CN116664561B - Intelligent detection system and method for welding quality AI of heat exchanger tube head - Google Patents
Intelligent detection system and method for welding quality AI of heat exchanger tube head Download PDFInfo
- Publication number
- CN116664561B CN116664561B CN202310935556.9A CN202310935556A CN116664561B CN 116664561 B CN116664561 B CN 116664561B CN 202310935556 A CN202310935556 A CN 202310935556A CN 116664561 B CN116664561 B CN 116664561B
- Authority
- CN
- China
- Prior art keywords
- tube head
- shooting
- layer
- tube
- axis
- 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
- 238000001514 detection method Methods 0.000 title claims abstract description 67
- 238000003466 welding Methods 0.000 title claims abstract description 60
- 238000000034 method Methods 0.000 title claims abstract description 33
- 230000007547 defect Effects 0.000 claims abstract description 32
- 238000003062 neural network model Methods 0.000 claims abstract description 21
- 230000008569 process Effects 0.000 claims abstract description 19
- 230000001360 synchronised effect Effects 0.000 claims abstract description 16
- 238000013135 deep learning Methods 0.000 claims abstract description 11
- 238000012545 processing Methods 0.000 claims abstract description 10
- 210000002569 neuron Anatomy 0.000 claims description 54
- 230000004913 activation Effects 0.000 claims description 42
- 230000006870 function Effects 0.000 claims description 13
- 238000010606 normalization Methods 0.000 claims description 13
- 238000011176 pooling Methods 0.000 claims description 10
- 238000004364 calculation method Methods 0.000 claims description 5
- 230000009467 reduction Effects 0.000 claims description 5
- 238000004458 analytical method Methods 0.000 claims description 4
- 239000000284 extract Substances 0.000 claims description 3
- 238000003908 quality control method Methods 0.000 abstract description 4
- 238000012423 maintenance Methods 0.000 abstract description 3
- 238000010586 diagram Methods 0.000 description 6
- 238000005457 optimization Methods 0.000 description 4
- 230000008646 thermal stress Effects 0.000 description 3
- 238000013459 approach Methods 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 2
- 238000001816 cooling Methods 0.000 description 2
- 238000009826 distribution Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 239000012535 impurity Substances 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 238000007781 pre-processing Methods 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- 238000004140 cleaning Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 238000005336 cracking Methods 0.000 description 1
- 230000003247 decreasing effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000008030 elimination Effects 0.000 description 1
- 238000003379 elimination reaction Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000001105 regulatory effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000007789 sealing Methods 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 238000005493 welding type Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- 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
- 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
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- 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
- G06N3/048—Activation functions
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/42—Global feature extraction by analysis of the whole pattern, e.g. using frequency domain transformations or autocorrelation
- G06V10/422—Global feature extraction by analysis of the whole pattern, e.g. using frequency domain transformations or autocorrelation for representing the structure of the pattern or shape of an object therefor
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/54—Extraction of image or video features relating to texture
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- 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
- G01N2021/8887—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 based on image processing techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
- G06T2207/30152—Solder
-
- 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)
- General Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Multimedia (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computational Linguistics (AREA)
- Molecular Biology (AREA)
- Mathematical Physics (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Medical Informatics (AREA)
- Quality & Reliability (AREA)
- Signal Processing (AREA)
- Chemical & Material Sciences (AREA)
- Analytical Chemistry (AREA)
- Biochemistry (AREA)
- Immunology (AREA)
- Pathology (AREA)
- Image Analysis (AREA)
- Butt Welding And Welding Of Specific Article (AREA)
Abstract
The invention relates to the technical field of detection, in particular to an intelligent detection system and method for the welding quality AI of a heat exchanger tube head, comprising the following steps: the image acquisition unit performs rotary shooting around a set axis, and the shooting inclination faces the axis; the inclination angle of the image acquisition unit meets the requirement of full-length synchronous shooting, and the rotation of the image acquisition unit meets the requirement of full-range synchronous shooting in the circumferential direction; the power unit is used for providing the required motion power for the image acquisition unit; an image processing unit that processes the image information from the image acquisition unit; and the deep learning unit learns the processed image information through a neural network model and outputs a welding defect detection result of the tube head. The invention can collect, process and optimize the image and detect the defect based on deep learning, these advantages make the system detect the welding defect of the heat exchanger tube head fast and accurately, raise the detection efficiency and accuracy, offer the powerful support for quality control and maintenance.
Description
Technical Field
The invention relates to the technical field of detection, in particular to an AI intelligent detection system and method for welding quality of a heat exchanger tube head.
Background
In the heat exchanger, the tube head is a part of the tube plate and is used for connecting the pipeline and the inlet and outlet pipelines, the tube head is usually fixed on the tube plate by welding, and the following problems are easy to occur during welding:
during welding, the weld zone is prone to cracking due to thermal stresses and shrinkage during cooling, which may be caused by thermal stresses exceeding the material's capacity due to welding temperature gradients, or by improper handling and control during welding; in the welding process, if gas which is not completely removed exists or a welding area is polluted, air holes are formed, and the strength and the sealing performance of a welding joint are affected by the air holes; for the weld area, impurities, oxides, or other foreign substances may be present that affect the strength and integrity of the weld joint; in addition, thermal stresses and shrinkage caused by cooling during welding may cause deformation of the welded joint and adjacent parts, which may affect the overall structure and performance of the heat exchanger.
In order to avoid the defects, the quality control of the welding position of the tube head is important, and the welding defects are reduced by adopting proper welding technology, using proper welding materials and ensuring the cleaning and impurity removal of a welding area; on the basis of the measures, strict quality detection and inspection are key steps for ensuring the delivery quality of products.
At present, the manual detection mode is time-consuming and labor-consuming, and an operator is required to check each tube head one by one, so that the efficiency is low, and the cost is high; because of the intervention of human subjective factors, the accuracy of manual detection is limited to a certain extent, different operators may have different judgment standards, and the consistency and reliability of the results are reduced; the conventional method requires a highly experienced operator for detection, which means that training and expertise are required, increasing labor costs and technical thresholds.
Disclosure of Invention
The invention provides an AI intelligent detection system and method for the welding quality of a heat exchanger tube head, thereby effectively solving the problems pointed out in the background art.
In order to achieve the above purpose, the technical scheme adopted by the invention is as follows:
an intelligent heat exchanger tube head welding quality AI detection system, comprising:
the image acquisition unit performs rotary shooting around a set axis, wherein the shooting is inclined towards the axis, and the axis is parallel to the axis of the tube head; the inclination angle of the image acquisition unit meets the full length of the inner walls and the outer walls of the tube heads in the axis direction, which are set in the detection range, so that synchronous shooting is realized, and the rotation of the image acquisition unit meets the full range of the inner walls and the outer walls of the tube bodies in the circumferential direction, which are set in the detection range, so that synchronous shooting is realized;
the power unit is fixedly connected with the image acquisition unit and provides required motion power for the image acquisition unit;
an image processing unit that processes image information from the image acquisition unit;
and the deep learning unit learns the processed image information through a neural network model and outputs a welding defect detection result of the pipe head.
Further, the neural network model includes:
an input layer for receiving the processed image information as an input;
the convolution layer carries out convolution operation on the image information, extracts the characteristics of the image and outputs a characteristic map;
a pooling layer for reducing the space dimension of the feature map;
the full-connection layer is used for flattening the feature map output by the pooling layer and connecting the feature map to one or more full-connection layers so as to learn the advanced features of the image;
and the output layer is used for classifying and outputting welding defects according to task requirements.
Further, the neural network model further comprises a batch normalization layer, wherein the batch normalization layer is arranged behind the convolution layer and normalizes a batch of feature images of each tube plate.
An intelligent detection method for the welding quality AI of a heat exchanger tube head comprises the following steps:
determining a detection range and a set number of tube heads which are taken as shooting objects in the detection range;
determining an axis between the tube heads, wherein the axis is parallel to the axis of the tube head, the axis is used as a rotating shaft for rotating and shooting the tube heads, and the rotation satisfies the full-range synchronous shooting of the inner wall and the outer wall of the tube body in the circumferential direction, which are set in a set number in a detection range;
determining a shooting angle relative to the axis, wherein the shooting angle meets the full-length synchronous shooting of the inner wall and the outer wall of the tube head in the axis direction in a set number in a detection range;
performing 360-degree rotation shooting, and processing the acquired image information;
and learning the processed image information through a neural network model, and outputting a welding defect detection result of the tube head.
Further, learning the processed image information through a neural network model includes:
receiving the processed image information;
performing convolution operation on the image information, extracting the characteristics of the image and outputting a characteristic map;
reducing the space dimension of the feature map and outputting the feature map;
flattening the feature map after dimension reduction, and connecting the feature map to one or more fully connected layers to learn advanced features of the image;
and aiming at task requirements, carrying out classification output of welding defects.
Further, learning the processed image information through a neural network model, and further includes: and normalizing a batch of the characteristic graphs correspondingly output by each tube plate.
Further, flattening the feature map after dimension reduction, and connecting the feature map to a plurality of fully connected layers, including:
different activation functions are in one-to-one correspondence with a plurality of full connection layers;
the number of neurons of each fully-connected layer is determined according to the activation degree of the layer.
Further, determining the number of neurons of each fully-connected layer according to the activation degree of the layer comprises:
collecting the activation value of each neuron aiming at a set full-connection layer to obtain an activation value set;
calculating at least one of a maximum value, a minimum value, an average value and a standard value for the activation value set;
and determining the number of neurons of the full connection layer according to the calculation result.
Further, after determining the number of neurons, a revision procedure for the number of neurons is further included, including:
analyzing the correlation between the tube plate and tube head connection form and the number of neurons;
according to the analysis result of the correlation, a corresponding weight is given to the tube plate and the tube head connection form;
and automatically revising the quantity of the neurons according to the specific requirements of the corresponding weight and welding quality detection.
Further, after the corresponding weight is obtained, the number of neurons of each fully connected layer is synchronously revised.
By the technical scheme of the invention, the following technical effects can be realized:
the invention can collect, process and optimize the image and detect the defect based on deep learning, these advantages make the system detect the welding defect of the heat exchanger tube head fast and accurately, raise the detection efficiency and accuracy, offer the powerful support for quality control and maintenance.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are required to be used in the embodiments or the description of the prior art will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments described in the present invention, and other drawings may be obtained according to the drawings without inventive effort to those skilled in the art.
FIG. 1 is a frame diagram of an AI intelligent detection system for welding quality of a heat exchanger tube head;
FIG. 2 is a first schematic diagram of a shooting angle of an image acquisition unit and an image information transmission path;
FIG. 3 is a schematic diagram of a six-axis robot arm connected to an image acquisition unit;
FIG. 4 is a second schematic diagram of the shooting angle of the image acquisition unit and the image information transmission path;
FIG. 5 is a frame diagram of a neural network model;
FIG. 6 is an optimization view of FIG. 5;
FIG. 7 is a flow chart of a method for intelligently detecting the welding quality AI of a heat exchanger tube head;
FIG. 8 is a flowchart for learning the processed image information through a neural network model;
FIG. 9 is a flow chart for determining the number of neurons of a fully connected layer;
FIG. 10 is a flowchart of a revision procedure for the number of neurons;
reference numerals: 1. a tube sheet; 2. a tube head; 3. a six-axis mechanical arm; 4. an axis; 100. an image acquisition unit; 200. an image processing unit; 300. a deep learning unit; 310. an input layer; 320. a convolution layer; 330. pooling layers; 340. a full connection layer; 350. an output layer; 360. normalizing the layers in batches; 400. a power unit.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments.
Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. The terminology used in the description of the invention herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and/or" as used herein includes any and all combinations of one or more of the associated listed items.
Example 1
An intelligent heat exchanger tube head welding quality AI detection system, comprising: an image acquisition unit 100 that performs rotation shooting around a set axis 4, the shooting being inclined toward the axis 4, and the axis 4 being parallel to the axis 4 of the ferrule 2; the inclination angle of the image acquisition unit 100 meets the full length of the inner walls and the outer walls of the tube heads 2 in the direction of the axis 4 in a set number in the detection range, and the rotation of the image acquisition unit 100 meets the full range of the inner walls and the outer walls of the tube bodies in the direction of the circumference in the set number in the detection range, so that synchronous shooting is realized; the power unit 400 is fixedly connected with the image acquisition unit 100 and provides the image acquisition unit 100 with the required motion power, and can adopt a six-axis mechanical arm 3 for power output; an image processing unit 200 that processes image information from the image acquisition unit 100; the deep learning unit 300 learns the processed image information by the neural network model and outputs the welding defect detection result of the ferrule 2.
As shown in fig. 1 to 4, in the invention, the image acquisition unit 100 performs rotation shooting around the set axis 4 and inclines towards the axis 4, so that full-length synchronous shooting of the inner wall and the outer wall of the tube head 2 in the direction of the axis 4 is realized. Meanwhile, through the support of the rotary power device, full-range synchronous shooting of the inner wall and the outer wall of the tube head 2 in the circumferential direction is realized, and the efficient and comprehensive image acquisition mode ensures that complete and continuous image data of the tube head 2 are acquired. The image processing unit 200 processes the image information from the image acquisition unit 100, including preprocessing, enhancement, and optimization, and improves the image quality by denoising, contrast enhancement, brightness adjustment, color balance adjustment, and other processes, so that key features are highlighted, and clear and accurate image data is provided for subsequent defect detection. The deep learning unit 300 learns and analyzes the processed image information through a neural network model, and trains through a large number of marked image samples, the deep learning unit 300 can automatically learn and identify different types of welding defect characteristics, and the defect detection method based on the deep learning has high accuracy and robustness, and can rapidly and accurately output the welding defect detection result of the pipe head 2.
In summary, the technical advantages of the intelligent detection system for the welding quality AI of the heat exchanger tube head include efficient and comprehensive image acquisition, image processing optimization and defect detection based on deep learning, so that the system can rapidly and accurately detect the welding defect of the heat exchanger tube head 2, the detection efficiency and accuracy are improved, and powerful support is provided for quality control and maintenance.
In practice, the choice of axis 4 may coincide with axis 4 of cartridge 2, in which case the set number of cartridges 2 in the detection range may be one or more; alternatively, the selection of the axis 4 may be set offset from the axis of the tip 2, in which case the set number of tips 2 may be one or more within the same detection range; because of the high density of tube heads 2 relative to tube sheet 1, it is preferred that each four tube heads 2 be a group of a number selected for shooting, in which case the axis of rotation is located between the four tube heads 2. However, in either case, it is ensured that the set number of tube heads 2 required to perform image capturing cannot block each other in the rotation capturing process to affect the full-range image capturing, and that the full-length image capturing of the inner wall and the outer wall of the tube head 2 is required to be ensured in the full-range rotation process, so that omission of the defect position is avoided. In the implementation process, the key is that the axis 4, the set number of the tube heads 2 and the setting position of the image acquisition unit 100 relative to the axis are reasonably selected according to the specific layout of the tube heads 2 and the design of the system, so that the image acquisition unit can not be blocked mutually in the rotary shooting process, the images of the inner wall and the outer wall of the tube heads 2 can be completely acquired, and the adjustment of the inclination angle, the shooting distance and the distance from the axis 4 are key measures for realizing the full-range image acquisition.
As a preference of the above embodiment, as shown in fig. 5, the neural network model includes: an input layer 310 receiving the processed image information as input, typically represented as a multi-dimensional array, each element representing a pixel value or feature of the image; a convolution layer 320 for performing convolution operation on the image information, extracting features of the image, and outputting a feature map; the convolution layer 320 performs convolution operation on the input image information using a series of filters to generate feature maps, wherein each feature map corresponds to a filter, and represents the response degree of the filter to different features; pooling layer 330, reducing the spatial dimension of the feature map; reducing the number of parameters and the amount of computation, common pooling operations include maximum pooling (MaxPooling) and average pooling (AveragePooling), reducing the size of feature maps by selecting the maximum or average value in a local area, while preserving the main features; a full connection layer 340 flattening the feature map output by the pooling layer 330 and connecting to one or more full connection layers 340 to learn advanced features of the image; each neuron in the fully connected layer 340 is connected to all neurons of the previous layer, learning a high-level feature representation of the image by learning weights and biases; the output layer 350 performs classification output of welding defects according to task requirements, specifically may perform classification output by using different activation functions and output coding modes, and specifically, may perform multi-classification output by using a softmax activation function for a classification task of welding defects.
The neural network model structure can be subjected to proper training and optimization, can be subjected to feature extraction and classification according to the input image information, and can output classification results about welding defects. It should be noted that the specific network architecture and layer number, and the parameter setting of each layer should be adjusted and optimized according to the actual data set and task requirements, and meanwhile, the techniques of data preprocessing, regularization, dropout and the like may also be applied in the network to improve the performance and generalization capability.
In a specific scenario of the present invention, for each tube plate 1, a plurality of tube heads 2 will be corresponding, taking image acquisition for four tube heads 2 at a time as an example, a batch of processed images will be obtained for each tube plate 1, in the welding quality detection of tube plates 1 and tube heads 2, each tube plate 1 may have different shapes, sizes and defect characteristics, in order to further improve accuracy of defect detection, as a preferred embodiment of the present invention, as shown in fig. 6, the neural network model further includes a batch normalization layer 360, where after the batch normalization layer 360 is disposed on the convolution layer 320, a batch of feature maps of each tube plate 1 is normalized.
The batch normalization layer 360 performs normalization operation on the feature map of each tube sheet 1 batch to ensure that the feature map of each tube sheet 1 has similar statistical properties after normalization, so that data consistency can be maintained, and variability among the tube sheet 1 data is reduced. In the welding quality detection of the tube plates 1 and the tube heads 2, the characteristic diagrams of each tube plate 1 can be regulated to be similar in scale and distribution range by normalizing, so that the model can learn the common characteristics of the tube plates better, and the defect detection accuracy is improved; through the use of the batch normalization layer 360, the difference among the data of different tube plates 1 can be reduced, the numerical range of each characteristic is balanced, and the parameter updating of the model is more stable and reliable, so that the problems of gradient elimination and gradient explosion are facilitated to be relieved, the convergence process of the model is accelerated, and the stability of the model is improved.
For example, each tube sheet 1 contains several profiles, which for each tube sheet 1 we can consider as a matrix or tensor, assuming each profile has a size of [ H, W, C ], where H represents height, W represents width, and C represents the number of channels.
Before the batch normalization layer 360, the numerical range of each feature map may be different, there may be different mean and variance, and through the batch normalization layer 360, we normalize the feature maps within each tube sheet 1, i.e., normalize the feature maps for each tube sheet 1 batch. For example, for a batch of feature maps of tube sheet 1, the mean and variance of the batch are calculated, and then each pixel value in the feature map is subtracted by the mean and divided by the variance to achieve normalization.
Example two
An intelligent detection method for the welding quality AI of a heat exchanger tube head, as shown in FIG. 7, comprises the following steps:
s10: determining a detection range and a set number of tube heads which are taken as shooting objects in the detection range;
s20: determining an axis between the tube heads, wherein the axis is parallel to the axis of the tube head, and is used as a rotating shaft for rotating and shooting the tube heads, and rotating the inner wall and the outer wall of the tube body, which are set in a detection range, in a full-range synchronous shooting in the circumferential direction;
s30: determining a shooting angle relative to the axis, wherein the shooting angle meets the full length synchronous shooting of the inner wall and the outer wall of the tube head in the axis direction in a set number in the detection range;
s40: performing 360-degree rotation shooting, and processing the acquired image information;
s50: and learning the processed image information through a neural network model, and outputting a welding defect detection result of the tube head.
As a preferable mode of the above embodiment, as shown in fig. 8, learning the processed image information by a neural network model includes:
s51: receiving the processed image information;
s52: performing convolution operation on the image information, extracting the characteristics of the image and outputting a characteristic map;
s53: reducing the space dimension of the feature map and outputting the feature map;
s54: flattening the feature map after dimension reduction, and connecting the feature map to one or more fully connected layers to learn advanced features of the image;
s55: and aiming at task requirements, carrying out classification output of welding defects.
As a preferable mode of the above embodiment, learning the processed image information by the neural network model further includes: and normalizing a batch of characteristic graphs output by each tube plate.
In the above embodiments, the technical effects achieved are the same as those of the first embodiment, and will not be described here again.
The feature map after dimension reduction is flattened and connected to a plurality of full connection layers, and the feature map comprises:
s541: different activation functions are in one-to-one correspondence for a plurality of full connection layers;
and determining the number of neurons of each fully-connected layer based on the degree of activation of that layer.
In the implementation process, different activation functions have different calculation formulas and activation ranges, so that the activation degrees of the different activation functions are different; for example, the ReLU function activates high when the input is positive, and the output is positive; the Sigmoid function has a low degree of activation when the input is negative and zero output, and a degree of activation of 1 when the input approaches positive infinity and a degree of activation of 0 when the input approaches negative infinity. For each full connection layer, the activation degree of the neurons can be evaluated by calculating the activation function output of each neuron, and in particular in the embodiment, by selecting proper activation functions and the number of the neurons, each full connection layer can extract different characteristics in the tube plate and tube head images; for example, early fully connected layers may use activation functions with a greater degree of activation to extract low-level features such as edges and textures; whereas subsequent fully connected layers may use activation functions with a smaller degree of activation to extract higher level features such as shape, defects, etc.
In the preferred scheme, the number of neurons is optimized according to the activation degree of each full-connection layer, so that the flexibility, the resource utilization efficiency and the feature learning capacity of the model can be improved, the number of neurons can be dynamically adjusted according to the requirements of different levels by the optimizing method, the model can better adapt to the requirements of welding quality detection tasks of different tube plates and tube heads, and the detection performance and accuracy are improved.
Specifically, according to the activation degree of each fully-connected layer, the number of neurons of the layer is determined, as shown in fig. 9, after step S541, including:
s542: collecting the activation value of each neuron aiming at a set full-connection layer to obtain an activation value set; specifically, during the forward propagation of the model, the activation value of each neuron of each fully connected layer is recorded;
s543: calculating at least one of a maximum value, a minimum value, an average value and a standard value for the activation value set, wherein the statistics can reflect the overall condition of the activation degree of the layer of neurons;
s544: and determining the number of neurons of the full connection layer according to the calculation result. Depending on the range and distribution of activation values, it may be decided whether the number of neurons needs to be increased or decreased.
In the implementation process, if the average value of the activation value set is larger, which means that the activation degree of most neurons is higher, the number of neurons can be considered to be increased so as to increase the capacity and expression capacity of the model; conversely, if the average value of the set of activation values is smaller, indicating that the degree of activation of most neurons is lower, it may be considered to reduce the number of neurons to increase the computational efficiency of the model and prevent overfitting. The method for determining the quantity of the neurons according to the activation degree can enable the model to be more flexible and adapt to the requirements of different data and tasks, can improve the performance and generalization capability of the model by dynamically adjusting the quantity of the neurons, and can avoid the problems of over fitting or under fitting and the like.
As a preference of the above embodiment, after determining the number of neurons, a revision process of the number of neurons is further included, as shown in fig. 10, including:
a1: analyzing the correlation between the tube plate and tube head connection form and the number of neurons; the importance and influence degree of the welding quality detection can be determined by different tube plate and tube head connection forms, wherein the connection forms comprise welding forms, welding parameters and the like, and the connection forms are closely related to the number of neurons;
a2: according to the analysis result of the correlation, a corresponding weight is given to the tube plate and the tube head connection form; the weight can represent the importance and influence degree of the connection form in the tube plate and tube head welding quality detection, and the neuron number can be correspondingly revised according to the weight;
a3: and automatically revising the quantity of the neurons according to the specific requirements of the corresponding weight and welding quality detection. According to the preferred scheme, through analyzing the correlation between the tube plate and tube head connection forms and the quantity of neurons, corresponding weights are given to different tube plate and tube head connection forms, so that the quantity of neurons is finely adjusted, the importance and influence degree of different tube plate and tube head connection forms are achieved, the quantity of neurons can be adjusted in a targeted manner, and the method is more suitable for the detection requirements of the welding quality of specific tube plates and tube heads. In the implementation process, a fixed corresponding weight is given to each connection form according to the analysis result, the corresponding weight can be a fixed value which is empirically set, the importance of each connection form can be reflected into the revision of the neuron number according to the corresponding weight based on expert opinion or the previous research result, and specifically, the revised neuron number is obtained by multiplying the corresponding weight or taking the corresponding weight as an adjustment factor and combining with a calculation formula of the neuron number.
After the corresponding weight is obtained, the number of the neurons of each full-connection layer is synchronously revised.
For each fully connected layer, the original neuron number is multiplied by the corresponding weight to obtain the revised neuron number, so that the neuron number of each fully connected layer can be kept synchronous in the revising process according to the difference of the corresponding weights.
The foregoing has outlined and described the basic principles, features, and advantages of the present invention. It will be understood by those skilled in the art that the present invention is not limited to the embodiments described above, and that the above embodiments and descriptions are merely illustrative of the principles of the present invention, and various changes and modifications may be made without departing from the spirit and scope of the invention, which is defined in the appended claims. The scope of the invention is defined by the appended claims and equivalents thereof.
Claims (3)
1. An intelligent detection method for the welding quality AI of a heat exchanger tube head is characterized by comprising the following steps:
s10: determining a detection range and a set number of tube heads which are taken as shooting objects in the detection range;
s20: determining an axis between the tube heads, wherein the axis is parallel to the axis of the tube head, the axis is used as a rotating shaft for rotating and shooting the tube heads, and the rotation satisfies the full-range synchronous shooting of the inner wall and the outer wall of the tube body in the circumferential direction, which are set in a set number in a detection range;
s30: determining a shooting angle relative to the axis, wherein the shooting angle meets the full-length synchronous shooting of the inner wall and the outer wall of the tube head in the axis direction in a set number in a detection range;
s40: performing 360-degree rotation shooting, and processing the acquired image information;
s50: learning the processed image information through a neural network model, and outputting a welding defect detection result of the tube head;
step S50 includes:
s51: receiving the processed image information;
s52: performing convolution operation on the image information, extracting the characteristics of the image and outputting a characteristic map;
s53: reducing the space dimension of the feature map and outputting the feature map;
s54: flattening the feature map after dimension reduction, and connecting the feature map to one or more fully connected layers to learn advanced features of the image;
s55: aiming at task requirements, carrying out classified output of welding defects;
further comprises: normalizing a batch of feature images correspondingly output by each tube plate;
step S54 includes:
s541: different activation functions are in one-to-one correspondence with a plurality of full connection layers;
determining the number of neurons of each fully-connected layer according to the activation degree of the layer, wherein the method comprises the following steps:
s542: collecting the activation value of each neuron aiming at a set full-connection layer to obtain an activation value set;
s543: calculating at least one of a maximum value, a minimum value, an average value and a standard value for the activation value set;
s544: determining the number of neurons of the full connection layer according to a calculation result;
also included is a revision procedure for the number of neurons, including:
a1: analyzing the correlation between the tube plate and tube head connection form and the number of neurons;
a2: according to the analysis result of the correlation, a corresponding weight is given to the tube plate and the tube head connection form;
a3: and automatically revising the quantity of the neurons according to the specific requirements of the corresponding weight and welding quality detection.
2. The intelligent detection method for the welding quality AI of the heat exchanger tube head according to claim 1, wherein after the corresponding weight is obtained, the number of neurons of each fully connected layer is synchronously revised.
3. An intelligent detection system for the welding quality AI of a heat exchanger tube head, which adopts the intelligent detection method for the welding quality AI of the heat exchanger tube head according to claim 1, and is characterized by comprising the following steps:
the image acquisition unit performs rotary shooting around a set axis, wherein the shooting is inclined towards the axis, and the axis is parallel to the axis of the tube head; the inclination angle of the image acquisition unit meets the full length of the inner walls and the outer walls of the tube heads in the axis direction, which are set in the detection range, so that synchronous shooting is realized, and the rotation of the image acquisition unit meets the full range of the inner walls and the outer walls of the tube bodies in the circumferential direction, which are set in the detection range, so that synchronous shooting is realized;
the power unit is fixedly connected with the image acquisition unit and provides required motion power for the image acquisition unit;
an image processing unit that processes image information from the image acquisition unit;
the deep learning unit learns the processed image information through a neural network model and outputs a welding defect detection result of the tube head;
the neural network model includes:
an input layer for receiving the processed image information as an input;
the convolution layer carries out convolution operation on the image information, extracts the characteristics of the image and outputs a characteristic map;
a pooling layer for reducing the space dimension of the feature map;
the full-connection layer is used for flattening the feature map output by the pooling layer and connecting the feature map to one or more full-connection layers so as to learn the advanced features of the image;
the output layer is used for classifying and outputting welding defects according to task requirements;
the neural network model also comprises a batch normalization layer, wherein the batch normalization layer is arranged on the convolution layer and is used for normalizing a batch of feature images of each tube plate.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310935556.9A CN116664561B (en) | 2023-07-28 | 2023-07-28 | Intelligent detection system and method for welding quality AI of heat exchanger tube head |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310935556.9A CN116664561B (en) | 2023-07-28 | 2023-07-28 | Intelligent detection system and method for welding quality AI of heat exchanger tube head |
Publications (2)
Publication Number | Publication Date |
---|---|
CN116664561A CN116664561A (en) | 2023-08-29 |
CN116664561B true CN116664561B (en) | 2023-10-17 |
Family
ID=87728267
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310935556.9A Active CN116664561B (en) | 2023-07-28 | 2023-07-28 | Intelligent detection system and method for welding quality AI of heat exchanger tube head |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116664561B (en) |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105891215A (en) * | 2016-03-31 | 2016-08-24 | 浙江工业大学 | Welding visual detection method and device based on convolutional neural network |
CN112116560A (en) * | 2020-08-20 | 2020-12-22 | 华南理工大学 | Welding image defect identification method and device, storage medium and equipment |
CN113379737A (en) * | 2021-07-14 | 2021-09-10 | 西南石油大学 | Intelligent pipeline defect detection method based on image processing and deep learning and application |
CN114180044A (en) * | 2021-12-29 | 2022-03-15 | 西安交通大学 | Suspension type ultra-wide-angle pipeline inspection device and method |
CN114706358A (en) * | 2022-06-02 | 2022-07-05 | 张家港华程机车精密制管有限公司 | Method and system for processing welding process information of straight welded pipe |
CN115601296A (en) * | 2022-08-25 | 2023-01-13 | 淮南师范学院(Cn) | Surface defect detection device and detection method based on deep learning |
CN116008304A (en) * | 2022-12-12 | 2023-04-25 | 上海森松制药设备工程有限公司 | Weld joint detection method suitable for pipeline interior |
CN116416432A (en) * | 2023-04-12 | 2023-07-11 | 西南石油大学 | Pipeline weld image segmentation method based on improved UNet |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
FR3109325B1 (en) * | 2020-04-20 | 2022-03-18 | Air Liquide | Method and device for determining the performance of a welding process |
-
2023
- 2023-07-28 CN CN202310935556.9A patent/CN116664561B/en active Active
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105891215A (en) * | 2016-03-31 | 2016-08-24 | 浙江工业大学 | Welding visual detection method and device based on convolutional neural network |
CN112116560A (en) * | 2020-08-20 | 2020-12-22 | 华南理工大学 | Welding image defect identification method and device, storage medium and equipment |
CN113379737A (en) * | 2021-07-14 | 2021-09-10 | 西南石油大学 | Intelligent pipeline defect detection method based on image processing and deep learning and application |
CN114180044A (en) * | 2021-12-29 | 2022-03-15 | 西安交通大学 | Suspension type ultra-wide-angle pipeline inspection device and method |
CN114706358A (en) * | 2022-06-02 | 2022-07-05 | 张家港华程机车精密制管有限公司 | Method and system for processing welding process information of straight welded pipe |
CN115601296A (en) * | 2022-08-25 | 2023-01-13 | 淮南师范学院(Cn) | Surface defect detection device and detection method based on deep learning |
CN116008304A (en) * | 2022-12-12 | 2023-04-25 | 上海森松制药设备工程有限公司 | Weld joint detection method suitable for pipeline interior |
CN116416432A (en) * | 2023-04-12 | 2023-07-11 | 西南石油大学 | Pipeline weld image segmentation method based on improved UNet |
Non-Patent Citations (3)
Title |
---|
Hengyuan Hu等.Network Trimming: A Data-Driven Neuron Pruning Approach towards Efficient Deep Architectures.《arXiv》.2016,1-9. * |
Zhifen Zhang等.Weld image deep learning-based on-line defects detection using convolutional neural networks for Al alloy in robotic arc welding.《Journal of Manufacturing Processes》.2019,(第45期),208-216. * |
刘文婧等.焊缝缺陷图像智能分类研究.《组合机床与自动化加工技术》.2022,(第6期),150-154. * |
Also Published As
Publication number | Publication date |
---|---|
CN116664561A (en) | 2023-08-29 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111915572B (en) | Adaptive gear pitting quantitative detection system and method based on deep learning | |
Xu et al. | Surface defects detection and identification of lithium battery pole piece based on multi-feature fusion and PSO-SVM | |
CN111815564B (en) | Method and device for detecting silk ingots and silk ingot sorting system | |
CN108008006B (en) | Welding seam defect detection method, device, equipment and system | |
CN115797354B (en) | Method for detecting appearance defects of laser welding seam | |
CN113469951B (en) | Hub defect detection method based on cascade region convolutional neural network | |
CN112669305B (en) | Metal surface rust resistance test bench and rust resistance evaluation method | |
WO2023084543A1 (en) | System and method for leveraging neural network based hybrid feature extraction model for grain quality analysis | |
CN112258490A (en) | Low-emissivity coating intelligent damage detection method based on optical and infrared image fusion | |
CN111445471A (en) | Product surface defect detection method and device based on deep learning and machine vision | |
CN114663427B (en) | Boiler part size detection method based on image processing | |
Gao et al. | Detection of wheat unsound kernels based on improved ResNet | |
CN117152119A (en) | Profile flaw visual detection method based on image processing | |
CN111402249A (en) | Image evolution analysis method based on deep learning | |
CN116664561B (en) | Intelligent detection system and method for welding quality AI of heat exchanger tube head | |
Saleh et al. | Tire defect detection model using machine learning | |
CN111695582A (en) | Flutter texture detection method and device | |
Garcia et al. | Identification of visually similar vegetable seeds using image processing and fuzzy logic | |
CN117253192A (en) | Intelligent system and method for silkworm breeding | |
JP3648728B2 (en) | Steel surface degradation degree evaluation system using self-organizing feature map | |
CN111238927A (en) | Fatigue durability evaluation method and device, electronic equipment and computer readable medium | |
CN115618506A (en) | Method for predicting power of single-shaft combined cycle gas turbine | |
CN114549403A (en) | Multi-monomer intelligent precise geometric circle center detection method for side section of mechanical part | |
CN113077002A (en) | Machine olfaction visualization sensing data analysis method based on spatial heterodyne Raman spectrum | |
CN111767815A (en) | Tunnel water leakage identification method |
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 |