CN114782392A - Plastic part welding quality evaluation method based on ridge regression analysis convolution regularization - Google Patents
Plastic part welding quality evaluation method based on ridge regression analysis convolution regularization Download PDFInfo
- Publication number
- CN114782392A CN114782392A CN202210486848.4A CN202210486848A CN114782392A CN 114782392 A CN114782392 A CN 114782392A CN 202210486848 A CN202210486848 A CN 202210486848A CN 114782392 A CN114782392 A CN 114782392A
- Authority
- CN
- China
- Prior art keywords
- convolution
- regularization
- regression analysis
- ridge regression
- layer
- 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.)
- Pending
Links
Images
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
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a plastic part welding quality evaluation method based on ridge regression analysis convolution regularization, wherein a characteristic information extraction part controlled by a ridge regression equation eliminates common information, so that the extraction of the characteristics of a welding line region of a plastic fuel tank is realized, and the acquisition of useful information in the characteristic analysis process is increased; a Dropblock module is introduced between the convolution layer and the pooling layer by a characteristic analysis part of convolution regularization optimization, so that the phenomenon of overfitting is prevented when a network learns pictures with more common characteristics, optimal characteristics are obtained and used as identification characteristics, and 27.4% of useless characteristic graphs are abandoned. Therefore, the two modules act together to effectively improve the anti-overfitting capability of the plastic fuel tank welding line identification network, reduce the training time and ensure the robustness of the model.
Description
Technical Field
The invention relates to the technical field of plastic part weld joint evaluation, in particular to a plastic part welding quality evaluation method based on ridge regression analysis convolution regularization.
Background
In the technical field of industrial manufacturing, welding procedures are usually involved, problems of insufficient welding, uneven welding allowance, inaccurate welding pose and the like exist in a welding process required by welding plastic fuel tank accessories on a body, and finally, the quality and the service life of a product are affected.
At present, most manufacturers of plastic fuel tanks adopt manual detection or traditional destructive selective inspection methods, wherein the problems of low efficiency, easiness in omission, high labor cost, low automation degree and the like exist in visual detection, and the method cannot meet the requirements of large-scale, high-precision and quick detection. The machine vision detection technology has remarkable advantages in the aspects of efficiency, cost and the like, and mainly comprises monocular vision detection and binocular stereo vision detection. The method for evaluating the welding quality of the plastic fuel tank by adopting the deep learning based on the vision has higher precision and good adaptability. The traditional deep learning algorithm has more limitations on detection of the welding seam outline of the workpiece to be welded, and is easy to generate an overfitting phenomenon when a welding seam picture with a large amount of common characteristic information is trained, so that the accuracy of the algorithm is low.
In 2020, Zhang Zhi et al proposed an image recognition algorithm based on the combination of a weld recognition convolutional neural network and ridge regression analysis. The tire image recognition [ J ] based on the improved weld recognition convolutional neural network is realized through the learning research of the ridge regression analysis and the weld recognition convolutional neural network recognition algorithm, a new regular term is introduced into the original loss function of the algorithm, so that the ratio relation of two parts in the new loss function is changed, the amplitude of the jitter of the fitting curve of the characteristic information is reduced, the training accuracy and the recognition rate of the tire damage image are improved to a certain extent, but for a small sample plastic fuel tank sample set, more useless characteristic graphs exist in the characteristic transfer process, and the weld evaluation detection on a flexible production line for welding the plastic fuel tank is not suitable.
The method and the system (with the authorization notice number of CN108665452) for identifying the welding seam defects by establishing a data set of the welding seam defects of the pipeline and using a support vector machine based on two classifications to identify the welding seam defects are invented by Liqiang and the like of Guangdong Dapeng liquefied natural gas Limited company in 2019, and the defects of the welding seams of the pipeline can be identified.
Huke steel of the university of Zhejiang industry in 2019 and the like invented a welding visual detection method and device based on a convolutional neural network (patent grant publication No. CN 112365501B). The invention enhances the precision of automatic weld defect identification by designing an omnidirectional sensor and building a convolutional neural network model to identify a weld, and can realize the detection of the weld inside a steel pipe. The method has the defects that the existing original convolution neural network model is only used for detecting the welding seam inside the steel pipe, and the network model is large and is not suitable for detecting the welding seam of the plastic fuel tank.
Zhao Zhuang of Nanjing Physician university in 2021 invented a weldment outline detection algorithm (patent grant publication No. CN112365501B) based on convolutional neural network, which reduces the introduced texture information, enhances semantic information, concentrates on the weld of the target, improves the identification precision and further improves the performance of the algorithm by adopting layer jump connection and introducing an ASPP module and an attention module. The method has the defects that the network model is complex, the training time is long, the common characteristics are not read and identified, and the method is not suitable for detecting the welding line of the plastic fuel tank.
In summary, in order to realize the identification of the weld defects of the plastic fuel tank, the feature extraction and feature analysis part in the plastic fuel tank weld identification network needs to be improved. In order to solve the problems, the plastic fuel tank weld joint identification convolutional neural network is optimized, and the ridge regression analysis equation and the convolution regularization are combined to enhance the feature acquisition and analysis capacity of the plastic fuel tank weld joint identification network, so that the plastic part welding quality evaluation method based on ridge regression analysis convolution regularization is provided
Disclosure of Invention
The invention aims to overcome the defects of the prior art, provides a plastic part welding quality evaluation method based on ridge regression analysis convolution regularization, and can improve the precision of plastic fuel tank weld evaluation.
The purpose of the invention is realized by the following steps: a plastic part welding quality evaluation method based on ridge regression analysis convolution regularization comprises the following steps:
1) establishing and processing a plastic fuel tank welding seam evaluation sample set;
2) building a convolution regularization weld joint identification convolution neural network based on ridge regression analysis and training;
3) and outputting a detection result.
Further, the step 2) specifically includes:
2-1) constructing a weld joint identification convolutional neural network;
2-2) fusing ridge regression analysis in the weld recognition convolutional neural network, introducing a ridge regression analysis equation between the first layer convolutional layer and the pooling layer to extract characteristic information, finding out multiple co-linear information, and rejecting part of information; obtaining a new regression coefficient, and adjusting the cost function of the network through the new regression coefficient;
2-3) fusing convolution regularization in a weld joint identification convolution neural network, and introducing a Dropblock module between a second layer convolution layer and a pooling layer;
2-4) adding a Dropout layer before the full connection layer; removing feature graphs with more common features in network convolution operation;
2-5) importing the data sample set in the step 1) into the improved weld joint recognition network for training to obtain a trained network model.
Further, the ridge regression analysis equation in the step 2-2) is used for extracting the characteristic information of the welding seam of the plastic fuel tank and obtaining a new regression coefficient to adjust the cost function of the network.
Further, the DropBlock module in step 2-3) includes two important parameters: s and gamma, where s is the block size used to control the preparation for discarding, and the parameter gamma represents the probability that the discarded semantic information activates a block of cells; the calculation formula is as follows:
where the parameter γ indicates the probability that a block of semantic information activation cells is discarded, s is the size of the block used to control the preparation for discarding, f indicates the size of the feature map there, and p indicates the probability that a cell will remain active.
Further, the step 3) specifically includes: and (4) putting the detected picture information as an input value into the trained network model in the step 2.5) for recognition, and outputting a result.
By adopting the technical scheme, compared with the prior art, the invention has the beneficial effects that: on one hand, ridge regression analysis is introduced into a feature extraction part to obtain the characteristic information of the multi-collinearity box body part existing in a welding line picture of the plastic fuel tank, on the other hand, a DropBlock module is introduced between a convolution layer and a pooling layer to extract a deep network for multiple times, so that the feature maps with more common features in the network are reduced, and the ridge regression analysis and the DropBlock module are combined to enable the welding line defect identification network of the plastic fuel tank to obtain the optimal features and serve as identification features, so that the overfitting resistance of the welding line identification network of the plastic fuel tank is improved, 27.4% of useless feature maps are abandoned, the training time is shortened, and the robustness of the model is guaranteed.
Drawings
FIG. 1 is a flow chart of the present invention.
Detailed Description
The method for evaluating the welding quality of plastic parts based on ridge regression analysis convolution regularization as shown in fig. 1 comprises the following steps:
1) establishing and processing a plastic fuel tank weld evaluation sample set;
the method comprises the steps of obtaining a weld picture of the plastic fuel tank by using an industrial camera, intercepting a weld area of the plastic fuel tank, marking a sample, and establishing an experimental data sample set by using the pictures of the defective weld and the qualified weld of the conventional plastic fuel tank.
And simply preprocessing an experimental data sample set, wherein the preprocessing technology comprises horizontal turning, vertical turning, random rotation or scaling technology and the like, so that the simple expansion of the sample is realized.
2) Building a convolution regularization weld joint identification convolution neural network based on ridge regression analysis and training;
2-1) constructing a weld joint identification convolutional neural network;
the deep convolution neural network framework for weld joint identification constructed by the invention is composed of five convolution layers, three pooling layers and three full-connection layers. The convolution layer and the pooling layer mainly perform extraction operation of image characteristic information, the fully-connected layer converts the characteristic diagram into a characteristic vector, and the last fully-connected layer delivers an output result to the Softmax layer.
2-2) a ridge regression analysis equation is introduced between a first convolution layer and a pooling layer of the weld recognition deep convolution neural network framework to extract characteristic information, multiple co-linear information is found out, and a part of information is removed, so that a variable with insufficient representativeness is found out through the stability of ridge regression coefficients; regularization is a common method for preventing overfitting, and the general principle is to add a constraint term for parameters after a cost function, and the constraint term is called a regularization term;
according to the invention, the strong analysis capability of the ridge regression analysis on the common data is utilized after the first layer of the convolution layer, the data information of some plastic fuel tank welding line pictures is abandoned, the information with rich and characteristic representativeness is selected and extracted, and a new regression coefficient is obtained. And adjusting the cost function of the network by using the new regression coefficient so as to achieve the purpose of changing the proportion of the function. The overfitting phenomenon of the network model in the training process is reduced, the training capability of the model is further enhanced, and the identification accuracy of the welding line of the plastic fuel tank is improved.
2-3) then introducing a Dropblock module between the second layer of convolutional layers and the pooling layer; the calculation formula is as follows:
wherein, the parameter gamma represents the probability of discarded semantic information activating a cell block, s is the size of a square block used for controlling the preparation for discarding, f represents the size of a feature map at the place, and p represents the probability of keeping a certain cell active; the present invention sets p to 0.8.
The DropBlock module includes two important parameters: s and γ, where s is the block size used to control the ready-to-discard, the network will typically take 3, 5, 7, when block _ size is 1, DropBlock becomes the traditional Dropout module, when s is 7, the effect is best, and the parameter γ represents the probability that the discarded semantic information activates a cell block;
DropBlock randomly sets a certain communication area of the features to zero, and forces the network to learn the features of the rest areas, so that regularization of the convolution layer is realized, and overfitting resistance is improved. A DropBlock module is introduced into the network, part of weld joint features of the plastic fuel tank are abandoned randomly, and optimal features are obtained and serve as identification features through multiple extraction of a deep network, so that the overfitting resistance of the weld joint identification network of the plastic fuel tank is improved, and the robustness of the model is guaranteed.
2-4) adding a Dropout layer before the full connection layer, and removing a characteristic diagram with more common characteristics in network convolution operation;
2-5) importing the data sample set in the step 1) into the improved weld joint recognition network for training to obtain a trained network model.
3) Outputting a detection result, putting the detected picture information as an input value into the trained network model in the step 2.5) for identification, and outputting a result;
and acquiring the welding seam image information of the automobile plastic fuel tank through a vision extraction module of a punching welding robot of a flexible welding production line of the plastic fuel tank, and putting the acquired information as an input value into a trained convolutional neural network model for identification.
According to the method, through preprocessing a weld defect picture of the plastic fuel tank collected by an existing industrial camera, a weld recognition network model is established, ridge regression analysis is carried out after a first layer of convolution layer to find out multiple collinear information, a part of variables with insufficient representativeness are removed, on the basis, a DropBlock module is introduced after a second layer of convolution layer, part of weld features of the plastic fuel tank are abandoned randomly, the phenomenon that the network over-fits when learning pictures with more common features is avoided, the optimal features are obtained and used as recognition features, intelligent evaluation is carried out on the welding quality of the plastic fuel tank through a mode combining ridge regression analysis and convolution regularization, the over-fitting resistance of the weld recognition network of the plastic fuel tank is improved, and the robustness of the model is guaranteed.
The present invention is not limited to the above embodiments, and based on the technical solutions disclosed in the present invention, those skilled in the art can make some substitutions and modifications to some technical features without creative efforts based on the disclosed technical contents, and these substitutions and modifications are all within the protection scope of the present invention.
Claims (5)
1. A plastic part welding quality evaluation method based on ridge regression analysis convolution regularization is characterized by comprising the following steps of:
1) establishing and processing a plastic fuel tank welding seam evaluation sample set;
2) building a convolution regularization weld joint identification convolution neural network based on ridge regression analysis and training;
3) and outputting a detection result.
2. The ridge regression analysis convolution regularization based plastic part welding quality assessment method as claimed in claim 1, wherein said step 2) specifically comprises:
2-1) constructing a weld joint identification convolutional neural network;
2-2) fusing ridge regression analysis in the weld recognition convolutional neural network, introducing a ridge regression analysis equation between the first layer convolutional layer and the pooling layer to extract characteristic information, finding out multiple co-linear information, and rejecting part of information; obtaining a new regression coefficient, and adjusting the cost function of the network through the new regression coefficient;
2-3) fusing convolution regularization in a weld joint identification convolution neural network, and introducing a Dropblock module between the second layer convolution layer and the pooling layer;
2-4) adding a Dropout layer before the full connection layer; removing feature graphs with more common features in network convolution operation;
2-5) importing the data sample set in the step 1) into the improved weld joint recognition network for training to obtain a trained network model.
3. The plastic part welding quality assessment method based on ridge regression analysis convolution regularization as claimed in claim 2, wherein said ridge regression analysis equation in step 2-2) is a cost function for extracting weld characteristic information of the plastic fuel tank and obtaining new regression coefficients to adjust the network.
4. The ridge regression analysis convolution regularization based plastic part weld quality assessment method as claimed in claim 2, wherein said DropBlock module in step 2-3) includes two important parameters: s and gamma, where s is the size of the block to be discarded and the parameter gamma indicates the probability that the discarded semantic information activates the block; the calculation formula is as follows:
where the parameter γ indicates the probability that a block of semantic information activation cells is discarded, s is the size of the block used to control the preparation for discarding, f indicates the size of the feature map there, and p indicates the probability that a cell will remain active.
5. The ridge regression analysis convolution regularization based plastic part welding quality assessment method as claimed in claim 2, wherein said step 3) specifically comprises: and (3) putting the detected picture information as an input value into the trained network model in the step 2.5) for recognition, and outputting a result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210486848.4A CN114782392A (en) | 2022-05-06 | 2022-05-06 | Plastic part welding quality evaluation method based on ridge regression analysis convolution regularization |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210486848.4A CN114782392A (en) | 2022-05-06 | 2022-05-06 | Plastic part welding quality evaluation method based on ridge regression analysis convolution regularization |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114782392A true CN114782392A (en) | 2022-07-22 |
Family
ID=82434394
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210486848.4A Pending CN114782392A (en) | 2022-05-06 | 2022-05-06 | Plastic part welding quality evaluation method based on ridge regression analysis convolution regularization |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114782392A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116038112A (en) * | 2022-12-06 | 2023-05-02 | 西南石油大学 | Laser tracking large-scale curved plate fillet welding system and method |
-
2022
- 2022-05-06 CN CN202210486848.4A patent/CN114782392A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116038112A (en) * | 2022-12-06 | 2023-05-02 | 西南石油大学 | Laser tracking large-scale curved plate fillet welding system and method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111815572B (en) | Method for detecting welding quality of lithium battery based on convolutional neural network | |
| CN113239930B (en) | Glass paper defect identification method, system, device and storage medium | |
| CN112233090B (en) | Film flaw detection method based on improved attention mechanism | |
| CN107316301B (en) | Convenient online detection system for concentration of suspended particles in sewage | |
| CN105913415A (en) | Image sub-pixel edge extraction method having extensive adaptability | |
| CN108802041B (en) | Method for rapidly changing small sample set of screen detection | |
| CN114581782B (en) | Fine defect detection method based on coarse-to-fine detection strategy | |
| CN114782392A (en) | Plastic part welding quality evaluation method based on ridge regression analysis convolution regularization | |
| CN112950576B (en) | Power transmission line defect intelligent identification method and system based on deep learning | |
| CN116843999B (en) | Gas cylinder detection method in fire operation based on deep learning | |
| CN116612123B (en) | Visual detection method for peanut oil processing quality | |
| CN113628178A (en) | Method for detecting surface defects of steel products with balanced speed and precision | |
| CN110619418A (en) | Multi-feature water quality prediction method based on mixed model combination algorithm | |
| CN117218083A (en) | Improved YOLOv8 steel pipe weld defect detection method | |
| CN115423741A (en) | Structural defect detection method based on deep learning | |
| CN113780111B (en) | Pipeline connector defect accurate identification method based on optimized YOLOv3 algorithm | |
| CN114037684A (en) | Defect detection method based on yolov5 and attention mechanism model | |
| Chen et al. | Research on anti-interference detection of 3D-printed ceramics surface defects based on deep learning | |
| CN112967271A (en) | Casting surface defect identification method based on improved DeepLabv3+ network model | |
| CN112001919B (en) | Fabric defect detection method based on particle swarm algorithm and two-dimensional digital filtering | |
| CN113034446A (en) | Automatic transformer substation equipment defect identification method and system | |
| CN117269168A (en) | New energy automobile precision part surface defect detection device and detection method | |
| CN116342542A (en) | Lightweight neural network-based steel product surface defect detection method | |
| CN116205877A (en) | Intelligent detection method for weld joint surface defects based on deep reinforcement learning | |
| CN114018951A (en) | Plastic part weld joint detection method based on DCGAN (DCGAN-based gas diffusion analysis) significant feature sample expansion pretreatment |
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 |