
It realizes the unartificial continuous detection of surface defect, classification and record, and testing result is effectively analyzed, extracts useful information,
And then more precisely judge and assess cut deal surface quality situation, and carry out effectively to periodicity defect using control strategy
Ground control has very important realistic meaning to product quality, production efficiency and product core competitiveness is improved.Patent is " a kind of
The analysis method (201710659994.1) of the periodicity defect of strip " examines instrument by table and carries out defect recognition, and passes through three
The default defect characteristic in stage realizes the detection of periodicity defect.In summary, foregoing invention is not specifically for cut deal week
Phase property defect is detected, and also cannot achieve the periodicity defect real-time control based on testing result.Therefore, it has been disclosed that deliver
There is no the report for carrying out real-time online detection for cut deal periodicity defect and being controlled in document.
Summary of the invention
The cut deal periodicity defect control based on real-time detection result that the technical problem to be solved in the present invention is to provide a kind of
Method processed its object is to real-time detection cut deal periodicity defect and is realized periodicity defect control, is avoided because there is roll marks
And quality of lot accident caused by quasi-periodic defect is scratched, to improve Heavy Plate Production efficiency, product quality and economic benefit.
This method comprises the following steps:
(1) it builds surface defects detection system and obtains defect image: by the surface defect being mounted on Heavy Plate Production line
The defect image of detection system acquisition cut deal.
(2) detection network and initiation parameter are built: building the convolutional Neural for the detection of cut deal surface common deficiency
Network model and shot and long term memory network model for periodicity defect detection, input is added in shot and long term memory network algorithm
Door forgets door and out gate, and three doors are controlled information in network using sigmoid activation primitive and transmitted.And to two networks
Carry out initiation parameter setting.
(3) common deficiency data are obtained: extracting the feature of surface defect in cut deal image simultaneously using convolutional neural networks
Classification obtains the data such as defect type, position, size, quantity.
(4) off-line training generates periodicity defect detection model: by the roll marks of step (3) convolutional neural networks acquisition, drawing
Point defective data is injured as training dataset, the shot and long term memory network is trained, to generate periodicity defect
Detection model.
(5) cut deal periodicity defect real-time detection: the cut deal defective data that step (3) real-time online is obtained inputs
Into the periodicity defect detection model described in step (4) based on shot and long term memory network, cut deal periodicity defect is obtained
Testing result.
(6) when detecting periodicity defect, knot cut deal periodicity defect control strategy: will test by data processing end
Fruit is input to the auxiliary console, master station and MES production system of controlling terminal, determines periodicity defect controlling party by master station
Case eliminates periodicity defect or shutdown roll change operation by reconditioning process with clear, to realize the effective of cut deal periodicity defect
Control.
Further, the surface defects detection system that this method is related to is whole by Image Acquisition end, data processing end and control
End composition includes upper and lower surfaces detection unit, parallel computer processing system and all kinds of servers, console composition, cooler
Structure, network data exchange machine etc..Image Acquisition end is made of cut deal upper and lower surfaces detection unit etc., by Image Acquisition end reality
When obtain cut deal upper and lower surfaces image.
Further, it builds for the convolutional neural networks model of cut deal surface common deficiency detection and for periodically
The shot and long term memory network model of defects detection, the number such as cut deal common deficiency type of the present invention, position, size, quantity
It is extracted according to by convolutional neural networks, convolutional neural networks are collectively constituted by input layer, convolutional layer, pond layer and full articulamentum, packet
Containing 12 convolutional layers and 3 pond layers.Input gate is added in shot and long term memory network algorithm of the present invention, forgets door and defeated
It gos out, three doors are controlled information in network using sigmoid activation primitive and transmitted.And initiation parameter is carried out to two networks
Setting.
The shot and long term memory network detection cycle defect process is made of four steps.The first step determines any information
It should forget, be executed by the Sigmod layer of " forgeing door ", input ht−1And xt, then in Ct−1 Each neuron state output 0
Number between ~ 1." 1 " expression " this is fully retained ", " 0 " expression " forgeing this completely "；Second step decision will be in neuron
Any information is saved in cell, the Sigmod layer of " forgeing door " determines the numerical value to be updated, one new candidate of tanh layers of generation
Numerical value Ct, increase in neuron state；Third step is to old state multiplied by a ft, forget to fall to determine the letter to be forgotten before
Breath, then increases new candidate value it×Ct, measured by updating the value of each state in much degree；4th step determines output
Value determines that the neuron state of which part exports using Sigmod layers；Then neuron state is allowed to allow output valve by tanh(
Become between -1 ~ 1) layer and it is multiplied by the output of Sigmod thresholding, it is final only to export the result for wanting output.
Further, periodicity defect detection model is generated by off-line training, by the roll marks of convolutional neural networks acquisition, is drawn
It injures point defective data to be grouped as training dataset, the feature vector after grouping is inputted into length according to timestamp
Phase memory network.After the feature vector input shot and long term memory network of each group of addition label, the last one cell factory is taken
(cell) input Boolean (H) is result.The sample of input is represented when numerical value is 1 as periodicity defect sample, otherwise not
It is.
Further, cut deal periodicity defect testing result processing step are as follows: pass through cut deal surface defects detection system
System acquisition cut deal defect image, and the information such as type, position and the quantity of each defect are obtained using convolutional neural networks, it will
Disadvantages described above information input carries out off-line training into the shot and long term memory network and generates periodicity defect detection model, will
The collected cut deal defect image of real-time online forms defective data after handling via convolutional neural networks, and will be above-mentioned real-time
Defective data, which is input in periodicity defect detection model, obtains the testing result of cut deal periodicity defect.
Further, when detecting periodicity defect roll marks, scuffing and point defect, according to the position of defect, size
And quantity determines the control strategy of periodicity defect, when such as there is roll marks, then issues the instruction for shutting down roll change, by master station to disappear
Influence except roll marks to subsequent batches cut deal quality；When such as there is scuffing, it should also shutdown inspection and the equipment for eliminating scuffing
Factor, to realize effective control of cut deal periodicity defect.
The advantageous effects of the above technical solutions of the present invention are as follows:
Traditional cut deal surface periodic defect cannot achieve real-time online detection and effectively control, and the present invention is by building length
Short-term memory network model detects cut deal periodicity defect with real-time online, realizes cut deal on this basis and periodically lacks
Sunken effective control is avoided because of quality of lot accident caused by roll marks occur and scratching quasi-periodic defect, to improve cut deal
Production efficiency, product quality and economic benefit.
Detailed description of the invention
Fig. 1 is a kind of cut deal periodicity defect control method flow chart based on real-time detection result.
Fig. 2 is defects detection and periodicity defect control system schematic diagram, in figure: 1 is upper cooling-water machine, and 2 be camera, and 3 are
Light source, 4 be upper surface image acquisition system, and 5 be detected cut deal, and 6 be roller-way, and 7 be lower cooling-water machine, and 8 be lower surface figure
As acquisition system, 9 be concurrent computational system, and 10 be data processing system, and 11 be periodicity defect detection system, and 12 is raw for MES
Production system, 13 be master console, and 14 be auxiliary console.
Fig. 3 is shot and long term memory network memory internal module diagram.
Specific embodiment
To keep the technical problem to be solved in the present invention, technical solution and advantage clearer, below in conjunction with attached drawing and tool
Body embodiment is described in detail.
The present invention provides a kind of cut deal periodicity defect detection method based on shot and long term memory network, as shown in Figure 1
For this method overhaul flow chart, comprising the following steps:
S1: build surface defects detection system and obtain defect image: surface defects detection system is detected single by upper and lower surfaces
The composition such as first, parallel computer processing system and all kinds of servers, console composition, cooling body, network data exchange machine.It is logical
Crossing the system can get the real-time cut deal upper and lower surfaces image comprising defect information.
S2: detection network and initiation parameter are built: builds the convolutional Neural for the detection of cut deal surface common deficiency
Network model and shot and long term memory network model for periodicity defect detection are cut deal common deficiency type, position, big
The data such as small, quantity are extracted by convolutional neural networks, and convolutional neural networks are by input layer, convolutional layer, pond layer and full articulamentum
It collectively constitutes, includes 12 convolutional layers and 3 pond layers.Input gate is added in shot and long term memory network algorithm, forgets door and defeated
It gos out, so that the weight of its own circulation is variation, allows network to forget the information currently accumulated, so as to avoid ladder
The problem of degree disappears or gradient expands, sigmoid activation primitive is utilized to control the transmitting of information in network in three doors.
Shot and long term memory network memory internal module diagram, including a unit and three doors are illustrated in figure 2, respectively
It is input gate, out gate and forgetting door.These three doors all use differentiable sigmoid function, which can guarantee three
Door obtains optimal parameter by training.Three doors all signify the transmitting that neuronal messages are controlled by information gate, and how much is distribution
New information passes to Current neural member, distributes the information of how many Current neural members to next neuron.These three doors are all
Non-linear summation unit, they include the inside and outside activation primitive of block of memory, and by multiplying come control unit
Activation primitive.Black dot in figure represents multiplying, input gate and out gate multiplied by the input value and output valve of unit, loses
Door is forgotten multiplied by the state value before unit.F represents the activation primitive of door, and usually using sigmoid function, therefore door is sharp
The value of function living is between 0 to 1.G is the input function of unit, and h is the output function of unit, they are usually used
Tanh function or sigmoid function.Weight relationship between unit and door makes to be represented by dashed line, without using the generation of dotted line
Table does not have weight relationship.The block of memory, which is exported, to be provided to the output valve of network other structures by out gate multiplying.In length
In phase memory network model, sigmoid function is for generating the value between 0 to 1 in three doors.Tanh function is generally used
Data are handled in state and output.
S3: common deficiency data are obtained: extracts the feature of surface defect in cut deal image simultaneously using convolutional neural networks
Classification obtains the data such as defect type, position, size, quantity.
S4: off-line training generates periodicity defect detection model: by the roll marks of step (2) convolutional neural networks acquisition, drawing
Point defective data is injured as training dataset, the shot and long term memory network is trained, the specific steps of which are as follows:
(1) picture feature is extracted using convolutional neural networks, feature vector dimension is (12,120,64).According to shot and long term
The input format of memory network is { batchsize, timestep, dims }, and wherein batchsize is trained batch,
Timestep is the input of shot and long term memory network timing, and dims is the data to be trained of input.
(2) timestep in the input of second dimension 12 of feature vector, as shot and long term memory network, the 1st peacekeeping
It is that 3rd dimension multiplies the result is that the content that inputs of dims time point.Exchange the 1st peacekeeping the 2nd dimension position, reshape become 12,
12*64}.Be then entered into network image feature vector be one group { 12,12*64 }, { 12,12*64 } ... 12,
12*64 } } there is the feature of 10 step.Complete the conversion of Img2Text.
(3) shape of feature vector is { batch_size, w, h, kernals }.If this feature vector is divided into
Suitable number is done, each elongated rectangle vector is the input of each step of shot and long term memory network, i.e., each vector
It is X0,X2。。。XT,Xt+2。
(4) feature vector after grouping is inputted into shot and long term memory network according to timestamp.Each group of addition label
After feature vector inputs shot and long term memory network, taking the input Boolean (H) of the last one cell factory (cell) is result.When
The sample of input is represented when numerical value is 1 as periodicity defect sample, otherwise is not.Defect recognition network is carried out according to above-mentioned rule
Training, ultimately generates periodicity defect detection model.
S5: output cut deal periodicity defect real-time detection result: the real-time cut deal defective data that step (3) are obtained
It is input in the periodicity defect detection model described in step (4) based on shot and long term memory network, obtains cut deal periodicity
The real-time detection result of defect.
S6: when detecting periodicity defect, knot cut deal periodicity defect control strategy: will test by data processing end
Fruit is input to the auxiliary console, master station and MES production system of controlling terminal, determines periodicity defect controlling party by master station
Case.When detecting periodicity defect roll marks, scuffing and point defect, the period is determined according to the position of defect, size and quantity
Property defects controlling strategy, when such as there is roll marks, then by master station issue shut down roll change instruction, to eliminate roll marks to subsequent batch
The influence of secondary cut deal quality；When such as there is scuffing, it should also shutdown inspection and the apparatus factor for eliminating scuffing, with thick in realization
Effective control of plate periodicity defect.
The above is a preferred embodiment of the present invention, it is noted that for those skilled in the art
For, without departing from the principles of the present invention, several improvements and modifications can also be made, these improvements and modifications
It should be regarded as protection scope of the present invention.
Claims (6)
1. a kind of cut deal periodicity defect control method based on real-time detection result, characterized by the following steps:
(1) surface defects detection system and defect image is obtained: by the surface defects detection being mounted on Heavy Plate Production line
The defect image of system acquisition cut deal；
(2) detection network and initiation parameter: convolutional neural networks model and use for the detection of cut deal surface common deficiency
Input gate is added in the shot and long term memory network model of periodicity defect detection, in shot and long term memory network algorithm, forget door and
Out gate, three doors are controlled information in network using sigmoid activation primitive and transmitted, and carry out initialization ginseng to two networks
Number setting；
(3) cut deal surface common deficiency data are obtained: extracting surface defect in cut deal image using convolutional neural networks
Feature is simultaneously classified, and cut deal surface common deficiency type, position, size, incremental data are obtained；
(4) off-line training generate periodicity defect detection model: by step (3) convolutional neural networks obtain roll marks, scratch and
Point defective data is trained the shot and long term memory network as training dataset, to generate periodicity defect detection
Model；
(5) the cut deal defective data that step (3) real-time online obtains cut deal periodicity defect real-time detection: is input to step
Suddenly in the periodicity defect detection model described in (4) based on shot and long term memory network, the inspection of cut deal periodicity defect is obtained
Survey result；
(6) when detecting periodicity defect, it is defeated that result cut deal periodicity defect control method: will test by data processing end
Enter the auxiliary console to controlling terminal, master station and MES production system, periodicity defect control program determined by master station,
Periodicity defect or shutdown roll change operation are eliminated by reconditioning process with clear.
2. a kind of cut deal periodicity defect control method based on real-time detection result according to claim 1, special
Sign is: the surface defects detection system is made of Image Acquisition end, data processing end and controlling terminal, includes upper and lower table
Face detection unit, parallel computer processing system and all kinds of servers, console composition, cooling body, network data exchange machine
Deng；Image Acquisition end is made of cut deal upper and lower surfaces detection unit etc., and it is upper and lower to obtain cut deal in real time by Image Acquisition end
The image on surface.
3. a kind of cut deal periodicity defect control method based on real-time detection result according to claim 1, special
Sign is: described for the convolutional neural networks model of cut deal surface common deficiency detection and for periodicity defect detection
Shot and long term memory network model, convolutional neural networks of the present invention are total to by input layer, convolutional layer, pond layer and full articulamentum
It include 12 convolutional layers and 3 pond layers with composition；Input gate is added in the shot and long term memory network algorithm, forgets door
And out gate, three doors are controlled information in network using sigmoid activation primitive and transmitted；And two networks are initialized
Parameter setting；The shot and long term memory network treatment process is made of four steps:
(1) information for determining the forgetting is executed by the Sigmod layer of " forgeing door ", inputs ht-1And xt, then in Ct-1Each mind
Through the number between first state output 0~1, " 1 " expression " this is fully retained ", " 0 " expression " forgeing this completely "；
(2) information to save in neuronal cell is determined, the Sigmod layer of " forgeing door " determines the numerical value to be updated, tanh
Layer generates a new candidate values Ct ~, increase in neuron state；
(3) to old state multiplied by a ft, forget to fall to determine the information to be forgotten before, then increase new candidate value it*Ct
~, it is measured by updating the value of each state in much degree；
(4) it determines output valve, determines that the neuron state of which part exports using Sigmod layers；Then neuron state is allowed to pass through
It crosses tanh (output valve is allowed to become between -1~1) layer and is multiplied by the output of Sigmod thresholding, final only export wants output
As a result.
4. a kind of cut deal periodicity defect control method based on real-time detection result according to claim 1, special
Sign is: the cut deal surface common deficiency type, position, size, quantity are extracted by convolutional neural networks.
5. a kind of cut deal periodicity defect control method based on real-time detection result according to claim 1, special
Sign is: the periodicity defect detection model is generated by off-line training, roll marks, scuffing and the fiber crops that convolutional neural networks are obtained
Point defect data are grouped as training dataset, by the feature vector after grouping according to timestamp input shot and long term memory
Network takes the last one cell factory (cell) after the feature vector of each group of addition label inputs shot and long term memory network
Input Boolean (H) is as a result, representing the sample of input when numerical value is 1 as periodicity defect sample, otherwise be not.
6. a kind of cut deal periodicity defect control method based on real-time detection result according to claim 1, special
Sign is: the processing step of the testing result of the cut deal periodicity defect are as follows: passes through cut deal surface defects detection system
Real-time online acquires cut deal defect image, and obtains type, position and quantity of each defect etc. using convolutional neural networks
Information is obtained by disadvantages described above information input into the periodicity defect detection model based on shot and long term memory network
The testing result of cut deal periodicity defect.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910514268.XA CN110174409B (en) | 2019-06-14 | 2019-06-14 | Medium plate periodic defect control method based on real-time detection result |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910514268.XA CN110174409B (en) | 2019-06-14 | 2019-06-14 | Medium plate periodic defect control method based on real-time detection result |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110174409A true CN110174409A (en) | 2019-08-27 |
| CN110174409B CN110174409B (en) | 2021-01-12 |
Family
ID=67698504
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910514268.XA Active CN110174409B (en) | 2019-06-14 | 2019-06-14 | Medium plate periodic defect control method based on real-time detection result |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110174409B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113033974A (en) * | 2021-03-08 | 2021-06-25 | 华院计算技术(上海)股份有限公司 | Digital steel coil characteristic quality prediction and process parameter dynamic adjustment method and system based on improved LSTM network |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104267032A (en) * | 2014-08-18 | 2015-01-07 | 苏州克兰兹电子科技有限公司 | System of on-line detection of plate and strip metal surface defects |
| CN108021938A (en) * | 2017-11-29 | 2018-05-11 | 中冶南方工程技术有限公司 | A kind of Cold-strip Steel Surface defect online detection method and detecting system |
| CN109035226A (en) * | 2018-07-12 | 2018-12-18 | 武汉精测电子集团股份有限公司 | Mura defects detection method based on LSTM model |
| CN109724984A (en) * | 2018-12-07 | 2019-05-07 | 上海交通大学 | A kind of defects detection identification device and method based on deep learning algorithm |
-
2019
- 2019-06-14 CN CN201910514268.XA patent/CN110174409B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104267032A (en) * | 2014-08-18 | 2015-01-07 | 苏州克兰兹电子科技有限公司 | System of on-line detection of plate and strip metal surface defects |
| CN108021938A (en) * | 2017-11-29 | 2018-05-11 | 中冶南方工程技术有限公司 | A kind of Cold-strip Steel Surface defect online detection method and detecting system |
| CN109035226A (en) * | 2018-07-12 | 2018-12-18 | 武汉精测电子集团股份有限公司 | Mura defects detection method based on LSTM model |
| CN109724984A (en) * | 2018-12-07 | 2019-05-07 | 上海交通大学 | A kind of defects detection identification device and method based on deep learning algorithm |
Non-Patent Citations (4)
| Title |
|---|
| JINGLIN CHEN等: "An Ensemble of Convolutional Neural Networks for Image Classification Based on LSTM", 《2017 INTERNATIONAL CONFERENCE ON GREEN INFORMATICS》 * |
| LI YI等: "An End-to-End Steel Strip Surface Defects Recognition System Based on Convolutional Neural Networks", 《STEEL RESEARCH INTERNATIONAL》 * |
| TIANYUAN LIU等: "A Hybrid CNN–LSTM Algorithm for Online Defect Recognition of CO2 Welding", 《SENSORS》 * |
| 贺笛 等: "深度学习在中厚板表面缺陷识别中的应用", 《第十一届中国钢铁年会论文集-S18冶金自动化与智能管控》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113033974A (en) * | 2021-03-08 | 2021-06-25 | 华院计算技术(上海)股份有限公司 | Digital steel coil characteristic quality prediction and process parameter dynamic adjustment method and system based on improved LSTM network |
| CN113033974B (en) * | 2021-03-08 | 2022-02-18 | 华院计算技术(上海)股份有限公司 | Digital steel coil characteristic quality prediction and process parameter dynamic adjustment method and system based on improved LSTM network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110174409B (en) | 2021-01-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110598736B (en) | Power equipment infrared image fault positioning, identifying and predicting method | |
| CN109978835B (en) | Online assembly defect identification system and method thereof | |
| CN111949700A (en) | Intelligent safety guarantee real-time optimization method and system for petrochemical device | |
| CN114329940A (en) | Continuous casting billet quality prediction method based on extreme learning machine | |
| CN113033974B (en) | Digital steel coil characteristic quality prediction and process parameter dynamic adjustment method and system based on improved LSTM network | |
| CN109226282B (en) | Steel plate on-line solid solution post-rolling rapid cooling method based on Internet of things | |
| CN106682159A (en) | Threshold configuration method | |
| Yang et al. | Deep learning-based intelligent defect detection of cutting wheels with industrial images in manufacturing | |
| CN110174409A (en) | A kind of cut deal periodicity defect control method based on real-time detection result | |
| CN114418956A (en) | Method and system for detecting change of key electrical equipment of transformer substation | |
| CN116664846B (en) | Method and system for realizing 3D printing bridge deck construction quality monitoring based on semantic segmentation | |
| CN116346870B (en) | Industrial Internet of things system based on real-time monitoring of processing quality and control method | |
| Liu et al. | Disturbance robust abnormality diagnosis of fused magnesium furnaces using deep neural networks | |
| Daogang et al. | Anomaly identification of critical power plant facilities based on YOLOX-CBAM | |
| CN114782362B (en) | Welding intelligent detection system based on deep migration learning | |
| CN115908843A (en) | Superheat degree recognition model training method, recognition method, equipment and storage medium | |
| CN115446276A (en) | Continuous casting breakout early warning method for recognizing V-shaped bonding characteristics of crystallizer copper plate based on convolutional neural network | |
| CN113240096B (en) | Casting cylinder cover micro-structure prediction method based on rough set and neural network | |
| Zhang | Computer image processing and neural network technology for thermal energy diagnosis of boiler plants | |
| CN113674249A (en) | PCB printing quality detection method based on industrial internet | |
| Tao et al. | Utilization of both machine vision and robotics technologies in assisting quality inspection and testing | |
| Schnelle et al. | A real-time expert system for quality control | |
| Yang et al. | Economic statistical process control for over‐adjusted process mean | |
| CN109239074A (en) | A kind of green anode carbon block detection method based on machine vision | |
| CN111258996A (en) | Product quality multi-source deep fusion forecasting method for industrial big data |
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 |