SLM powder spreading defect detection method

Abstract

The invention relates to the technical field of metal additive manufacturing, in particular to an SLM powder laying defect detection method which is characterized by comprising the following steps that firstly, a strip-shaped light-emitting diode serves as a light source of a detection system and irradiates the surface of a powder laying layer; 2, performing side axis acquisition on the image of the surface of the powder laying layer by adopting an industrial camera and a lens to obtain the image of the surface of the powder laying layer; 3, performing perspective transformation processing on the obtained surface image of the powder spreading layer; 4, performing powder spreading defect extraction on the corrected surface image of the powder spreading layer by adopting an image processing algorithm; and step 5, distinguishing the extracted different types of powder spreading defects by adopting a multi-layer sensor. The method is high in recognition rate and detection speed, the forming efficiency of selective laser melting equipment can be met, and the SLM forming process can be monitored to guarantee the quality of finished products.

Description

Technical Field

[0001] The present invention relates to the technical field of metal additive manufacturing, and in particular to a method for detecting SLM powder spreading defects. Background Art

[0002] SLM forming technology is a forming technology in the field of metal additive manufacturing. In order to improve the quality of formed parts, it is necessary to monitor the selective laser melting forming process. At present, the research on selective laser melting forming process monitoring is mostly focused on the monitoring of the molten pool, but it is impossible to control the quality of the formed parts in real time. As an indispensable link in the SLM forming process, the quality of powder laying directly affects the quality of the formed parts. Therefore, it is very necessary to monitor the SLM forming process to ensure the quality of the finished product. Summary of the invention

[0003] The technical problem to be solved by the present invention is to provide a method for detecting SLM powder spreading defects with high recognition rate and fast detection speed, which can not only meet the forming efficiency of the selective laser melting equipment, but also monitor the SLM forming process to ensure the quality of the finished product.

[0004] In order to solve the above technical problems, the technical solution adopted by the present invention is as follows: a method for detecting SLM powder spreading defects comprises the following steps:

[0005] Step 1: Use a strip light emitting diode as the light source of the detection system and illuminate the surface of the powder layer;

[0006] Step 2: Use an industrial camera and lens to collect the image of the surface of the powder layer from the side axis to obtain the surface image of the powder layer;

[0007] Step 3: Perform perspective transformation processing on the acquired surface image of the powder layer to correct the linear distortion of the surface image of the powder layer to avoid image distortion;

[0008] Step 4: Use image processing algorithm to extract powdering defects from the corrected powdering layer surface image, and include the following steps: a. Eliminate the influence of light; b. Threshold segmentation; c. Image denoising; Powdering defects include cladding layer increase defects, impurity accumulation defects and scraper damage defects, impurity accumulation defects and scraper damage defects are collectively referred to as streak defects;

[0009] Step 5: Use a multi-layer perceptron to distinguish the different types of powder spreading defects extracted.

[0010] Furthermore, the pixel size of the industrial camera is 3.45μm×3.45μm, the sensor area is 2 / 3, the resolution is 2448×2448, the frame rate is 23.5fps, the camera interface is Gigabit Ethernet, the lens format size is 2 / 3, the focal length is 8mm, the aperture range is 1.4-16, and the minimum object distance is 0.1mm.

[0011] Furthermore, the correction process of perspective transformation is to project each pixel of the acquired surface image of the powder layer onto a new plane at one time. The specific formula is as follows:

[0012]

[0013] Where (u, v) is the original coordinate of P, (u corr ,v corr ,z corr ) is P corr The coordinates of (h 11 ,h 12 ;h 21 ,h 22 ) is the linear transformation matrix, (h 13 ;h 23 ) is the translation matrix, (h 31 ,h 32 ) is the perspective matrix;

[0014] Since the final corrected surface image of the powder layer is located on a two-dimensional plane, P corr The coordinates of divided by z corr , the transformation formula is as follows:

[0015]

[0016] In the formula, (u ' corr ,v' corr ,1) is the final corrected coordinate.

[0017] Furthermore, the specific steps to eliminate the influence of light are as follows: a. Calculate the average gray value of the entire powder layer surface image; b. Use a 200×200 filter template to average filter the powder layer surface image; c. Subtract the original powder layer surface image from the average filtered powder layer surface image, and add the average gray value of the entire powder layer surface image; the above steps can be expressed as the following formula:

[0018] G out =G in -G mean +vIn the formula, G out is the output image, G in is the input image, G meanis the average image after filtering, and v is the average gray value of the entire image.

[0019] Furthermore, the threshold segmentation is to set a high threshold to segment the brighter part of the powdering defect, and set a low threshold to segment the darker part of the powdering defect. Among them, the defect of the cladding layer appears brighter in the surface image of the powdering layer, and the stripe defect appears darker in the surface image of the powdering layer. The formula of the threshold segmentation is as follows:

[0020]

[0021] Where G'(i,j) is the image after threshold segmentation, G(i,j) is the image after uneven illumination processing, H is the high threshold, and L is the low threshold.

[0022] Furthermore, in order to select the appropriate threshold under different lighting environments, the high threshold is set to u+k1σ and the low threshold is set to u-k2σ, where u is the grayscale average of the image, σ is the grayscale standard deviation, and k1 and k2 are both magnification factors.

[0023] Furthermore, image denoising includes performing denoising processing on the surface image of the powder layer using a morphological filter for small noise and a connected domain filter for large noise, wherein the surface image of the powder layer is opened by using a 3×3 structural element of the morphological filter to retain a 3×3 pixel area, and then the area with an area less than 30 pixels is deleted by the connected domain filter, thereby extracting defects added by the cladding layer; the surface image of the powder layer is opened by using a 3×3 structural element and a 1×5 structural element of the morphological filter respectively, and then the two open operation images are merged, and then the merged surface image of the powder layer is opened by using a 2×20 structural element of the morphological filter, and finally, the area with an area less than 1500 pixels or a width less than 150 pixels is deleted by the connected domain filter, thereby extracting stripe defects.

[0024] Furthermore, the multilayer perceptron uses the minimum pixel value, pixel standard deviation, defect height, and defect roundness value of the stripe defect as feature vectors to classify the impurity accumulation defect and scraper damage defect contained in the stripe defect. Therefore, the number of input layer neurons of the network structure of the multilayer perceptron is set to 4, the number of output layer neurons is set to 2, the maximum-minimum method is used for data normalization, the activation function is selected as softmax, and the number of hidden layer neurons is determined by an empirical formula. The specific empirical formula is as follows:

[0025] H = log2n

[0026] Where H is the number of neurons in the hidden layer, and n is the number of neurons in the input layer.

[0027] Furthermore, before the multilayer perceptron classifies the strip defects, multiple impurity accumulation defect images, multiple scraper damage defect images, multiple cladding layer increase defect images and multiple qualified images are selected as training samples to train the multilayer perceptron to verify the classification effect of the multilayer perceptron.

[0028] From the above description, it can be seen that the SLM powder defect detection method provided by the present invention has the following beneficial effects: by using an industrial camera and a lens to collect images of the surface of the powder layer to obtain the surface image of the powder layer, and designing a corresponding image processing algorithm to extract the powder defects in the surface image of the powder layer, and establishing a multi-layer sensor to distinguish different types of powder defects. Therefore, if powder defects are detected during the powdering process, the processing is stopped and corresponding adjustments are made. The detection method has a high recognition rate and a fast detection speed. It can not only meet the molding efficiency of the selective laser melting equipment (SLM), but also monitor the SLM formation process to ensure the quality of the finished product. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 This is a schematic diagram of the installation location of the industrial camera.

[0030] Figure 2 Schematic diagram of the installation position of the strip light emitting diode.

[0031] Figure 3 Schematic diagram of powder coating defects on the surface image of the powder coating layer, A is the defect of increased cladding layer, B is the defect of impurity accumulation, and C is the defect of scraper damage.

[0032] Figure 4 Schematic diagram of perspective transformation effect.

[0033] Figure 5 Flowchart for powder coating defect extraction.

[0034] Figure 6 This is the original powder layer surface image.

[0035] Figure 7 This is the surface image of the powder layer after eliminating the influence of light.

[0036] Figure 8 This is the surface image of the high threshold powder layer after threshold segmentation.

[0037] Fig. 9 This is the surface image of the low threshold powder layer after threshold segmentation.

[0038] Fig.10 The surface image of the powder layer with defects added by the cladding layer is extracted after image denoising.

[0039] Fig.11The surface image of the powder layer with stripe defects is extracted after image denoising.

[0040] Fig.12 This is the original powder layer surface image.

[0041] Fig.13 The extracted dust pavement surface image with stripe defects.

[0042] In the figure: 1-industrial camera; 2-laser galvanometer; 3-selective laser melting equipment; 4-strip light emitting diode; 5-vibrating mirror lens. DETAILED DESCRIPTION

[0043] The present invention is further described below through specific implementation modes.

[0044] The method for detecting SLM powder spreading defects of the present invention comprises the following steps:

[0045] Step 1: Use a strip light emitting diode as the light source of the detection system and illuminate the surface of the powder layer;

[0046] Step 2: Use the industrial camera 1 and the lens to collect the image of the surface of the powder layer from the side axis to obtain the surface image of the powder layer;

[0047] Step 3: Perform perspective transformation processing on the acquired surface image of the powder layer to correct the linear distortion of the surface image of the powder layer to avoid image distortion;

[0048] Step 4: extracting powdering defects from the corrected powdering layer surface image using an image processing algorithm, and the steps include: a. eliminating the influence of light; b. threshold segmentation; c. image denoising; the powdering defects include cladding layer increase defects, impurity accumulation defects and scraper damage defects, and the impurity accumulation defects and the scraper damage defects are collectively referred to as streak defects;

[0049] Step 5: Use a multi-layer perceptron to distinguish the different types of powder spreading defects extracted.

[0050] The image of the powder layer surface is collected by using an industrial camera 1 and a lens to obtain the surface image of the powder layer, and a corresponding image processing algorithm is designed to extract powder defects in the surface image of the powder layer, and a multi-layer sensor is established to distinguish different types of powder defects. Therefore, if powder defects are detected during the powdering process, the processing is stopped and corresponding adjustments are made. The detection method has a high recognition rate and a fast detection speed. It can not only meet the molding efficiency of the selective laser melting equipment 3 (SLM), but also monitor the SLM formation process to ensure the quality of the finished product.

[0051] Accordingly, the installation position of the strip light emitting diode (LED) is as follows Figure 2As shown, there is at least one strip-shaped light emitting diode 4 (LED) installed on the inner top surface of the cover of the selective laser melting device 3 (SLM) and located on one side of the vibration mirror lens 5 .

[0052] like Figure 3 As shown in the figure, since spheroidization often causes unevenness on the surface of parts, resulting in local high phenomena, during the diffusion process of metal powder, it is difficult for metal powder to cover the local high area, thus forming a defect of increasing the cladding layer; due to the excessive energy of the molten pool, the liquid metal evaporates to produce black smoke impurities. During the laser scanning process, due to the influence of the laser scanning direction and the laser scanning speed, a part of the black smoke impurities falling on the metal surface cannot be brought into the gas circulation system by the airflow generated by the fan, resulting in the accumulation of black smoke impurities. The scraper will drive such impurities, thus forming an impurity accumulation defect; due to the long-term friction between the scraper and the higher part of the coating, part of the scraper surface is missing. When the powder dispersing device is activated, part of the metal powder driven by the scraper will leak out from the position where the scraper belt is missing, thus forming a scraper damage defect.

[0053] In addition, the impurity accumulation defect and the scraper damage defect appear in the form of stripes on the surface image of the powder coating layer, and are therefore collectively referred to as stripe defects.

[0054] like Figure 1 As shown, since the laser galvanometer 2 has been directly installed on the selective laser melting device 3, the image of the powder surface can only be collected through the side axis. Since the molding bin size of the SLM selective laser melting device 3 used in the detection system of this application is 120×120×100 (mm), in order to improve the resolution of the system, the detection system must be able to distinguish details of at least 0.5 mm, and each detail needs to be composed of 4-8 pixels. In this application, the detection system takes 6 pixels corresponding to a 0.5 mm powder spreading defect. Therefore, the number of pixels in the length and width directions of the molding chamber needs to be greater than 1440, and due to the shooting angle of the industrial camera 1, the powder bed of the molding chamber will appear in the image in the shape of a diamond, so the final number of pixels in each direction of the industrial camera 1 needs to be greater than 1440×√2≈2037. For this reason, the pixel size of the industrial camera 1 is 3.45μm×3.45μm, the sensor area is 2 / 3, the resolution is 2448×2448, the frame rate is 23.5fps, and the camera interface is Gigabit Ethernet.

[0055] Since the distance from the industrial camera 1 to the surface of the powder layer is 215 mm, the magnification is β=6.6 / 120√2=0.0389, and the focal length of the objective lens is 215 / (1+1 / β)=8.05 mm. Therefore, the format size of the lens is 2 / 3, the focal length is 8 mm, the aperture range is 1.4-16, and the minimum object distance is 0.1 mm.

[0056] Preferably, the camera of the industrial camera 1 adopts a Hikvision MV-CA050-10GM camera with an IMX 264 sensor, and the lens adopts a FL-CC0814A-2M Ricoh 8mm lens.

[0057] Since the industrial camera 1 mainly adopts the side axis method, the collected surface image of the powder layer has certain linear distortion, which easily causes image distortion and increases the difficulty of powder defect detection. Therefore, corresponding measures need to be taken for correction. Specifically, the correction process of the perspective transformation is to project each pixel of the acquired surface image of the powder layer onto a new plane at one time. The specific formula is as follows:

[0058]

[0059] Where (u, v) is the original coordinate of P, (u corr ,v corr ,z corr ) is P corr The coordinates of (h 11 ,h 12 ;h 21 ,h 22 ) is the linear transformation matrix, (h 13 ;h 23 ) is the translation matrix, (h 31 ,h 32 ) is the perspective matrix;

[0060] Since the final corrected surface image of the powder layer is located on a two-dimensional plane, P corr The coordinates of divided by z corr , the transformation formula is as follows:

[0061]

[0062] In the formula, (u ' corr ,v' corr ,1) is the final corrected coordinate.

[0063] From the above formula, we can know that only 4 points are needed to list 8 equations and find the unknown quantities of 8 correction transformation matrices. Correspondingly, the perspective transformation effect is as follows: Figure 4 shown.

[0064] from Figure 3It can be seen that the image of the powdering surface is affected by the strip light emitting diode (LED) light, resulting in uneven image illumination. Therefore, measures should be taken to eliminate the influence of illumination to avoid increasing the difficulty of later image processing. The specific steps of eliminating the influence of light are as follows: a. Calculate the average grayscale value of the entire powdering layer surface image; b. Use a 200×200 filter template to average filter the powdering layer surface image; c. Subtract the original powdering layer surface image from the average filtered powdering layer surface image, and add the average grayscale value of the entire powdering layer surface image; the above steps can be expressed as the following formula:

[0065] G out =G in -G mean +v

[0066] In the formula, G out is the output image, G in is the input image, G mean is the average image after filtering, and v is the average gray value of the entire image.

[0067] from Figure 3 It can be seen that the brightness of the cladding layer increase defect and the stripe defect are inconsistent, wherein the cladding layer increase defect appears brighter in the powder layer surface image, and the stripe defect appears darker in the powder layer surface image. Therefore, the threshold segmentation is to set a high threshold to segment the brighter part of the powder defect, and set a low threshold to segment the darker part of the powder defect. The formula of the threshold segmentation is as follows:

[0068]

[0069] Where G'(i,j) is the image after threshold segmentation, G(i,j) is the image after uneven illumination processing, H is the high threshold, and L is the low threshold.

[0070] In order to select the appropriate threshold under different lighting environments, the high threshold is set to u+k1σ and the low threshold is set to u-k2σ, where u is the grayscale average of the image, σ is the grayscale standard deviation, and k1 and k2 are both magnification factors.

[0071] Since the SLM selective laser melting device 3 prints fine metal powder, the fine metal powder spread on the powder bed is in a loose state, so the reflection state of light is different, which will cause a lot of noise in the surface image of the powder layer. It is necessary to perform noise reduction on the surface image of the powder layer. Accordingly, the image denoising includes using a morphological filter for small noise and a connected domain filter for large noise to perform noise reduction on the surface image of the powder layer, wherein the surface image of the powder layer is opened by using the 3×3 structural element of the morphological filter to retain the pixel area containing 3×3, and then the area with an area less than 30 pixels is deleted by the connected domain filter, so as to extract the defect of the cladding layer increase; the surface image of the powder layer is opened by using the 3×3 structural element and the 1×5 structural element of the morphological filter, respectively, and then the two open operation images are merged, and then the merged surface image of the powder layer is opened by using the 2×20 structural element of the morphological filter. Finally, the area with an area less than 1500 pixels or a width less than 150 pixels is deleted by the connected domain filter, so as to extract the stripe defect.

[0072] Correspondingly, the powder spreading defect extraction process is as follows Figure 5 As shown in Figure 2, the powder spreading defect extraction effect is as follows: Figures 6 to 13 shown, and Figure 6 The original powder layer surface image contains defects added by the cladding layer and impurity accumulation defects. Fig.12 This is the surface image of the original powder coating layer containing scraper damage defects.

[0073] The multilayer perceptron uses the minimum pixel value, pixel standard deviation, defect height, and defect roundness value of the stripe defect as feature vectors to classify the impurity accumulation defect and the scraper damage defect contained in the stripe defect. Therefore, the number of input layer neurons of the network structure of the multilayer perceptron (MLP) is set to 4, the number of output layer neurons is set to 2, the maximum-minimum method is used for data normalization, the activation function is selected as softmax, and the number of hidden layer neurons is determined by an empirical formula. The specific empirical formula is as follows:

[0074] M = log2n

[0075] Where H is the number of neurons in the hidden layer, and n is the number of neurons in the input layer.

[0076] Therefore, the number of neurons in the hidden layer is determined to be 2 by an empirical formula.

[0077] Before the multilayer perceptron classifies the strip defects, a plurality of impurity accumulation defect images, a plurality of scraper damage defect images, a plurality of cladding layer increase defect images and a plurality of qualified images are selected as training samples to train the multilayer perceptron to verify the classification effect of the multilayer perceptron.

[0078] Preferably, the training samples include 40 impurity accumulation defect images, 40 scraper damage defect images, 20 cladding layer increase defect images and 20 qualified images.

[0079] The confusion matrix of the multilayer perceptron classification results is shown in the following table:

[0080]

[0081] The above are only some specific implementation methods of the present invention, but the design concept of the present invention is not limited thereto. Any non-substantial changes to the present invention using this concept shall be deemed as an infringement of the protection scope of the present invention.

Claims

1. A method for detecting SLM powder spreading defects, characterized by: The steps include: Step 1: Use a strip light emitting diode as the light source of the detection system and illuminate the surface of the powder layer; Step 2: Use an industrial camera and lens to collect the image of the surface of the powder layer from the side axis to obtain the surface image of the powder layer; Step 3: Perform perspective transformation processing on the acquired surface image of the powder layer to correct the linear distortion of the surface image of the powder layer to avoid image distortion; Step 4: extracting powdering defects from the corrected powdering layer surface image using an image processing algorithm, and the steps include: a. eliminating the influence of light; b. threshold segmentation; c. image denoising; the powdering defects include cladding layer increase defects, impurity accumulation defects and scraper damage defects, and the impurity accumulation defects and the scraper damage defects are collectively referred to as streak defects; Step 5: Use a multi-layer perceptron to distinguish the different types of powder spreading defects extracted.

2. The method for detecting SLM powder spreading defects according to claim 1, characterized in that: The pixel size of the industrial camera is 3.45μm×3.45μm, the sensor area is 2 / 3, the resolution is 2448×2448, the frame rate is 23.5fps, the camera interface is Gigabit Ethernet, the format size of the lens is 2 / 3, the focal length is 8mm, the aperture range is 1.4-16, and the minimum object distance is 0.1mm.

3. The method for detecting SLM powder spreading defects according to claim 1, characterized in that: The correction process of the perspective transformation is to project each pixel of the obtained surface image of the powder layer onto a new plane at one time. The specific formula is as follows: Where (u, v) is the original coordinate of P, (u corr ,v corr ,z corr ) is P corr The coordinates of (h 11 ,h 12 ;h 21 ,h 22 ) is the linear transformation matrix, (h 13 ;h 23 ) is the translation matrix, (h 31 ,h 32 ) is the perspective matrix; Since the final corrected surface image of the powder layer is located on a two-dimensional plane, P corr The coordinates of divided by z corr , the transformation formula is as follows: In the formula, (u ' corr ,v' corr ,1) is the final corrected coordinate.

4. The method for detecting SLM powder spreading defects according to claim 1, characterized in that: The specific steps of eliminating the influence of light are as follows: a. Calculate the average gray value of the entire powder layer surface image; b. Use a 200×200 filter template to average filter the powder layer surface image; c. Subtract the original powder layer surface image from the average filtered powder layer surface image, and add the average gray value of the entire powder layer surface image; the above steps can be expressed as the following formula: G out =G in -G mean +v In the formula, G out is the output image, G in is the input image, G mean is the average image after filtering, and v is the average gray value of the entire image.

5. The method for detecting SLM powder spreading defects according to claim 1, characterized in that: The threshold segmentation is to set a high threshold to segment the brighter part of the powdering defect, and set a low threshold to segment the darker part of the powdering defect, wherein the cladding layer increase defect appears brighter in the powdering layer surface image, and the stripe defect appears darker in the powdering layer surface image. The formula of the threshold segmentation is as follows: Where G'(i,j) is the image after threshold segmentation, G(i,j) is the image after uneven illumination processing, H is the high threshold, and L is the low threshold.

6. The method for detecting SLM powder spreading defects according to claim 5, characterized in that: In order to select the appropriate threshold under different lighting environments, the high threshold is set to u+k1σ and the low threshold is set to u-k2σ, where u is the grayscale average of the image, σ is the grayscale standard deviation, and k1 and k2 are both magnification factors.

7. The method for detecting SLM powder spreading defects according to claim 1, characterized in that: The image denoising includes performing denoising on the surface image of the powder layer using a morphological filter for small noise and a connected domain filter for large noise, wherein the surface image of the powder layer is opened by using a 3×3 structural element of the morphological filter to retain a 3×3 pixel area, and then the area with an area less than 30 pixels is deleted by the connected domain filter, thereby extracting defects of the cladding layer; the surface image of the powder layer is opened by using the 3×3 structural element and the 1×5 structural element of the morphological filter respectively, and then the two opened operation images are merged, and then the merged surface image of the powder layer is opened by using the 2×20 structural element of the morphological filter, and finally, the area with an area less than 1500 pixels or a width less than 150 pixels is deleted by the connected domain filter, thereby extracting stripe defects.

8. The method for detecting SLM powder spreading defects according to claim 1, characterized in that: The multilayer perceptron uses the minimum pixel value, pixel standard deviation, defect height, and defect roundness value of the stripe defect as feature vectors to classify the impurity accumulation defect and the scraper damage defect contained in the stripe defect. Therefore, the number of input layer neurons of the network structure of the multilayer perceptron is set to 4, the number of output layer neurons is set to 2, the maximum-minimum method is used for data normalization, the excitation function is selected as softmax, and the number of hidden layer neurons is determined by an empirical formula. The specific empirical formula is as follows: H = log2n Where H is the number of neurons in the hidden layer, and n is the number of neurons in the input layer.

9. The method for detecting SLM powder spreading defects according to claim 1, characterized in that: Before the multilayer perceptron classifies the strip defects, a plurality of impurity accumulation defect images, a plurality of scraper damage defect images, a plurality of cladding layer increase defect images and a plurality of qualified images are selected as training samples to train the multilayer perceptron to verify the classification effect of the multilayer perceptron.

Images