CN116559196A - Arc additive manufacturing defect detection system and method - Google Patents
Arc additive manufacturing defect detection system and method Download PDFInfo
- Publication number
- CN116559196A CN116559196A CN202310498199.4A CN202310498199A CN116559196A CN 116559196 A CN116559196 A CN 116559196A CN 202310498199 A CN202310498199 A CN 202310498199A CN 116559196 A CN116559196 A CN 116559196A
- Authority
- CN
- China
- Prior art keywords
- information
- neural network
- additive manufacturing
- training
- defect detection
- 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.)
- Granted
Links
- 230000007547 defect Effects 0.000 title claims abstract description 79
- 239000000654 additive Substances 0.000 title claims abstract description 52
- 230000000996 additive effect Effects 0.000 title claims abstract description 52
- 238000004519 manufacturing process Methods 0.000 title claims abstract description 47
- 238000001514 detection method Methods 0.000 title claims abstract description 40
- 238000000034 method Methods 0.000 title claims abstract description 37
- 230000008569 process Effects 0.000 claims abstract description 19
- 230000001360 synchronised effect Effects 0.000 claims abstract description 8
- 238000012549 training Methods 0.000 claims description 48
- 238000013528 artificial neural network Methods 0.000 claims description 47
- 230000006870 function Effects 0.000 claims description 21
- 230000004913 activation Effects 0.000 claims description 18
- 238000013527 convolutional neural network Methods 0.000 claims description 16
- 238000012360 testing method Methods 0.000 claims description 15
- 230000015654 memory Effects 0.000 claims description 13
- 238000012795 verification Methods 0.000 claims description 11
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 6
- 230000015572 biosynthetic process Effects 0.000 claims description 6
- 239000000463 material Substances 0.000 claims description 6
- 230000007787 long-term memory Effects 0.000 claims description 5
- 230000009467 reduction Effects 0.000 claims description 5
- 230000000694 effects Effects 0.000 claims description 3
- 238000000605 extraction Methods 0.000 claims description 3
- 238000001914 filtration Methods 0.000 claims description 3
- 230000014509 gene expression Effects 0.000 claims description 3
- 238000003062 neural network model Methods 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 3
- 230000035945 sensitivity Effects 0.000 claims description 3
- 125000004122 cyclic group Chemical group 0.000 claims description 2
- 238000011156 evaluation Methods 0.000 claims description 2
- 238000011056 performance test Methods 0.000 claims description 2
- 238000005516 engineering process Methods 0.000 description 5
- 238000010924 continuous production Methods 0.000 description 3
- 239000000758 substrate Substances 0.000 description 3
- 238000010586 diagram Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 239000002184 metal Substances 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000006403 short-term memory Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008094 contradictory effect Effects 0.000 description 1
- 230000001808 coupling effect Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 238000003475 lamination Methods 0.000 description 1
- 238000002844 melting Methods 0.000 description 1
- 230000008018 melting Effects 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 238000003466 welding Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/95—Investigating the presence of flaws or contamination characterised by the material or shape of the object to be examined
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/62—Systems in which the material investigated is excited whereby it emits light or causes a change in wavelength of the incident light
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N25/00—Investigating or analyzing materials by the use of thermal means
- G01N25/72—Investigating presence of flaws
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
- G01N29/04—Analysing solids
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
- G01N29/44—Processing the detected response signal, e.g. electronic circuits specially adapted therefor
- G01N29/4481—Neural 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
- G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P10/00—Technologies related to metal processing
- Y02P10/25—Process efficiency
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Theoretical Computer Science (AREA)
- Evolutionary Computation (AREA)
- Analytical Chemistry (AREA)
- Pathology (AREA)
- Chemical & Material Sciences (AREA)
- Immunology (AREA)
- Biochemistry (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Acoustics & Sound (AREA)
- Signal Processing (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Investigating Or Analyzing Materials By The Use Of Magnetic Means (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to a defect detection system and method for arc additive manufacturing, and belongs to the field of arc additive manufacturing. The problems that most of the technical schemes can only detect surface defects of arc material-increasing components, and defects such as internal air holes and unfused defects cannot be detected are solved. The system comprises a magneto-optical imager, an infrared camera, a high-speed video camera, a synchronous controller, a computer and an acoustic sensor, wherein the magneto-optical imager, the infrared camera, the high-speed video camera and the acoustic sensor are electrically connected with the computer through the synchronous controller. The invention is based on multi-mode information in the arc material-increasing process, mainly comprises surface morphology information, temperature field information, acoustic information and magneto-optical information, realizes the detection of defects such as air holes, unfused and the like in the interior, and has high detection precision.
Description
Technical Field
The invention relates to a defect detection system and method for arc additive manufacturing, and belongs to the field of arc additive manufacturing.
Background
The arc additive manufacturing technology is an advanced manufacturing technology for obtaining a three-dimensional solid metal component by taking an arc as a heat source and layering metal wires layer by layer according to a specified path after melting. The arc additive manufacturing technology has the advantages of high lamination efficiency, high material utilization rate, low equipment cost and the like, is particularly suitable for the integrated forming of medium-large complex components, and has wide application prospects in the fields of aerospace, aviation, ships and the like. However, the arc additive manufacturing process is a multi-physical field coupling process, and is easily affected by various factors in the additive manufacturing process, so that various defects such as air holes, unfused, poor surface forming and the like are generated, and the performance and reliability of the arc additive component are difficult to ensure. Therefore, it is a very critical technology to realize effective and high-precision detection of arc additive component defects.
In recent years, with the continuous improvement of a large amount of data and computer computing power brought by social data, deep learning has been rapidly developed, and convolutional neural networks are widely used for defect detection in additive manufacturing processes. However, the prior art has the following drawbacks:
1. most of the technical schemes can only detect surface defects of the arc additive components, and the defects of internal air holes, unfused and the like can not be detected;
2. most of the technical schemes only use one kind of information in the arc material-increasing process to detect defects, the arc material-increasing process is a complex process with multi-parameter coupling effect, the influence degree of different defects on the same signal can be the same when different defects are formed, and certain defects can have no influence on some signal intensities, so that the defects in the arc material-increasing process are difficult to effectively detect only by one kind of signal;
3. arc material increase is a dynamic continuous process, and defect generation is also a dynamic continuous process, but most of the current technologies only use information at one moment to detect defects in the arc material increase process, and fail to fully use information on a time sequence to detect defects.
Therefore, it is desirable to provide a system and a method for detecting arc additive manufacturing defects to solve the above-mentioned problems.
Disclosure of Invention
The present invention addresses the deficiencies of the arc additive manufacturing techniques described above, and provides an arc additive manufacturing defect detection system and method, a brief overview of which is provided below in order to provide a basic understanding of certain aspects of the invention. It should be understood that this summary is not an exhaustive overview of the invention. It is not intended to identify key or critical elements of the invention or to delineate the scope of the invention.
The technical scheme of the invention is as follows:
the arc additive manufacturing defect detection system comprises a magneto-optical imager, an infrared camera, a high-speed video camera, a synchronous controller, a computer and an acoustic sensor, wherein the magneto-optical imager, the infrared camera, the high-speed video camera and the acoustic sensor are electrically connected with the computer through the synchronous controller.
The arc additive manufacturing defect detection method adopts the arc additive manufacturing defect detection system, and comprises the following steps:
s1: the method comprises the steps of obtaining multi-mode information in the arc material adding process by using a magneto-optical imager, an infrared camera, a high-speed camera and an acoustic sensor, dividing the obtained continuous information into frames to obtain signals of each frame, and marking information data and dividing the information data into training data, testing data and verification data;
s2: building a convolutional neural network and a cyclic neural network framework, and setting corresponding model super parameters;
s3: respectively inputting the obtained information into a convolutional neural network, training and optimizing a model by adopting an Adam algorithm, and extracting surface appearance information, temperature field information, acoustic signals and magnetic signal characteristics;
s4: inputting the surface morphology information, the temperature field information, the acoustic signal and the magnetic signal characteristics of the time sequence extracted in the S3 into a long-period memory neural network, optimizing a model by adopting an Adam algorithm, extracting the surface morphology information, the temperature field information, the acoustic signal and the magnetic signal characteristics of the time sequence direction, inputting the obtained surface morphology, the temperature field, the acoustic signal and the magnetic signal characteristics of the time sequence direction into a fully-connected layer neural network, optimizing the fully-connected layer, outputting the probability of air holes, unfused and poor surface formation, and predicting and identifying additive defects;
s5: inputting test data and verification data into a training model, detecting the effectiveness and accuracy of the S1, S2, S3 and S4 training models, and verifying the performance of the training models;
s6: the multi-mode information of the arc additive manufacturing process is collected in real time, the multi-mode information is input into a trained neural network model, whether defects are generated in the arc additive manufacturing process or not is output, and therefore online defect detection and positioning of the arc additive manufacturing process are achieved.
Preferably: in S1, before dividing a training set, a test set and a verification set, firstly, filtering the extracted information picture to remove irrelevant information pictures, then intercepting the information picture, omitting the irrelevant information at the edge, obtaining a main information part, rotating, amplifying and mirroring the processed information picture, and expanding the data scale.
Preferably: the super parameters required to be determined by the neural network in S2, S3, S4 and S5 include learning rate, regularization parameters, learning round number epoch, activation function form and small batch data mini-batch size.
Preferably: in S2, when training the neural network, setting an initial learning rate to be 0.1, then performing trial with multiplying power reduction, setting data mini-batch to be 64, and determining training iteration times by adopting an early stopping method;
the convolutional neural network adopts a ReLu activation function, the long-term memory neural network forgetting gate adopts a Sigmoid activation function, the output gate adopts a Softsign activation function, the output layer adopts a Softmax function for processing, and the three activation function expressions are as follows:
ReLu:y=max(0,x)
Sigmoid:
Softsign:
Softmax:
preferably: and S3, carrying out weight initialization by adopting Kaiming during training of the neural network, inputting the collected multi-mode data into a convolution layer and a downsampling layer of the convolution neural network, carrying out feature extraction of information related to defect formation, and carrying out data dimension reduction on the extracted feature information.
Preferably: s4, inputting the length time sequence multi-mode information characteristic X (t) of the related defects extracted by the convolutional neural network into a long-short-period memory neural network, inputting a gate (input gate), a forgetting gate (for gate) and an output gate (output gate) in a memory block (memory block) of the long-short-period memory neural network, jointly determining that information which is irrelevant to the defects in the multi-mode information X (t) of a certain time sequence is forgotten by utilizing a Sigmoid and Softsign activation function, not being used for defect detection, outputting multi-mode data Y (t) related to the defects by the output gate (output gate), calculating the information characteristic of Y (t) through a full-connection layer, and outputting the probability of various defects through Softmax;
and calculating the difference between the output value and the target value to perform error counter propagation, sequentially solving the error of each layer of neural network, updating the weight of the neural network through gradient descent, and ending the training process when the error is smaller than or equal to the expected value, wherein the training process is accelerated by using the NVIDIA Geforce 820M GPU.
Preferably: and S5, performing performance test on the trained model by using a test set and a verification set which are not used for training, avoiding the occurrence of under fitting and over fitting of the trained model, comprehensively evaluating the sensitivity and the accuracy of defect detection by adopting F-Score during performance evaluation, and intuitively representing the effect of detecting the defects by adopting an AUC curve.
The invention has the following beneficial effects:
the invention is based on multi-mode information in the arc material-increasing process, mainly comprises surface morphology information, temperature field information, acoustic information and magneto-optical information, realizes the detection of defects such as air holes, unfused and the like in the interior, and has high detection precision;
when the neural network is trained, the data enhancement method is adopted, so that the data volume required by the neural network training is greatly reduced, and the cost for acquiring experimental data is effectively saved; detecting defects of the arc material-increasing process by combining multi-mode information;
the invention adopts a convolutional neural network and a long-term and short-term memory network to extract and process the multi-mode information of space dimension and the multi-mode information of time dimension respectively, and performs defect detection on a dynamic continuous process.
Drawings
FIG. 1 is a schematic diagram of an arc additive manufacturing defect detection system;
FIG. 2 is a schematic diagram of an arc additive manufacturing defect detection method;
fig. 3 is a flow chart of neural network training.
In the figure: 1-magneto-optical imager, 2-infrared camera, 3-high-speed camera, 4-synchronous controller, 5-computer, 6-acoustic sensor, 7-base plate, 8-welder, 9-component.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the present invention more apparent, the present invention is described below by means of specific embodiments shown in the accompanying drawings. It should be understood that the description is only illustrative and is not intended to limit the scope of the invention. In addition, in the following description, descriptions of well-known structures and techniques are omitted so as not to unnecessarily obscure the present invention.
The first embodiment is as follows: referring to fig. 1, an arc additive manufacturing defect detection system of the present embodiment includes a magneto-optical imager 1, an infrared camera 2, a high-speed camera 3, a synchronization controller 4, a computer 5, and an acoustic sensor 6, where the magneto-optical imager 1, the infrared camera 2, the high-speed camera 3, and the acoustic sensor 6 are electrically connected to the computer 5 through the synchronization controller 4, the acoustic sensor 6 is mounted at the bottom of a substrate 7, the substrate 7 is used for placing a member 9, the magneto-optical imager 1, the infrared camera 2, and the high-speed camera 3 are mounted above the substrate 7, and a welding gun 8 welds the build 9.
The second embodiment is as follows: 1-3, an arc additive manufacturing defect detection method of the present embodiment adopts an arc additive manufacturing defect detection system, which includes the following steps:
s1: the method comprises the steps of obtaining multi-mode information in the arc material adding process by using a magneto-optical imager 1, an infrared camera 2, a high-speed camera 3 and an acoustic sensor 6, dividing the obtained continuous information into frames to obtain signals of each frame, and then marking information data and dividing the information data into training, testing and verifying data;
s2: building a convolutional neural network and a long-term and short-term memory neural network framework, and setting corresponding model super parameters;
s3: the obtained four kinds of information are respectively input into a convolutional neural network, training and optimizing are carried out on the model by adopting an Adam algorithm, and the characteristics of the surface appearance information, the temperature field information, the acoustic signal and the magnetic signal are extracted;
s4: inputting the surface morphology information, the temperature field information, the acoustic signals and the magnetic signal characteristics of the time sequence extracted in the S3 into a long-period memory neural network, optimizing a model by adopting an Adam algorithm, extracting the surface morphology information, the temperature field information, the acoustic signals and the magnetic signal characteristics of the time sequence direction, inputting the obtained surface morphology, the temperature field, the acoustic signals and the magnetic signal characteristics of the time sequence direction into a full-connection layer neural network, optimizing the full-connection layer, outputting air holes, unfused probability and probability of poor surface formation, and predicting and identifying additive defects, wherein the neural network comprises a convolutional neural network and a circulating neural network, and the convolutional neural network and the circulating neural network comprise the full-connection layer neural network;
s5: inputting test data and verification data into a training model, detecting the effectiveness and accuracy of the S1, S2, S3 and S4 training models, and verifying the performance of the training models;
s6: the multi-mode information of the arc additive manufacturing process is collected in real time, the multi-mode information is input into a trained neural network model, whether defects are generated in the additive manufacturing process or not is output, and accordingly online detection and positioning of the defects in the arc additive manufacturing process are achieved, namely the model trained by S1-S5 is used for the actual additive manufacturing process.
And a third specific embodiment: referring to fig. 1-3, a defect detection method for arc additive manufacturing in this embodiment is described, in S1, a magneto-optical imager 1, an infrared camera 2, a high-speed camera 3, and an acoustic sensor 6 synchronously transmit multi-mode information collected in real time during an arc additive manufacturing process to a computer 5, a synchronous controller 4 is used for information receiving control, and the synchronous controller 4 synchronously controls the infrared camera, an acoustic emission device, a high-number camera, and the magneto-optical imager, and synchronously transmits the multi-mode information collected in real time during the arc additive manufacturing process to the computer; before dividing the training set, the test set and the verification set, firstly filtering the extracted information picture to remove the irrelevant information picture, then intercepting the information picture, omitting the irrelevant information at the edge, obtaining the main information part, rotating, amplifying and mirroring the processed information picture, and expanding the data scale; the picture data is then marked and pressed 8:1:1 is divided into training, testing and verifying data, the marked pictures also need to be imported into txt files, the training data and the data labels are read when the marked pictures are used for training a neural network, the training, testing and verifying data need to be further converted into tensor (tensor) forms, a training model is realized based on a Pytorch framework 1.10.0, and the Python interpreter version is 3.9.7.
The specific embodiment IV is as follows: 1-3, in the arc additive manufacturing defect detection method of the present embodiment, the super parameters to be determined by the neural network in S2, S3, S4, S5 include learning rate, regularization parameters, learning round number epoch, activation function form and small batch data mini-batch size.
Fifth embodiment: 1-3, in S2, when training the neural network, the initial learning rate is set to 0.1, then continuous multiplying power is reduced for trial, small batch data mini-batch is set to 64, the training iteration number is determined by adopting an early stop method, and whether training is stopped is judged by verifying whether training errors are improved in the last 10 steps;
the convolutional neural network adopts a ReLu activation function, the long-term memory neural network forgetting gate adopts a Sigmoid activation function, the output gate adopts a Softsign activation function, the output layer adopts a Softmax function for processing, and the three activation function expressions are as follows:
ReLu:y=max(0,x)
Sigmoid:
Softsign:
Softmax:
specific embodiment six: referring to fig. 1-3, in S3, weighting initialization is performed by adopting Kaiming during training of the neural network, so as to avoid incapability of training of the neural network due to gradient disappearance or gradient explosion during training, the collected multi-mode data is input into a convolution layer and a downsampling layer of the convolution neural network, feature extraction of information related to defect formation is performed, and data dimension reduction is performed on the extracted feature information, so that overfitting of the neural network is avoided.
Seventh embodiment: referring to fig. 1-3, in S4, a certain length time series multi-mode information feature X (t) of a related defect extracted by a convolutional neural network is input into a long-short-period memory neural network, an input gate (input gate), a forgetting gate (for gate) and an output gate (output gate) in a memory block (memory block) of the long-short-period memory neural network jointly determine that information irrelevant to the defect in the multi-mode information X (t) of the certain time series should be forgotten by using Sigmoid and Softsign activation functions, and is not used for defect detection, and finally the output gate (output gate) outputs multi-mode data Y (t) related to the defect, and finally the information feature of Y (t) is calculated through a full connection layer, and finally the probabilities of various defects are output through Softmax;
and calculating the difference between the output value and the target value to perform error counter propagation, sequentially solving the error of each layer of neural network, updating the weight of the neural network through gradient descent, and ending the training process when the error is smaller than or equal to the expected value, wherein the training process is as shown in fig. 3, and the training process is accelerated by using the NVIDIA Geforce 820M GPU.
Eighth embodiment: referring to fig. 1-3, in a method for detecting defects in arc additive manufacturing according to this embodiment, in S5, performance testing is performed on a trained model using a test set and a verification set that are not used for training, so that under-fitting and over-fitting of the trained model are avoided, sensitivity and accuracy of defect detection are comprehensively assessed by using F-Score in performance assessment, and effects of detecting defects by using a neural network are intuitively represented by using an AUC curve.
It should be noted that, in the above embodiments, as long as the technical solutions that are not contradictory can be arranged and combined, those skilled in the art can exhaust all the possibilities according to the mathematical knowledge of the arrangement and combination, so the present invention does not describe the technical solutions after the arrangement and combination one by one, but should be understood that the technical solutions after the arrangement and combination have been disclosed by the present invention.
The above description is only of the preferred embodiments of the present invention and is not intended to limit the present invention, but various modifications and variations can be made to the present invention by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (8)
1. An arc additive manufacturing defect detection system, characterized in that: the system comprises a magneto-optical imager (1), an infrared camera (2), a high-speed camera (3), a synchronous controller (4), a computer (5) and an acoustic sensor (6), wherein the magneto-optical imager (1), the infrared camera (2), the high-speed camera (3) and the acoustic sensor (6) are electrically connected with the computer (5) through the synchronous controller (4).
2. The arc additive manufacturing defect detection method is characterized by comprising the following steps of: an arc additive manufacturing defect detection system employing the method of claim 1, comprising the steps of:
s1: the method comprises the steps of acquiring multi-mode information in an arc material adding process by using a magneto-optical imager (1), an infrared camera (2), a high-speed camera (3) and an acoustic sensor (6), dividing the acquired continuous information into frames to acquire signals of each frame, and then marking information data and dividing the information data into training data, test data and verification data;
s2: building a convolutional neural network and a cyclic neural network framework, and setting corresponding model super parameters;
s3: respectively inputting the obtained information into a convolutional neural network, training and optimizing a model by adopting an Adam algorithm, and extracting surface appearance information, temperature field information, acoustic signals and magnetic signal characteristics;
s4: inputting the surface morphology information, the temperature field information, the acoustic signals and the magnetic signal characteristics of the time sequence extracted in the S3 into a circulating neural network, optimizing a model by adopting an Adam algorithm, extracting the surface morphology information, the temperature field information, the acoustic signals and the magnetic signal characteristics of the time sequence direction, inputting the obtained surface morphology, the temperature field, the acoustic signals and the magnetic signal characteristics of the time sequence direction into a fully-connected layer neural network, optimizing the fully-connected layer, outputting the probability of air holes, unfused and poor surface formation, and predicting and identifying additive defects;
s5: inputting test data and verification data into a training model, detecting the effectiveness and accuracy of the S1, S2, S3 and S4 training models, and verifying the performance of the training models;
s6: the multi-mode information of the arc additive manufacturing process is collected in real time, the multi-mode information is input into a trained neural network model, whether defects are generated in the arc additive manufacturing process or not is output, and therefore online defect detection and positioning of the arc additive manufacturing process are achieved.
3. The arc additive manufacturing defect detection method of claim 2, wherein: in S1, before dividing a training set, a test set and a verification set, firstly, filtering the extracted information picture to remove irrelevant information pictures, then intercepting the information picture, omitting the irrelevant information at the edge, obtaining a main information part, rotating, amplifying and mirroring the processed information picture, and expanding the data scale.
4. A method of arc additive manufacturing defect detection according to claim 3, wherein: the super parameters required to be determined by the neural network in S2, S3, S4 and S5 include learning rate, regularization parameters, learning round number epoch, activation function form and small batch data mini-batch size.
5. The arc additive manufacturing defect detection method of claim 4, wherein: in S2, when training the neural network, setting an initial learning rate to be 0.1, then performing trial with multiplying power reduction, setting data mini-batch to be 64, and determining training iteration times by adopting an early stopping method;
the convolutional neural network adopts a ReLu activation function, the long-term memory neural network forgetting gate adopts a Sigmoid activation function, the output gate adopts a Softsign activation function, the output layer adopts a Softmax function for processing, and the three activation function expressions are as follows:
ReLu:y=max(0,x)
Sigmoid:
Softsign:
Softmax:
6. the arc additive manufacturing defect detection method of claim 2 or 4, wherein: and S3, carrying out weight initialization by adopting Kaiming during training of the neural network, inputting the collected multi-mode data into a convolution layer and a downsampling layer of the convolution neural network, carrying out feature extraction of information related to defect formation, and carrying out data dimension reduction on the extracted feature information.
7. The arc additive manufacturing defect detection method of claim 2 or 4, wherein: s4, inputting the length time series multi-mode information characteristic X (t) of the related defects extracted by the convolutional neural network into a long-short-period memory neural network, wherein an input gate (input gate), a forgetting gate (for gate) and an output gate (output gate) in a memory block (memoblock) of the long-short-period memory neural network jointly determine that information which is irrelevant to the defects in the multi-mode information X (t) of a certain time series is forgotten by utilizing a Sigmoid and Softsign activation function, and the information is not used for defect detection, and finally the output gate (output gate) outputs multi-mode data Y (t) related to the defects, and finally the information characteristic of Y (t) is calculated through a full connection layer and finally the probabilities of various defects are output through Softmax;
and calculating the difference between the output value and the target value to perform error counter propagation, sequentially solving the error of each layer of neural network, updating the weight of the neural network through gradient descent, and ending the training process when the error is smaller than or equal to the expected value, wherein the training process is accelerated by using the NVIDIA Geforce 820M GPU.
8. The arc additive manufacturing defect detection method of claim 2 or 4, wherein: and S5, performing performance test on the trained model by using a test set and a verification set which are not used for training, avoiding the occurrence of under fitting and over fitting of the trained model, comprehensively evaluating the sensitivity and the accuracy of defect detection by adopting F-Score during performance evaluation, and intuitively representing the effect of detecting the defects by adopting an AUC curve.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310498199.4A CN116559196B (en) | 2023-05-06 | 2023-05-06 | Arc additive manufacturing defect detection system and method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310498199.4A CN116559196B (en) | 2023-05-06 | 2023-05-06 | Arc additive manufacturing defect detection system and method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN116559196A true CN116559196A (en) | 2023-08-08 |
CN116559196B CN116559196B (en) | 2024-03-12 |
Family
ID=87501144
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310498199.4A Active CN116559196B (en) | 2023-05-06 | 2023-05-06 | Arc additive manufacturing defect detection system and method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116559196B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116893177A (en) * | 2023-09-11 | 2023-10-17 | 四川大学 | Method for monitoring internal defects in laser additive manufacturing process based on CCD and pyrometer |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108333219A (en) * | 2018-03-19 | 2018-07-27 | 长沙理工大学 | A kind of online lossless detection method for band large-scale metal component laser gain material manufacturing process |
CN111831895A (en) * | 2019-04-19 | 2020-10-27 | 天津科技大学 | Network public opinion early warning method based on LSTM model |
US20210089003A1 (en) * | 2017-12-20 | 2021-03-25 | Moog Inc. | Convolutional neural network evaluation of additive manufacturing images, and additive manufacturing system based thereon |
CN113828798A (en) * | 2021-09-23 | 2021-12-24 | 西安交通大学 | Local heat treatment method and system for electric arc additive manufacturing structural part |
CN114782322A (en) * | 2022-03-28 | 2022-07-22 | 华中科技大学 | YOLOv5 model arc additive manufacturing molten pool defect detection method |
CN115861187A (en) * | 2022-11-15 | 2023-03-28 | 哈尔滨工程大学 | Laser deposition additive manufacturing online monitoring system and method |
-
2023
- 2023-05-06 CN CN202310498199.4A patent/CN116559196B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20210089003A1 (en) * | 2017-12-20 | 2021-03-25 | Moog Inc. | Convolutional neural network evaluation of additive manufacturing images, and additive manufacturing system based thereon |
CN108333219A (en) * | 2018-03-19 | 2018-07-27 | 长沙理工大学 | A kind of online lossless detection method for band large-scale metal component laser gain material manufacturing process |
CN111831895A (en) * | 2019-04-19 | 2020-10-27 | 天津科技大学 | Network public opinion early warning method based on LSTM model |
CN113828798A (en) * | 2021-09-23 | 2021-12-24 | 西安交通大学 | Local heat treatment method and system for electric arc additive manufacturing structural part |
CN114782322A (en) * | 2022-03-28 | 2022-07-22 | 华中科技大学 | YOLOv5 model arc additive manufacturing molten pool defect detection method |
CN115861187A (en) * | 2022-11-15 | 2023-03-28 | 哈尔滨工程大学 | Laser deposition additive manufacturing online monitoring system and method |
Non-Patent Citations (3)
Title |
---|
TIANYUAN LIU ET AL.: "A hybrid CNN–LSTM algorithm for online defect recognition of CO2 welding", 《SENSORS》, vol. 18, pages 4369 - 1 * |
VIGNEASHWARA PANDIYAN ET AL.: "Deep learning-based monitoring of laser powder bed fusion process on variable time-scales using heterogeneous sensing and operando X-ray radiography guidance", 《ADDITIVE MANUFACTURING》, vol. 58, pages 103007 - 1 * |
董世运 等: "《激光增材再制造技术》", 西安电子科技大学出版社, pages: 275 - 276 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116893177A (en) * | 2023-09-11 | 2023-10-17 | 四川大学 | Method for monitoring internal defects in laser additive manufacturing process based on CCD and pyrometer |
CN116893177B (en) * | 2023-09-11 | 2023-11-14 | 四川大学 | Method for monitoring internal defects in laser additive manufacturing process based on CCD and pyrometer |
Also Published As
Publication number | Publication date |
---|---|
CN116559196B (en) | 2024-03-12 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Yang et al. | Pipeline magnetic flux leakage image detection algorithm based on multiscale SSD network | |
Khan et al. | Batteries state of health estimation via efficient neural networks with multiple channel charging profiles | |
Gamboa et al. | Validation of the rapid detection approach for enhancing the electronic nose systems performance, using different deep learning models and support vector machines | |
Wang et al. | Thermal pattern contrast diagnostic of microcracks with induction thermography for aircraft braking components | |
Sun et al. | Development of a physics-informed doubly fed cross-residual deep neural network for high-precision magnetic flux leakage defect size estimation | |
CN116559196B (en) | Arc additive manufacturing defect detection system and method | |
CN107703496B (en) | Interactive multimode Bernoulli filtering maneuvering weak target tracking-before-detection method | |
CN110516735B (en) | Natural gas pipeline event classification method based on LSTM network and Adam algorithm | |
CN112113755B (en) | Mechanical fault intelligent diagnosis method based on deep convolution-kurtosis neural network | |
CN111753877B (en) | Product quality detection method based on deep neural network migration learning | |
Li et al. | An improved AIC onset-time picking method based on regression convolutional neural network | |
Peng et al. | An aero-engine RUL prediction method based on VAE-GAN | |
CN112947080B (en) | Scene parameter transformation-based intelligent decision model performance evaluation system | |
CN117809164A (en) | Substation equipment fault detection method and system based on multi-mode fusion | |
Li et al. | PCB Defect Detection Based on Improved YOLOX Algorithm | |
CN113218998B (en) | Eddy current thermal imaging defect identification method based on global Moran index | |
Yu et al. | Thermal energy diagnosis of boiler plant by computer image processing and neural network technology | |
Watanabe et al. | Convolutional neural network (CNNs) based image diagnosis for failure analysis of power devices | |
CN113052255B (en) | Intelligent detection and positioning method for reactor | |
CN115130496A (en) | Pipeline pressure signal anomaly detection method based on Bagging and RM-LOF integrated single classifier | |
Huang et al. | Research on surface defect detection of glass wafer based on visual inspection | |
CN114578197A (en) | Transformer bushing partial discharge three-dimensional positioning method and system based on planar UHF sensor | |
Peng et al. | High-Precision Surface Crack Detection for Rolling Steel Production Equipment in ICPS | |
Wei et al. | Classification of wellhead device defects based on CNN-GA-BP model | |
CN110057742B (en) | On-line detection method for porosity of lithium ion battery pole piece and application thereof |
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 |