CN117725470A - Method for classifying and identifying defects of laser cladding cracks and pores based on acoustic emission signals - Google Patents
Method for classifying and identifying defects of laser cladding cracks and pores based on acoustic emission signals Download PDFInfo
- Publication number
- CN117725470A CN117725470A CN202410177713.9A CN202410177713A CN117725470A CN 117725470 A CN117725470 A CN 117725470A CN 202410177713 A CN202410177713 A CN 202410177713A CN 117725470 A CN117725470 A CN 117725470A
- Authority
- CN
- China
- Prior art keywords
- signal
- laser cladding
- crack
- acoustic emission
- sample
- 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
- 238000004372 laser cladding Methods 0.000 title claims abstract description 153
- 238000000034 method Methods 0.000 title claims abstract description 148
- 230000007547 defect Effects 0.000 title claims abstract description 142
- 239000011148 porous material Substances 0.000 title claims abstract description 33
- 230000008569 process Effects 0.000 claims abstract description 102
- 238000012549 training Methods 0.000 claims abstract description 28
- 238000002474 experimental method Methods 0.000 claims abstract description 27
- 238000012360 testing method Methods 0.000 claims abstract description 22
- 238000007781 pre-processing Methods 0.000 claims abstract description 18
- 238000001514 detection method Methods 0.000 claims abstract description 15
- 238000000354 decomposition reaction Methods 0.000 claims description 36
- 238000005253 cladding Methods 0.000 claims description 35
- 239000000843 powder Substances 0.000 claims description 23
- 230000006870 function Effects 0.000 claims description 17
- 238000013135 deep learning Methods 0.000 claims description 16
- 238000010586 diagram Methods 0.000 claims description 15
- 238000010606 normalization Methods 0.000 claims description 11
- 238000004519 manufacturing process Methods 0.000 claims description 9
- 238000012545 processing Methods 0.000 claims description 9
- 239000000758 substrate Substances 0.000 claims description 9
- 238000004364 calculation method Methods 0.000 claims description 7
- 230000003287 optical effect Effects 0.000 claims description 7
- 238000005520 cutting process Methods 0.000 claims description 6
- 238000005457 optimization Methods 0.000 claims description 6
- 238000005516 engineering process Methods 0.000 claims description 5
- 230000035515 penetration Effects 0.000 claims description 5
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 claims description 4
- 239000003795 chemical substances by application Substances 0.000 claims description 4
- 238000004140 cleaning Methods 0.000 claims description 4
- 239000012459 cleaning agent Substances 0.000 claims description 4
- 230000000149 penetrating effect Effects 0.000 claims description 4
- 238000005507 spraying Methods 0.000 claims description 4
- 230000009466 transformation Effects 0.000 claims description 4
- 229910000640 Fe alloy Inorganic materials 0.000 claims description 3
- 229910000831 Steel Inorganic materials 0.000 claims description 3
- 238000007667 floating Methods 0.000 claims description 3
- 238000002844 melting Methods 0.000 claims description 3
- 230000008018 melting Effects 0.000 claims description 3
- 238000005070 sampling Methods 0.000 claims description 3
- 239000010959 steel Substances 0.000 claims description 3
- 239000003344 environmental pollutant Substances 0.000 claims description 2
- 239000012216 imaging agent Substances 0.000 claims description 2
- 230000010354 integration Effects 0.000 claims description 2
- 231100000719 pollutant Toxicity 0.000 claims description 2
- 238000004321 preservation Methods 0.000 claims description 2
- 239000010410 layer Substances 0.000 description 83
- 239000002184 metal Substances 0.000 description 4
- 239000010953 base metal Substances 0.000 description 3
- 238000007431 microscopic evaluation Methods 0.000 description 3
- 239000002344 surface layer Substances 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 229910045601 alloy Inorganic materials 0.000 description 2
- 239000000956 alloy Substances 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 230000004927 fusion Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 208000037656 Respiratory Sounds Diseases 0.000 description 1
- 239000000654 additive Substances 0.000 description 1
- 230000000996 additive effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000011248 coating agent Substances 0.000 description 1
- 238000000576 coating method Methods 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000009659 non-destructive testing Methods 0.000 description 1
- 238000005498 polishing Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000007711 solidification Methods 0.000 description 1
- 230000008023 solidification Effects 0.000 description 1
- 230000035882 stress Effects 0.000 description 1
- 230000008646 thermal stress Effects 0.000 description 1
- 239000011800 void material Substances 0.000 description 1
Abstract
The invention relates to the field of laser cladding defect detection, and particularly discloses a method for classifying and identifying laser cladding crack and pore defects based on acoustic emission signals, wherein the acoustic emission signals of laser cladding processes under different process parameters are obtained through laser cladding orthogonal experiments; repeating experiments by using laser cladding process parameters which simultaneously generate cracks and air hole defects, and obtaining a data set through a two-dimensional threshold dividing method and data preprocessing; training and testing in a network model; and finally, after the actually collected signals are subjected to data preprocessing, sending the signals into a trained laser cladding defect identification network model based on a multilayer perceptron, and identifying the number of cracks and air hole defects generated in the section of laser cladding original acoustic emission signals and the duration time of each defect. The invention can rapidly and accurately identify whether cracks and air holes are generated in the laser cladding process, and judge the number of the cracks and the air hole defects generated in the laser cladding process and the duration time of each crack and each air hole.
Description
Technical Field
The invention relates to the technical field of laser cladding defect detection, in particular to a laser cladding crack and pore defect classification and identification method based on acoustic emission signals.
Background
The laser cladding technology is a special application form of metal additive manufacturing technology, and the principle is that alloy powder which is prepared according to a certain proportion is melted by a high-energy laser beam, meanwhile, the high-energy laser beam acts on the surface layer of the base metal to enable the irradiation area of the surface layer to be melted, finally, a coating taking the alloy powder material as a main body is formed on the surface of the base metal, and the melted powder metal and the base metal form stable metallurgical bonding. The technological process of the laser cladding technology is a process of coupling multiple physical fields, and various uncertain and unstable factors exist in the forming process to influence the final cladding result.
In the laser cladding process, the energy density of the high-energy laser beam is extremely high, the size of a molten pool is small, the laser cladding has the process of sudden temperature rise and drop, the solidification rate of the molten pool is extremely high, so that a large temperature difference exists between the molten pool and the surrounding environment, and the generated thermal stress is large; in addition, the metal matrix and the metal powder with special properties are used, so that the cladding layer is extremely easy to crack and air hole defects. Wherein, the metallurgical defect in the cladding layer can obviously reduce the fatigue strength of the cladding layer; macroscopic defects occurring outside the cladding layer can directly lead to scrapping of parts. Especially, the defects of micro cracks, air holes and the like in the cladding layer are often subjected to alternating stress for a long time when in subsequent assembly and use, so that the original micro defects on the surface layer and in the cladding layer are gradually expanded, and finally, the scrapping of parts and even accidents are caused. Therefore, in order to ensure the stability and uniformity of the laser cladding process, the occurrence of defects in the cladding process needs to be monitored.
In the laser cladding process, due to the differences of factors such as type, size, shape and position, each defect can generate an acoustic signal with unique characteristics, the acoustic signal can monitor the state of the laser cladding process and the defect, and meanwhile, the corresponding relation between the signal characteristics and the defect can be established by extracting the acoustic signal characteristics. The use of acoustic emissions to detect defect generation during laser cladding is therefore a convenient and efficient method. The deep learning is widely applied to the field of nondestructive testing as a rapid development method at present, and provides an important technical means for deep research on defect detection in the laser cladding process. Therefore, the state information of a molten pool in the laser cladding process is recorded through the acoustic emission signals, and the generation of cracks and air hole defects in the laser cladding process can be effectively classified and identified through the method of combining deep learning after the acoustic emission signals are processed by collecting the acoustic emission signals in the laser cladding process.
The method for acquiring, processing and analyzing the data set in the defect detection of the laser cladding process based on the deep learning is the core of the field, and aiming at the use of the supervised deep learning in the defect detection of the laser cladding process, the first problem is to acquire a large number of accurate and effective data sets for training of a deep learning network, and the generation of cracks and air holes in the laser cladding is a random process with unstable quantity and size. Thus, there is a need for a method that can accurately and efficiently extract a set of acoustic emission signals that contain normal, cracks, and pores that can be used for deep learning network training data sets.
Through acoustic emission detection equipment, gather abundant dynamic data in the laser cladding process, can reflect abundant characteristics such as molten pool state, cladding layer quality according to monitoring data, effective classification discernment is carried out to laser cladding crack gas pocket acoustic emission signal to the rethread degree of depth study, finally carries out real-time feedback adjustment to each technological parameter in the laser cladding process, improves cladding layer quality, reduces crackle, gas pocket defect, finally realizes high accuracy, high-quality laser cladding, has important realistic meaning.
Disclosure of Invention
In order to solve the problems in the prior art, the invention provides a classification and identification method for the defects of laser cladding cracks and air holes based on acoustic emission signals, and solves the problems in the background art.
In order to achieve the above purpose, the present invention provides the following technical solutions: a method for classifying and identifying defects of laser cladding cracks and pores based on acoustic emission signals comprises the following steps:
s1, selecting laser cladding powder and a substrate, designing laser cladding orthogonal experiments with different technological parameters, and performing acoustic emission detection in a laser cladding process to obtain acoustic emission signals;
s2, performing linear cutting on the cladding sample after the laser cladding orthogonal experiment, observing and analyzing internal defects of the sample by adopting an optical microscope, and comparing the number and positions of cracks and air hole defects which are actually generated with the corresponding relation between the acquired acoustic emission signals, and selecting normal cladding process parameters, crack generation process parameters, air hole generation process parameters and crack and air hole defect generation process parameters;
S3, repeating experiments by using laser cladding process parameters which simultaneously generate cracks and air hole defects, and collecting original acoustic emission signal data required by training of a subsequent deep learning identification method;
s4, manufacturing a data set, extracting a normal signal sample, a crack signal sample and a pore signal sample from the simultaneously generated crack and pore acoustic emission signals by using a two-dimensional threshold value dividing method, and manufacturing the data set 1;
s5, data preprocessing, namely performing 6-layer wavelet decomposition on three cladding signal samples in the manufactured data set 1, extracting detail components of the three signals, calculating the energy value of each detail component, and finally performing normalization processing to obtain detail energy characteristic signals of the three signals to manufacture a data set 2;
s6, establishing a defect recognition model based on deep learning, designing a laser cladding defect recognition network model based on a multilayer perceptron MLP, training and testing a network by using the data set 2, and finally obtaining the laser cladding defect recognition network model based on the multilayer perceptron MLP;
s7, after the collected laser cladding original acoustic emission signals are preprocessed by the data in the step S5, the preprocessed laser cladding original acoustic emission signals are sent into a trained laser cladding defect recognition network model based on a multilayer perceptron MLP, and a normal signal sample, a crack signal sample and an air hole signal sample are recognized through the network model;
And identifying the cracks, the number of the air hole defects and the duration of each defect generated in the acquired laser cladding original acoustic emission signal by using a laser cladding crack air hole defect duration and number identification algorithm.
Preferably, in step S1, the selected laser cladding powder and the substrate specifically include: the cladding powder is 30wt% WCp/Fe alloy powder, and the cladding substrate is 45 steel; the laser cladding orthogonal experiment for designing different technological parameters specifically comprises a laser power, a scanning speed and a powder feeding speed, and the parameter setting specifically comprises the following steps: setting the laser power range to 1100W-1400W, the scanning speed to 6mm/s-12mm/s, the powder feeding speed to 10g/min-12g/min, setting a fixed threshold value to 45dB, a floating threshold value to 6dB, a front amplifying to 20dB and a sampling frequency to 1MSPS by using a 16-channel Express8 acoustic emission system, and keeping the straight line distance from the center of a melting channel to an acoustic emission sensor to be more than 100mm in the acquisition process and keeping the contact temperature of the R15a acoustic emission sensor to be always less than 175 ℃.
Preferably, in step S3, repeated experiments are performed using laser cladding process parameters that generate cracks and air hole defects simultaneously, and the repeated experimental results observe the occurrence of cracks by using a penetration detection technology, where the specific steps of penetration detection include:
Repeated experiments are carried out by using laser cladding process parameters which simultaneously generate cracks and air hole defects, and firstly, cleaning the surface of a workpiece by using a cleaning agent, so that the surface of the workpiece is ensured to be free from pollutants and the permeation channel is kept clean and tidy; uniformly spraying the surface of the processed clean workpiece by using a penetrating agent, and standing for 5-15min; then cleaning the penetrating agent on the surface of the workpiece by using a cleaning agent again, uniformly shaking the imaging agent, and uniformly spraying at a position 150-300 mm away from the surface of the workpiece to be detected; after the developer is sprayed, the workpiece is kept stand for 10-20min, and then crack defects generated by the workpiece can be displayed.
Preferably, in step S4, a two-dimensional threshold value dividing method is used to extract a normal signal sample, a crack signal sample and a pore signal sample from the simultaneously generated crack and pore acoustic emission signals, where the two-dimensional threshold value dividing method specifically includes:
s4.1, using the obtained acoustic emission signals which simultaneously generate cracks and air holes, and using the acquired time domain signalsA(t) Representing every 1024 data points as one signal sample, the energy value of each sample is calculated using equation (1)E x Acquiring an acoustic emission signal energy diagram collected under the process parameters of simultaneously generating cracks and air holes, taking the maximum energy value of a normal acoustic emission signal part in the acoustic emission signal energy diagram of simultaneously generating the cracks and the air holes as a first threshold value, carrying out threshold value division on the collected acoustic emission signal, wherein the acoustic emission signal is a normal signal sample smaller than the threshold value and a defect signal sample larger than the threshold value;
；
Wherein,E x representing the energy value of each sample,A(t) Representing the collectionThe collected time domain signal, dt, represents integration over time;
s4.2, using the defect signal samples obtained in step S4.1, calculating the energy value of each sample using equation (1)E x The energy of the defect sample is obtained, and as the duration of each air hole is different from the duration of each crack, the duration of each air hole is less than 10ms in the single-pass laser cladding process through statistics, so that the duration of 10ms is used as a second threshold value, the second threshold value is used for threshold value division of the defect signal sample, the air hole signal sample is smaller than the second threshold value, and the crack signal sample is larger than the second threshold value;
s4.3, the normal signal sample, the crack signal sample and the air hole signal sample which are extracted through threshold division in the steps S4.1 and S4.2 are used for forming the data set 1.
Preferably, in step S5, the data preprocessing performs 6-layer wavelet decomposition on three kinds of cladding signal samples in the manufactured data set 1, extracts detailed components of three kinds of signals, calculates an energy value of each detailed component, and finally performs normalization processing to obtain detailed energy characteristics of the three kinds of signals, so as to manufacture the data set 2, which specifically includes the following steps:
S5.1, carrying out 6 layers of wavelet decomposition on each sample by using a normal signal sample, a crack signal sample and an air hole signal sample in the data set 1 manufactured in the step S4, wherein the selected wavelet basis function is Daubechies 8, and the number of wavelet decomposition layers is 6; in wavelet decomposition, first, a first layer of wavelet decomposition decomposes an original sample signal into an approximate component cA1 and a detail component cD1, a second layer of wavelet decomposition decomposes the approximate component cA1 obtained from the first layer into an approximate component cA2 and a detail component cD2 of the second layer, each layer of wavelet decomposition after that decomposes the approximate component of the previous layer, and 6 layers of wavelet decomposition are performed in total to obtain 6 groups of detail components, namely cD1, cD2, cD3, cD4, cD5 and cD6, respectively; respectively acquiring detail components after 6 layers of wavelet decomposition, and solving the energy value of each detail component by using a formula (1) to obtain detail energy signals eD1, eD2, eD3, eD4, eD5 and eD6 after 6 layers of wavelet decomposition of each sample;
s5.2, using the detail energy signals eD1, eD2, eD3, eD4, eD5 and eD6 processed in the step S5.1, using a formula (2) to calculate the sum of 6 detail energy signals, calculating the ratio of each detail energy signal in the sum, completing normalization processing of the detail energy signals, and finally obtaining detail energy characteristics fD1, fD2, fD3, fD4, fD5 and fD6 of each sample, and combining the energy characteristics of each sample into a data set 2;
；
Wherein,X new representing the ratio of each detailed energy signal in the sum after normalization calculation,representing summing up 6 detailed energy signals,/->Representing taking each of the detailed energy signals involved in the calculation in turn.
Preferably, in step S6, the laser cladding defect recognition network model based on the multi-layer perceptron MLP includes: an input layer with an input size of 6 dimensions, three hidden layers for respectively carrying out dimension transformation on input signals, a Dropout function for preventing the whole model from being overfitted, an output layer for outputting an identification result, and a cross entropy loss function for recording model loss;
the network model specifically operates as: firstly, using a signal sample with 6 dimensions of each sample characteristic obtained after data preprocessing as an input signal, increasing the dimension of sample data from 6 dimensions to 128 dimensions after passing through a first layer of hiding layer, changing the dimension of the signal from 128 dimensions to 128 dimensions after passing through a second hiding layer, reducing the dimension of the signal from 128 dimensions to 64 dimensions through a third hiding layer, then randomly discarding a part of data through a Dropout layer to prevent the overfitting of a model, finally reducing the dimension of the signal from 64 dimensions to 3 dimensions through an output layer, respectively representing the recognition results of three different cladding signals of normal, crack and air hole, finally recording the loss of the model by using a cross entropy loss function, performing back propagation intelligent optimization by using an Adam optimizer, and performing automatic optimization by using a formula (3):
；
Wherein,for learning rate parameters, ++>Is a weight coefficient>For loss function->Representing the partial derivative.
Preferably, in step S6, the data set 2 is used to train and test the network, which specifically includes: data set 2 was run at 4:1 is divided into a training set and a testing set, and is sent into a laser cladding defect identification network model based on a multilayer perceptron MLP, the network learning rate is set to 0.001, the batch size is set to 64, and the training round is 200 rounds.
Preferably, in step S7, the identifying, by the laser cladding crack and air hole defect duration and number identifying algorithm, the crack, the air hole number and the duration generated in the section of the laser cladding original acoustic emission signal specifically includes the following steps:
firstly, judging signal samples identified by a laser cladding defect identification network according to a network output sequence, and judging which of a normal signal sample, a crack signal sample and an air hole signal sample the current signal sample belongs to;
in the judging process, if a defect, namely a crack and an air hole signal sample is judged, recording is started on the defect sample, when the algorithm identifies the 1 st crack signal sample, a complete crack signal timer starts to record, if a non-crack signal sample is identified in the recording process, the algorithm continuously identifies 3 signal samples backwards, if any signal sample in the 3 signal samples which are identified subsequently appears as a crack signal sample, the 3 signal samples are all included in the complete crack signal, and then the complete crack signal timer continuously records backwards; if all 3 signal samples which are identified subsequently are non-crack signal samples, ending the recording by the complete crack signal timer, storing all the signal samples recorded at the present time as 1 complete crack signal, and storing the number of the recorded samples as the duration time of the complete crack signal;
After the preservation is finished, clearing the complete crack timer, and continuing to judge the algorithm backwards; the process of identifying the complete air hole signal is the same as that of identifying the complete crack signal, and finally, the process is finished until all acquired laser cladding original acoustic emission signal samples are identified;
and finally, the number of the stored complete crack timers and the number of the stored complete pore timers are the number of cracks and pore defects generated in the laser cladding process, and the number of samples in each complete crack and complete pore timer is the duration time of each crack and pore defect in the laser cladding process.
The beneficial effects of the invention are as follows: according to the method, acoustic emission signals of the laser cladding process under different technological parameters are obtained through designing laser cladding orthogonal experiments of different technological parameters; microscopic analysis is carried out through a linear cutting and optical microscope, the relation between the number and the position of cracks and air holes which are actually generated and the acquired acoustic emission signals is compared, and normal cladding process parameters, crack generation process parameters, air hole generation process parameters and crack and air hole defect generation process parameters are selected from the relations; repeated experiments are carried out by using laser cladding process parameters which simultaneously generate cracks and air hole defects, and original acoustic emission signal data required by training of a subsequent deep learning identification method are acquired; obtaining a data set 1 required by a deep learning training and testing model by using a two-dimensional threshold dividing method, and obtaining a processed data set 2 by using a method of extracting a fine energy characteristic signal by using 6 layers of wavelet decomposition; sending the manufactured data set into a built laser cladding defect recognition network model based on a multilayer perceptron MLP for training and testing to obtain a laser cladding crack and air hole defect recognition model based on acoustic emission signals; finally, after the actually collected laser cladding original acoustic emission signals are subjected to data preprocessing, sending the actually collected laser cladding original acoustic emission signals into a trained laser cladding defect identification network model based on a multilayer perceptron MLP, identifying a normal signal sample, a crack signal sample and an air hole signal sample through the network model, and identifying the number of cracks and air hole defects generated in the laser cladding original acoustic emission signals and the duration of each defect through a laser cladding crack air hole defect duration and number identification algorithm; finally, the method provided by the invention can be used for rapidly and accurately identifying whether cracks and air holes are generated in the laser cladding process and judging the number of the cracks and air hole defects and the duration time of each defect.
Drawings
FIG. 1 is a schematic overall flow chart of the method of the present invention;
FIG. 2 is a graph of orthogonal experimental parameters of laser cladding for different process parameters (laser power, scanning speed, powder feeding speed) in the example;
FIG. 3 is a schematic diagram of an experimental device for acquiring acoustic emission signals of a laser cladding process by using an acoustic emission system in an embodiment;
FIG. 4 is a waveform diagram of an acoustic emission signal of a crack and an air hole generated simultaneously in an embodiment;
FIG. 5 is a schematic view of an optical microscope for observing the concurrent generation of cracks and air hole melt channels in the example;
FIG. 6 is a waveform diagram and an energy diagram of a normal, crack, and blow hole acoustic emission signal in an embodiment;
FIG. 7 is a graph showing the results of repeated experiments using laser cladding process parameters that simultaneously produce cracks and pinhole defects in the examples;
FIG. 8 is a schematic diagram of PT test results of repeated experiments using laser cladding process parameters that simultaneously produce cracks and air hole defects in the examples;
FIG. 9 is a flow chart of a two-dimensional thresholding method in an embodiment;
FIG. 10 is a schematic diagram of the results of two-dimensional thresholding using the example for simultaneous crack and void generation;
FIG. 11 is a flow chart of data preprocessing in an embodiment;
FIG. 12 is a waveform diagram of acoustic emission signals of a normal sample, a crack sample and an air hole sample and a detailed energy characteristic signal schematic diagram in the embodiment;
FIG. 13 is a schematic diagram of a laser cladding defect recognition network model based on a multilayer perceptron MLP;
FIG. 14 is a schematic diagram of training results of an MLP network control group and a laser cladding defect recognition network model based on a multi-layer perceptron MLP in an embodiment;
FIG. 15 is a diagram showing the different cases of the laser cladding crack and air hole defect duration and quantity recognition algorithm in the embodiment;
FIG. 16 is a graph showing the identification result of the number of cracks and air holes of the original acoustic emission signal of the laser cladding in the embodiment.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
Examples
Referring to fig. 1 to 16, the present invention provides a technical solution: a method for classifying and identifying laser cladding crack and pore defects based on acoustic emission signals is shown in fig. 1, and specifically comprises the following steps:
1) Selecting laser cladding powder and a substrate: the cladding powder is 30wt% WCp/Fe alloy powder, and the cladding substrate is 45 steel substrate;
2) Designing laser cladding orthogonal experiments of different process parameters (laser power, scanning speed and powder feeding speed), and performing acoustic emission detection in the laser cladding process to obtain acoustic emission signals; setting a laser power range to 1100W-1400W, a scanning speed to 6mm/s-12mm/s, a powder feeding speed to 10g/min-12g/min, wherein specific orthogonal experimental parameters are shown in figure 2, setting a fixed threshold value to 45dB, a floating threshold value to 6dB, a front amplifying to 20dB and a sampling frequency to 1MSPS by using a 16-channel Express8 acoustic emission system of American physical acoustic company, acquiring acoustic emission signals of laser cladding processes of different process parameters (laser power, scanning speed and powder feeding speed) by using an R15a acoustic emission sensor, acquiring acoustic emission signal devices of the laser cladding processes by using an acoustic emission system, wherein the straight line distance from the center of a melting channel to the acoustic emission sensor is kept to be more than 100mm in the acquisition process, and the contact temperature of the R15a acoustic emission sensor is kept to be always less than 175 ℃ (the highest temperature allowed by equipment);
3) After laser cladding orthogonal experiments with different process parameters, linear cutting is carried out on a cladding sample, polishing is carried out on the sample after linear cutting, the processed sample is placed under an optical microscope to observe and analyze internal defects of the sample, it can be found that the sample actually generates a crack and two air holes, the generation time of the two air holes is almost the same, and the generation time and the quantity of the cracks and the air holes are mutually matched can be determined by comparing acoustic emission signals acquired by the sample with an acoustic emission signal acquired by the sample as shown in fig. 4 and an optical microscope observation as shown in fig. 5. Therefore, normal cladding process parameters, crack generation process parameters, pore generation process parameters and process parameters for generating cracks and pore defects simultaneously can be selected from different process parameter laser cladding orthogonal experiments based on the microscopic analysis; the acoustic emission signals of the normal, crack and air hole can be determined from the laser cladding orthogonal experiments with different technological parameters, and the acquired acoustic emission signals are used A(t) Representing, the energy value of each sample is calculated every 1024 data points as one sampleE x Using equation (1), energy values of the three acoustic emission signals are calculated, and a normal energy signal, a crack energy signal, and a pore energy signal can be obtained, as shown in fig. 6;
；
wherein,E x representing the energy value of each sample,A(t) Representing the acquired time domain signal, dt represents integrating over time.
4) Repeated experiments are carried out by using laser cladding process parameters which simultaneously generate cracks and air hole defects, the experimental results are shown in fig. 7, the crack generation condition in the cladding results is observed by using penetration detection (PT detection), the results are shown in fig. 8, and the original acoustic emission signal data required by subsequent deep learning training is acquired by using an acoustic emission system;
5) The method for preparing the data set comprises the steps of preparing a data set 1 by using a two-dimensional threshold dividing method, as shown in fig. 9, extracting a normal signal sample, a crack signal sample and a pore signal sample from the acoustic emission signals of cracks and pores generated at the same time of repeated experiment acquisition, wherein the dividing result is as shown in fig. 10, and specifically comprises the following steps:
5.1 Using the simultaneous crack and pore acoustic emission signals obtained in step 4), calculating the energy value for each sample using equation (1) E x The method comprises the steps of obtaining energy signals for generating cracks and air holes simultaneously, counting acoustic emission signals acquired by all repeated experiments, and finding that the maximum energy value of all normal cladding signals is smaller than 0.5, so that 0.5 is used as a first threshold value, the first threshold value is used for carrying out threshold value division on the energy signals for generating the cracks and the air holes simultaneously, the energy signals smaller than the energy threshold value are normal signal samples, and the energy signals larger than the energy threshold value are defect (including the cracks and the air holes) signal samples;
5.2 Using the defect signal samples obtained in step 5.1), calculating the energy value of each sample using formula (1)E x The energy signal of the defect sample is obtained, and as the duration of each air hole is different from that of each crack, the duration of each air hole is found to be less than 10ms in the single-channel laser cladding process through statistics, the duration of 10ms is used as a second threshold value, the second threshold value is used for threshold value division of the defect signal sample, the air hole signal sample is smaller than the second threshold value, and the crack signal sample is larger than the second threshold value;
5.3 Using the normal signal sample, the crack signal sample and the air hole signal sample extracted after the threshold division in the steps 5.1) and 5.2) for two times, wherein 800 samples are used for forming a data set 1 together, namely 2400 samples are used for each sample;
6) The data preprocessing, the data preprocessing flow is shown in fig. 11, the 6-layer wavelet decomposition is carried out on three signal samples in the manufactured data set 1, the detail components of the three signals are extracted, the energy value of each detail component is calculated, and finally the normalization processing is carried out to obtain the energy characteristic signals of the three signals, and the data set 2 is manufactured as shown in fig. 12;
6.1 Using the normal signal sample, the crack signal sample and the air hole signal sample in the data set 1 manufactured in the step 5), carrying out 6 layers of wavelet decomposition on each sample, wherein the selected wavelet basis function is Daubechies 8, the number of wavelet decomposition layers is 6, firstly, the first layer of wavelet decomposition is used for decomposing an original sample signal into an approximate component (low-frequency coefficient) cA1 and a detail component (high-frequency coefficient) cD1, the second layer of wavelet decomposition is used for decomposing the approximate component cA1 obtained in the first layer into an approximate component cA2 and a detail component cD2 of the second layer, and then each layer of wavelet decomposition is used for decomposing the approximate component of the upper layer, and the total 6 layers of wavelet decomposition are carried out to obtain 6 groups of detail components which are cD1, cD2, cD3, cD4, cD5 and cD6 respectively; respectively acquiring detail components after 6 layers of wavelet decomposition, and solving the energy value of each detail component by using a formula (1) to obtain detail energy signals eD1, eD2, eD3, eD4, eD5 and eD6 after 6 layers of wavelet decomposition of each sample;
6.2 Using the detail energy signals eD1, eD2, eD3, eD4, eD5 and eD6 after 6-layer wavelet decomposition obtained in the step 6.1), firstly obtaining the sum of 6 detail energy signals by using a formula (2), then calculating the ratio of each detail energy signal in the sum, completing normalization processing of the detail energy signals, finally obtaining detail energy characteristic signals fD1, fD2, fD3, fD4, fD5 and fD6 of each sample, and combining the energy characteristic signals of each sample to form a data set 2;
；
wherein,X new representing the ratio of each detailed energy signal in the sum after normalization calculation,representing summing up 6 detailed energy signals,/->Representing taking each of the detailed energy signals involved in the calculation in turn.
7) And establishing a defect recognition model based on deep learning, designing a laser cladding defect recognition network model based on a multilayer perceptron MLP, respectively training and testing a network by using a data set 1 and a data set 2, comparing the recognition accuracy of the network, and finally proving that the data set 2 after 6 layers of wavelet decomposition and energy characteristic signal extraction has the network model test accuracy of 97.78%.
A multi-layer perceptron network model control group is designed for training and testing the data set 1, and the network model comprises: the method comprises the steps of carrying out dimension transformation on input signals by 1 input Layer with 1024 dimensions and 3 hidden layers, reducing the input signals from 1024 dimensions to 512 dimensions by a first hidden Layer (Linear Layer 1), reducing the input signals from 512 dimensions to 512 dimensions by a second hidden Layer (Linear Layer 2), reducing the input signals from 512 dimensions to 128 dimensions by a third hidden Layer (Linear Layer 3), reducing the input signals from 128 dimensions to 3 dimensions by an output Layer (Linear Layer 4), inserting a Relu function in the middle of the first three hidden layers for activation treatment, setting a Dropout function behind the third hidden Layer for preventing overfitting of the whole model, outputting 3 nodes at the output Layer, representing the identification results of three different fusion signals of normal, crack and air hole which are finally output, recording the loss of the model by using a Cross entropy loss function (Cross-Entropy Loss Function), and carrying out reverse propagation intelligent optimization by using an Adam optimizer. The back propagation algorithm is automatically optimized using equation (3):
；
Wherein,for learning rate parameters, ++>Is a weight coefficient>For loss function->Representing the partial derivative, so that the weight coefficient of the connecting layer can be automatically optimized according to the deviation of each calculation result.
Designing a laser cladding defect identification network model based on a multilayer perceptron MLP, wherein the network structure is shown in fig. 13 and is used for training and testing a data set 2, and the network model comprises: the method comprises the steps of carrying out dimension transformation on input signals by 1 input Layer with the size of 6 dimensions and 3 hidden layers, respectively, increasing the signal from 6 dimensions to 128 dimensions by a first hidden Layer (Linear Layer 1), changing the signal from 128 dimensions to 128 dimensions by a second hidden Layer (Linear Layer 2), reducing the signal from 128 dimensions to 64 dimensions by a third hidden Layer (Linear Layer 3), reducing the signal from 64 dimensions to 3 dimensions by an output Layer (Linear Layer 4), inserting a Relu function in the middle of the first three hidden layers for activation treatment, setting a Dropout function in the third hidden Layer for preventing overfitting of the whole model, setting 3 nodes in the output Layer, respectively representing the identification results of three different fusion-coated signals of normal, crack and air hole which are finally recording the loss of the model by using a Cross entropy loss function (Cross-Entropy Loss Function), and carrying out reverse propagation intelligent optimization by using an Adam optimizer. The back propagation algorithm is automatically optimized to use the formula (3);
Using the data set 1 and the data set 2 obtained in step 5), 6), both data sets were compared according to 4:1 is divided into a training set and a testing set, a laser cladding defect identification network model based on a multi-layer perceptron MLP is designed, the network learning rate is set to be 0.001, the batch size is set to be 64, the training round is 200, the multi-layer perceptron network model comparison group uses the data set 1 for training and testing, and the highest accuracy of the multi-layer perceptron network model comparison group testing is 94.44%; the laser cladding defect identification network model based on the multilayer perceptron MLP is trained and tested by using the data set 2, and the highest accuracy of the model test is 97.78%; comparing the training results of the data set 1 obtained by data preprocessing and the data set 2 obtained by data preprocessing by using two network models, the model training and testing results are shown in fig. 14, and it can be proved that the recognition accuracy of the network can be effectively improved by data preprocessing (namely, extracting the detailed energy characteristic signals after 6 layers of wavelet decomposition), and the laser cladding defect recognition network model based on the multilayer perceptron MLP can effectively recognize the generation of cracks and air holes in the laser cladding process.
8) The number of generated cracks and air hole defects and the duration time of each crack and air hole defect are identified, and the specific flow is as follows:
8.1 The number of generated cracks and air hole defects and the duration time of each crack and air hole defect are identified, firstly, the collected laser cladding original acoustic emission signals are sent to a trained laser cladding defect identification network model based on a multi-layer perceptron MLP through the data preprocessing link of the step 6), and a normal signal sample, a crack signal sample and an air hole signal sample are identified through the network model;
8.2 Identifying the cracks, the number of the air holes and the duration time generated in the section of laser cladding original acoustic emission signal by a laser cladding crack air hole defect duration time and number identification algorithm; as shown in fig. 15, the algorithm recognition situation is that, first, a signal sample identified by the laser cladding defect recognition network is judged according to the network output sequence, which of a normal signal sample, a crack signal sample and an air hole signal sample the current signal sample belongs to is judged, and if the signal sample is judged to be a defect (crack or air hole) signal sample in the judging process, the recording of the defect sample is started, and the following description is given by taking the identification and recording of 1 complete crack signal as an example: when the algorithm identifies the 1 st crack signal sample, the complete crack signal timer starts to record, if the non-crack signal sample is identified in the recording process, the algorithm continuously identifies 3 signal samples backwards, if any signal sample in the subsequently identified 3 signal samples is also a crack signal sample, the 3 signal samples are all summarized into the complete crack signal, and then the complete crack signal timer continuously records backwards; if all 3 signal samples which are subsequently identified are non-crack signal samples, ending the recording by a complete crack signal timer, storing all the signal samples recorded at the time as 1 complete crack signal, storing the number of the recorded samples as the duration time of the complete crack signal (each signal sample represents 1ms time), resetting the complete crack timer after the storing is finished, and continuing to judge backwards by an algorithm; and the process of identifying the complete air hole signal is the same as that of identifying the complete crack signal, and finally, the process is finished until all acquired laser cladding original acoustic emission signal samples are identified. And finally, the number of the stored complete crack timers and the number of the stored complete pore timers are the number of cracks and pore defects generated in the laser cladding process, and the number of samples in each complete crack and complete pore timer is the duration time of each crack and pore defect in the laser cladding process. The number of the generated cracks and the generated air hole defects and the duration time of each crack and each air hole defect can be rapidly and accurately identified through the laser cladding crack air hole defect duration time and the number identification algorithm. Finally, the algorithm detects that 4 crack signals are generated in total in the laser cladding original acoustic emission signal used in the embodiment, the duration time is respectively 33ms, 34ms, 33ms and 16ms,2 air hole signals are respectively 3ms and 4ms, and the specific recognition result is shown in fig. 16; finally, the method provided by the invention can be used for rapidly and accurately identifying whether cracks and air holes are generated in the laser cladding process and judging the number of defects generated in the laser cladding process and the duration time of each defect.
In the embodiment, firstly, acoustic emission signals of laser cladding processes under different process parameters are obtained through laser cladding orthogonal experiments of the different process parameters; microscopic analysis is carried out through a linear cutting and optical microscope, the relation between the number and the position of cracks and air holes which are actually generated and the acquired acoustic emission signals is compared, and normal cladding process parameters, crack generation process parameters, air hole generation process parameters and crack and air hole defect generation process parameters are selected from the relations; repeated experiments are carried out by using laser cladding process parameters which simultaneously generate cracks and air hole defects, and original acoustic emission signal data required by training of a subsequent deep learning identification method are acquired; obtaining a data set 1 required by a deep learning training and testing model by using a two-dimensional threshold dividing method, and obtaining a processed data set 2 by performing data preprocessing by using a method of extracting detailed energy characteristics by using 6 layers of wavelet decomposition; then the processed data sets 1 and 2 are sent into a built control group multi-layer perceptron network model and a laser cladding defect recognition network model based on a multi-layer perceptron MLP for training and testing, a laser cladding crack gas hole recognition model based on acoustic emission is obtained, and finally, the test accuracy of the two models is respectively 94.44% and 97.78%, which means that the two designed network models can better recognize normal cladding signals, crack cladding signals and gas hole cladding signals generated in the laser cladding process; compared with the network error rate of a control group, the network error rate of the laser cladding defect identification network model based on the multilayer perceptron MLP is reduced by 60.07%, so that the wavelet energy feature extraction method is proved to be capable of effectively improving the network identification accuracy; preprocessing the data of the step 6) of the actually collected laser cladding original acoustic emission signals, sending the data into a trained laser cladding defect identification network model based on a multilayer perceptron MLP, identifying a normal signal sample, a crack signal sample and an air hole signal sample through the network model, and identifying the number of cracks and air hole defects generated in the laser cladding original acoustic emission signals and the duration of each defect through a laser cladding crack air hole defect duration and number identification algorithm; finally, the method provided by the invention can be used for rapidly and accurately identifying whether cracks and air holes are generated in the laser cladding process, and judging the number of the cracks and air hole defects generated in the laser cladding process and the duration time of each crack and air hole.
Although the present invention has been described with reference to the foregoing embodiments, it will be apparent to those skilled in the art that modifications may be made to the embodiments described, or equivalents may be substituted for elements thereof, and any modifications, equivalents, improvements and changes may be made without departing from the spirit and principles of the present invention.
Claims (8)
1. The method for classifying and identifying the defects of the laser cladding cracks and the air holes based on the acoustic emission signals is characterized by comprising the following steps:
s1, selecting laser cladding powder and a substrate, designing laser cladding orthogonal experiments with different technological parameters, and performing acoustic emission detection in a laser cladding process to obtain acoustic emission signals;
s2, performing linear cutting on the cladding sample after the laser cladding orthogonal experiment, observing and analyzing internal defects of the sample by adopting an optical microscope, and comparing the number and positions of cracks and air hole defects which are actually generated with the corresponding relation between the acquired acoustic emission signals, and selecting normal cladding process parameters, crack generation process parameters, air hole generation process parameters and crack and air hole defect generation process parameters;
S3, repeating experiments by using laser cladding process parameters which simultaneously generate cracks and air hole defects, and collecting original acoustic emission signal data required by training of a subsequent deep learning identification method;
s4, manufacturing a data set, extracting a normal signal sample, a crack signal sample and a pore signal sample from the simultaneously generated crack and pore acoustic emission signals by using a two-dimensional threshold value dividing method, and manufacturing the data set 1;
s5, data preprocessing, namely performing 6-layer wavelet decomposition on three cladding signal samples in the manufactured data set 1, extracting detail components of the three signals, calculating the energy value of each detail component, and finally performing normalization processing to obtain detail energy characteristic signals of the three signals to manufacture a data set 2;
s6, establishing a defect recognition model based on deep learning, designing a laser cladding defect recognition network model based on a multilayer perceptron MLP, training and testing a network by using the data set 2, and finally obtaining the laser cladding defect recognition network model based on the multilayer perceptron MLP;
s7, after the collected laser cladding original acoustic emission signals are preprocessed by the data in the step S5, the preprocessed laser cladding original acoustic emission signals are sent into a trained laser cladding defect recognition network model based on a multilayer perceptron MLP, and a normal signal sample, a crack signal sample and an air hole signal sample are recognized through the network model;
And identifying the cracks, the number of the air hole defects and the duration of each defect generated in the acquired laser cladding original acoustic emission signal by using a laser cladding crack air hole defect duration and number identification algorithm.
2. The method for classifying and identifying the defects of the laser cladding cracks and the air holes based on the acoustic emission signals according to claim 1 is characterized in that: in step S1, the selected laser cladding powder and the substrate specifically include: the cladding powder is 30wt% WCp/Fe alloy powder, and the cladding substrate is 45 steel; the laser cladding orthogonal experiment for designing different technological parameters specifically comprises a laser power, a scanning speed and a powder feeding speed, and the parameter setting specifically comprises the following steps: setting the laser power range to 1100W-1400W, the scanning speed to 6mm/s-12mm/s, the powder feeding speed to 10g/min-12g/min, setting a fixed threshold value to 45dB, a floating threshold value to 6dB, a front amplifying to 20dB and a sampling frequency to 1MSPS by using a 16-channel Express8 acoustic emission system, and keeping the straight line distance from the center of a melting channel to an acoustic emission sensor to be more than 100mm in the acquisition process and keeping the contact temperature of the R15a acoustic emission sensor to be always less than 175 ℃.
3. The method for classifying and identifying the defects of the laser cladding cracks and the air holes based on the acoustic emission signals according to claim 1 is characterized in that: in step S3, repeated experiments are performed by using laser cladding process parameters for generating cracks and air hole defects simultaneously, and the repeated experimental results observe the generation condition of the cracks by adopting a penetration detection technology, wherein the specific steps of penetration detection include:
Repeated experiments are carried out by using laser cladding process parameters which simultaneously generate cracks and air hole defects, and firstly, cleaning the surface of a workpiece by using a cleaning agent, so that the surface of the workpiece is ensured to be free from pollutants and the permeation channel is kept clean and tidy; uniformly spraying the surface of the processed clean workpiece by using a penetrating agent, and standing for 5-15min; then cleaning the penetrating agent on the surface of the workpiece by using a cleaning agent again, uniformly shaking the imaging agent, and uniformly spraying at a position 150-300 mm away from the surface of the workpiece to be detected; after the developer is sprayed, the workpiece is kept stand for 10-20min, and then crack defects generated by the workpiece can be displayed.
4. The method for classifying and identifying the defects of the laser cladding cracks and the air holes based on the acoustic emission signals according to claim 1 is characterized in that: in step S4, a two-dimensional threshold value dividing method is used to extract a normal signal sample, a crack signal sample and a pore signal sample from the simultaneously generated crack and pore acoustic emission signals, and the two-dimensional threshold value dividing method specifically includes:
s4.1, using the obtained acoustic emission signals which simultaneously generate cracks and air holes, and using the acquired time domain signalsA(t) Representing every 1024 data points as one signal sample, the energy value of each sample is calculated using equation (1) E x Acquiring an acoustic emission signal energy diagram collected under the process parameters of simultaneously generating cracks and air holes, taking the maximum energy value of a normal acoustic emission signal part in the acoustic emission signal energy diagram of simultaneously generating the cracks and the air holes as a first threshold value, carrying out threshold value division on the collected acoustic emission signal, wherein the acoustic emission signal is a normal signal sample smaller than the threshold value and a defect signal sample larger than the threshold value;
；
wherein,E x representing the energy value of each sample,A(t) Representing the acquired time domain signal, dt representing integration over time;
s4.2, using the defect signal samples obtained in step S4.1, calculating the energy value of each sample using equation (1)E x The energy of the defect sample is obtained, because each air hole and the duration of the crack are different, the duration of the air hole is smaller than 10ms in the single-pass laser cladding process, so the duration of 10ms is used as a second threshold value, the second threshold value is used for threshold value division of the defect signal sample, the air hole signal sample smaller than the second threshold value is a large air hole signal sampleAt a second threshold value, a crack signal sample;
s4.3, the normal signal sample, the crack signal sample and the air hole signal sample which are extracted through threshold division in the steps S4.1 and S4.2 are used for forming the data set 1.
5. The method for classifying and identifying the defects of the laser cladding cracks and the air holes based on the acoustic emission signals according to claim 1 is characterized in that: in step S5, the data preprocessing, performing 6-layer wavelet decomposition on three kinds of cladding signal samples in the manufactured data set 1, extracting detailed components of three kinds of signals, calculating an energy value of each detailed component, and finally performing normalization processing to obtain detailed energy characteristics of the three kinds of signals, so as to manufacture the data set 2, which comprises the following specific steps:
s5.1, carrying out 6 layers of wavelet decomposition on each sample by using a normal signal sample, a crack signal sample and an air hole signal sample in the data set 1 manufactured in the step S4, wherein the selected wavelet basis function is Daubechies 8, and the number of wavelet decomposition layers is 6; in wavelet decomposition, first, a first layer of wavelet decomposition decomposes an original sample signal into an approximate component cA1 and a detail component cD1, a second layer of wavelet decomposition decomposes the approximate component cA1 obtained from the first layer into an approximate component cA2 and a detail component cD2 of the second layer, each layer of wavelet decomposition after that decomposes the approximate component of the previous layer, and 6 layers of wavelet decomposition are performed in total to obtain 6 groups of detail components, namely cD1, cD2, cD3, cD4, cD5 and cD6, respectively; respectively acquiring detail components after 6 layers of wavelet decomposition, and solving the energy value of each detail component by using a formula (1) to obtain detail energy signals eD1, eD2, eD3, eD4, eD5 and eD6 after 6 layers of wavelet decomposition of each sample;
S5.2, using the detail energy signals eD1, eD2, eD3, eD4, eD5 and eD6 processed in the step S5.1, using a formula (2) to calculate the sum of 6 detail energy signals, calculating the ratio of each detail energy signal in the sum, completing normalization processing of the detail energy signals, and finally obtaining detail energy characteristics fD1, fD2, fD3, fD4, fD5 and fD6 of each sample, and combining the energy characteristics of each sample into a data set 2;
；
wherein,X new representing the ratio of each detailed energy signal in the sum after normalization calculation,representing summing up 6 detailed energy signals,/->Representing taking each of the detailed energy signals involved in the calculation in turn.
6. The method for classifying and identifying the defects of the laser cladding cracks and the air holes based on the acoustic emission signals according to claim 1 is characterized in that: in step S6, the laser cladding defect recognition network model based on the multi-layer perceptron MLP includes: an input layer with an input size of 6 dimensions, three hidden layers for respectively carrying out dimension transformation on input signals, a Dropout function for preventing the whole model from being overfitted, an output layer for outputting an identification result, and a cross entropy loss function for recording model loss;
The network model specifically operates as: firstly, using a signal sample with 6 dimensions of each sample characteristic obtained after data preprocessing as an input signal, increasing the dimension of sample data from 6 dimensions to 128 dimensions after passing through a first layer of hiding layer, changing the dimension of the signal from 128 dimensions to 128 dimensions after passing through a second hiding layer, reducing the dimension of the signal from 128 dimensions to 64 dimensions through a third hiding layer, then randomly discarding a part of data through a Dropout layer to prevent the overfitting of a model, finally reducing the dimension of the signal from 64 dimensions to 3 dimensions through an output layer, respectively representing the recognition results of three different cladding signals of normal, crack and air hole, finally recording the loss of the model by using a cross entropy loss function, performing back propagation intelligent optimization by using an Adam optimizer, and performing automatic optimization by using a formula (3):
；
wherein,for learning rate parameters, ++>Is a weight coefficient>For loss function->Representing the partial derivative.
7. The method for classifying and identifying the defects of the laser cladding cracks and the air holes based on the acoustic emission signals according to claim 1 is characterized in that: in step S6, training and testing the network using the data set 2 specifically includes: data set 2 was run at 4:1 is divided into a training set and a testing set, and is sent into a laser cladding defect identification network model based on a multilayer perceptron MLP, the network learning rate is set to 0.001, the batch size is set to 64, and the training round is 200 rounds.
8. The method for classifying and identifying the defects of the laser cladding cracks and the air holes based on the acoustic emission signals according to claim 1 is characterized in that: in step S7, the identification algorithm of the duration and the number of the air hole defects of the laser cladding crack identifies the crack, the number of air holes and the duration generated in the original acoustic emission signal of the laser cladding, and specifically includes the following steps:
firstly, judging signal samples identified by a laser cladding defect identification network according to a network output sequence, and judging which of a normal signal sample, a crack signal sample and an air hole signal sample the current signal sample belongs to;
in the judging process, if a defect, namely a crack and an air hole signal sample is judged, recording is started on the defect sample, when the algorithm identifies the 1 st crack signal sample, a complete crack signal timer starts to record, if a non-crack signal sample is identified in the recording process, the algorithm continuously identifies 3 signal samples backwards, if any signal sample in the 3 signal samples which are identified subsequently appears as a crack signal sample, the 3 signal samples are all included in the complete crack signal, and then the complete crack signal timer continuously records backwards; if all 3 signal samples which are identified subsequently are non-crack signal samples, ending the recording by the complete crack signal timer, storing all the signal samples recorded at the present time as 1 complete crack signal, and storing the number of the recorded samples as the duration time of the complete crack signal;
After the preservation is finished, clearing the complete crack timer, and continuing to judge the algorithm backwards; the process of identifying the complete air hole signal is the same as that of identifying the complete crack signal, and finally, the process is finished until all acquired laser cladding original acoustic emission signal samples are identified;
and finally, the number of the stored complete crack timers and the number of the stored complete pore timers are the number of cracks and pore defects generated in the laser cladding process, and the number of samples in each complete crack and complete pore timer is the duration time of each crack and pore defect in the laser cladding process.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410177713.9A CN117725470B (en) | 2024-02-08 | Method for classifying and identifying defects of laser cladding cracks and pores based on acoustic emission signals |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410177713.9A CN117725470B (en) | 2024-02-08 | Method for classifying and identifying defects of laser cladding cracks and pores based on acoustic emission signals |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117725470A true CN117725470A (en) | 2024-03-19 |
| CN117725470B CN117725470B (en) | 2024-04-26 |
Family
ID=
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2012000648A1 (en) * | 2010-06-28 | 2012-01-05 | Precitec Kg | Method for closed-loop controlling a laser processing operation and laser material processing head using the same |
| WO2014180870A1 (en) * | 2013-05-06 | 2014-11-13 | Vrije Universiteit Brussel | Effective structural health monitoring |
| WO2019207562A1 (en) * | 2018-04-26 | 2019-10-31 | Scanmaster (Irt) Ltd | Improved characterization and classification of spot welds by ultrasonic diagostic techniques |
| US20200166909A1 (en) * | 2018-11-20 | 2020-05-28 | Relativity Space, Inc. | Real-time adaptive control of manufacturing processes using machine learning |
| WO2023111542A1 (en) * | 2021-12-13 | 2023-06-22 | The University Of Liverpool | Defect identification in additive manufacturing based on time series in-process parameter data |
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2012000648A1 (en) * | 2010-06-28 | 2012-01-05 | Precitec Kg | Method for closed-loop controlling a laser processing operation and laser material processing head using the same |
| WO2014180870A1 (en) * | 2013-05-06 | 2014-11-13 | Vrije Universiteit Brussel | Effective structural health monitoring |
| WO2019207562A1 (en) * | 2018-04-26 | 2019-10-31 | Scanmaster (Irt) Ltd | Improved characterization and classification of spot welds by ultrasonic diagostic techniques |
| US20200166909A1 (en) * | 2018-11-20 | 2020-05-28 | Relativity Space, Inc. | Real-time adaptive control of manufacturing processes using machine learning |
| WO2023111542A1 (en) * | 2021-12-13 | 2023-06-22 | The University Of Liverpool | Defect identification in additive manufacturing based on time series in-process parameter data |
Non-Patent Citations (5)
| Title |
|---|
| HAYTHEM GAJA等: "Defect classification of laser metal deposition using logistic regression and artificial neural networks for pattern recognition", 《THE INTERNATIONAL JOURNAL OF ADVANCED MANUFACTURING TECHNOLOGY》, vol. 94, 10 August 2017 (2017-08-10), pages 315 - 326, XP036395097, DOI: 10.1007/s00170-017-0878-9 * |
| MINGKE LI等: "Crack Formation Mechanisms and Control Methods of Laser Cladding Coatings: A Review", 《COATINGS》, vol. 13, no. 6, 17 June 2023 (2023-06-17), pages 1 - 26 * |
| 史墨可等: "基于GA-BP神经网络的宽带激光熔覆裂纹缺陷预测", 《热加工工艺》, 16 January 2024 (2024-01-16), pages 1 - 6 * |
| 师陆冰等: "齿轮钢表面激光熔覆Ni基涂层工艺参数优化", 《金属热处理》, vol. 48, no. 11, 25 November 2023 (2023-11-25), pages 266 - 275 * |
| 李凯强: "镍基合金激光熔覆过程裂纹检测及诊断方法", 《中国优秀硕士学位论文全文数据库 (中国优秀硕士学位论文全文数据库 (工程科技辑))》, no. 1, 15 January 2022 (2022-01-15), pages 022 - 769 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Gaja et al. | Defects monitoring of laser metal deposition using acoustic emission sensor | |
| Gaja et al. | Defect classification of laser metal deposition using logistic regression and artificial neural networks for pattern recognition | |
| CN109360204B (en) | Inner defect detection method of multilayer metal lattice structure material based on Faster R-CNN | |
| Huang et al. | A neural network and multiple regression method for the characterization of the depth of weld penetration in laser welding based on acoustic signatures | |
| Ito et al. | Detection and location of microdefects during selective laser melting by wireless acoustic emission measurement | |
| Rieder et al. | On-and offline ultrasonic characterization of components built by SLM additive manufacturing | |
| CN103630543A (en) | Judgment method for detecting defects of wave absorption coating by using pulse infrared thermal waves | |
| CN106841403A (en) | A kind of acoustics glass defect detection method based on neutral net | |
| CN109940305A (en) | Laser Deep Penetration Welding intelligent monitor system and method based on plasma electric signal | |
| CN112157368B (en) | Laser non-penetration welding seam penetration nondestructive testing method | |
| CN117725470B (en) | Method for classifying and identifying defects of laser cladding cracks and pores based on acoustic emission signals | |
| CN106198749A (en) | A kind of data fusion method of multiple sensor based on Metal Crack monitoring | |
| Behrens et al. | Acoustic emission technique for online monitoring during cold forging of steel components: a promising approach for online crack detection in metal forming processes | |
| Taheri | Nondestructive evaluation and in-situ monitoring for metal additive manufacturing | |
| CN117725470A (en) | Method for classifying and identifying defects of laser cladding cracks and pores based on acoustic emission signals | |
| Selvarajan et al. | Fuzzy logic optimization with regression analysis on EDM machining parameters of Si 3N 4-TiN ceramic composites | |
| CN105312770B (en) | Laser welding mode determination method based on plasma electric signals | |
| CN100525727C (en) | Method for measuring and identifying bat kind using with bat echo location acoustic | |
| CN116165280A (en) | Method and system for monitoring internal defects of laser cladding process in real time | |
| CN109444265B (en) | Laser ultrasonic vibration detection device and method | |
| Kim et al. | Multi-branch deep fusion network-based automatic detection of weld defects using non-destructive ultrasonic test | |
| CN115266951A (en) | Method and system for monitoring internal defects in selective laser melting process in real time on line | |
| Allocca et al. | Chaos Theory Approach as Advanced Technique for GDI Spray Analysis | |
| Meng et al. | Spectral analysis for crack detection during TIG welding using acoustic emission techniques | |
| Gaja et al. | Defects classification of laser metal deposition using acoustic emission sensor |
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 |