CN115310486B - Intelligent welding quality detection method - Google Patents
Intelligent welding quality detection method Download PDFInfo
- Publication number
- CN115310486B CN115310486B CN202210951200.XA CN202210951200A CN115310486B CN 115310486 B CN115310486 B CN 115310486B CN 202210951200 A CN202210951200 A CN 202210951200A CN 115310486 B CN115310486 B CN 115310486B
- Authority
- CN
- China
- Prior art keywords
- welding
- welding quality
- vibration
- channel
- module
- 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
Landscapes
- Investigating Or Analyzing Materials By The Use Of Ultrasonic Waves (AREA)
Abstract
The invention discloses an intelligent detection method for welding quality, which comprises the following steps: step one: arranging an excitation point and a plurality of vibration sensors on a welded product; step two: the welding product is vibrated, vibration signals passing through the welding part are collected, and the vibration signals collected by the vibration sensors form a multi-channel vibration signal; acquiring a classification label of the welding quality of the current welding product, and forming welding quality data by the multi-channel vibration signals and the classification label together; step three: judging whether the number of the obtained welding quality data reaches a set threshold value or not; if yes, a data set is obtained, and a fifth step is executed; if not, executing the fourth step; step four: arranging excitation points and vibration sensors at the same positions of another welding product, and executing the second step; step five: constructing an artificial intelligence algorithm model, and training the artificial intelligence algorithm model by utilizing the data set; step six: and (5) carrying out on-line detection on welding quality of the welding product by using an artificial intelligence algorithm model.
Description
Technical Field
The invention belongs to the technical field of welding, and particularly relates to an intelligent welding quality detection method.
Background
The welding process is used as the main connecting process method and is widely applied to important commodities such as high-speed trains, automobiles and the like. For example, there are more than 5000 welds on a car. Therefore, the quality of welding is directly related to the quality of commodity and even the safety problem. At present, detection means for welding quality can be mainly divided into manual judgment and equipment detection. The manual detection is mainly judged through experience of a welder, and the welding quality is judged through observing the welding quality and the flatness of a welding area; the welding quality detection method based on manual experience is often inaccurate in judgment result due to non-uniform standards. The equipment detection is mainly used for detecting welding parts through equipment such as an ultrasonic flaw detector; the welding quality detection method based on equipment is low in efficiency and high in equipment cost.
Chinese patent publication No. CN113909667B discloses a welding quality evaluation method of an ultrasonic welder based on vibration data, by setting a first vibration acceleration sensor under a welding seat and a second vibration acceleration sensor on a side surface of the welding seat, in a welding process, vibration data in the welding process is collected by the first vibration acceleration sensor and the second vibration acceleration sensor, a confidence interval of vibration index is obtained by using vibration data of normal welding, and the vibration index of vibration data of current welding is compared with the confidence interval, thereby evaluating the welding quality.
The welding quality evaluation method of the ultrasonic welding machine based on the vibration data can realize the evaluation requirement of the welding quality to a certain extent, but can only evaluate the standardized welding operation, and the vibration data in the welding process are collected. When unfavorable working conditions such as improper input pressure, overlapping of transfer sheets, interference of welding heads and the like occur in the welding process, the welding quality can be evaluated by reflecting the unfavorable working conditions into vibration data, and the welding quality is not directly measured and evaluated for the finally formed welding structure. Because there is no necessarily correlation between the occurrence of the adverse conditions and the final welding quality, there is a large error in the evaluation result of the method.
Disclosure of Invention
In view of the above, the invention aims to provide an intelligent detection method for welding quality, which can directly detect a finally formed welding product and can effectively improve detection precision and reliability.
In order to achieve the above purpose, the present invention provides the following technical solutions:
an intelligent detection method for welding quality comprises the following steps:
step one: arranging an excitation point and a plurality of vibration sensors on a welded product;
step two: outputting exciting force with a set waveform to an exciting point by using an exciter so as to enable a welded product to vibrate, collecting vibration signals of a welded part passing through the welded product by using vibration sensors, and forming a multi-channel vibration signal by using the vibration signals collected by the vibration sensors;
acquiring a classification label of the welding quality of the current welding product, and forming welding quality data of the current welding product by the multi-channel vibration signal and the classification label together;
step three: judging whether the number of the obtained welding quality data reaches a set threshold value or not; if yes, obtaining a data set consisting of a set number of welding quality data, and executing a step five; if not, executing the fourth step;
step four: arranging excitation points and vibration sensors at the same positions of the same welding product, and executing the second step;
step five: constructing an artificial intelligence algorithm model, and training the artificial intelligence algorithm model by utilizing the data set;
step six: and inputting a multichannel vibration signal into the artificial intelligent algorithm model obtained through training, outputting a classification label corresponding to the welding quality of the welding product, and detecting the welding quality of the welding product on line.
Further, the waveform of the exciting force is a sine wave, a square wave, a triangular wave or a rectangular wave.
Further, the data set is divided into a training set, a verification set and a test set for training, verifying and testing the artificial intelligence algorithm model, respectively.
Further, the artificial intelligence algorithm model includes a deep learning model, a machine learning model, or a reinforcement learning model.
Further, the artificial intelligence algorithm model adopts a multi-channel-scale-receptive field fusion characteristic extraction model.
Further, the multichannel-scale-receptive field fusion feature extraction model comprises an initialization module, a feature fusion layer and a classifier, wherein a parallel residual error calibration module and a multi-receptive field module are arranged between the initialization module and the feature fusion layer;
the initialization model is used for extracting preliminary data characteristics of the multichannel vibration signals, and inputting the extracted preliminary data characteristics into the residual error calibration module and the multi-receptive field module;
the residual error calibration module realizes further excavation of the data characteristics of the multichannel vibration signals through fine granularity characteristic extraction and characteristic attention calibration on multiple scales;
the multi-receptive field module acquires effective information stored in the signal through different receptive field sizes on the multi-channel;
the feature fusion layer is used for fusing the features extracted by the residual calibration module and the multi-receptive field module and inputting the fused features into the classifier;
and classifying the fused features by the separator to obtain a classification label of the welding quality of the current welding product.
Further, the feature fusion layer fuses the features extracted by the residual calibration module and the multi-receptive field module by adopting a linear fusion method:
F＝αF R +βF M
wherein F represents the fused features; f (F) R Representing the features extracted by the residual calibration module; f (F) M Representing the extracted characteristics of the multi-receptive field module; alpha and beta represent the learnable parameters, respectively.
The invention has the beneficial effects that:
according to the intelligent detection method for welding quality, the vibration exciting points and the plurality of vibration sensors are arranged on a welding product, the vibration exciter is used for outputting the vibration exciting force with the set waveform to the vibration exciting points, so that the welding product vibrates, a vibration signal can change through a welding part in the process of propagating in the welding product, the vibration signal carries relevant information of the quality condition of the welding part at the moment, and then the signal is collected by the vibration sensors; different vibration signals can be captured by the vibration sensors at different positions, and welding quality information carried by the multi-channel vibration signals formed by the vibration signals is more abundant; the artificial intelligent algorithm model acquires welding quality information from the multi-channel vibration signal, removes invalid information in the multi-channel vibration signal, then judges the welding quality condition of the current welding piece through the classifier, and outputs a classification label of the welding quality, such as qualification or disqualification, so that the technical purpose of online detection of the welding quality of the welding product is achieved;
specifically, in each measurement, the arrangement positions of the excitation point and the vibration sensor on the welded product are the same, and meanwhile, the excitation force excited by the vibration exciter is the same in each measurement, so that the consistency of each measurement can be ensured, and the detection precision and reliability can be effectively improved.
The invention also has the following advantages:
(1) The invention can realize the rapid quality detection of the welding product before the welding product is put into use, and a complex welding quality detection device is not required to be designed;
(2) The invention can be suitable for any welding products, including complex curved surface welding pieces and assembly welding pieces, namely, the invention has the advantages of wide application range and high portability, and solves the problem that the welding quality of the complex welding pieces is difficult to judge by manual experience;
(3) The invention is suitable for the online measurement of welding products with different welding modes, including but not limited to spot welding, arc welding and the like;
(4) According to the invention, an artificial intelligent algorithm model is adopted, and can automatically extract the characteristics in the multichannel vibration signals and perform data characteristic mining; in addition, the artificial intelligence algorithm model has the advantages of wide coverage, good adaptability and data driving: the artificial intelligence algorithm can be mapped to any function; because the artificial intelligence algorithm model is data-driven, a very good performance effect can be achieved as long as the data volume is enough, and in addition, the design work of characteristics or kernel functions is avoided based on the data-driven artificial intelligence algorithm model, the degree of dependence on expert experience is reduced, and the efficiency is greatly improved.
Drawings
In order to make the objects, technical solutions and advantageous effects of the present invention more clear, the present invention provides the following drawings for description:
FIG. 1 is a flow chart of the intelligent detection method of welding quality of the invention;
FIG. 2 is a physical view of an automobile welding body in white;
FIG. 3 is a flow chart of data acquisition;
FIG. 4 is a three-dimensional schematic view of a rear door of a vehicle body obtained by welding;
FIG. 5 is a schematic view of the location and number of vibration points and vibration sensors disposed on a rear door of a vehicle body;
FIG. 6 is a schematic structural diagram of a multi-channel-scale-receptive field fusion feature extraction model;
fig. 7 is a schematic structural diagram of a residual calibration module.
Detailed Description
The present invention will be further described with reference to the accompanying drawings and specific examples, which are not intended to limit the invention, so that those skilled in the art may better understand the invention and practice it.
As shown in fig. 1, the welding quality intelligent detection method of the embodiment includes the following steps:
step one: an excitation point and a plurality of vibration sensors are arranged on the welded product. Specifically, for the same welding product, the positions of the excitation points measured each time are the same, and the arrangement positions and the number of the vibration sensors are the same.
Step two: and outputting exciting force with a set waveform to the exciting point by using the exciter so as to enable the welded product to vibrate, collecting vibration signals of the welded part passing through the welded product by using the vibration sensors, and forming a multi-channel vibration signal by using the vibration signals collected by the vibration sensors. And obtaining a classification label of the welding quality of the current welding product, and forming the welding quality data of the current welding product by the multi-channel vibration signal and the classification label together.
Specifically, for the same welding product, the exciting force waveforms generated by each exciter must be the same, so that the welding quality information carried in the signals received by the vibration sensor is more obvious. The excitation force signal can be selected from but not limited to sine wave, square wave, triangular wave, rectangular wave, etc. Of course, for different types of welded products, different excitation force waveforms may be selected. For example, as shown in fig. 2, for welding a white automobile body, an excitation point is selected at a certain position of the white automobile body, a plurality of excitation sensors are arranged at other positions of the white automobile body, and the welding quality condition of the whole white automobile body can be obtained by analyzing the characteristics in multichannel signal data formed by the excitation sensors.
Specifically, as shown in fig. 3, in this embodiment, the process of collecting the vibration signal is as follows:
(1) Selecting a certain welding product, selecting corresponding exciting point positions, the number and positions of vibration sensors on the welding product, and setting the magnitude and waveform of exciting force;
(2) Generating an exciting force signal;
(3) The signal acquisition device acquires an exciting force signal;
(4) Converting the exciting force signal from a digital signal to an analog signal through a D/A converter, amplifying the power, transmitting the amplified signal to a vibration exciter, and generating exciting force with specific waveform by using the vibration exciter;
(5) The vibration sensor acquires a vibration signal after passing through a welding part;
(6) And converting the analog signals acquired by the vibration sensor into digital signals through A/D conversion to obtain vibration signals.
Step three: judging whether the number of the obtained welding quality data reaches a set threshold value or not; if yes, obtaining a data set consisting of a set number of welding quality data, and executing a step five; if not, executing the fourth step. In this way, a sufficient amount of weld quality data may be included in the data set to enhance the performance effects of the artificial intelligence algorithm model.
Step four: arranging excitation points and vibration sensors at the same positions of the same welding product, and executing the second step;
step five: and constructing an artificial intelligence algorithm model, and training the artificial intelligence algorithm model by utilizing the data set. In this embodiment, the data set is divided into a training set, a verification set, and a test set for training, verifying, and testing the artificial intelligence algorithm model, respectively.
Specifically, the artificial intelligence algorithm model may employ a deep learning model, a machine learning model, a reinforcement learning model, or the like. The artificial intelligence algorithm model of this embodiment employs a multi-channel-scale-receptive field fusion feature extraction model (MMDFF model). Specifically, the multi-channel-scale-receptive field fusion feature extraction model of this embodiment includes an initialization Module (initialization Module), a feature fusion layer (fusion layer) and a classifier (classifier), and a parallel residual calibration Module (ResCASB Module) and a multi-receptive field Module (MD Module) are disposed between the initialization Module and the feature fusion layer.
Specifically, the initialization model is used for extracting preliminary data features of the multichannel vibration signals, and inputting the extracted preliminary data features into the residual calibration module and the multi-receptive field module. The initialization module of this embodiment is composed of two layers of one-dimensional convolution.
The residual calibration module realizes further excavation of the multichannel vibration signal data characteristics through fine granularity characteristic extraction and characteristic attention calibration on multiple scales. In the residual error calibration module, the system mainly comprises two parts of multi-scale residual error and attention calibration.
The multi-receptive field module obtains effective information stored in the signal through different receptive field sizes on the multiple channels.
The feature fusion layer is used for fusing the features extracted by the residual error calibration module and the multi-receptive field module and inputting the fused features into the classifier. The feature fusion layer fuses the features extracted by the residual calibration module and the multi-receptive field module by adopting a linear fusion method:
F＝αF R +βF M
wherein F represents the fused features; f (F) R Representing the features extracted by the residual calibration module; f (F) M Representing the extracted characteristics of the multi-receptive field module; alpha and beta represent the learnable parameters, respectively.
And classifying the fused features by the separator to obtain a classification label of the welding quality of the current welding product.
Step six: and inputting a multichannel vibration signal into the artificial intelligent algorithm model obtained through training, outputting a classification label corresponding to the welding quality of the welding product, and detecting the welding quality of the welding product on line.
A specific embodiment of the welding quality intelligent detection method of the present embodiment will be described below with reference to a body-in-white rear door assembly.
As shown in fig. 4, a three-dimensional schematic view of a rear door of a vehicle body obtained by welding is shown. Fig. 5 is a schematic view of the arrangement positions of the excitation points and the vibration sensors. As shown in fig. 5, a point on the weldment was selected as the excitation application point, and 8 vibration sensors were placed in total at the edge end and the tip end of the door of the weldment.
The bit machine generates exciting force signals, the exciting signals select sine waves, and the frequency is 100Hz; the exciting force signal is acquired by a signal acquisition device; converting the exciting force signal from a digital signal to an analog signal through a D/A converter, amplifying the power, and transmitting the amplified signal to a vibration exciter to generate sine waveform exciting force; the vibration exciter continuously works for 10s, and the acquisition time of all vibration sensors is set to be 10s. Ensuring that the vibration sensor can receive all vibration signals. And carrying out A/D conversion on the analog signals acquired by the vibration sensor, and converting the analog signals into digital signals to obtain vibration signals. And finally, combining the processed vibration signals into an eight-channel vibration signal.
In this embodiment, 10000 pieces of weldment data are collected in total, and a training data set is set, and the ratio of the verification data set to the test data set is 6:2:2. And training an artificial intelligence algorithm model by using the training data set and the verification data set, and judging the performance of the model by using the test data set. As shown in fig. 6, a multi-channel-scale-receptive field fusion feature extraction model (MMDFF Module) is selected in this embodiment to extract eight-channel vibration signal data features. The MMDFF Module structure is shown in fig. 6, and includes an initialization Module, a parallel residual calibration Module (rescalsb Module) and a multi-receptive field Module (MD Module), a feature fusion layer, and a classifier.
Specifically, the initialization module performs preliminary data feature extraction on the eight-channel vibration signal, the initialization module is composed of two layers of one-dimensional convolutions, and the convolution kernel sizes of the initialization module are respectively 7 and 5. The primarily extracted eight-channel vibration signal data features are then input in parallel to a residual calibration module and a multi-receptive field module.
As shown in fig. 7, the residual calibration module mainly performs fine-grained feature extraction and feature attention calibration on multiple scales, so as to further mine eight-channel vibration signal data features. In the residual error calibration module, the system mainly comprises two parts of multi-scale residual error and attention calibration. For multi-scale residuals, when an eight-channel vibration signal data feature is entered, it will first pass through a layer of k=1 convolution layer, and then divide the feature map into 4 parts. The first part of the circuit is very simple, x 1 Directly transfer to y without processing 1 The method comprises the steps of carrying out a first treatment on the surface of the Second part of the circuit, x 2 After the convolution of k=3, the signal is divided into two lines, and one line is continuously propagated forward to y 2 Another is transmitted to x 3 Thus, the third line obtains the information of the second line; a third line, a fourth line, and so on. After the data features pass through the multi-scale residual errors, the extraction of global and local information can be realized. For the feature attention module, the Global Average Pooling (GAP) and Global Maximum Pooling (GMP) are adopted to compress the space dimension of the input features, the Global Maximum Pooling (GMP) encodes the most significant part, another important clue of channel information is represented, the two are fused to infer finer channel attention weight values, and finally effective information of the data features is obtainedThe message is retained. The feature mentioned in this module is marked F R 。
The multi-receptive field module obtains effective information stored in the signals mainly through different receptive field sizes on multiple channels. When the eight-channel vibration signal data feature enters the multi-receptive field module, the eight-channel vibration signal data feature firstly passes through a convolution layer with k=1, then the feature is input into three channels, the convolution kernel of each channel is 3, and the hole sizes of the eight channels are 1,2 and 3 respectively. In each channel, after the eight-channel vibration signals are subjected to convolution feature extraction of different receptive fields, feature fusion is carried out in a layer of the flame, so that effective information under the different receptive fields is effectively reserved. The feature mentioned in this module is marked F M 。
And fusing the characteristics extracted by the residual calibration module and the multi-receptive field module in a characteristic fusion layer. The characteristic fusion layer adopts a linear fusion mode, and the specific formula is as follows:
F＝αF R +βF M
wherein F represents the fused features; f (F) R Representing the features extracted by the residual calibration module; f (F) M Representing the extracted characteristics of the multi-receptive field module; alpha and beta represent the learnable parameters, respectively.
And inputting the feature representation after feature fusion into a classifier to classify, and finally obtaining a classification label of the welding quality of the current welding piece.
And performing online welding piece quality detection by using the trained artificial intelligent algorithm model.
The above-described embodiments are merely preferred embodiments for fully explaining the present invention, and the scope of the present invention is not limited thereto. Equivalent substitutions and modifications will occur to those skilled in the art based on the present invention, and are intended to be within the scope of the present invention. The protection scope of the invention is subject to the claims.
Claims (4)
1. An intelligent detection method for welding quality is characterized in that: the method comprises the following steps:
step one: arranging an excitation point and a plurality of vibration sensors on a welded product;
step two: outputting exciting force with a set waveform to an exciting point by using an exciter so as to enable a welded product to vibrate, collecting vibration signals of a welded part passing through the welded product by using vibration sensors, and forming a multi-channel vibration signal by using the vibration signals collected by the vibration sensors;
acquiring a classification label of the welding quality of the current welding product, and forming welding quality data of the current welding product by the multi-channel vibration signal and the classification label together;
step three: judging whether the number of the obtained welding quality data reaches a set threshold value or not; if yes, obtaining a data set consisting of a set number of welding quality data, and executing a step five; if not, executing the fourth step;
step four: arranging excitation points and vibration sensors at the same positions of the same welding product, and executing the second step;
step five: constructing an artificial intelligence algorithm model, and training the artificial intelligence algorithm model by utilizing the data set;
the artificial intelligence algorithm model adopts a multichannel-scale-receptive field fusion characteristic extraction model; the multichannel-scale-receptive field fusion feature extraction model comprises an initialization module, a feature fusion layer and a classifier, wherein a parallel residual error calibration module and a multi-receptive field module are arranged between the initialization module and the feature fusion layer;
the initialization module consists of two layers of one-dimensional convolutions and is used for extracting preliminary data characteristics of the multichannel vibration signals and inputting the extracted preliminary data characteristics into the residual error calibration module and the multi-receptive field module;
the residual error calibration module consists of two parts of multi-scale residual error and attention calibration, and further excavation of multi-channel vibration signal data characteristics is realized through fine granularity characteristic extraction and characteristic attention calibration on multiple scales;
the multi-scale residual comprises a convolution layer with k=1 that divides the feature map into 4 parts; wherein the first part x of the feature map 1 Directly to y 1 The method comprises the steps of carrying out a first treatment on the surface of the Second of the feature mapPart x 2 After the convolution of k=3, the signal is divided into two lines, and one line is continuously propagated forward to y 2 Another is transmitted to x 3 The method comprises the steps of carrying out a first treatment on the surface of the Third part x of the feature map 3 After obtaining the second part of information, the second part of information is divided into two lines after being convolved by K=3, and one line is continuously transmitted to y forwards 3 Another is transmitted to x 4 The method comprises the steps of carrying out a first treatment on the surface of the Fourth part x of the feature map 4 After obtaining the third part of information, the third part of information is forward propagated to y through K=3 convolution 4 ；y 1 、y 2 、y 3 And y 4 After passing through a convolution layer with K=1, outputting;
the attention calibration adopts Global Average Pooling (GAP) and Global Maximum Pooling (GMP) to compress the space dimension of the input features, the Global Maximum Pooling (GMP) encodes the most significant part, another important clue for the channel information is represented, the two important clues are fused to deduce finer channel attention weight, and finally effective information of the data features is reserved;
the multi-receptive field module acquires effective information stored in the signal through different receptive field sizes on the multi-channel; after the channel vibration signal data characteristics enter the multi-receptive field module, the characteristics firstly pass through a layer of convolution layer with K=1, then the characteristics are input into three channels, the convolution kernel of each channel is 3, and the cavity sizes of the channels are 1,2 and 3 respectively; in each channel, after the eight-channel vibration signal is subjected to convolution feature extraction of different receptive fields, feature fusion is carried out in a layer of the flame, so that effective information under the different receptive fields is effectively reserved;
the feature fusion layer is used for fusing the features extracted by the residual calibration module and the multi-receptive field module and inputting the fused features into the classifier; the characteristic fusion layer adopts a linear fusion mode, and the specific formula is as follows:
F＝αF R +βF M
wherein F represents the fused features; f (F) R Representing the features extracted by the residual calibration module; f (F) M Representing the extracted characteristics of the multi-receptive field module; alpha and beta respectively represent a learnable parameter;
the classifier classifies the fused features to obtain a classification label of the welding quality of the current welding product;
step six: and inputting a multichannel vibration signal into the artificial intelligent algorithm model obtained through training, outputting a classification label corresponding to the welding quality of the welding product, and detecting the welding quality of the welding product on line.
2. The welding quality intelligent detection method as defined in claim 1, wherein: the waveform of the exciting force is sine wave, square wave, triangular wave or rectangular wave.
3. The welding quality intelligent detection method as defined in claim 1, wherein: the data set is divided into a training set, a verification set and a test set which are respectively used for training, verifying and testing the artificial intelligence algorithm model.
4. The welding quality intelligent detection method as defined in claim 1, wherein: the artificial intelligence algorithm model includes a deep learning model, a machine learning model, or a reinforcement learning model.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210951200.XA CN115310486B (en) | 2022-08-09 | 2022-08-09 | Intelligent welding quality detection method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210951200.XA CN115310486B (en) | 2022-08-09 | 2022-08-09 | Intelligent welding quality detection method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115310486A CN115310486A (en) | 2022-11-08 |
| CN115310486B true CN115310486B (en) | 2023-09-26 |
Family
ID=83861512
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210951200.XA Active CN115310486B (en) | 2022-08-09 | 2022-08-09 | Intelligent welding quality detection method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115310486B (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110261483A (en) * | 2019-07-31 | 2019-09-20 | 三一汽车起重机械有限公司 | Weld defect detection method, device, detection device and readable storage medium storing program for executing |
| CN111815572A (en) * | 2020-06-17 | 2020-10-23 | 深圳市大德激光技术有限公司 | Method for detecting welding quality of lithium battery based on convolutional neural network |
| CN112184686A (en) * | 2020-10-10 | 2021-01-05 | 深圳大学 | Segmentation algorithm for detecting laser welding defects of safety valve of power battery |
| CN112465790A (en) * | 2020-12-03 | 2021-03-09 | 天津大学 | Surface defect detection method based on multi-scale convolution and trilinear global attention |
| CN113255778A (en) * | 2021-05-28 | 2021-08-13 | 广汽本田汽车有限公司 | Welding spot quality detection method and device based on multi-model fusion and storage medium |
| CN113378725A (en) * | 2021-06-15 | 2021-09-10 | 山东大学 | Cutter fault diagnosis method, equipment and storage medium based on multi-scale-channel attention network |
| CN114119637A (en) * | 2021-11-29 | 2022-03-01 | 大连大学 | Brain white matter high signal segmentation method based on multi-scale fusion and split attention |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111739075B (en) * | 2020-06-15 | 2024-02-06 | 大连理工大学 | Deep network lung texture recognition method combining multi-scale attention |
-
2022
- 2022-08-09 CN CN202210951200.XA patent/CN115310486B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110261483A (en) * | 2019-07-31 | 2019-09-20 | 三一汽车起重机械有限公司 | Weld defect detection method, device, detection device and readable storage medium storing program for executing |
| CN111815572A (en) * | 2020-06-17 | 2020-10-23 | 深圳市大德激光技术有限公司 | Method for detecting welding quality of lithium battery based on convolutional neural network |
| CN112184686A (en) * | 2020-10-10 | 2021-01-05 | 深圳大学 | Segmentation algorithm for detecting laser welding defects of safety valve of power battery |
| CN112465790A (en) * | 2020-12-03 | 2021-03-09 | 天津大学 | Surface defect detection method based on multi-scale convolution and trilinear global attention |
| CN113255778A (en) * | 2021-05-28 | 2021-08-13 | 广汽本田汽车有限公司 | Welding spot quality detection method and device based on multi-model fusion and storage medium |
| CN113378725A (en) * | 2021-06-15 | 2021-09-10 | 山东大学 | Cutter fault diagnosis method, equipment and storage medium based on multi-scale-channel attention network |
| CN114119637A (en) * | 2021-11-29 | 2022-03-01 | 大连大学 | Brain white matter high signal segmentation method based on multi-scale fusion and split attention |
Non-Patent Citations (1)
| Title |
|---|
| 基于全局时空感受野的高效视频分类方法;王辉涛 等;小型微型计算机***(第08期);第1768-1775页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115310486A (en) | 2022-11-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106198725B (en) | A kind of butt weld defect detecting system and detection method based on feature guided wave | |
| US6857553B1 (en) | Method and apparatus for in-process sensing of manufacturing quality | |
| CN104730109B (en) | A kind of resistance spot welding quality of welding spot detection method based on dynamic resistance curve | |
| CN106447040B (en) | Mechanical equipment health state evaluation method based on Heterogeneous Multi-Sensor Data fusion | |
| CN102162577B (en) | Pipeline defect surface integrity detection device and detection method | |
| CN106271036A (en) | Ultrasonic metal welding method for evaluating quality, device and ultrasonic metal bonding machine | |
| CN103760230B (en) | Based on the weld defects giant magnetoresistance eddy current detection method of BP neural network | |
| KR102007494B1 (en) | System for inspecting welding quality of weld zone using ultrasonic | |
| CN106274977A (en) | The direct current Magnetic Flux Leakage Testing System of a kind of trigger-type drainage pattern and method thereof | |
| CN103760231B (en) | Weld defect giant magnetoresistance eddy current detection method based on decision tree and detection device | |
| CN104677751B (en) | Quality detection method for resistance-spot-welding spots on basis of calculation of thermal effect of welding process | |
| CN105913059A (en) | Vehicle VIN code automatic identifying system and control method therefor | |
| JP2007256153A (en) | System for detecting railway vehicle truck abnormality | |
| KR102230886B1 (en) | Projection welding defect detection system based on image recognition | |
| Summerville et al. | Nugget diameter in resistance spot welding: a comparison between a dynamic resistance based approach and ultrasound C-scan | |
| JPH059744B2 (en) | ||
| CN115310486B (en) | Intelligent welding quality detection method | |
| CN103713042A (en) | Eddy-current welding defect detection method based on k-nearest neighbor algorithm | |
| CN102419242B (en) | For the apparatus and method of testing vehicle air bag control unit | |
| CN103760229B (en) | Based on the weld defects giant magnetoresistance eddy current detection method of support vector machine | |
| CN105548349A (en) | Rectangular probe pulsed eddy current detecting method for realizing defect reconstruction technology | |
| CN103675103B (en) | A kind of method utilizing resistance spot welding process Acoustic Emission Signal Energy equivalent quantitatively to detect nugget crackle | |
| CN106624313A (en) | On-line detecting system and method for spot welding | |
| CN206662523U (en) | A kind of spot welding on-line detecting system | |
| Ling et al. | Input electrical impedance as signature for nondestructive evaluation of weld quality during ultrasonic welding of plastics |
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 |