CN117437199A - Resistance spot welding quality detection method and system based on feature fusion - Google Patents
Resistance spot welding quality detection method and system based on feature fusion Download PDFInfo
- Publication number
- CN117437199A CN117437199A CN202311433968.9A CN202311433968A CN117437199A CN 117437199 A CN117437199 A CN 117437199A CN 202311433968 A CN202311433968 A CN 202311433968A CN 117437199 A CN117437199 A CN 117437199A
- Authority
- CN
- China
- Prior art keywords
- signal
- dynamic power
- signals
- imf
- welding
- 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
- 238000003466 welding Methods 0.000 title claims abstract description 97
- 238000001514 detection method Methods 0.000 title claims abstract description 57
- 230000004927 fusion Effects 0.000 title claims abstract description 28
- 238000000034 method Methods 0.000 claims abstract description 48
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 23
- 230000008569 process Effects 0.000 claims abstract description 19
- 238000004364 calculation method Methods 0.000 claims abstract description 8
- 239000011159 matrix material Substances 0.000 claims description 12
- 238000004422 calculation algorithm Methods 0.000 claims description 11
- 238000012549 training Methods 0.000 claims description 11
- 238000001914 filtration Methods 0.000 claims description 7
- 230000006870 function Effects 0.000 claims description 7
- 238000000926 separation method Methods 0.000 claims description 3
- 238000005259 measurement Methods 0.000 description 8
- 238000004519 manufacturing process Methods 0.000 description 5
- 238000013528 artificial neural network Methods 0.000 description 4
- 241000288673 Chiroptera Species 0.000 description 3
- 239000013598 vector Substances 0.000 description 3
- 238000000429 assembly Methods 0.000 description 2
- 230000000712 assembly Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 230000006698 induction Effects 0.000 description 2
- 239000002184 metal Substances 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000005855 radiation Effects 0.000 description 2
- 238000012216 screening Methods 0.000 description 2
- 238000006467 substitution reaction Methods 0.000 description 2
- 241001391944 Commicarpus scandens Species 0.000 description 1
- 229910001209 Low-carbon steel Inorganic materials 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 229910052799 carbon Inorganic materials 0.000 description 1
- 238000010835 comparative analysis Methods 0.000 description 1
- 238000012790 confirmation Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 238000009658 destructive testing Methods 0.000 description 1
- 238000005265 energy consumption Methods 0.000 description 1
- 230000014509 gene expression Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000005304 joining Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000009659 non-destructive testing Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000010355 oscillation Effects 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000007637 random forest analysis Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012163 sequencing technique Methods 0.000 description 1
- 238000005507 spraying Methods 0.000 description 1
- 230000026676 system process Effects 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
- 238000009827 uniform distribution Methods 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/277—Analysis of motion involving stochastic approaches, e.g. using Kalman filters
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
- G06V10/806—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- 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
- G06T2207/30168—Image quality inspection
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Multimedia (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Databases & Information Systems (AREA)
- Quality & Reliability (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Investigating Or Analyzing Materials Using Thermal Means (AREA)
Abstract
The invention discloses a resistance spot welding quality detection method and a system based on feature fusion, wherein the method comprises the following steps: collecting welding current signals, electrode voltage signals and temperature signals in a welding process; calculating a dynamic power signal from the welding current signal and the electrode voltage signal; taking fitting errors into consideration, carrying out improved EMD (empirical mode decomposition) on the dynamic power signals to obtain main component signals; fusing the temperature signal and the main component signal and converting the fused temperature signal and the main component signal into an image to obtain a signal image; and inputting the signal image into a pre-trained detection model to finish quality detection. The system comprises: the system comprises a signal acquisition module, a dynamic power calculation module, a signal decomposition module, a signal fusion module and a detection module. By using the invention, the quality of the resistance spot welding can be comprehensively and real-timely detected. The invention can be widely applied to the field of welding detection.
Description
Technical Field
The invention relates to the field of welding detection, in particular to a resistance spot welding quality detection method and system based on feature fusion.
Background
Resistance spot welding is a spot joining technique widely used in the manufacture of sheet structures. The working principle is that a certain pressure is applied between the two electrodes and the workpiece to be welded, and part of metal is melted by utilizing resistance heat generated when a large current passes through the workpiece, so that a welding spot is formed. The resistance spot welding has the advantages of high production efficiency, low cost, high automation degree and the like, and is widely applied to the production of roof body assemblies and the like in the automobile manufacturing.
Traditional destructive testing has lower detection efficiency and can cause waste of welding materials. In recent years, many scholars have conducted research on nondestructive testing of resistance spot welding quality. However, the existing nondestructive detection method for the quality of the resistance spot welding can not realize comprehensive online detection of all welding spots in the production process, and still belongs to post-welding spot inspection, and has certain locality and time delay.
Disclosure of Invention
In view of the above, in order to solve the technical problem that the existing resistance welding quality detection method cannot comprehensively detect in real time, the invention provides a resistance spot welding quality detection method based on feature fusion, which comprises the following steps:
collecting welding current signals, electrode voltage signals and temperature signals in a welding process;
calculating a dynamic power signal from the welding current signal and the electrode voltage signal;
taking fitting errors into consideration, carrying out improved EMD (empirical mode decomposition) on the dynamic power signals to obtain main component signals;
fusing the temperature signal and the main component signal and converting the fused temperature signal and the main component signal into an image to obtain a signal image;
and inputting the signal image into a pre-trained detection model to finish quality detection.
In some embodiments, the step of performing an improved EMD decomposition of the dynamic power signal to obtain a principal component signal taking into account the fitting error specifically includes:
calculating each maximum value and each minimum value of the dynamic power signal, and defining a rightmost extreme point, a leftmost extreme point and an error set;
calculating left monotonicity of the rightmost extreme point, comparing the left monotonicity with left monotonicity of each point in the error set, and deleting the corresponding point if the monotonicity is different;
calculating the right monotonicity of the leftmost extreme point, comparing the right monotonicity with the left monotonicity of each corresponding point in the error set, and deleting the corresponding point if the monotonicity is different;
predicting the next extreme value according to the rest extreme points to obtain predicted points;
fitting according to the predicted points to obtain an upper envelope curve and a lower envelope curve of the dynamic power signal;
calculating local mean values of the upper envelope curve and the lower envelope curve;
the dynamic power signal is used as an original signal and subtracted from the local mean value to obtain a first component signal;
based on the first component signal, combining with the judgment of the basic condition to obtain IMF;
and selecting the IMF by adopting a correlation coefficient, and separating to obtain a main component signal.
In this embodiment, an extremum endpoint prediction based on extremum point correlation is provided to improve the EMD decomposition to overcome the accumulated fitting error.
In some embodiments, the determination of the base condition comprises:
the number of the extreme values of the first component signal is equal to or different from the number of the zero crossing points by not more than 1;
the mean value of the upper and lower envelopes determined by any local extreme point is equal to zero.
In this embodiment, the IMF determination condition is determined, and when the component signal meets the condition, an IMF signal is obtained by confirmation.
In some embodiments, the step of obtaining IMF based on the first component signal in combination with the determination of the basic condition specifically includes:
the dynamic power signal is used as an original signal and subtracted from the local part to obtain a first component signal;
if the first component signal meets the judgment of the basic condition, taking the first component signal as an IMF;
if the first component signal does not meet the judgment of the basic condition, taking the first component signal as a new original signal, and circularly subtracting the judgment step until the IMF is obtained;
subtracting the dynamic power signal from the IMF to obtain a residual function;
and taking the residual function as a new original signal and repeating the subtraction judging step until the next IMF is obtained.
In this embodiment, the multi-component signal is decomposed into a set of single-component signals, each IMF representing an oscillation mode with instantaneous frequency.
In some embodiments, the step of selecting the IMF by using a correlation coefficient and separating to obtain a principal component signal specifically includes:
calculating a correlation coefficient between the dynamic power signal and each IMF;
finding out the IMF with the largest correlation coefficient, and recording the serial number of the IMF as M;
the sum of IMFs with a sequence number smaller than M is used as a separation noise signal, and the sum of IMFs with a sequence number greater than M is used as a main component signal.
In this embodiment, since the correlation coefficient between the IMF containing noise and the original signal is weak, the IMF of the noise signal can be screened out by this method.
In some embodiments, the step of fusing and converting the temperature signal and the principal component signal into an image to obtain a signal image specifically includes:
fusing the temperature signal and the main component signal based on a Kalman filtering algorithm to obtain a fused signal;
based on the fusion signal, constructing an image matrix by taking a time step as the width of the image, and converting time sequence data into image data to obtain a signal image.
In the embodiment, the data are fused through Kalman filtering, and the fused signals consider measurement noise and system dynamics, so that the method has higher accuracy and stability.
In some embodiments, the specific training steps of the detection model are as follows:
constructing a detection model based on a residual error network;
training the detection model by using a pre-constructed training set, and optimizing the Dropout rate initial value and the learning rate of the detection model by using a bat algorithm until the detection accuracy reaches a preset value.
In the embodiment, the fusion signal is converted into the image data to be used as input, and the initial value of the Dropout rate and the learning rate of ResNet are optimized by adopting a bat algorithm, so that model errors are reduced, and the model has higher accuracy in the on-line detection and identification of the resistance spot welding quality.
The invention also provides a resistance spot welding quality detection system based on feature fusion, which comprises:
the signal acquisition module is used for acquiring welding current signals, electrode voltage signals and temperature signals in the welding process;
the dynamic power calculation module is used for calculating a dynamic power signal according to the welding current signal and the electrode voltage signal;
the signal decomposition module is used for carrying out improved EMD decomposition on the dynamic power signal in consideration of fitting errors to obtain a main component signal;
the signal fusion module is used for fusing the temperature signal and the main component signal and converting the fused signals into images to obtain signal images;
and the detection module is used for inputting the signal image into a pre-trained detection model to finish quality detection.
Based on the scheme, the invention provides a resistance spot welding quality detection method and a system based on feature fusion, wherein firstly, a dynamic power signal in a welding process is calculated through current and voltage; extracting principal component signals using an improved EMD decomposition to overcome fitting errors of a conventional EMD decomposition; the main component signals of the dynamic power and the temperature signals around the welding spot are fused through Kalman filtering fusion, so that the effective measurement data of different sensors are fully utilized, and the accuracy and the stability are higher; finally, intelligent recognition is carried out on the resistance spot welding quality by using a residual neural network (BA-ResNet model) optimized by a bat algorithm, so that the traditional manual detection method is changed.
Drawings
FIG. 1 is a flow chart of steps of a method for detecting quality of resistance spot welding based on feature fusion of the present invention;
FIG. 2 is a schematic diagram of a portion of the residual network of the present invention;
fig. 3 is a schematic structural diagram of a detection device according to an embodiment of the present invention.
Detailed Description
The following description of the embodiments of the present application will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are only some, but not all, of the embodiments of the present application. All other embodiments, which can be made by one of ordinary skill in the art without undue burden from the present disclosure, are within the scope of the present disclosure.
For convenience of description, only a portion related to the present invention is shown in the drawings. Embodiments and features of embodiments in this application may be combined with each other without conflict.
It should be appreciated that "system," "apparatus," "unit" and/or "module" as used in this application is one method for distinguishing between different components, elements, parts, portions or assemblies at different levels. However, if other words can achieve the same purpose, the word can be replaced by other expressions.
As used in this application and in the claims, the terms "a," "an," "the," and/or "the" are not specific to the singular, but may include the plural, unless the context clearly dictates otherwise. In general, the terms "comprises" and "comprising" merely indicate that the steps and elements are explicitly identified, and they do not constitute an exclusive list, as other steps or elements may be included in a method or apparatus. The inclusion of an element defined by the phrase "comprising one … …" does not exclude the presence of additional identical elements in a process, method, article, or apparatus that comprises an element.
In the description of the embodiments of the present application, "plurality" means two or more than two. The following terms "first", "second" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include one or more such feature.
Additionally, flowcharts are used in this application to describe the operations performed by systems according to embodiments of the present application. It should be appreciated that the preceding or following operations are not necessarily performed in order precisely. Rather, the steps may be processed in reverse order or simultaneously. Also, other operations may be added to or removed from these processes.
Referring to fig. 1, a flowchart of an alternative example of a method for detecting quality of resistance spot welding based on feature fusion according to the present invention may be applied to a computer device, and the method according to the present embodiment may include, but is not limited to, the following steps:
step S1, acquiring welding current signals, electrode voltage signals and temperature signals in a welding process;
in the embodiment, a twisted pair and a rogowski coil are respectively arranged at an electrode of a resistance spot welder and are used for collecting welding current and electrode voltage signals in a welding process; an annular heat radiation infrared temperature sensor is arranged above the upper electrode to collect temperature signals and is used for synchronously collecting annular temperatures around welding spots, and an annular array temperature average value is taken as the temperature signals.
S2, calculating a dynamic power signal according to the welding current signal and the electrode voltage signal;
the calculation formula of the dynamic power is as follows:
P(t)＝U(t)×I(t)
wherein P (t) is dynamic power, U (t) is welding voltage, I (t) is welding current, and t is welding time.
S3, taking fitting errors into consideration, and performing improved EMD (empirical mode decomposition) on the dynamic power signals to obtain main component signals;
s4, fusing the temperature signal and the main component signal and converting the fused signals into images to obtain signal images;
and S5, inputting the signal image into a pre-trained detection model to finish quality detection.
In some possible embodiments, the step of performing an improved EMD decomposition of the dynamic power signal to obtain a principal component signal taking into account fitting errors specifically includes:
step S3.1, calculating the maxima e of the dynamic power signal S (t) max (i) And minimum value e min (i) Wherein e is max (end) and e min (end) defining an error set E for the rightmost extreme point;
E＝{|e max (i)-e max (end)|，|e min (i)-e min (end)|},i＝1,2,…,n
s3.2, calculating left monotonicity of the rightmost extreme point, comparing the left monotonicity with left monotonicity of each point in the error set, and deleting the corresponding point if the monotonicity is different;
s3.3, predicting the next extreme value according to the rest extreme points to obtain predicted points;
the remaining m maxima points are set up as equations:
A m e max (j 1 )+A m-1 e max (j 1 -1)+…+A 0 e max (j 1 -m)＝e max (j 1 +1)
A m e max (j 2 )+A m-1 e max (j 2 -1)+…+A 0 e max (j 2 -m)＝e max (j 2 +1)
A m e max (j m )+A m-1 e max (j m -1)+…+A 0 e max (j m -m)＝e max (j m +1)
the next maximum of the rightmost maximum point may be predicted as follows:
A m e max (end)+A m-1 e max (end-1)+…+A 0 e max (end-m)＝e max (end+1)
similarly can be applied to e min (end+1)，e max (start-1) and e min (start-1) prediction, wherein e min (end+1) represents the next predicted point, e, of the rightmost minimum point max (start-1) the previous predicted point e representing the leftmost maximum point min (start-1) represents the previous predicted point of the leftmost minimum point.
S3.4, fitting according to predicted points to obtain an upper envelope curve and a lower envelope curve of the dynamic power signal;
and connecting predicted extreme points by using a cubic spline curve, and fitting an upper envelope curve and a lower envelope curve of the signal s (t).
In EMD decomposition, each IMF requires multiple "screening" processes, and each screening process requires the calculation of a local average of the signal from the upper and lower envelopes. The upper (lower) envelope is obtained by spline interpolation fitting of local maximum (small) values of the signal, and the endpoint of the original signal cannot generally judge whether the endpoint is an extreme point, if the endpoint of the original signal is regarded as the extreme point, the upper and lower envelope may diverge at two ends of the data sequence, the divergence gradually inwards along with the operation, so that the whole data sequence is affected, the extreme points obtained by preliminary calculation are all located inside the signal, the left extension of the leftmost extreme point and the right extension of the rightmost extreme point in the existing extreme point set are obtained through monotonicity prediction of the internal extreme point, and compared with the endpoint of the original signal, the extreme point predicted by the original EMD decomposition method is used for replacing the endpoint of the original signal.
Step S3.5, calculating the local mean m of the upper envelope and the lower envelope 1 (t)；
S3.6, taking the dynamic power signal as an original signal and subtracting the local mean value to obtain a first component signal;
the formula of this step is denoted as h 1 (t)＝s(t)-m 1 (t)。
Step S3.7, based on the first component signal, combining with the judgment of the basic condition to obtain the IMF;
wherein each IMF of the plurality must satisfy two basic conditions, the determination of which includes: (1) The number of extremum values and the number of zero crossings of the first component signal must be equal or differ by not more than 1. (2) The mean of the upper and lower envelopes, determined by any local extreme point, must be equal to zero.
S3.7.1 if the first component signal satisfies the basic condition, determining that the first component signal h 1 (t) as IMF;
s3.7.2 if the first component signal does not satisfy the basic condition, determining that the first component signal h 1 (t) as a new original signal, the subtraction judging step is circulated until IMF is obtained, denoted as C 1 (t)；
S3.7.3 subtracting the dynamic power signal from the IMF to obtain a residual function;
the formula of this step is denoted as r 1 (t)＝s(t)-C 1 (t)。
S3.6.5, taking the residual function as a new original signal and repeating the subtraction judging step until the next IMF is obtained.
The above is followed until the final residual signal is monotonic, and the improved EMD decomposition is completed.
And S3.7, selecting the IMF by adopting a correlation coefficient, and separating to obtain a main component signal.
The calculation method of the correlation coefficient comprises the following steps:
wherein x is i And C i The i-th sample of the original signal and the extracted IMF, respectively.
The correlation coefficient is calculated between the original signal and the extracted IMF, if the original signal and the ith IMF are different, the correlation coefficient rho is weak, if the original signal and the ith IMF are quasi-identical, the correlation coefficient rho is strong, so that the rho coefficient between the IMF containing noise and the original signal is weak, and the IMF of the noise signal can be screened out by the method. And respectively calculating the correlation coefficient of each IMF and the original signal, finding out the IMF with the largest correlation coefficient, recording the IMF with the sequence number of M, wherein the sum of the IMFs with the sequence numbers of less than M is a separation noise signal, and the sum of all the IMFs from M to the tail is a main component signal. The dynamic power principal component signal separated through the above steps is regarded as a measured value of the power signal.
It is difficult to determine whether the value at the end point is an extreme point of the sequence in the conventional EMD decomposition, and if the end point value is directly regarded as the extreme point, the upper and lower envelopes at the end point may diverge, which results in a fitting error in the decomposition, and the error is accumulated as the IMF is decomposed. Therefore, this embodiment proposes an improvement of the EMD decomposition based on extreme endpoint prediction of the extreme point correlation.
In some possible embodiments, the step of fusing and converting the temperature signal and the principal component signal into an image to obtain a signal image specifically includes:
s4.1, fusing the temperature signal and the main component signal based on a Kalman filtering algorithm to obtain a fused signal;
and adopting Kalman filtering to fuse the power signal and the temperature signal, and setting the power and the temperature as system state vectors, wherein a system prediction equation is as follows:
in the method, in the process of the invention,is a system state vector, P k Representing power, T k The temperature is represented by F, which is the state transition matrix, here F is the identity matrix.
Updating the state covariance estimate using the state process noise covariance matrix Q:
wherein P is k-1 The system state covariance at the moment k-1 is expressed, and Q is the system process noise covariance matrix.
Calculating Kalman gain:
wherein K is k For Kalman gain, H is the measurement matrix, R is the measurement noise covariance matrix, where H is a 2×2 identity matrix.
Updating the state estimation:
wherein x is k For best estimate at time k, z k Is the measurement at time k.
Updating the state covariance after the state estimate is complete:
the steps are circulated to obtain filtering fusion data of power and temperature at each moment, and the fused signals consider measurement noise and system dynamics, so that the method has higher accuracy and stability.
And S4.2, constructing an image matrix by taking a time step as the width of the image based on the fusion signal, and converting the time sequence data into image data to obtain a signal image.
After the kalman filtered data signal is obtained, each row represents a time step and each column represents a system state (power and temperature) because the output of the kalman state estimation is a two-dimensional vector time sequence containing power and temperature. Converting time series data into image data, determining time steps as the width of the image, creating an image matrix, wherein the dimension is [ width, time steps, 2], and in the matrix, a first channel is used for representing power, a second channel is used for representing temperature, and obtaining the image data.
In some possible embodiments, the specific training steps of the detection model are as follows:
constructing a detection model based on a residual error network;
the structure of the detection model refers to fig. 2, conv is convolution, BN is batch normalization, reLu is a correction linear unit, max pool is maximum value pooling, average pool is mean pool, FC is a full connection layer, residual modules are in a dashed line frame, and the number of the residual modules can be adjusted according to specific conditions.
Training the detection model by using a pre-constructed training set, and optimizing the Dropout rate initial value and the learning rate of the detection model by using a bat algorithm until the detection accuracy reaches a preset value.
The bat algorithm steps are as follows:
first, a set of bats is initialized in a random fashion in a hyper-parameter space, where each bat contains two hyper-parameters: a Dropout rate initial value (denoted D) and a learning rate (denoted γ). The initial maximum pulse volume is set to A 0 The maximum pulse rate is R 0 Search pulse frequency f i Is in the range f min To f max The attenuation coefficient of the pulse loudness is alpha, the enhancement coefficient of the search frequency is gamma, and the search precision epsilon or the maximum iteration number item is set max 。
For each bat, a residual neural network is constructed and trained using the bat's hyper-parameter settings (Dropout rate initial value and learning rate). Then, the fitness value is evaluated according to the accuracy of the residual network model on the validation set.
Initializing bat position x i Finding the minimum solution X of the current error according to the approach degree of the target value * Updating the hyper-parameters of the bat:
f i ＝f min +(f max -f min )β
in the above, f i Is the search pulse frequency of bat i, beta is the pulse frequency of [0,1 ]]Is used to determine the random number of the uniform distribution,dropout rate initial values of bat i at time t and t-1, respectively, +.>Respectively representing the learning rate, X of bat i at time t and t-1 D* Is the optimal initial value of the Dropout rate, X in all bats currently γ* Is the optimal learning rate in all bats currently.
Then generating uniformly distributed random number rand E [0,1 ]]If rand>Pulse emissivity r i Randomly disturbing the current solution to generate a new solution and performing out-of-range processing; if rand<Pulse loudness A i And f (x) i )<f(X * ) Then a new solution is accepted and loudness and pulse rate are calculated according to the following equation:
in the above-mentioned method, the step of,and->Indicating the pulse loudness of bat i at time t+1 and t, respectively, +.>Representing the pulse rate of bat i at time t + 1.
And sequencing the hyper-parameters of the bat according to the fitness value to determine a hyper-parameter combination with the minimum current error, repeating the iteration steps until the preset optimal solution condition is met or the maximum iteration times are reached, finally outputting a global optimal value and the optimal hyper-parameter combination, and training a residual neural network detection model.
In addition, a construction method of the training set is also provided, a large number of welding experiments are carried out to prepare sample points by respectively changing four main welding parameters of welding current, welding time, electrode pressure and welding position through control variables, and the welding points can be divided into four categories of virtual welding, normal welding, splashing welding and overlapping welding according to different welding qualities. The nugget diameter of the virtual welding spot is obviously smaller than the welding requirement and even no nugget is formed, and the splashing welding spot is formed by spraying molten metal from the joint surface of the welding piece or between the electrode and the surface of the welding piece in the welding process, so that the surface of the welding piece is burnt, blackened and uneven, the appearance quality of the welding spot is reduced, the welding energy consumption is increased, and the surface of the electrode is continuously worn. The welding spots which are overlapped with the existing welding spot nugget area by more than one half are regarded as overlapped welding spots, and although the overlapped welding spots can also generate splashing in the welding process, the welding spots are different from common splashing welding spots in that nuggets are formed at the original positions before a large welding current passes through, so that larger splashing is easy to occur, the tensile strength is relatively larger, and the dynamic signal change rule is different from that of the splashing welding spots.
The low carbon steel with the size of 100mm multiplied by 50mm multiplied by 0.8mm is adopted for experiments, and the prediction capability of the algorithm model is superior to that of a random forest algorithm, a BP neural network and a convolutional neural network based on dynamic resistance and dynamic power signals through comparative analysis, so that higher detection accuracy can be achieved.
In conclusion, the method has the advantages of convenience in use, good reliability, high detection precision and capability of detecting in real time.
Based on the method, the invention also provides a structure of the detection device in the practical application scene, referring to fig. 3, the rogowski coil is matched with the integral amplifying circuit to realize the measurement of welding current, and in order to reduce the induction noise generated by secondary current in the welding process, twisted pair wires are selected to reduce the induction noise to a certain extent by reducing the surrounding area of the wires. In a vehicle body resistance welder mainly comprising welding tongs, the voltage of a secondary loop is measured by selecting a method of embedding a wire at the output end of a secondary rectifying diode of a transformer in consideration of the fact that the exposed wire is easy to break and damage or is damaged by welding spatter. The method can avoid the exposure and damage of the lead wires and ensure the stability and accuracy of voltage measurement. And simultaneously, an annular heat radiation infrared temperature sensor is arranged on the welding tongs, and the average temperature of the annular periphery of the welding spot is measured. When welding is carried out, the method is adopted to synchronously acquire and obtain a welding current, voltage and temperature change sequence in the welding process.
A resistance spot welding quality detection system based on feature fusion, comprising:
the signal acquisition module is used for acquiring welding current signals, electrode voltage signals and temperature signals in the welding process;
the dynamic power calculation module is used for calculating a dynamic power signal according to the welding current signal and the electrode voltage signal;
the signal decomposition module is used for carrying out improved EMD decomposition on the dynamic power signal in consideration of fitting errors to obtain a main component signal;
the signal fusion module is used for fusing the temperature signal and the main component signal and converting the fused signals into images to obtain signal images;
and the detection module is used for inputting the signal image into a pre-trained detection model to finish quality detection.
The content in the method embodiment is applicable to the system embodiment, the functions specifically realized by the system embodiment are the same as those of the method embodiment, and the achieved beneficial effects are the same as those of the method embodiment.
While the preferred embodiment of the present invention has been described in detail, the invention is not limited to the embodiment, and various equivalent modifications and substitutions can be made by those skilled in the art without departing from the spirit of the invention, and these modifications and substitutions are intended to be included in the scope of the present invention as defined in the appended claims.
Claims (8)
1. The resistance spot welding quality detection method based on feature fusion is characterized by comprising the following steps of:
collecting welding current signals, electrode voltage signals and temperature signals in a welding process;
calculating a dynamic power signal from the welding current signal and the electrode voltage signal;
taking fitting errors into consideration, carrying out improved EMD (empirical mode decomposition) on the dynamic power signals to obtain main component signals;
fusing the temperature signal and the main component signal and converting the fused temperature signal and the main component signal into an image to obtain a signal image;
and inputting the signal image into a pre-trained detection model to finish quality detection.
2. The method for detecting quality of resistance spot welding based on feature fusion according to claim 1, wherein the step of performing improved EMD decomposition on the dynamic power signal in consideration of fitting errors to obtain a main component signal specifically comprises:
calculating each maximum value and each minimum value of the dynamic power signal, and defining a rightmost extreme point, a leftmost extreme point and an error set;
calculating left monotonicity of the rightmost extreme point, comparing the left monotonicity with left monotonicity of each point in the error set, and deleting the corresponding point if the monotonicity is different;
calculating the right monotonicity of the leftmost extreme point, comparing the right monotonicity with the left monotonicity of each corresponding point in the error set, and deleting the corresponding point if the monotonicity is different;
predicting the next extreme value according to the rest extreme points to obtain predicted points;
fitting according to the predicted points to obtain an upper envelope curve and a lower envelope curve of the dynamic power signal;
calculating local mean values of the upper envelope curve and the lower envelope curve;
the dynamic power signal is used as an original signal and subtracted from the local mean value to obtain a first component signal;
based on the first component signal, combining with the judgment of the basic condition to obtain IMF;
and selecting the IMF by adopting a correlation coefficient, and separating to obtain a main component signal.
3. The method for detecting the quality of resistance spot welding based on feature fusion according to claim 2, wherein the determination of the basic condition comprises:
the number of the extreme values of the first component signal is equal to or different from the number of the zero crossing points by not more than 1;
the mean value of the upper and lower envelopes determined by any local extreme point is equal to zero.
4. A method for detecting quality of resistance spot welding based on feature fusion according to claim 3, wherein the step of obtaining IMF based on the first component signal in combination with the determination of the basic condition specifically comprises:
if the first component signal meets the judgment of the basic condition, taking the first component signal as an IMF;
if the first component signal does not meet the judgment of the basic condition, taking the first component signal as a new original signal, and circularly subtracting the judgment step until the IMF is obtained;
subtracting the dynamic power signal from the IMF to obtain a residual function;
and taking the residual function as a new original signal and repeating the subtraction judging step until the next IMF is obtained.
5. The method for detecting quality of resistance spot welding based on feature fusion according to claim 4, wherein the step of selecting the IMF by using a correlation coefficient and separating to obtain a main component signal comprises the steps of:
calculating a correlation coefficient between the dynamic power signal and each IMF;
finding out the IMF with the largest correlation coefficient, and recording the serial number of the IMF as M;
the sum of IMFs with a sequence number smaller than M is used as a separation noise signal, and the sum of IMFs with a sequence number greater than M is used as a main component signal.
6. The method for detecting the quality of resistance spot welding based on feature fusion according to claim 1, wherein the step of fusing the temperature signal and the principal component signal and converting the fused signals into an image to obtain a signal image specifically comprises the steps of:
fusing the temperature signal and the main component signal based on a Kalman filtering algorithm to obtain a fused signal;
based on the fusion signal, constructing an image matrix by taking a time step as the width of the image, and converting time sequence data into image data to obtain a signal image.
7. The method for detecting the quality of resistance spot welding based on feature fusion according to claim 1, wherein the specific training steps of the detection model are as follows:
constructing a detection model based on a residual error network;
training the detection model by using a pre-constructed training set, and optimizing the Dropout rate initial value and the learning rate of the detection model by using a bat algorithm until the detection accuracy reaches a preset value.
8. A resistance spot welding quality detection system based on feature fusion, comprising:
the signal acquisition module is used for acquiring welding current signals, electrode voltage signals and temperature signals in the welding process;
the dynamic power calculation module is used for calculating a dynamic power signal according to the welding current signal and the electrode voltage signal;
the signal decomposition module is used for carrying out improved EMD decomposition on the dynamic power signal in consideration of fitting errors to obtain a main component signal;
the signal fusion module is used for fusing the temperature signal and the main component signal and converting the fused signals into images to obtain signal images;
and the detection module is used for inputting the signal image into a pre-trained detection model to finish quality detection.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311433968.9A CN117437199A (en) | 2023-11-01 | 2023-11-01 | Resistance spot welding quality detection method and system based on feature fusion |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311433968.9A CN117437199A (en) | 2023-11-01 | 2023-11-01 | Resistance spot welding quality detection method and system based on feature fusion |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117437199A true CN117437199A (en) | 2024-01-23 |
Family
ID=89553075
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311433968.9A Pending CN117437199A (en) | 2023-11-01 | 2023-11-01 | Resistance spot welding quality detection method and system based on feature fusion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117437199A (en) |
-
2023
- 2023-11-01 CN CN202311433968.9A patent/CN117437199A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Pal et al. | Artificial neural network modeling of weld joint strength prediction of a pulsed metal inert gas welding process using arc signals | |
| Wang et al. | A tutorial on deep learning-based data analytics in manufacturing through a welding case study | |
| Lin | The use of the Taguchi method with grey relational analysis and a neural network to optimize a novel GMA welding process | |
| Lei et al. | Real-time weld geometry prediction based on multi-information using neural network optimized by PCA and GA during thin-plate laser welding | |
| Chen et al. | A spectroscopic method based on support vector machine and artificial neural network for fiber laser welding defects detection and classification | |
| Thekkuden et al. | Investigation of feed-forward back propagation ANN using voltage signals for the early prediction of the welding defect | |
| Wan et al. | Quality monitoring based on dynamic resistance and principal component analysis in small scale resistance spot welding process | |
| CN111738369A (en) | Weld penetration state and penetration depth real-time prediction method based on visual characteristics of molten pool | |
| Chen et al. | Prediction of weld bead geometry of MAG welding based on XGBoost algorithm | |
| CN117593298B (en) | Laser welding quality detection system based on machine vision | |
| JP2003516860A (en) | Method and apparatus for quality control of laser butt welded sheet metal or band seams | |
| Garcia-Allende et al. | Spectral processing technique based on feature selection and artificial neural networks for arc-welding quality monitoring | |
| CN113828947B (en) | BP neural network laser welding seam forming prediction method based on double optimization | |
| Kao et al. | Laser cladding quality monitoring using coaxial image based on machine learning | |
| CN114861498A (en) | Resistance spot welding quality on-line detection method fused with multi-sensing time sequence signal mechanism model | |
| Lee et al. | Development of real-time diagnosis framework for angular misalignment of robot spot-welding system based on machine learning | |
| Sarkar et al. | Machine learning method to predict and analyse transient temperature in submerged arc welding | |
| Oh et al. | A study on intelligent algorithm to control welding parameters for lap-joint | |
| Yu et al. | Monitoring of butt weld penetration based on infrared sensing and improved histograms of oriented gradients | |
| CN117437199A (en) | Resistance spot welding quality detection method and system based on feature fusion | |
| Nogay et al. | Classification of operation cases in electric arc welding wachine by using deep convolutional neural networks | |
| Peng et al. | Real-time defect detection scheme based on deep learning for laser welding system | |
| Lim et al. | Estimation of weld pool sizes in GMA welding process using neural networks | |
| Baek et al. | Optimization of weld penetration prediction based on weld pool image and deep learning approach in gas tungsten arc welding | |
| Jamnikar et al. | Comprehensive molten pool condition-process relations modeling using CNN for wire-feed laser additive manufacturing |
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 |