Selective laser melting 3D printing process monitoring method based on multi-source information fusion

Through multi-source information fusion and CNN network to identify radiation abnormal information of melting ducts and melt pools, a monitoring model was established, which solved the problem of difficult product quality during the 3D printing of selected laser melting, and achieved a more comprehensive and accurate monitoring effect.

CN118875320APending Publication Date: 2024-11-01SOUTHEAST UNIV +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410813068.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-06-23
Publication Date
2024-11-01

AI Technical Summary

Technical Problem

The prior art is difficult to effectively monitor the product quality during the 3D printing process of selective laser melting, especially under high-speed laser scanning, it is difficult to obtain real printing process information under the influence of multiple factors, and it is difficult to achieve accurate detection by a single monitoring method.

Method used

Using a monitoring method based on multi-source information fusion, multi-dimensional printing data is obtained through coaxial radiation acquisition device and industrial camera layer shooting device, CNN network is used to identify radiation abnormality information of melt duct and melt pool, and an abnormality monitoring model is established based on actual surface information. , realize real-time evaluation and monitoring.

Benefits of technology

It realizes comprehensive, authentic and accurate monitoring of the molding quality of the selected laser melting 3D printing process, improves the monitoring effect, and is suitable for lightweight network model applications in actual industrial sites.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118875320A_ABST
    Figure CN118875320A_ABST
Patent Text Reader

Abstract

The invention discloses a selective laser melting 3D printing process monitoring method based on multi-source information fusion. The selective laser melting 3D printing process monitoring method comprises the steps that multi-dimensional printing data of a printing layer of a test printing piece in the 3D printing process is obtained; identifying one-dimensional radiation abnormal information, two-dimensional radiation abnormal information and two-dimensional picture abnormal information; establishing an anomaly monitoring model by using the actual surface information, the one-dimensional radiation anomaly information, the two-dimensional radiation anomaly information and the two-dimensional picture anomaly information; and multi-dimensional printing data of the printing layers of the printed product in the 3D printing process are obtained, and then real-time evaluation monitoring is conducted on the surface information of all the printing layers of the printed product through the abnormity monitoring model. According to the 3D printing process monitoring method, the monitoring effect of obtaining the multi-dimensional printing data is more comprehensive, real and accurate; by establishing the abnormality monitoring model for printing abnormality monitoring, the lightweight network model is more convenient and effective for production and use in an actual industrial field.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of selective laser melting process monitoring, and in particular relates to a selective laser melting 3D printing process monitoring method based on multi-source information fusion. Background Art

[0002] With the continuous advancement of science and technology, additive manufacturing technology (AM) has become an important force in promoting the development of modern industry. Among them, selective laser melting (SLM) technology, as an advanced metal additive manufacturing technology, is gradually showing its huge potential and value. Since the birth of SLM technology in the early 1990s, after decades of development, it has made significant progress in equipment performance, material types, manufacturing processes, etc. At present, SLM technology can realize the rapid prototyping of various materials such as metals, ceramics, and polymers, and the dimensional accuracy and surface quality of the formed parts have also been greatly improved. In addition, with the continuous development of computer-aided design (CAD) and computer-aided manufacturing (CAM) technology, the automation and intelligence level of SLM technology is also gradually improving.

[0003] Although selective laser melting technology has shown broad application prospects in many fields, it still faces some difficulties and challenges. The most prominent problem is that the quality of printed products is difficult to monitor effectively. Currently, the printing quality of products is mainly detected by industrial CT scanning of the workpiece after printing. However, the price of this industrial CT is very expensive, and there is no way to repair and eliminate the existing defects and problems in the workpiece by testing the workpiece that has been printed. Therefore, it is very valuable to study and implement process monitoring technology for the quality of printed products based on the characteristics of selective laser melting technology.

[0004] During the selective laser melting printing process, the laser scanning speed can reach an average of 800 to 1300 mm / s. Under this high-speed laser scanning, the local metal powder will melt and cool in a very short time. The selective laser melting process is a complex process affected by many factors, such as the combination of typical process parameters such as laser power, scanning speed and scanning spacing, the flow direction and airflow intensity of the gas in the chamber during the replacement of the protective gas in the chamber, the scanning path and scanning strategy of the printed workpiece, the splashing of metal powder and the remelting of the solidified metal. These factors will affect the monitoring of the selective laser melting process. For this high-speed changing and multi-factor affected monitoring target, there is currently no mature monitoring solution to monitor the quality of the printed product. At the same time, due to the complexity of the printing process, it is also difficult to effectively and truly obtain the information of the workpiece during the printing process using a single monitoring method. Summary of the invention

[0005] The purpose of the invention is to provide a selective laser melting 3D printing process monitoring method based on multi-source information fusion, which can monitor the molding quality of the printing process.

[0006] Technical solution: The method for monitoring the selective laser melting 3D printing process based on multi-source information fusion of the present invention comprises the following steps:

[0007] Step 1, obtaining multi-dimensional printing data of the printing layer of the test print during the 3D printing process, the multi-dimensional printing data including one-dimensional radiation data of each melt path obtained by a coaxial radiation acquisition device and a two-dimensional layer image of the melt pool surface obtained by an industrial camera layer image device;

[0008] Step 2, identifying one-dimensional radiation anomaly information of each melt path, two-dimensional radiation anomaly information of the melt pool surface, and two-dimensional image anomaly information of the melt pool surface from the multi-dimensional printing data of the test print;

[0009] Step 3, obtaining the actual surface information of the printed layer of the test print, using the actual surface information, one-dimensional radiation anomaly information, two-dimensional radiation anomaly information and two-dimensional image anomaly information for fusion analysis, and establishing an anomaly monitoring model for printing anomaly monitoring;

[0010] Step 4, obtaining multi-dimensional printing data of the printing layer of the printed product during the 3D printing process, and then identifying the one-dimensional radiation anomaly information of each melt path, the two-dimensional radiation anomaly information of the molten pool surface, and the two-dimensional image anomaly information of the molten pool surface from the multi-dimensional printing data of the printed product, and then using the one-dimensional radiation anomaly information, the two-dimensional radiation anomaly information, and the two-dimensional image anomaly information to perform real-time evaluation and monitoring of the surface information of each printing layer of the printed product through an anomaly monitoring model.

[0011] Furthermore, in steps 2 and 4, the specific steps of identifying the one-dimensional radiation anomaly information of each melt path, the two-dimensional radiation anomaly information of the molten pool surface, and the two-dimensional image anomaly information of the molten pool surface are:

[0012] Firstly, the trained CNN network is used to identify abnormal data points in the one-dimensional radiation data of each melting channel to obtain one-dimensional radiation anomaly information;

[0013] Secondly, coordinate correspondence and numerical compression are performed on the one-dimensional radiation data of each melting channel to obtain a two-dimensional radiation heat map, and then the abnormal area in the two-dimensional radiation heat map is identified to obtain two-dimensional radiation abnormality information;

[0014] Finally, the abnormal area in the two-dimensional layer image is identified to obtain the abnormal information of the two-dimensional image.

[0015] Furthermore, the specific training steps of the CNN network are:

[0016] First, the surface depth information of each printed layer of the test print is obtained using an ultra-depth-of-field microscope;

[0017] Secondly, the one-dimensional molten pool radiation data of each melt path of each printing layer of the test print obtained by the coaxial radiation acquisition device is subjected to wavelet analysis to obtain the wavelet time-frequency information graph of the radiation data;

[0018] Thirdly, the wavelet time-frequency information graph is equidistantly windowed, and based on the surface depth information obtained by the ultra-depth-of-field microscope, the wavelet time-frequency information graph in each window is printed and annotated to obtain a three-classification data set, which includes flatness data, concave hole data, and convex hole data. The concave hole data includes the number of concave holes and the coordinates of the concave hole positions, and the convex hole data includes the number of convex holes and the coordinates of the convex positions. Then, the molding quality of the melt channel is calculated based on the three-classification data set.

[0019] Finally, the CNN network is trained using the flat data, concave hole data, and convexity data in the three-category data set to obtain a CNN network for identifying abnormal data points in one-dimensional radiation data.

[0020] Furthermore, when the wavelet time-frequency information graph is equidistantly windowed, the window length of the equidistantly windowed graph is set to:

[0021]

[0022] Where s is the actual concave and convex size of the printed surface, in μm, t is the sampling speed of the coaxial radiation acquisition device, in μs / sampling point, v is the printing speed, in μm / μs, and n is the length coefficient of the window.

[0023] Furthermore, the specific calculation formula for calculating the molding quality of the melt channel based on the three-classification data set is:

[0024]

[0025] Where N is the total number of printed dots in the window, num1 is the number of concave holes, and num2 is the number of convex holes.

[0026] Furthermore, the specific steps of coordinate correspondence and numerical compression of the one-dimensional radiation data of each melt path are as follows:

[0027] First, the radiation value of each data point is mapped to the corresponding spatial coordinate value by equidistant sampling of the one-dimensional radiation data:

[0028] First, determine that the melt channel number of the radiation value of the current data point is x, the data number is N, and the data number corresponding to the starting point of the melt channel numbered x is N0, so the coordinates of the current data point are obtained as (x, N-N0);

[0029] Then calculate the molten pool radiation data of the current data point as the molten pool radiation data corresponding to the position (N-N0)×t×vμm of the molten channel with serial number x, where t is the sampling speed of the coaxial radiation acquisition device and v is the printing speed;

[0030] Secondly, the molten pool radiation data of the current data point is numerically compressed: the molten pool radiation data of the current data point is normalized according to the formula Linear normalization is performed, T represents the molten pool radiation data, T MIN is the minimum value of the molten pool radiation data in the same printing surface, T MAX is the maximum value of the molten pool radiation data in the same printing surface, and T0 is represented by the normalized result, that is, the gray value of the current data point;

[0031] Finally, the above coordinate correspondence and numerical compression steps are repeated to complete the coordinate correspondence and numerical compression of each data point of each melt path in the molten pool, and a two-dimensional radiation heat map of the molten pool surface can be obtained.

[0032] Furthermore, the specific steps for identifying abnormal areas in the two-dimensional radiation heat map are as follows:

[0033] First, the two-dimensional radiation heat map is subjected to graphics processing to obtain a grayscale image of the two-dimensional radiation heat map. The graphics processing includes extreme value filtering, corrosion and expansion processing, and grayscale processing.

[0034] Secondly, customize a two-dimensional filtering algorithm for binary separation processing;

[0035] Finally, a two-dimensional filtering algorithm is used to filter the grayscale image of the two-dimensional radiation heat map to obtain the abnormal feature information of the two-dimensional radiation heat map, and the abnormal feature information includes the coordinate information and area information of the abnormal area.

[0036] Furthermore, the specific steps for identifying abnormal areas in the two-dimensional layer image are as follows:

[0037] First, the field of view of the two-dimensional layer image is transformed so that the viewing angle of the two-dimensional layer image is changed to be directly above;

[0038] Secondly, the layer shot image after the field of view transformation is subjected to graphics processing to obtain a grayscale image of the layer shot image after the field of view transformation. The graphics processing includes extreme value filtering processing, corrosion and expansion processing, and grayscale processing;

[0039] Again, customize a two-dimensional filtering algorithm for binary separation processing;

[0040] Finally, a two-dimensional filtering algorithm is used to filter the grayscale image of the layer image after the field of view transformation to obtain the abnormal feature information of the two-dimensional layer image, which includes the coordinate information and area information of the abnormal area.

[0041] Furthermore, the algorithm formula of the customized two-dimensional filtering algorithm is:

[0042]

[0043] Wherein, Pix is ​​the pixel value of the corresponding coordinate after filtering, and Data is the grayscale value of the corresponding coordinate in the layer shooting image obtained by the camera layer shooting device.

[0044] Furthermore, in step 3, the specific steps of using the actual surface information, the one-dimensional radiation anomaly information, the two-dimensional radiation anomaly information and the two-dimensional image anomaly information for fusion analysis are as follows:

[0045] Firstly, the surface depth information of the test print is collected using an ultra-depth-of-field microscope, and then the information of the abnormal area actually existing in the surface information is extracted to obtain the surface abnormal area information, and the data set E is constructed using the surface abnormal area information;

[0046] Secondly, collect one-dimensional radiation anomaly information, two-dimensional radiation anomaly information and two-dimensional image anomaly information, and use the one-dimensional radiation anomaly information to construct a data set D1, use the two-dimensional radiation anomaly information to construct a data set D2, and use the two-dimensional image anomaly information to construct a data set D3;

[0047] Next, the corresponding relationship between the data sets D1, D2 and D3 and the data set E is established as: {E} = A{D1, D2, D3} + {B};

[0048] Finally, after repeated tests of each printing layer, multiple groups of E and corresponding D1, D2 and D3 are obtained. The coefficient matrix A and intercept matrix B in the corresponding relationship are obtained by using the multivariate linear regression model method. The output is the corresponding relationship {E}=A{D1,D2,D3}+{B} as the anomaly monitoring model. The anomaly monitoring model obtains the monitoring result data set E through the data sets D1, D2, D3 and the coefficient matrix A and intercept matrix B.

[0049] Compared with the prior art, the present invention has the following beneficial effects: by acquiring multi-dimensional printing data, the monitoring effect of multiple data sources will be more comprehensive, true and accurate compared with the existing single sensor monitoring method; by establishing an abnormality monitoring model for printing abnormality monitoring, a lightweight network model is achieved, which is more convenient and effective for production use in actual industrial sites. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 is a flow chart of the method of the present invention;

[0051] Figure 2 It is a schematic structural diagram of the coaxial radiation collection device of the present invention;

[0052] Figure 3 It is a schematic diagram of the structure of the industrial camera layer shooting device of the present invention;

[0053] Figure 4 It is a finite element simulation schematic diagram of the present invention;

[0054] Figure 5 A surface depth information map obtained by the ultra-depth microscope of the present invention;

[0055] Figure 6 It is a schematic diagram of the one-dimensional radiation anomaly information identification process of the present invention;

[0056] Figure 7 It is a schematic diagram of the process of identifying two-dimensional radiation anomaly information and two-dimensional image anomaly information of the present invention. DETAILED DESCRIPTION

[0057] The technical solution of the present invention is described in detail below with reference to the accompanying drawings, but the protection scope of the present invention is not limited to the embodiments.

[0058] Embodiment 1:

[0059] like Figure 1 As shown, the selective laser melting 3D printing process monitoring method based on multi-source information fusion disclosed in the present invention includes the following steps:

[0060] Step 1, obtaining multi-dimensional printing data of the printing layer of the test print during the 3D printing process, the multi-dimensional printing data including one-dimensional radiation data of each melt path obtained by a coaxial radiation acquisition device and a two-dimensional layer image of the melt pool surface obtained by an industrial camera layer image device;

[0061] Step 2, identifying one-dimensional radiation anomaly information of each melt path, two-dimensional radiation anomaly information of the melt pool surface, and two-dimensional image anomaly information of the melt pool surface from the multi-dimensional printing data of the test print;

[0062] Step 3, obtaining the actual surface information of the printed layer of the test print, using the actual surface information, one-dimensional radiation anomaly information, two-dimensional radiation anomaly information and two-dimensional image anomaly information for fusion analysis, and establishing an anomaly monitoring model for printing anomaly monitoring;

[0063] Step 4, obtaining multi-dimensional printing data of the printing layer of the printed product during the 3D printing process, and then identifying the one-dimensional radiation anomaly information of each melt path, the two-dimensional radiation anomaly information of the molten pool surface, and the two-dimensional image anomaly information of the molten pool surface from the multi-dimensional printing data of the printed product, and then using the one-dimensional radiation anomaly information, the two-dimensional radiation anomaly information, and the two-dimensional image anomaly information to perform real-time evaluation and monitoring of the surface information of each printing layer of the printed product through an anomaly monitoring model.

[0064] By acquiring multi-dimensional printing data, compared with the existing single sensor monitoring method, the monitoring effect of multiple data sources will be more comprehensive, real and accurate; by establishing an abnormal monitoring model for printing abnormality monitoring, a lightweight network model is realized, which is more convenient and effective for production use in actual industrial sites.

[0065] Further, such as Figure 2 As shown, in step 1, the coaxial radiation collection device includes a high-energy laser generator 41, a semi-transparent and semi-reflective mirror 42, a galvanometer 43, a field lens 44, a narrow-band filter 46, a silicon PIN photodiode 47, a data processing module 48 and an industrial computer 49; the high-energy laser generator 41, the semi-transparent and semi-reflective mirror 42, the galvanometer 43 and the field lens 44 sequentially constitute a laser emission path, the high-energy laser generator 41 emits laser light, the laser light passes through the semi-transparent and semi-reflective mirror 42 in one direction, and then is deflected by the galvanometer 43 and focused by the field lens 44, and then transmitted to the molding platform 45, and the metal powder on the surface of the molding platform 45 is laser-melted to form a molten pool;

[0066] The radiation beam path is formed in sequence by the field lens 44, the galvanometer 43, the semi-transparent and semi-reflective mirror 42, the narrow-band filter 46 and the silicon PIN photodiode 47. The beam after the molten pool is excited and the reflected beam of the laser on the surface of the molten pool pass through the field lens 44 and the galvanometer 43, and then return to the semi-transparent and semi-reflective mirror 42. At this time, the beam contains the molten pool information, and the beam of this optical path cannot penetrate the semi-transparent and semi-reflective mirror 42, and can only be mirror-reflected on the surface of the semi-transparent and semi-reflective mirror 42. After the reflected beam passes through the narrow-band filter 46, the characteristic light of the wavelength λ containing the characteristic information of the molten pool reaches the silicon PIN photodiode 47, and the silicon PIN photodiode 47 senses the beam radiated by the molten pool; the data processing module 48 collects the sensing signal of the silicon PIN photodiode 47; the industrial computer 49 processes the signal collected by the data processing module 48 to obtain one-dimensional molten pool radiation data; Figure 3 As shown, the industrial camera layer shooting device includes a fill light 51 and an industrial camera 52; the industrial camera 52 shoots each layer of the molten pool on the surface of the forming platform 45 to obtain a layer shot image; the fill light 51 provides fill light when the industrial camera 52 is shooting.

[0067] Furthermore, in steps 2 and 4, the specific steps of identifying the one-dimensional radiation anomaly information of each melt path, the two-dimensional radiation anomaly information of the molten pool surface, and the two-dimensional image anomaly information of the molten pool surface are:

[0068] Firstly, the trained CNN network is used to identify abnormal data points in the one-dimensional radiation data of each melting channel to obtain one-dimensional radiation anomaly information;

[0069] Secondly, coordinate correspondence and numerical compression are performed on the one-dimensional radiation data of each melting channel to obtain a two-dimensional radiation heat map, and then the abnormal area in the two-dimensional radiation heat map is identified to obtain two-dimensional radiation abnormality information;

[0070] Finally, the abnormal area in the two-dimensional layer image is identified to obtain the abnormal information of the two-dimensional image.

[0071] Furthermore, the specific training steps of the CNN network are:

[0072] First, the surface depth information of each printed layer of the test print is obtained using an ultra-depth-of-field microscope, and then the same material properties and printing parameters are imported for finite element simulation analysis, such as Figure 4 As shown in (a) and (b), Figure 4 (a) is a particle powder bed observed under an electron microscope, and (b) is a powder bed model close to the actual printing state established using the discrete particle method. The finite element simulation module is used to perform discrete particle modeling on the actual printing process to obtain the simulation results of the printed product.

[0073] Secondly, the one-dimensional molten pool radiation data of each melt path of each printing layer of the test print obtained by the coaxial radiation acquisition device is subjected to wavelet analysis to obtain the wavelet time-frequency information graph of the radiation data;

[0074] Thirdly, the wavelet time-frequency information graph is equidistantly windowed, and based on the surface depth information obtained by the ultra-depth-of-field microscope, the wavelet time-frequency information graph in each window is printed and annotated to obtain a three-classification data set, which includes flatness data, concave hole data, and convex hole data. The concave hole data includes the number of concave holes and the coordinates of the concave hole positions, and the convex hole data includes the number of convex holes and the coordinates of the convex positions. Then, the molding quality of the melt channel is calculated based on the three-classification data set.

[0075] Finally, the CNN network is trained using the flat data, concave hole data, and convexity data in the three-category dataset. Figure 6 As shown, preferably, the simulation results obtained by the finite element simulation module are introduced into the CNN network as simulation constraints to obtain a CNN network for identifying abnormal data points in one-dimensional radiation data. The finite element simulation results under ideal conditions are introduced as constraint data to eliminate the influence of extreme values.

[0076] Furthermore, when the wavelet time-frequency information graph is equidistantly windowed, the window length of the equidistantly windowed graph is set to:

[0077]

[0078] Where s is the actual concave and convex size of the printed surface, in μm, t is the sampling speed of the coaxial radiation acquisition device, in μs / sampling point, v is the printing speed, in μm / μs, and n is the length coefficient of the window.

[0079] Furthermore, the specific calculation formula for calculating the molding quality of the melt channel based on the three-classification data set is:

[0080]

[0081] Where N is the total number of printed dots in the window, num1 is the number of concave holes, and num2 is the number of convex holes.

[0082] The molding quality of the melt channel can be quantitatively evaluated by calculating the molding quality value of the melt channel.

[0083] Furthermore, the specific steps of coordinate correspondence and numerical compression of the one-dimensional radiation data of each melt path are as follows:

[0084] First, the radiation value of each data point is mapped to the corresponding spatial coordinate value by equidistant sampling of the one-dimensional radiation data:

[0085] First, determine that the melt channel number of the radiation value of the current data point is x, the data number is N, and the data number corresponding to the starting point of the melt channel numbered x is N0, so the coordinates of the current data point are obtained as (x, N-N0);

[0086] Then calculate the molten pool radiation data of the current data point as the molten pool radiation data corresponding to the position (N-N0)×t×vμm of the molten channel with serial number x, where t is the sampling speed of the coaxial radiation acquisition device and v is the printing speed;

[0087] Secondly, the molten pool radiation data of the current data point is numerically compressed: the molten pool radiation data of the current data point is normalized according to the formula Linear normalization is performed, T represents the molten pool radiation data, T MIN is the minimum value of the molten pool radiation data in the same printing surface, T MAX is the maximum value of the molten pool radiation data in the same printing surface, and T0 is represented by the normalized result, that is, the gray value of the current data point;

[0088] Finally, the above coordinate correspondence and numerical compression steps are repeated to complete the coordinate correspondence and numerical compression of each data point of each melt path in the molten pool, and a two-dimensional radiation heat map of the molten pool surface can be obtained.

[0089] Further, such as Figure 7 As shown in Figure 1, the specific steps for identifying abnormal areas in a two-dimensional radiation heat map are:

[0090] First, the two-dimensional radiation heat map is subjected to graphics processing to obtain a grayscale image of the two-dimensional radiation heat map. The graphics processing includes extreme value filtering, corrosion and expansion processing, and grayscale processing.

[0091] Secondly, customize a two-dimensional filtering algorithm for binary separation processing;

[0092] Finally, a two-dimensional filtering algorithm is used to filter the grayscale image of the two-dimensional radiation heat map to obtain the abnormal feature information of the two-dimensional radiation heat map, and the abnormal feature information includes the coordinate information and area information of the abnormal area.

[0093] Further, such as Figure 7 As shown in FIG. 1 , the specific steps for identifying abnormal areas in the two-dimensional layer image are as follows:

[0094] First, the field of view of the two-dimensional layer image is transformed so that the viewing angle of the two-dimensional layer image is changed to be directly above;

[0095] Secondly, the layer shot image after the field of view transformation is subjected to graphics processing to obtain a grayscale image of the layer shot image after the field of view transformation. The graphics processing includes extreme value filtering processing, corrosion and expansion processing, and grayscale processing;

[0096] Again, customize a two-dimensional filtering algorithm for binary separation processing;

[0097] Finally, a two-dimensional filtering algorithm is used to filter the grayscale image of the layer image after the field of view transformation to obtain the abnormal feature information of the two-dimensional layer image, which includes the coordinate information and area information of the abnormal area.

[0098] Furthermore, extreme value filtering processing is to use extreme value filters to remove extreme values ​​from the two-dimensional radiation heat map and the layer image after field of view transformation; corrosion and expansion processing is to corrode and expand the two-dimensional radiation heat map and the layer image after extreme value removal, so as to enhance the local information in the two-dimensional radiation heat map and the layer image; grayscale processing is to perform grayscale processing on the two-dimensional radiation heat map and the layer image after the local information is enhanced, so as to obtain the grayscale image of the two-dimensional radiation heat map and the layer image.

[0099] Furthermore, the algorithm formula of the customized two-dimensional filtering algorithm is:

[0100]

[0101] Wherein, Pix is ​​the pixel value of the corresponding coordinate after filtering, and Data is the grayscale value of the corresponding coordinate in the layer shooting image obtained by the camera layer shooting device.

[0102] The two-dimensional filtering algorithm can be used to realize the binary separation processing of the two-dimensional radiation heat map and the layer shot map, and then the envelope information of the abnormal area in the grayscale image after the binary separation processing can be obtained, and the centroid coordinates and area of ​​the abnormal area can be calculated. The centroid coordinates are used as the coordinate information of the abnormal area, and the area is used as the area information of the abnormal area.

[0103] Furthermore, in step 3, the specific steps of using the actual surface information, the one-dimensional radiation anomaly information, the two-dimensional radiation anomaly information and the two-dimensional image anomaly information for fusion analysis are as follows:

[0104] First, the surface depth information of the test print is collected using an ultra-depth-of-field microscope, such as Figure 5 As shown, the information of the abnormal area actually existing in the surface information is extracted to obtain the surface abnormal area information, and the data set E is constructed using the surface abnormal area information;

[0105] Secondly, collect one-dimensional radiation anomaly information, two-dimensional radiation anomaly information and two-dimensional image anomaly information, and use the one-dimensional radiation anomaly information to construct a data set D1, use the two-dimensional radiation anomaly information to construct a data set D2, and use the two-dimensional image anomaly information to construct a data set D3;

[0106] Next, the corresponding relationship between the data sets D1, D2 and D3 and the data set E is established as: {E} = A{D1, D2, D3} + {B};

[0107] Finally, after repeated tests of each printing layer, multiple groups of E and corresponding D1, D2 and D3 are obtained. The coefficient matrix A and intercept matrix B in the corresponding relationship are obtained by using the multivariate linear regression model method, and the output is the corresponding relationship {E}=A{D1,D2,D3}+{B} as the abnormality monitoring model. The abnormality monitoring model obtains the monitoring result data set E through the data sets D1, D2, D3 and the coefficient matrix A and intercept matrix B. That is, the abnormality monitoring model can locate the coordinates and area of ​​the abnormal area of ​​each SLM metal 3D printing according to the detection information of the printing.

[0108] As described above, although the present invention has been shown and described with reference to specific preferred embodiments, it should not be construed as limiting the present invention itself. Various changes in form and details may be made without departing from the spirit and scope of the present invention as defined in the appended claims.

Claims

1. A selective laser melting 3D printing process monitoring method based on multi-source information fusion, It is characterized in that The steps include: Step 1, obtaining multi-dimensional printing data of the printing layer of the test print during the 3D printing process, the multi-dimensional printing data including one-dimensional radiation data of each melt path obtained by a coaxial radiation acquisition device and a two-dimensional layer image of the melt pool surface obtained by an industrial camera layer image device; Step 2, identifying one-dimensional radiation anomaly information of each melt path, two-dimensional radiation anomaly information of the melt pool surface, and two-dimensional image anomaly information of the melt pool surface from the multi-dimensional printing data of the test print; Step 3, obtaining the actual surface information of the printed layer of the test print, using the actual surface information, one-dimensional radiation anomaly information, two-dimensional radiation anomaly information and two-dimensional image anomaly information for fusion analysis, and establishing an anomaly monitoring model for printing anomaly monitoring; Step 4, obtaining multi-dimensional printing data of the printing layer of the printed product during the 3D printing process, and then identifying the one-dimensional radiation anomaly information of each melt path, the two-dimensional radiation anomaly information of the molten pool surface, and the two-dimensional image anomaly information of the molten pool surface from the multi-dimensional printing data of the printed product, and then using the one-dimensional radiation anomaly information, the two-dimensional radiation anomaly information, and the two-dimensional image anomaly information to perform real-time evaluation and monitoring of the surface information of each printing layer of the printed product through an anomaly monitoring model.

2. According to claim 1, the method for monitoring the selective laser melting 3D printing process based on multi-source information fusion, It is characterized in that In steps 2 and 4, the specific steps of identifying the one-dimensional radiation anomaly information of each melt path, the two-dimensional radiation anomaly information of the molten pool surface, and the two-dimensional image anomaly information of the molten pool surface are as follows: Firstly, the trained CNN network is used to identify abnormal data points in the one-dimensional radiation data of each melting channel to obtain one-dimensional radiation anomaly information; Secondly, coordinate correspondence and numerical compression are performed on the one-dimensional radiation data of each melting channel to obtain a two-dimensional radiation heat map, and then the abnormal area in the two-dimensional radiation heat map is identified to obtain two-dimensional radiation abnormality information; Finally, the abnormal area in the two-dimensional layer image is identified to obtain the abnormal information of the two-dimensional image.

3. According to claim 2, the method for monitoring the selective laser melting 3D printing process based on multi-source information fusion, It is characterized in that The specific training steps of the CNN network are: First, the surface depth information of each printed layer of the test print is obtained using an ultra-depth-of-field microscope; Secondly, the one-dimensional molten pool radiation data of each melt path of each printing layer of the test print obtained by the coaxial radiation acquisition device is subjected to wavelet analysis to obtain the wavelet time-frequency information graph of the radiation data; Thirdly, the wavelet time-frequency information graph is equidistantly windowed, and based on the surface depth information obtained by the ultra-depth-of-field microscope, the wavelet time-frequency information graph in each window is printed and annotated to obtain a three-classification data set, which includes flatness data, concave hole data, and convex hole data. The concave hole data includes the number of concave holes and the coordinates of the concave hole positions, and the convex hole data includes the number of convex holes and the coordinates of the convex positions. Then, the molding quality of the melt channel is calculated based on the three-classification data set. Finally, the CNN network is trained using the flat data, concave hole data, and convexity data in the three-category data set to obtain a CNN network for identifying abnormal data points in one-dimensional radiation data.

4. According to claim 3, the method for monitoring the selective laser melting 3D printing process based on multi-source information fusion, It is characterized in that When the wavelet time-frequency information graph is equidistantly windowed, the window length of the equidistant windowing is set as: Where s is the actual concave and convex size of the printed surface, in μm, t is the sampling speed of the coaxial radiation acquisition device, in μs / sampling point, v is the printing speed, in μm / μs, and n is the length coefficient of the window.

5. According to claim 3, the method for monitoring the selective laser melting 3D printing process based on multi-source information fusion, It is characterized in that The specific calculation formula for calculating the molding quality of the melt channel based on the three-classification data set is: Where N is the total number of printed dots in the window, num1 is the number of concave holes, and num2 is the number of convex holes.

6. The method for monitoring the selective laser melting 3D printing process based on multi-source information fusion according to claim 2, It is characterized in that The specific steps for coordinate correspondence and numerical compression of the one-dimensional radiation data of each melt channel are as follows: First, the radiation value of each data point is mapped to the corresponding spatial coordinate value by equidistant sampling of the one-dimensional radiation data: First determine the melt number of the radiation value of the current data point is x, the data number is N, and the data number corresponding to the starting point of the melt number x is N. 0 , so the coordinates of the current data point are (x,NN 0 ); Then calculate the molten pool radiation data of the current data point as (NN 0 )×t×vμm corresponding to the molten pool radiation data, t is the sampling speed of the coaxial radiation acquisition device, and v is the printing speed; Secondly, the molten pool radiation data of the current data point is numerically compressed: the molten pool radiation data of the current data point is normalized according to the formula Linear normalization is performed, T represents the molten pool radiation data, T MIN is the minimum value of the molten pool radiation data in the same printing surface, T MAX is the maximum value of the molten pool radiation data in the same printing surface, T 0 It is expressed as the normalized result, that is, the gray value of the current data point; Finally, the above coordinate correspondence and numerical compression steps are repeated to complete the coordinate correspondence and numerical compression of each data point of each melt path in the molten pool, and a two-dimensional radiation heat map of the molten pool surface can be obtained.

7. The method for monitoring the selective laser melting 3D printing process based on multi-source information fusion according to claim 2, It is characterized in that The specific steps for identifying abnormal areas in two-dimensional radiation heat maps are: First, the two-dimensional radiation heat map is subjected to graphics processing to obtain a grayscale image of the two-dimensional radiation heat map. The graphics processing includes extreme value filtering, corrosion and expansion processing, and grayscale processing. Secondly, customize a two-dimensional filtering algorithm for binary separation processing; Finally, a two-dimensional filtering algorithm is used to filter the grayscale image of the two-dimensional radiation heat map to obtain the abnormal feature information of the two-dimensional radiation heat map, and the abnormal feature information includes the coordinate information and area information of the abnormal area.

8. The method for monitoring the selective laser melting 3D printing process based on multi-source information fusion according to claim 2, It is characterized in that The specific steps for identifying abnormal areas in two-dimensional layer images are as follows: First, the field of view of the two-dimensional layer image is transformed so that the viewing angle of the two-dimensional layer image is changed to be directly above; Secondly, the layer shot image after the field of view transformation is subjected to graphics processing to obtain a grayscale image of the layer shot image after the field of view transformation. The graphics processing includes extreme value filtering processing, corrosion and expansion processing, and grayscale processing; Again, customize a two-dimensional filtering algorithm for binary separation processing; Finally, a two-dimensional filtering algorithm is used to filter the grayscale image of the layer image after the field of view transformation to obtain the abnormal feature information of the two-dimensional layer image, which includes the coordinate information and area information of the abnormal area.

9. The method for monitoring the selective laser melting 3D printing process based on multi-source information fusion according to claim 7 or 8, It is characterized in that The algorithm formula of the customized two-dimensional filtering algorithm is: Wherein, Pix is ​​the pixel value of the corresponding coordinate after filtering, and Data is the grayscale value of the corresponding coordinate in the layer shooting image obtained by the camera layer shooting device.

10. The method for monitoring the selective laser melting 3D printing process based on multi-source information fusion according to claim 1, It is characterized in that In step 3, the specific steps of using the actual surface information, one-dimensional radiation anomaly information, two-dimensional radiation anomaly information and two-dimensional image anomaly information for fusion analysis are as follows: Firstly, the surface depth information of the test print is collected using an ultra-depth-of-field microscope, and then the information of the abnormal area actually existing in the surface information is extracted to obtain the surface abnormal area information, and the data set E is constructed using the surface abnormal area information; Secondly, collect one-dimensional radiation anomaly information, two-dimensional radiation anomaly information and two-dimensional image anomaly information, and use the one-dimensional radiation anomaly information to construct a data set D1, use the two-dimensional radiation anomaly information to construct a data set D2, and use the two-dimensional image anomaly information to construct a data set D3; Next, the corresponding relationship between the data sets D1, D2 and D3 and the data set E is established as: {E} = A{D1, D2, D3} + {B}; Finally, after repeated tests of each printing layer, multiple groups of E and corresponding D1, D2 and D3 are obtained. The coefficient matrix A and intercept matrix B in the corresponding relationship are obtained by using the multivariate linear regression model method. The output is the corresponding relationship {E}=A{D1,D2,D3}+{B} as the anomaly monitoring model. The anomaly monitoring model obtains the monitoring result data set E through the data sets D1, D2, D3 and the coefficient matrix A and intercept matrix B.