CN114119595A - GMAW welding quality on-line monitoring and evaluating method based on integrated deep learning - Google Patents

GMAW welding quality on-line monitoring and evaluating method based on integrated deep learning Download PDF

Info

Publication number
CN114119595A
CN114119595A CN202111479871.2A CN202111479871A CN114119595A CN 114119595 A CN114119595 A CN 114119595A CN 202111479871 A CN202111479871 A CN 202111479871A CN 114119595 A CN114119595 A CN 114119595A
Authority
CN
China
Prior art keywords
welding
quality
deep learning
signals
gmaw
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
Application number
CN202111479871.2A
Other languages
Chinese (zh)
Inventor
陈伟光
陈华斌
朱抗洪
侯震
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shanghai Jiaotong University
Original Assignee
Shanghai Jiaotong University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Shanghai Jiaotong University filed Critical Shanghai Jiaotong University
Priority to CN202111479871.2A priority Critical patent/CN114119595A/en
Publication of CN114119595A publication Critical patent/CN114119595A/en
Pending legal-status Critical Current

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
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G06F18/253—Fusion techniques of extracted features
    • 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
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • 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
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • 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
    • 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
    • G06T2207/30168—Image quality inspection

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Computational Linguistics (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Quality & Reliability (AREA)
  • Image Analysis (AREA)

Abstract

The invention relates to an integrated deep learning-based GMAW welding quality on-line monitoring and evaluating method, which comprises the following steps: collecting welding information including a welding pool image, an electric arc sound signal, a welding voltage signal and a welding current signal in the welding process; preprocessing the welding information and extracting characteristics; connecting and fusing the acquired characteristics of different signal sources to construct a fused characteristic vector; constructing a quality basic detection model, inputting the fused feature vector into the quality basic detection model, and outputting the welding quality in the current state; and obtaining the welding quality in the current state by adopting different quality basic detection models, and obtaining a final decision result according to the results of the different quality basic detection models. Compared with the prior art, the method has the advantages of realizing on-line monitoring, improving the accuracy of detection quality and the like.

Description

GMAW welding quality on-line monitoring and evaluating method based on integrated deep learning
Technical Field
The invention relates to the technical field of robot welding, in particular to an integrated deep learning-based GMAW welding quality online monitoring and evaluating method.
Background
Welding quality is a primary consideration in welding manufacture, and welding defects and imperfections can reduce the strength of the welded component, affecting the reliability and integrity of the welded structure. More seriously, the welded product is scrapped due to serious welding defects. Up to now, the quality inspection of the welded product is mainly through the nondestructive inspection after welding, which undoubtedly increases the production period and the production cost.
At present, in practical industrial production, such as engineering machinery manufacturing industry, welding quality inspection is necessary in the welding process. However, due to the complex production scenario in the welding process, the collection and processing of the welding information become more difficult, and the difficulty of quality detection in the welding process is increased. In industrial actual production, for example, in the engineering machinery manufacturing industry, the bogie production of rail trains, the real-time monitoring and evaluation of welding quality has the following problems:
(1) the welding scene is complex, the change of the welding track is large, the welding pool obtained by shooting has a deflection translation state, and strong arc light exists in the welding process, so that the influence on feature extraction is large. Conventional weld pool geometry does not fully describe the change in weld state.
(2) The research on different welding defects and defects in GMAW (gas metal arc welding) is not complete enough, and the defects and defects existing in the welding process cannot be evaluated completely.
Chinese patent publication No. CN112207482A discloses a multivariate information monitoring and control system and method for welding quality control, which simultaneously uses multiple kinds of sensing information to perform real-time detection of welding quality, however, the characteristics extracted for the molten pool image in the patent are extraction of characteristic parameters such as melt width, half melt length, molten pool back-supporting angle, molten pool area, etc., these geometric characteristics are easily affected by the production environment, and some characteristics cannot be extracted in the molten pool image of some welding defects. Meanwhile, a specific method for quality detection is not described, whether the quality detection can be realized or not is questioned, and the patent does not consider the wire clamping state which is easy to appear in GMAW welding, so that the method is not suitable for monitoring and evaluating the welding quality in the welding of engineering machinery manufacturing such as a boom of a bucket rod.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provide an integrated deep learning-based online GMAW welding quality monitoring and evaluating method.
The purpose of the invention can be realized by the following technical scheme:
an integrated deep learning-based GMAW welding quality on-line monitoring and evaluation method comprises the following steps:
s1: collecting welding information including welding pool images, electric arc sound signals, welding voltage signals and welding current signals in the welding process.
Preferably, a CCD camera is adopted to collect welding pool images in the welding process, a voltage and current sensor is adopted to collect welding voltage signals and welding current signals in the welding process, and a sound sensor is adopted to collect arc sound signals in the welding process.
Preferably, a 660nm filter is added to the CCD camera.
Further, before extracting the characteristics of the welding signals of different signal sources, the method also comprises a data alignment step:
aligning welding signals of different signal sources, setting a welding pool image corresponding to a recognizable welding quality state, and corresponding f2/f1 welding voltage signals, welding current signals and arc sound signal data points to a welding pool image, namely corresponding to a welding state according to the acquisition frequency f1 of the welding pool image and the acquisition frequency f2 of the welding voltage signals, the welding current signals and the arc sound signals.
S2: preprocessing the collected welding information, and extracting the characteristics of the welding signals of different signal sources. In the step, the feature extraction processes of different signal source information are carried out in parallel.
The feature extraction of this step includes:
s21: extracting voltage and current characteristics; characteristic values include, but are not limited to, standard deviation, crest factor, skewness, and kurtosis.
S22: extracting the characteristics of the sound signals: the sound signal time domain feature extraction includes but is not limited to standard deviation, crest factor, skewness and kurtosis. The sound signal frequency domain signal is obtained through Fourier transform, and the sound signal frequency domain characteristics include, but are not limited to, different frequency band frequency intensities.
S23: extraction of molten pool characteristics: the molten pool characteristics are obtained by extracting the fully connected layer in the convolutional network. The convolution neural network model structure and weight value extracted by the characteristics are the model stored after training. The training process can be expressed as follows:
1) data set making, network model training and specific layer feature result output;
2) manufacturing a data set, wherein the image data of the molten pool comes from the shooting of a CCD camera in the welding process; and marking the molten pool images in different welding quality states, wherein the welding quality states include but are not limited to welding leakage, incomplete penetration, welding deviation, wire clamping, misalignment, slag inclusion, surface pores and the like. The marked weld pool image can be as follows 7: 2: 1, training set, validation set and test set.
3) And (3) training a network model, wherein the network input is a molten pool image, and the output is different welding quality states. And storing the trained model for extracting the image characteristics of the molten pool.
4) And extracting the specific layer features, and reading the trained model. The molten pool image is used as input, a specific full-connection layer is selected as output, and the characteristic vector of the molten pool image is extracted and obtained.
Preferably, the S23 may perform feature extraction of the weld puddle image by using an adjusted VGG-16 network pre-trained on ImagNet, where the adjusted VGG-16 network includes 13 convolutional layers and 3 fully-connected layers, and shields the front five-round convolutional pooling process, trains the fully-connected layers, performs feature extraction of pictures on the trained network, and obtains data of the FC8 layer, that is, obtains the weld puddle image features.
Further, before the obtained characteristics of different signal sources are connected and fused, the method also comprises the step of standardizing the different characteristics.
S3: and connecting and fusing the acquired characteristics of different signal sources to construct a fused characteristic vector.
The fusion method comprises the following steps:
xf＝(zv,zs)
in the formula, zfIs a fused feature vector, zv、zsFeature vectors from different sources.
S4: and constructing a quality basic detection model, inputting the fused feature vector into the quality basic detection model, and outputting the welding quality in the current state.
The quality basic detection model is a classification model which is obtained after training of a basic model. The base model includes, but is not limited to, SVM, RF, or XGboost.
S5: and obtaining the welding quality in the current state by adopting different quality basic detection models, and obtaining a final decision result according to the results of the different quality basic detection models.
The final decision result can be obtained by preferentially adopting a voting mode, and the decision process is as follows:
s51: a vote is generated. Three base models are trained separately, and the results of the models are saved. Wherein the output of the model is the classification probabilities of the different classifications. The type of the maximum prediction probability is the current classification recognition result.
S52: and (5) voting. And outputting classification probability to each identifiable welding quality state by the three basic models, and voting classification results of different models to a final decision-making model.
S53: and (5) a decision making process. And the decision-making model superposes the classification probabilities given by different models according to the classification numbers and then calculates the average. And traversing the calculated values of each type, and selecting the class with the highest classification probability as the final result.
The models used in the scheme are all stored models after offline training. And in the real-time welding detection process, the trained model parameters are read in advance, and corresponding characteristics are extracted and welding quality is detected.
In the real-time welding detection process, the above process is a detection period. In the actual detection, a detection period is carried out by taking a molten pool image acquired by collection as a reference. During the welding process, the image acquisition frequency can be between 3Hz and 10 Hz.
In the actual welding production process, such as the welding of a bucket arm in engineering machinery, the invention is verified, and the welding quality state can be accurately and completely identified.
Compared with the prior art, the GMAW welding quality online monitoring and evaluating method based on the integrated deep learning at least has the following beneficial effects:
1) compared with the traditional welding quality detection means after welding, the method can detect the welding quality on line, can prompt the occurrence of welding defects and defects in real time, and can repair the welding defects in time without waiting for the repair of a large section of welding seam after welding;
2) according to the invention, by collecting multi-sensing information in the welding process, including a molten pool image, arc sound, welding voltage and welding current, the welding quality state can be represented from multiple dimensions, and the defect of insufficient single sensing information can be well compensated;
3) the method utilizes the convolution neural network method to extract the image characteristics of the molten pool, and can better adapt to the conditions of displacement, deflection and scaling of the image in the actual complex welding production process;
4) the method integrates the quality detection results of three different basic models, obtains the final decision result in a voting mode, and improves the accuracy of quality detection.
Drawings
FIG. 1 is a flow misintention of an integrated deep learning-based GMAW welding quality online monitoring and evaluation method of the invention;
FIG. 2 is a block diagram of a welding system in accordance with an embodiment of the present invention;
fig. 3 is a network structure diagram according to an embodiment of the invention.
Detailed Description
The invention is described in detail below with reference to the figures and specific embodiments. It is to be understood that the embodiments described are only a few embodiments of the present invention, and not all embodiments. All other embodiments, which can be obtained by a person skilled in the art without any inventive step based on the embodiments of the present invention, shall fall within the scope of protection of the present invention.
Examples
The invention relates to a GMAW welding quality on-line monitoring and evaluating method based on integrated deep learning, which can meet the application of welding quality monitoring and evaluation in engineering machinery manufacturing, such as welding of a boom of a bucket rod, can effectively solve the problem that the characteristics of a molten pool are difficult to extract in a complex welding scene and the current situation that welding defects and defects cannot be completely covered in the current research, and can completely describe and identify the typical defects and defects of a GMAW robot: welding leakage, incomplete penetration, welding deviation, wire clamping, misalignment, slag inclusion, surface pores and the like.
As shown in fig. 1, the specific implementation flow of the online GMAW welding quality monitoring and evaluating method based on integrated deep learning of the present invention includes:
1. collection of welding process information
The welding system and the welding information acquisition system of the embodiment are shown in fig. 2, and mainly comprise a welding robot, a welding workpiece, a welding power supply, a CCD camera, a sound sensor, a hall sensor, an industrial personal computer and an information acquisition card. Wherein the welding workpiece comprises a protective gas cylinder, a welding gun, a wire feeder and the like. The sound sensor may preferably be an arc sound sensor. The CCD camera and the arc sound sensor are mounted near the welding gun. However, during the welding process, the temperature near the welding torch is high. In order to ensure that the camera cannot be damaged due to overheating in welding work, cooling water is adopted to cool the camera. In order to reduce the influence of arc on image acquisition, a 660nm filter is arranged in front of the camera. The acquired image of the molten pool is transmitted to an acquisition system control machine. The sound signals collected by the arc sound sensor pass through a signal conditioner and are collected by a data acquisition card (information acquisition card). The voltage and current signals in the welding process are collected through the Hall sensor, and are collected by the data acquisition card after being adjusted by the acquisition circuit. The electric arc sound signals and the voltage and current signals collected by the data collection card are transmitted to the collection system controller. Before welding, the welding acquisition frequency is set. Wherein, the image acquisition frequency is 5Hz, and the welding voltage, the welding current and the electric arc sound acquisition frequency are all 10000 Hz. And after each system is adjusted, real-time welding data acquisition is carried out.
And marking the molten pool images in different welding quality states, wherein the welding quality states include but are not limited to welding leakage, incomplete penetration, welding deviation, wire clamping, misalignment, slag inclusion, surface pores and the like. Preferably, the marked molten pool image is as follows: 2: 1, making a training set, a verification set and a test set.
2. Multi-source information feature extraction
Before feature extraction, data from different sources need to be aligned. Due to the fact that information acquisition frequencies are different in the welding process, data needs to be aligned. In the invention, a picture is defined to correspond to an identifiable welding quality state, and 2000 voltage signal, current signal and sound signal data points correspond to a picture, namely correspond to a welding state according to the acquisition frequency of a molten pool image.
And (3) extracting the characteristics of the molten pool image: the invention adopts VGG-16 pre-trained on Imagnet to extract the characteristics of the molten pool image. The VGG-16 network is composed of 13 convolutional layers and 3 fully-connected layers. In this embodiment, an adjustment is performed on the basic structure of the original VGG-16 network, the full connection layer is mainly modified, and the adjusted network structure is shown in fig. 3. And (5) shielding the front five-wheel convolution pooling process by the adjusted network, and training a full connection layer. The platforms used were Tensorflow 2.3 and Python 3.8. And (3) carrying out picture feature extraction on the trained network, and mainly extracting data of an FC8 layer to obtain a molten pool image feature vector.
For voltage current feature extraction: in this embodiment, windowing and framing are performed according to 2000 data points, and features such as an average value, a peak-to-peak value, a standard deviation, a root mean square, a form factor, a peak factor, skewness, and kurtosis are mainly extracted.
For sound signal feature extraction: the time domain features are calculated by windowing and framing according to 2000 data points, and features such as an average value, a peak-to-peak value, a standard deviation, a root mean square, a form factor, a peak factor, skewness, kurtosis and the like are mainly extracted. The frequency domain characteristics were averaged according to different frequency ranges { <1000Hz, 1000Hz-2000Hz, 2000Hz-3000Hz, 3000Hz-4000Hz, >4000Hz }.
3. Multi-source information feature fusion
The data after feature extraction has different feature ranges and different dimensions. In order to eliminate the influence of different characteristic dimensions and facilitate the comparability between indexes, the obtained data characteristics are standardized. Scaling the original index to an interval between 0 and 1. Normalization is performed on features of the same dimension in different kinds of data. The normalization formula is as follows:
Figure BDA0003394891040000061
wherein x isnorFor normalized eigenvalues, xmaxIs the maximum value, x, of the sample data under the same characteristicminAnd x is the data of a certain characteristic value in the sample data.
The extracted features come from different information sources, and a plurality of one-dimensional feature vectors are obtained under the condition that the welding quality can be identified, wherein the one-dimensional feature vectors are respectively a molten pool image feature vector, a voltage signal feature vector, a current signal feature vector, a sound signal time domain feature vector and a sound signal frequency domain feature vector. A single eigenvector does not distinguish well between the different weld quality states. In order to improve the accuracy of weld quality identification, feature vectors from different information sources are fused. The fusion method is to connect the feature vectors of different information sources in series to obtain a one-dimensional feature vector capable of identifying the welding quality state. The fusion method comprises the following steps:
xf＝(zv,zs)
wherein z isfIs a fused feature vector, zv、zsFeature vectors from different sources.
4. Building a basic test model
And the fused feature vector is used as input to enter a classification model, and the output is the welding quality in the current state. The classification model will build a data-driven model using the following dataset D as input:
Figure BDA0003394891040000071
wherein x isiThe data feature vector corresponding to the ith welding state, i.e. the fused feature vector, is composed of a plurality of features. y isiThe method is characterized in that the method is a label corresponding to the ith welding state, namely the welding states of 1-welding leakage, 2-incomplete penetration, 3-welding deviation, 4-wire clamping, 5-misalignment, 6-slag inclusion, 7-surface pore space and the like.
And selecting SVM, RF and XGboost as basic models. This is because SVM is suitable for solving the classification problem of small samples and has good generalization ability, but needs further optimization on multi-classification. The RF is capable of balancing errors for unbalanced data sets and is more suitable for use in weld detection situations. The XGBoost considers the case where the training data is a sparse value, and may specify a default direction of the branch for a missing value or a specified value. The embodiment uses scimit-leann and Python to train classification models of SVM, RF and XGboost. The trained model can be used for detecting the welding quality state.
5. Synthesizing the detection result to obtain the final decision
In order to integrate the advantages of multiple models, the embodiment uses SVM, RF, XGBoost models as basic models, and then obtains the final decision result by voting the results of different classification models, where the specific decision flow is as follows:
first, generating a vote. Three base models are trained separately, and the results of the models are saved. Wherein the output of the model is the classification probabilities of the different classifications. The type of the maximum prediction probability is the current classification recognition result. The mapping of class number and welding quality state is set as { 0-normal, 1-weld leakage, 2-incomplete penetration, 3-weld deviation, 4-wire clamping, 5-misalignment, 6-slag inclusion, 7-surface porosity }.
And step two, a voting process. And outputting classification probability to each identifiable welding quality state by the three basic models, and voting classification results of different models to a final decision-making model.
And thirdly, a decision making process. And the decision-making model superposes the classification probabilities given by different models according to the classification numbers and then calculates the average. And traversing the calculated values of each type, and selecting the class with the highest classification probability as the final result.
Compared with the traditional welding quality detection means, the method can detect the welding quality on line and replace the nondestructive detection after welding to a certain extent. Through collecting multi-sensing information in the welding process, including a molten pool image, arc sound, welding voltage and welding current, the welding quality state can be represented from multiple dimensions, and the defect that single sensing information is insufficient can be well overcome.
While the invention has been described with reference to specific embodiments, the invention is not limited thereto, and those skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope of the invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims (10)

1. An integrated deep learning-based GMAW welding quality on-line monitoring and evaluation method is characterized by comprising the following steps:
collecting welding information including a welding pool image, an electric arc sound signal, a welding voltage signal and a welding current signal in the welding process;
preprocessing collected welding information, and extracting characteristics of welding signals of different signal sources;
connecting and fusing the acquired characteristics of different signal sources to construct a fused characteristic vector;
constructing a quality basic detection model, inputting the fused feature vector into the quality basic detection model, and outputting the welding quality in the current state;
and obtaining the welding quality in the current state by adopting different quality basic detection models, and obtaining a final decision result according to the results of the different quality basic detection models.
2. The integrated deep learning-based GMAW welding quality on-line monitoring and evaluation method according to claim 1, characterized in that a CCD camera is used for collecting images of a welding pool in the welding process, a voltage and current sensor is used for collecting welding voltage signals and welding current signals in the welding process, and a sound sensor is used for collecting arc sound signals in the welding process.
3. The GMAW welding quality on-line monitoring and evaluating method based on the integrated deep learning of claim 1, wherein collected welding voltage signals and welding current signals are subjected to preprocessing including zero drift removal, and after preprocessing, statistical characteristics are extracted; preprocessing the collected arc sound signals and then extracting sound time domain characteristics and sound frequency domain characteristics; and extracting the characteristics of the welding pool image by adopting a convolution neural network for the collected welding pool image.
4. The integrated deep learning-based online GMAW welding quality monitoring and evaluation method of claim 3, further comprising a data alignment step before feature extraction of the welding signals of different signal sources:
aligning welding signals of different signal sources, setting a welding pool image corresponding to a recognizable welding quality state, and corresponding f2/f1 welding voltage signals, welding current signals and arc sound signal data points to a welding pool image, namely corresponding to a welding state according to the acquisition frequency f1 of the welding pool image and the acquisition frequency f2 of the welding voltage signals, the welding current signals and the arc sound signals.
5. The GMAW welding quality on-line monitoring and evaluating method based on the integrated deep learning of claim 3, characterized in that after a welding pool image is collected, the welding pool image in different welding quality states including but not limited to welding leakage, incomplete penetration, welding deviation, wire sticking, misalignment, slag inclusion and surface porosity is subjected to labeling in different quality states, the labeled welding pool image is divided into a training set, a verification set and a test set, and a convolutional neural network is adopted to extract the welding pool image characteristics.
6. The GMAW welding quality on-line monitoring and evaluating method based on the integrated deep learning of claim 5, wherein the adjusted VGG-16 network pre-trained on Imagnet is adopted for feature extraction of a welding pool image, the adjusted VGG-16 network comprises 13 convolution layers and 3 full-connection layers, the front five-round convolution pooling process is shielded, the full-connection layer is trained, feature extraction of pictures is carried out on the trained network, and data of FC8 layers is obtained, namely the welding pool image features are obtained.
7. The integrated deep learning-based online GMAW welding quality monitoring and evaluation method according to claim 1, further comprising the step of normalizing the different acquired characteristics of the different signal sources before performing connection fusion on the acquired characteristics.
8. The integrated deep learning-based GMAW welding quality online monitoring and evaluation method as claimed in claim 1, wherein the quality basis detection model is a classification model obtained after training of a basis model.
9. The integrated deep learning-based online GMAW weld quality monitoring and evaluation method of claim 8, wherein the base model includes, but is not limited to, SVM, RF or XGBoost.
10. The integrated deep learning-based GMAW welding quality online monitoring and evaluation method as claimed in claim 1, wherein the final decision result is obtained by voting the results of different quality basic detection models.
CN202111479871.2A 2021-12-06 2021-12-06 GMAW welding quality on-line monitoring and evaluating method based on integrated deep learning Pending CN114119595A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111479871.2A CN114119595A (en) 2021-12-06 2021-12-06 GMAW welding quality on-line monitoring and evaluating method based on integrated deep learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111479871.2A CN114119595A (en) 2021-12-06 2021-12-06 GMAW welding quality on-line monitoring and evaluating method based on integrated deep learning

Publications (1)

Publication Number Publication Date
CN114119595A true CN114119595A (en) 2022-03-01

Family

ID=80366911

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111479871.2A Pending CN114119595A (en) 2021-12-06 2021-12-06 GMAW welding quality on-line monitoring and evaluating method based on integrated deep learning

Country Status (1)

Country Link
CN (1) CN114119595A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115106615A (en) * 2022-08-30 2022-09-27 苏芯物联技术(南京)有限公司 Welding deviation real-time detection method and system based on intelligent working condition identification
CN115255566A (en) * 2022-09-26 2022-11-01 苏芯物联技术(南京)有限公司 Welding deviation real-time intelligent detection method based on high-quality time domain characteristics
CN117020361A (en) * 2023-08-30 2023-11-10 鹤山市福维德智能设备有限公司 Automatic detection device and detection method for mechanical manufacture

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115106615A (en) * 2022-08-30 2022-09-27 苏芯物联技术(南京)有限公司 Welding deviation real-time detection method and system based on intelligent working condition identification
CN115255566A (en) * 2022-09-26 2022-11-01 苏芯物联技术(南京)有限公司 Welding deviation real-time intelligent detection method based on high-quality time domain characteristics
CN117020361A (en) * 2023-08-30 2023-11-10 鹤山市福维德智能设备有限公司 Automatic detection device and detection method for mechanical manufacture
CN117020361B (en) * 2023-08-30 2024-02-09 鹤山市福维德智能设备有限公司 Automatic detection device and detection method for mechanical manufacture

Similar Documents

Publication Publication Date Title
CN114119595A (en) GMAW welding quality on-line monitoring and evaluating method based on integrated deep learning
US10803573B2 (en) Method for automated detection of defects in cast wheel products
CN113362326B (en) Method and device for detecting defects of welding spots of battery
Dai et al. Deep learning assisted vision inspection of resistance spot welds
CN111815572B (en) Method for detecting welding quality of lithium battery based on convolutional neural network
CN109886298B (en) Weld quality detection method based on convolutional neural network
CN104992449B (en) Information identification and surface defect online test method based on machine vision
CN103499585B (en) Based on noncontinuity lithium battery film defect inspection method and the device thereof of machine vision
He et al. Autonomous detection of weld seam profiles via a model of saliency-based visual attention for robotic arc welding
CN111815555A (en) Metal additive manufacturing image detection method and device combining anti-neural network with local binary
CN111369516A (en) Transformer bushing heating defect detection method based on infrared image recognition
CN116309277A (en) Steel detection method and system based on deep learning
CN115239728A (en) Fire-fighting equipment identification method
CN116704733A (en) Aging early warning method and system for aluminum alloy cable
Hong et al. Real-time quality monitoring of ultrathin sheets edge welding based on microvision sensing and SOCIFS-SVM
Hong et al. Filter-PCA-based process monitoring and defect identification during climbing helium arc welding process using DE-SVM
CN116703807A (en) System and method for analyzing weld quality
Tang et al. A new method to assess fiber laser welding quality of stainless steel 304 based on machine vision and hidden Markov models
CN114092478A (en) Anomaly detection method
CN109741311B (en) Aluminum alloy fusion welding back face fusion width detection method with false edge
CN115953387A (en) Radiographic image weld defect detection method based on deep learning
Soares et al. Computer vision system for weld bead geometric analysis
CN110287968B (en) Detection method for welding line flaw detection negative image counterfeiting based on LBP texture
CN112129774A (en) Online detection method for welding unfused defect
CN113919396A (en) Vibration signal and image characteristic machine tool cutter wear state monitoring method based on semi-supervised learning

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