In-situ quality evaluation system and method for laser powder bed fusion processed components

Abstract

The present invention discloses an in-situ quality evaluation system and method for laser powder bed fusion processing components. The in-situ quality evaluation method for laser powder bed fusion processing components includes the following steps: obtaining the molten pool radiation information corresponding to the target processing layer and performing filtering and noise reduction processing; extracting the time / frequency domain characteristics of the molten pool radiation information corresponding to the target processing layer one by one; based on the extracted time / frequency domain characteristics, using different unsupervised learning anomaly detection algorithms to analyze and solve whether an anomaly occurs in the melt channel; merging the solution results of different unsupervised learning anomaly detection algorithms, calculating the performance index and using the performance index to quantitatively evaluate the processing quality. The present invention extracts the time / frequency domain characteristics of the molten pool radiation information, uses different unsupervised learning anomaly detection algorithms to analyze and solve whether an anomaly occurs in the melt channel to quantitatively evaluate the processing quality, and solves the problem that the laser powder bed fusion quality analysis relies heavily on the quality of the data set and prior knowledge.

Description

Technical Field

[0001] The present invention relates to the field of additive manufacturing technology, and in particular to an in-situ quality evaluation system and method for a laser powder bed melting processed component. Background Art

[0002] Laser powder bed fusion (LPBF) is an additive manufacturing technology that uses a laser beam to melt and solidify metal powder layer by layer at extremely high speeds to form the desired parts. It can be used in a wide range of materials, integrates design and manufacturing, and has obvious advantages in the refined manufacturing of complex structures. It has been widely used in the manufacturing process of complex components of various equipment.

[0003] During the LPBF forming process, the laser interacts with the metal powder, causing dramatic changes in the melt pool temperature and phase. The forming process is highly complex, and random defects are easily generated, which in turn affect the microstructure and overall performance of the part. In recent years, powder bed fusion process monitoring and quality analysis methods have received increasing attention. LPBF process monitoring technology, which uses information from the forming process and combines it with intelligent prediction methods to predict forming defects, allows for rapid preliminary quality assessments during the forming process, enabling timely process adjustments and ensuring quality assurance.

[0004] However, existing laser powder bed melting quality analysis methods, when predicting internal defects and forming properties of parts based on melt pool radiation data, all use supervised learning methods similar to convolutional neural networks. These methods first collect process data, manually label this data to establish a dataset, and then train the model. The effectiveness of the model's predictions depends primarily on the quality of the dataset, and any degradation of the dataset significantly impacts the effectiveness of the predictions. Furthermore, the amount of melt pool radiation light signal data for a complete project is enormous, and the factors influencing the melt pool radiation light signal are complex. Abnormal data in the signal requires prior knowledge to be identified, thus relying heavily on prior knowledge.

[0005] In summary, existing laser powder bed melting quality analysis methods have the problems of high dependence on the quality of the data set and high dependence on prior knowledge. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide an in-situ quality evaluation system and method for components processed by laser powder bed melting.

[0007] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0008] A system for evaluating in-situ quality of components processed by laser powder bed fusion, comprising: a melt pool radiation acquisition device for acquiring melt pool radiation information of each processing layer during a printing process; a computer for acquiring the melt pool radiation information of each processing layer acquired by the melt pool radiation acquisition device, acquiring the melt pool radiation information corresponding to a target processing layer from the melt pool radiation information of each processing layer and performing filtering and noise reduction processing, extracting the time / frequency domain features of the melt pool radiation information corresponding to the target processing layer melt channel by melt channel, analyzing and solving whether an anomaly occurs in the melt channel using different unsupervised learning anomaly detection algorithms, merging the solution results of different unsupervised learning anomaly detection algorithms, calculating a performance index, and using the performance index to quantitatively evaluate the processing quality.

[0009] A method for in-situ quality evaluation of components processed by laser powder bed fusion is disclosed, comprising the following steps: obtaining molten pool radiation information corresponding to a target processing layer and performing filtering and noise reduction processing; extracting time / frequency domain features of the molten pool radiation information corresponding to the target processing layer for each melt path; analyzing and solving whether an anomaly occurs in the melt path based on the extracted time / frequency domain features using different unsupervised learning anomaly detection algorithms; and merging the solution results of different unsupervised learning anomaly detection algorithms, calculating a performance index, and using the performance index to quantitatively evaluate the processing quality.

[0010] The beneficial technical effect of the present invention is that the above-mentioned in-situ quality evaluation system and method for laser powder bed fusion processing components extracts the time / frequency domain characteristics of the molten pool radiation information, and based on the extracted time / frequency domain characteristics, uses different unsupervised learning anomaly detection algorithms to analyze and solve whether the melt path has abnormalities to quantitatively evaluate the processing quality, thereby solving the problem that the quality analysis of laser powder bed fusion is highly dependent on the quality of the data set and on prior knowledge. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 Schematic diagram of the structure of the in-situ quality evaluation system for laser powder bed fusion processing components of the present invention;

[0012] Figure 2 Schematic diagram of the process of the in-situ quality evaluation method of the laser powder bed fusion processing component of the present invention.

[0013] Description of reference numerals:

[0014] 1-forming chamber, 2-powder, 3-powder spreading mechanism, 4-feeding mechanism, 5-forming platform, 6-forming part, 7-laser, 8-laser beam, 9-dichroic mirror, 10-laser deflection and focusing system, 11-coaxial optical path molten pool radiation, 12-spatial molten pool radiation, 13-spatial molten pool radiation acquisition unit, 14-signal acquisition card, 15-computer, 100-coaxial optical path molten pool radiation acquisition unit, 201-first semi-transparent and semi-reflective mirror, 202-first focusing filter Mirror group, 203-visible light radiation sensor, 301-second semi-transparent and semi-reflective mirror, 302-second focusing filter group, 303-near-infrared radiation sensor, 401-third semi-transparent and semi-reflective mirror, 402-third focusing filter group, 403-colorimetric temperature sensor, 501-fourth semi-transparent and semi-reflective mirror, 502-fourth focusing filter group, 503-laser reflection light sensor, 601-reflecting mirror, 6021-fifth focusing filter group, 603-infrared radiation sensor. DETAILED DESCRIPTION

[0015] In order to enable those skilled in the art to more clearly understand the objectives, technical solutions and advantages of the present invention, the present invention is further described below with reference to the accompanying drawings and embodiments.

[0016] The invention provides an in-situ quality evaluation system for a laser powder bed melting processed component.

[0017] like Figure 1 As shown, in one embodiment of the present invention, the in-situ quality evaluation system for laser powder bed fusion processing components includes a molten pool radiation acquisition device and a computer 15, and the molten pool radiation acquisition device is communicatively connected to the computer 15 via a signal acquisition card 14; the molten pool radiation acquisition device is used to collect the molten pool radiation information of each processing layer during the part processing process; the computer 15 is used to obtain the molten pool radiation information of each processing layer collected by the molten pool radiation acquisition device, obtain the molten pool radiation information corresponding to the target processing layer from the molten pool radiation information of each processing layer and perform filtering and noise reduction processing, extract the time / frequency domain characteristics of the molten pool radiation information corresponding to the target processing layer for each molten channel, use different unsupervised learning anomaly detection algorithms to analyze and solve whether an anomaly occurs in the molten channel, merge the solution results of different unsupervised learning anomaly detection algorithms, calculate the performance index and use the performance index to quantitatively evaluate the processing quality.

[0018] In this embodiment, the molten pool radiation collection device includes a spatial molten pool radiation collection unit 13 and a coaxial optical path molten pool radiation collection unit 100; the spatial molten pool radiation collection unit 13 is disposed at the upper top of the forming chamber 1 and is used to collect spatial molten pool radiation information of one or more wavelength bands for each processing layer; the coaxial optical path molten pool radiation collection unit 100 is disposed in the laser processing optical path and is used to collect coaxial optical path molten pool radiation information of one or more wavelength bands for each processing layer. In other embodiments, the molten pool radiation collection device may also include only the spatial molten pool radiation collection unit 13, or only the coaxial optical path molten pool radiation collection unit 100.

[0019] In this embodiment, the coaxial optical path melt pool radiation collection unit 100 includes a visible light radiation monitoring subunit, a near-infrared radiation monitoring subunit, a colorimetric temperature measurement subunit, a laser reflection monitoring subunit, and an infrared radiation monitoring subunit. In other embodiments, the coaxial optical path melt pool radiation collection unit 100 may also include any one or any combination of the visible light radiation monitoring subunit, the near-infrared radiation monitoring subunit, the colorimetric temperature measurement subunit, the laser reflection monitoring subunit, and the infrared radiation monitoring subunit.

[0020] The visible light radiation monitoring subunit includes a first semi-transparent and semi-reflective mirror 201, a first focusing filter group 202 and a visible light radiation sensor 203, wherein the first semi-transparent and semi-reflective mirror 201 can reflect radiation in the 400nm-750nm band and can transmit radiation in other bands, and the first focusing filter group 202 is used to focus and combine the radiation in the 400nm-750nm band and transmit it to the visible light radiation sensor 203. The visible light radiation sensor 203 is used to collect radiation in the 400nm-750nm band for in-situ monitoring of the status information of the metal plume and plasma cloud generated during the laser powder bed melting process.

[0021] The near-infrared radiation monitoring subunit includes a second semi-transparent and semi-reflective mirror 301, a second focusing filter group 302 and a near-infrared radiation sensor 303, wherein the second semi-transparent and semi-reflective mirror 301 can reflect radiation in the 750nm-900nm band and can transmit radiation in other bands, the second focusing filter group 302 is used to focus and combine the radiation in the 750nm-900nm band and transmit it to the near-infrared radiation sensor 303, and the near-infrared radiation sensor 303 is used to collect radiation in the 750nm-900nm band for in-situ monitoring of the status information of the plasma cloud and molten pool thermal radiation generated in the laser powder bed melting process, and can also be used to reflect the depth of penetration.

[0022] The colorimetric temperature measurement subunit includes a third semi-transparent and semi-reflective mirror 401, a third focusing filter group 402 and a colorimetric temperature sensor 403, wherein the third semi-transparent and semi-reflective mirror 401 can reflect radiation in the 900nm-950nm band and can transmit radiation in other bands, the third focusing filter group 402 is used to focus and combine the radiation in the 900nm-950nm band and transmit it to the colorimetric temperature sensor, and the colorimetric temperature sensor 403 is used to collect radiation in the 900nm-950nm band for in-situ measurement of the temperature change of the molten pool during laser powder bed melting processing.

[0023] The laser reflection monitoring subunit includes a fourth semi-transparent and semi-reflective mirror 501, a fourth focusing filter group 502 and a laser reflection light sensor 503. The fourth semi-transparent and semi-reflective mirror 501 can reflect radiation in the 1030nm-1080nm band and can transmit radiation in other bands. The fourth focusing filter group 502 is used to focus and combine the radiation in the 1030nm-1080nm band and transmit it to the laser reflection light sensor 503. The laser reflection light sensor 503 is used to collect radiation in the 1030nm-1080nm band to monitor the reflected laser formed by the laser that is not absorbed when the laser interacts with the powder, thereby reflecting the laser defocus state and the roughness and thickness changes of the processed surface during the laser powder bed melting process.

[0024] The infrared radiation monitoring subunit includes a reflector 601, a fifth focusing filter group 602 and an infrared radiation sensor 603, wherein the reflector 601 can reflect radiation in a band greater than or equal to 1200nm, the fifth focusing filter group 602 is used to focus and combine radiation in a band greater than or equal to 1200nm and transmit it to the infrared radiation sensor 603, and the infrared radiation sensor 603 is used to collect radiation in a band greater than or equal to 1200nm for monitoring the surface and surrounding thermal information of the molten pool, plume, spatter, etc. during the laser powder bed melting process.

[0025] The spatial molten pool radiation collection unit 13 includes any one or any combination of a visible light radiation sensor, a near-infrared radiation sensor, a colorimetric temperature sensor, a laser reflection light sensor, and an infrared radiation sensor. In specific implementation, if it is necessary to monitor the spatial molten pool radiation of a certain band, a corresponding sensor can be selected from the visible light radiation sensor, the near-infrared radiation sensor, the colorimetric temperature sensor, the laser reflection light sensor, and the infrared radiation sensor and installed on the upper top of the forming chamber 1; if it is necessary to monitor the spatial molten pool radiation of several bands, several corresponding sensors can be selected from the visible light radiation sensor, the near-infrared radiation sensor, the colorimetric temperature sensor, the laser reflection light sensor, and the infrared radiation sensor and installed in a distributed manner on the upper top of the forming chamber 1.

[0026] based on Figure 1 The laser powder bed fusion processing component in-situ quality evaluation system shown in the figure provides a laser powder bed fusion processing component in-situ quality evaluation method, and the laser powder bed fusion processing component in-situ quality evaluation method is applied to a computer.

[0027] like Figure 2 As shown, in one embodiment of the present invention, the in-situ quality evaluation method for a laser powder bed fusion processed component includes steps S10 to S40:

[0028] S10: Obtain the molten pool radiation information corresponding to the target processing layer and perform filtering and noise reduction processing.

[0029] In specific implementations, a laser powder bed fusion system equipped with a melt pool radiation collection device is used to print parts, collecting melt pool radiation information during the printing process. During the printing process, the build platform 5 within the build chamber 1 descends by one layer thickness, the feeding mechanism 4 provides a layer of powder 2, and the powder spreading mechanism 3 then moves to provide a layer of powder to the build platform. Laser 7 then emits a laser beam 8, which passes through a dichroic mirror 9 and a laser deflection and focusing system 10 to the focal plane of the build platform 5, i.e., the surface of the built part 6. The laser interacts with the powder to form a melt pool, generating melt pool radiation.

[0030] The molten pool radiation collection device can collect molten pool radiation information through two ways: one is that the spatial molten pool radiation 12 generated by the molten pool is freely transmitted to the spatial molten pool radiation collection unit 13, and is collected by the sensor inside the spatial molten pool radiation collection unit 13; the other is that the coaxial optical path molten pool radiation 11 generated by the molten pool is continuously reflected and transmitted to the various subunits of the coaxial optical path molten pool radiation collection unit 100 through the laser deflection and focusing system 10 and the dichroic mirror 9, and is collected by the various subunits of the coaxial optical path molten pool radiation collection unit 100.

[0031] It should be noted that during the printing process, the wavelength of the laser beam 8 emitted by the laser 7 is 1064 nm. The dichroic mirror 9 can transmit the 1064 nm laser beam wavelength and fully reflect the remaining wavelengths. When collecting the molten pool radiation information, the spatial molten pool radiation collection unit 13 can be used alone to collect spatial molten pool radiation information of one or more wavelengths for each processing layer, or the coaxial optical path molten pool radiation collection unit 100 can be used alone to collect coaxial optical path molten pool radiation information of one or more wavelengths for each processing layer, or both the spatial molten pool radiation collection unit 13 and the coaxial optical path molten pool radiation collection unit 100 can be used simultaneously to collect spatial molten pool radiation information of one or more wavelengths and coaxial optical path molten pool radiation information of one or more wavelengths.

[0032] The molten pool radiation information collected by the molten pool radiation collection device is transmitted to the signal acquisition card 14. The computer 15 is communicated with the signal acquisition card 14 and can obtain the molten pool radiation information of each processing layer collected by the molten pool radiation collection device for subsequent in-situ calculation and analysis.

[0033] Laser powder bed fusion printing usually processes thousands of layers of a part. In order to quickly evaluate, the target processing layer can be selected and the molten pool radiation information corresponding to the target processing layer can be used as the raw data for in-situ calculation and analysis. The target processing layer can be selected by the following strategy: randomly select any n processing layers as the target processing layer; or use the processing risk layer determined by the evaluation as the target processing layer, wherein the processing risk layer is obtained through technical evaluation (including personnel experience evaluation or computational simulation evaluation). Of course, in some embodiments that do not have high requirements for evaluation timeliness, all processing layers of the part processing can also be used as the target processing layer.

[0034] After selecting the target processing layer, the melt pool radiation information corresponding to the target processing layer is obtained from the collected melt pool radiation information of each processing layer and filtered and denoised. The filtering and denoising methods used in the filtering and denoising process include, but are not limited to, time domain filtering and denoising methods, frequency domain filtering and denoising methods, time-frequency domain integrated filtering and denoising methods, and machine learning denoising methods.

[0035] S20. Extracting the time / frequency domain characteristics of the molten pool radiation information corresponding to the target processing layer, melt channel by melt channel.

[0036] First, the molten pool radiation information corresponding to each target processing layer after filtering and noise reduction is divided and numbered through the feature process identifier, laser switch and position vector to form an index relationship of "number-feature-melt channel-intensity signal".

[0037] Then, for the molten pool radiation information corresponding to each target processing layer after segmentation and numbering, the time / frequency domain features corresponding to each molten channel are extracted one by one. In an embodiment of the present invention, the time / frequency domain features extracted in this step include 16 time domain features p1-p16 and 12 frequency domain features p17-p28, as shown in Table 1. Among them, the 16 time domain features include 10 dimensional features and 6 dimensionless features. The 10 dimensional features are mean, maximum, minimum, peak-to-peak value, median, variance, standard deviation, mean absolute deviation, root mean square and root amplitude, which mainly reflect the changes in the signal amplitude and energy of the molten pool radiation; the 6 dimensionless features are skewness, kurtosis, peak factor, waveform factor, pulse factor and margin factor, which mainly reflect the distribution of the molten pool radiation signal in the time domain. The frequency domain feature p17 reflects the magnitude of the molten pool radiation energy in the frequency domain. The frequency domain features p18-p20, p22, and p26-p28 characterize the degree of dispersion or concentration of the spectrum. The frequency domain features p21 and p23-p25 reflect the changes in the main frequency band position.

[0038]

[0039]

[0040] Table 1: Definition of time domain features and frequency domain features extracted by the present invention

[0041] Where x(n) is the time domain signal, n = 1, 2, ..., N; N is the total number of samples of any radiation signal on each melt channel. s(k) is the signal spectrum, k = 1, 2, ..., K; K is the number of spectrum lines of any radiation signal on each melt channel. k is the frequency value of the kth spectrum line.

[0042] S30. Based on the extracted time / frequency domain features, different unsupervised learning anomaly detection algorithms are used to analyze and determine whether an anomaly occurs in the melt channel.

[0043] Because the causes and manifestations of abnormal melt paths during LPBF are often complex, and because each single algorithm model exhibits a degree of singularity, and different algorithms have their own strengths and weaknesses, no single algorithm can provide the optimal solution. Previous studies have shown that combinations of multiple classifiers perform better than single ones, and these combinations have been used in computer vision and in rotating machinery fault diagnosis.

[0044] Based on the extracted time / frequency domain features, the present invention uses different unsupervised learning anomaly detection algorithms to analyze and solve the problem of whether the melt channel is abnormal. Then, the solution results of different algorithms are combined according to certain principles to obtain the optimal solution to the problem.

[0045] The present invention can employ various unsupervised learning-based anomaly detection algorithms to analyze and determine whether a melt channel anomaly has occurred. As shown in Table 2, these unsupervised learning-based anomaly detection algorithms primarily include: ABOD and COPOD based on probability models; PCA based on linear models; KNN, HBOS, LOF, and CBLOF based on neighbor degree models; IForest, FB, and LODA based on ensemble methods; and AutoEncoder and VAE based on neural networks.

[0046]

[0047]

[0048] Table 2: Anomaly detection algorithms based on unsupervised learning

[0049] S40. Merge the solution results of different unsupervised learning anomaly detection algorithms, calculate the performance index, and use the performance index to quantitatively evaluate the processing quality.

[0050] After step S30, n different solution results can be obtained, which are recorded as O n (X), O n (X) is the solution of the nth anomaly detection algorithm model, representing the number of abnormal melt paths predicted by the nth anomaly detection algorithm model, where n = 1, 2, 3, ..., n. X is the input analysis data, representing the time / frequency domain characteristics corresponding to each melt path within the target processing layer.

[0051] The present invention uses n different anomaly detection algorithm models to predict the number of abnormal melt channels in the target processing layer, obtaining n different prediction results (solution results). The solution results of the different algorithms are then merged according to certain principles to obtain the final abnormal melt channel number MCOT. Since the averaging method is the simplest and most efficient multi-classifier combination strategy, the embodiment of the present invention selects the averaging method as the merging strategy for the solution results of multiple different anomaly detection algorithms. That is, the mean (mean) of the solution results of each different anomaly detection algorithm is taken to obtain the final abnormal melt channel number MCOT:

[0052] MCOT=Mean(O1(X),O2(X),O3(X),…,O n (X)).

[0053] In other embodiments of the present invention, an extreme value method, a weighted average method, a weighted extreme value method, etc. may also be used as a strategy for merging the solution results of multiple different anomaly detection algorithms to calculate the final abnormal melt channel number MCOT.

[0054] It is known that the more normal melt channels there are, the better the printing quality will be, and vice versa. Based on this principle, the present invention proposes a calculation method for measuring the quality performance index. Since the more abnormal melt channels there are, the worse the quality performance will be, the performance index pp should be expressed as the inverse of MCOT. Therefore, the performance index pp can be calculated using the following formula:

[0055]

[0056] The in-situ quality evaluation method for laser powder bed fusion-processed components of the present invention extracts the time / frequency domain characteristics of the molten pool radiation information. Based on the extracted time / frequency domain characteristics, different unsupervised learning anomaly detection algorithms are used to analyze and solve whether anomalies occur in the melt path to quantitatively evaluate the processing quality, thereby solving the problem that laser powder bed fusion quality analysis is highly dependent on the quality of the data set and prior knowledge.

[0057] The above description is merely a preferred embodiment of the present invention and does not constitute any form of limitation to the present invention. Those skilled in the art may make various equivalent changes and improvements based on the above embodiment. Any equivalent changes or modifications made within the scope of the claims shall fall within the scope of protection of the present invention.

Claims

1. An in-situ quality evaluation system for laser powder bed fusion processed components, characterized in that: Includes: A molten pool radiation collection device is used to collect molten pool radiation information of each processing layer during the printing process; a computer for acquiring the molten pool radiation information of each processing layer collected by the molten pool radiation collection device, acquiring the molten pool radiation information corresponding to the target processing layer from the molten pool radiation information of each processing layer and performing filtering and noise reduction processing, extracting the time / frequency domain features of the molten pool radiation information corresponding to the target processing layer one by one, and using different unsupervised learning anomaly detection algorithms to analyze and solve whether an anomaly occurs in the melt channel, wherein the different unsupervised learning anomaly detection algorithms include ABOD algorithm, COPOD algorithm, PCA algorithm, KNN algorithm, HBOS algorithm, LOF algorithm, CBLOF algorithm, IForest algorithm, FB algorithm, LODA algorithm, AutoEncoder algorithm and VAE algorithm, using an averaging method as a merging strategy to merge the solution results of different unsupervised learning anomaly detection algorithms, calculating a performance index and using the performance index to quantitatively evaluate the processing quality; Among them, the specific process of using the averaging method as a merging strategy to merge the solution results of different unsupervised learning anomaly detection algorithms is to take the average of the solution results of each different anomaly detection algorithm to obtain the final number of abnormal melt channels.

2. The in-situ quality evaluation system for laser powder bed fusion processed components according to claim 1, characterized in that: The molten pool radiation collection device includes: The spatial molten pool radiation collection unit is arranged on the upper top of the forming chamber and is used to collect the spatial molten pool radiation information of one or more bands of each processing layer; The coaxial optical path molten pool radiation collection unit is arranged in the laser processing optical path and is used to collect the coaxial optical path molten pool radiation information of one band or multiple bands of each processing layer.

3. The in-situ quality evaluation system for laser powder bed fusion processed components according to claim 2, characterized in that: The coaxial optical path molten pool radiation collection unit includes any one or any combination of a visible light radiation monitoring subunit, a near infrared radiation monitoring subunit, a colorimetric temperature measurement subunit, a laser reflection monitoring subunit, and an infrared radiation monitoring subunit; The near-infrared radiation monitoring subunit includes a second semi-transparent and semi-reflective mirror, a second focusing filter group, and a near-infrared radiation sensor. The second semi-transparent and semi-reflective mirror can reflect radiation in the 750nm-900nm band and can transmit radiation in other bands. The second focusing filter group is used to focus and combine the radiation in the 750nm-900nm band and transmit it to the near-infrared radiation sensor. The near-infrared radiation sensor is used to collect radiation in the 750nm-900nm band. The infrared radiation monitoring subunit includes a reflector, a fifth focusing filter group and an infrared radiation sensor. The reflector can reflect radiation in a band greater than or equal to 1200nm. The fifth focusing filter group is used to focus and combine radiation in a band greater than or equal to 1200nm and transmit it to the infrared radiation sensor. The infrared radiation sensor is used to collect radiation in a band greater than or equal to 1200nm.

4. The in-situ quality evaluation system for laser powder bed fusion processed components according to claim 3, characterized in that: The visible light radiation monitoring subunit includes a first semi-transparent and semi-reflective mirror, a first focusing filter group, and a visible light radiation sensor. The first semi-transparent and semi-reflective mirror can reflect radiation in the 400nm-750nm band and can transmit radiation in other bands. The first focusing filter group is used to focus and combine the radiation in the 400nm-750nm band and transmit it to the visible light radiation sensor. The visible light radiation sensor is used to collect radiation in the 400nm-750nm band. The colorimetric temperature measurement subunit includes a third semi-transparent and semi-reflective mirror, a third focusing filter group, and a colorimetric temperature measurement sensor. The third semi-transparent and semi-reflective mirror can reflect radiation in the 900nm-950nm band and can transmit radiation in other bands. The third focusing filter group is used to focus and combine the radiation in the 900nm-950nm band and transmit it to the colorimetric temperature measurement sensor. The colorimetric temperature measurement sensor is used to collect radiation in the 900nm-950nm band. The laser reflection monitoring subunit includes a fourth semi-transparent and semi-reflective mirror, a fourth focusing filter group and a laser reflection light sensor. The fourth semi-transparent and semi-reflective mirror can reflect radiation in the 1030nm-1080nm band and can transmit radiation in other bands. The fourth focusing filter group is used to focus and combine the radiation in the 1030nm-1080nm band and transmit it to the laser reflection light sensor. The laser reflection light sensor is used to collect radiation in the 1030nm-1080nm band.

5. A method for in-situ quality evaluation of components processed by laser powder bed fusion, characterized in that: The steps include: Obtain the molten pool radiation information corresponding to the target processing layer and perform filtering and noise reduction processing; Extract the time / frequency domain characteristics of the molten pool radiation information corresponding to the target processing layer one by one; Based on the extracted time / frequency domain features, different unsupervised learning anomaly detection algorithms are used to analyze and determine whether the melt channel has an anomaly. These include the ABOD algorithm, the COPOD algorithm, the PCA algorithm, the KNN algorithm, the HBOS algorithm, the LOF algorithm, the CBLOF algorithm, the IForest algorithm, the FB algorithm, the LODA algorithm, the AutoEncoder algorithm, and the VAE algorithm. The averaging method is used as a merging strategy to merge the solution results of different unsupervised learning anomaly detection algorithms, and the performance index is calculated and used to quantitatively evaluate the processing quality; Among them, the specific process of using the averaging method as a merging strategy to merge the solution results of different unsupervised learning anomaly detection algorithms is to take the average of the solution results of each different anomaly detection algorithm to obtain the final number of abnormal melt channels.

6. The in-situ quality evaluation method for a laser powder bed fusion processed component according to claim 5, wherein: The target processing layer is selected by adopting the following strategies: randomly selecting any n processing layers as the target processing layer; or taking the processing risk layer determined by evaluation as the target processing layer; or taking all processing layers of the part processing as the target processing layer.

7. The in-situ quality evaluation method for a laser powder bed fusion-processed component according to claim 5, wherein: The time / frequency domain features include 16 time domain features and 12 frequency domain features.

8. The in-situ quality evaluation method for a laser powder bed fusion processed component according to claim 5, wherein: The performance index is calculated using the following formula: ； in, O n (X) is the solution of the nth anomaly detection algorithm model, n = 1, 2, 3, ... n, MCOT is the number of abnormal melt channels, is the performance index.

9. The in-situ quality evaluation method for a laser powder bed fusion processed component according to any one of claims 5 to 8, characterized in that: The molten pool radiation information includes: spatial molten pool radiation information of one band or multiple bands; and / or coaxial optical path molten pool radiation information of one band or multiple bands.

Images