Part surface defect online monitoring and quality control method in laser powder bed melting process
By using an industrial camera and convolutional neural network to monitor and control surface defects of parts in real time during the laser powder bed melting process, the problem of lack of real-time monitoring and feedback control in the prior art is solved. This enables quantitative monitoring and adaptive adjustment of surface defects of parts, thereby improving production efficiency and part quality.
Patent Information
- Application Number
- CN202410480829.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-04-22
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2044-04-22
AI Technical Summary
Existing laser powder bed melting technology lacks effective online real-time monitoring and feedback control methods during the manufacturing process, making it difficult to detect and mitigate surface defects in parts in a timely manner, thus affecting part quality and production efficiency.
An industrial camera is used to acquire powder-coated images of the part surface layer by layer. Defects are segmented and quantified using the Unet semantic segmentation model. Combined with the melting state recognition of laser scanning images, a three-classification four-layer convolutional neural network is used to determine the melting state, thereby realizing feedback control for adjusting process parameters layer by layer.
It enables quantitative monitoring and adaptive adjustment of surface defects in parts, reduces manual intervention, improves the intelligence of the production process and the stability of part quality, and enhances production efficiency.
Smart Images

Figure CN118926551A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of metal additive manufacturing monitoring, and in particular to a method for monitoring defect lines on a part surface and controlling quality during a laser powder bed melting process. Background Art
[0002] Laser Powder Bed Fusion (LPBF) is a leading additive manufacturing technology widely used in the manufacturing industry. Based on the CAD model, the technology uses laser to melt and solidify metal powder materials layer by layer to build the physical structure of the part layer by layer. Compared with traditional manufacturing processes, LPBF has many advantages such as high design freedom, short product cycle and high material utilization rate. It has been widely used in fields such as medical implants and aerospace engine design. However, during the LPBF molding process, a series of defects such as spheroidization, spattering, swelling and surface roughness are accompanied. These defects seriously reduce the quality of the inner surface of the part, adversely affect the powder distribution in the subsequent layers, and damage the fusion between the layers. In particular, the collision between the powder spreading vehicle and the rough inner surface may cause damage to the parts and the recoating mechanism, which will cause a huge waste of manpower and materials.
[0003] Online real-time in-situ monitoring can detect defects in a timely manner and issue early warnings on the degree of defects, while feedback control can dynamically adjust the process parameters of subsequent layers based on the monitored signals. This automated process does not require human intervention and can monitor the printing status of parts in real time and adjust the parameters of subsequent layers in a timely manner, effectively reducing material waste and improving part surface quality, thereby enhancing the intelligence level of LPBF equipment and promoting the development of intelligent manufacturing.
[0004] In the existing monitoring technology, such as the 3D printing process fault diagnosis method and device with patent publication number CN109968671B, the method monitors the temperature of the current layer in real time and uses a machine learning model to determine whether there is a fault in the printing of the current layer during the fused deposition modeling printing process, and provides fault type information; however, the metal additive manufacturing process is more complicated, especially the LPBF process needs to be carried out in a closed cabin and under the protection of inert gas, which makes monitoring and control very difficult;
[0005] In addition, for example, the online monitoring method for laser selective melting defects with patent publication number CN116740001A uses industrial cameras to collect images after powder spreading and laser scanning, and realizes online defect classification by combining the collected images and process parameter information; for example, the online monitoring method, device, equipment and medium for laser powder bed melting defects with patent publication number CN117390433A. This method converts air-borne acoustic emission signals into a spectrum diagram, and characterizes the defect information characteristics by calculating the signal frequency amplitude density ratio of the low frequency band, medium frequency band and high frequency band of the spectrum; for example, the online monitoring system and online monitoring method for powder additive manufacturing defects with patent publication number CN115582559A use acoustic emission technology to realize online monitoring of LPBF process defects, and use acoustic signals to obtain part surface defects in real time; however, the above monitoring methods have not yet realized the work of realizing layer-by-layer feedback control to alleviate the degree of defects based on monitoring signals. Summary of the invention
[0006] In order to overcome the defects and shortcomings of the prior art, the present invention provides a method for monitoring and controlling the surface defect lines of parts in a laser powder bed melting process. The present invention identifies the melting state layer by layer and quantifies the degree of defects based on an industrial camera. Based on the monitored signal, a feedback control method is used to adjust the process parameters of subsequent layers layer by layer, thereby improving part quality and production efficiency.
[0007] In order to achieve the above object, the present invention adopts the following technical solutions:
[0008] The present invention provides a method for monitoring and controlling the surface defect line of a part in a laser powder bed melting process, comprising the following steps:
[0009] During the processing, the powder spreading image of the part surface and the image after laser scanning are collected layer by layer, and the powder spreading image of the part surface is enhanced to obtain the enhanced powder spreading image of the part surface;
[0010] Construct and train a Unet semantic segmentation model for part surface defect segmentation, input the enhanced part surface powder spreading image into the trained Unet semantic segmentation model, segment the surface defect area, and generate a mask image;
[0011] Quantify the surface defects and obtain the surface defect quantification index Ap;
[0012] The image after laser scanning is input into a three-classification four-layer convolutional neural network for judging the melting state of the part surface to obtain a signal M;
[0013] The surface defect quantitative index Ap and signal M are input into the proportional control system to calculate the process parameters required for the next layer, and the process parameters are adjusted layer by layer until the final forming.
[0014] As a preferred technical solution, the image enhancement process adopts limited contrast histogram equalization process.
[0015] As a preferred technical solution, the Unet semantic segmentation model includes a feature extraction part and an upsampling part;
[0016] The feature extraction part extracts features from the powder spreading image of the part surface through the convolution operation of the convolution module, and uses maximum pooling downsampling to reduce the size of the feature map;
[0017] The upsampling part uses a transposed convolution operation to upsample the feature map to enlarge the size of the feature map;
[0018] Use skip connections to concatenate the feature maps obtained by the maximum pooling downsampling operation with the feature maps of the upsampling operation;
[0019] The feature map obtained by the last upsampling and the feature map obtained by the first feature extraction are concatenated, the feature map is processed by the convolution operation of the convolution module, the number of channels is adjusted by 1x1 convolution, and a single-channel feature for classification is obtained. The sigmoid function is used to process the feature map to obtain a pixel-by-pixel classification result of the same size as the original input image, and the final mask map is generated to achieve semantic segmentation of the image.
[0020] As a preferred technical solution, the convolution module includes two convolution layers, namely a first convolution layer with a kernel function size of 3×3, a step size of 1, and a padding of 1, and a second convolution layer with a kernel function size of 3×3, a step size of 1, and a padding of 1. Each convolution layer is followed by a batch normalization operation and a leaky ReLu activation. The input of the second convolution layer is the output of the first convolution layer, and the input of the first convolution layer is a feature map obtained by downsampling the output of the previous convolution module through maximum pooling.
[0021] As a preferred technical solution, the feature extraction part performs feature extraction on the part surface powder image through the convolution operation of the convolution module, and performs feature extraction on the part surface powder image through multiple feature extraction operations, and the number of convolution kernels of the convolution module corresponding to each feature extraction operation is different.
[0022] As a preferred technical solution, the upsampling part uses a transposed convolution with a kernel size of 2×2 and a step size of 2 to perform an upsampling operation on the feature map. The number of transposed convolutions used in each upsampling operation is different. The new feature map obtained is channel-joined with the feature map of the same size obtained by the feature extraction part through a jump connection to obtain a fused feature map.
[0023] The convolution module performs convolution operations on the fused feature maps, and the number of convolution modules used each time is different.
[0024] As a preferred technical solution, the image after laser scanning is input into a three-classification four-layer convolutional neural network for judging the melting state of the part surface to obtain a signal M, which specifically includes:
[0025] The image after laser scanning undergoes four feature extraction operations in sequence. Each feature extraction operation includes convolution, batch normalization, Leaky ReLU activation and maximum pooling downsampling operation. The extracted features undergo global average pooling operation and 3 convolution layers with a kernel size of 1×1, and the probability value of the melting state is output through the softmax function.
[0026] As a preferred technical solution, the surface defects are quantified to obtain a quantitative index of surface defects, which is specifically expressed as:
[0027]
[0028] Where R and C are the width and length of the original image, respectively. II[·] is an indicator function used to determine whether the predicted probability of a pixel exceeds a predefined threshold α. If it exceeds the threshold α, II(P(r,c)) = 1, otherwise it is 0.
[0029] As a preferred technical solution, the probability value of the melting state is output through the softmax function, and the maximum probability value is selected as the final result to obtain the M value. The calculation formula is expressed as:
[0030]
[0031] Among them, P 欠熔化 , P 正常 , P 过熔化 They represent the probability values of under-melting, normal melting, and over-melting respectively.
[0032] As a preferred technical solution, the surface defect quantitative index Ap and signal M are input into the proportional control system to calculate the process parameters required for the next layer, and the process parameters are adjusted layer by layer, specifically including:
[0033] P 下一层 =P 当前层 +M×Kp×Error-(1-M)Kp×Error
[0034] Among them, P 下一层 is the laser power required for the next layer, P 当前层 is the laser power of the current layer, M is a Boolean value used to indicate the current melting state, that is, if the current melting state is over-melting, M=1, otherwise M=0, Kp is the proportional coefficient, and Error indicates the error between the current monitoring signal Ap and the expected value.
[0035] Compared with the prior art, the present invention has the following advantages and beneficial effects:
[0036] (1) Compared with the limitation of the prior art that the online monitoring of laser powder bed defects based on vision can only classify defects but cannot quantify defects layer by layer, the present invention uses an industrial camera to collect the images of the surface of the part after powdering during the processing layer by layer, and directly inputs them into the semantic segmentation model for defect segmentation to obtain the quantitative index of surface defects; it can automatically determine the melting state of the part according to the image after laser scanning, and adaptively adjust the direction of the laser process parameters according to the melting state, which can effectively alleviate the surface defects of the subsequent layer parts, making the production process more intelligent and adaptive;
[0037] (2) Traditional online monitoring systems may require manual intervention to adjust process parameters, while the present invention utilizes the information of online monitoring and feedback control theory to achieve automatic adjustment of process parameters during the LPBF process, thereby reducing the need for manual intervention, improving production efficiency and ensuring the stability of part quality. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] Figure 1 It is a schematic flow chart of the method for monitoring and controlling the surface defect line of a part in the laser powder bed melting process of the present invention;
[0039] Figure 2 It is a schematic diagram of the overall device of the present invention for taking layer-by-layer images of the surface of a part after powder coating and laser scanning using an industrial camera;
[0040] Figure 3 (a) is a schematic diagram of the original image of the powder coating on the part surface;
[0041] Figure 3 (b) is a schematic diagram of the powder spreading image on the surface of the enhanced part;
[0042] Figure 3 (c) is a schematic diagram of the mask image after defect segmentation;
[0043] Figure 4 It is a schematic diagram of the architecture of the Unet semantic segmentation model for part surface defect segmentation of the present invention;
[0044] Figure 5 (a) is a schematic diagram of the architecture of the convolution module of the Unet semantic segmentation model of the present invention;
[0045] Figure 5 (b) is a schematic diagram of the architecture of the maximum pooling downsampling of the Unet semantic segmentation model of the present invention;
[0046] Figure 5 (c) is a schematic diagram of the architecture of the transposed convolution upsampling of the Unet semantic segmentation model of the present invention;
[0047] Figure 6 (a) is a schematic diagram of an image of an under-melted state of a part surface after laser scanning according to the present invention;
[0048] Figure 6 (b) is a schematic diagram of the image of the normal melting state of the part surface after laser scanning of the present invention;
[0049] Figure 6 (c) is a schematic diagram of an image of an over-melted state of a part surface after laser scanning according to the present invention;
[0050] Figure 7 Schematic diagram of the architecture of a four-layer convolutional neural network for identifying the melting state of parts according to the present invention. DETAILED DESCRIPTION
[0051] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.
[0052] Example
[0053] like Figure 1 As shown, this embodiment provides a method for monitoring and controlling the surface defect line of a part in a laser powder bed melting process, comprising the following steps:
[0054] S1: During the processing, the powder spreading image of the part surface and the image after laser scanning are collected layer by layer, and the powder spreading image of the part surface is enhanced to obtain the enhanced powder spreading image of the part surface;
[0055] like Figure 2 As shown, according to the part position information, an industrial camera is used to take layer-by-layer images of the part surface after powder coating and laser scanning during the manufacturing process;
[0056] After the powder is applied, the clarity of the background and the target in the original image is insufficient, such as Figure 3 As shown in (a), in order to enhance the contrast between the background and the target and reduce the influence of lighting, the image enhancement process uses limited contrast histogram equalization (CLAHE). CLAHE effectively reduces image noise and better distinguishes the defect area from the background area. Figure 3 (b) as shown.
[0057] S2: Input the enhanced part surface powder spreading image into the trained surface defect online monitoring neural network model to segment the surface defect area and generate a mask image, such as Figure 3 (c)
[0058] In this embodiment, the surface defect online monitoring neural network model adopts the Unet semantic segmentation model for part surface defect segmentation, which includes a feature extraction part and an upsampling part, and the feature map obtained by the corresponding feature extraction and the feature map obtained by upsampling are channel-joined through jump connections. Specifically, the feature extraction part extracts features from the part surface powder image through convolution operation, and uses the maximum pooling downsampling to reduce the size of the feature map and increase the abstract level of the feature; the upsampling part uses the transposed convolution operation to upsample the feature map, enlarges the size of the feature map, and uses the jump connection to channel-join the feature map obtained by the maximum pooling downsampling operation and the feature map of the upsampling operation to retain more detail information; finally, the feature map obtained by the last upsampling and the feature map obtained by the first feature extraction are channel-joined, the feature map is processed by the convolution module, and then the number of channels is adjusted by 1x1 convolution to obtain a single-channel feature for classification, and then the feature is processed by the sigmoid function to obtain a pixel-by-pixel classification result of the same size as the original input image, and the final mask map is generated to achieve semantic segmentation of the image.
[0059] The input of the feature extraction part is the original part surface image, and the output is a feature map with lower resolution and high-level feature representation; the input of the upsampling part is the feature map of the final output of the feature extraction part, and the output is a feature map with higher resolution and more semantics. The network structure is as follows Figure 4 As shown, it has the following characteristics:
[0060] (1) The input of the surface defect online monitoring neural network model for online segmentation and quantification of surface defects is the powder spreading image of the part surface. In order to adapt to the semantic segmentation network that has been trained in this embodiment, the input image needs to be grayscaled and resized to 224×224;
[0061] (2) The feature extraction part uses convolution modules and maximum pooling operations to extract image features and achieve dimension reduction of feature size. Specifically, the convolution module is first used to extract image features, and then the maximum pooling operation is used to retain important information and reduce the size of the feature map;
[0062] (3) The upsampling part uses transposed convolution upsampling and convolution modules to increase the resolution of features and improve the network's expressiveness. A transposed convolution operation with a kernel size of 2×2 and a step size of 2 is used. Two blank pixels are inserted between adjacent elements on the feature map and a convolution operation is performed to enlarge the image, thereby increasing the size of the feature map. The size of the feature map after transposed convolution is:
[0063] H 输出 =(H 输入 -1)×step length+K-2×filling
[0064] Among them, K is the convolution kernel size. Subsequently, the convolution module further processes the enlarged feature map to extract richer and more complex features, thereby enhancing the expressive ability of the network. This combination enables the network to better capture the detailed information in the image and improves the accuracy and performance of surface defect segmentation.
[0065] (4) Both the feature extraction part and the upsampling part use the same type of convolutional module, such as Figure 5 As shown in (a), each convolution module includes two convolution layers, namely, a first convolution layer with a kernel function size of 3×3, a stride of 1, and a padding of 1, and a second convolution layer with a kernel function size of 3×3, a stride of 1, and a padding of 1. Each convolution layer is followed by a batch normalization operation and a leaky ReLu activation, and the number of convolution kernels used in the two convolution layers is consistent; the first convolution layer of this embodiment is used to extract low-level features of the input, such as the edge and surface texture of the defect, and the second convolution layer is used to extract high-level features of the input, such as the shape of the defect; the batch normalization layer is used to normalize the input of each channel to avoid some gradient disappearance or gradient explosion; the leaky ReLu activation is used to increase the nonlinear expression ability of the neural network, which helps the network learn more complex functional relationships.
[0066] The upsampling part performs a transposed convolution upsampling operation on the feature map, such as Figure 5 As shown in (c), the resolution of the feature map is doubled by using a transposed convolution kernel with a parameter of 2×2, a stride of 2, and a padding of 0. Then, the upsampled feature map is concatenated with the downsampled feature map of the feature extraction part through a skip connection, and the concatenated feature map is convolved using a convolution module.
[0067] The second convolution input of each convolution module is the output of the first convolution layer. Specifically, the input of the first convolution module in the feature extraction part is the powder image of the part surface; the extracted features of the first convolution layer are used as the input of the second convolution layer, and the input of the subsequent convolution module is the reduced-dimensional feature map obtained after the previous convolution module is down-sampled by maximum pooling; the input of the upsampling part is the output of the last convolution module in the feature extraction part, and the input of the convolution module in the upsampling part is the new feature map obtained by splicing two feature map channels. These two feature maps are the feature map obtained by upsampling and the features output by the convolution module in the feature extraction part.
[0068] In the feature extraction part, the maximum pooling downsampling is used, such as Figure 5 As shown in (b), the convolution kernel with parameters of 2×2, stride of 2, and padding of 0 is used to reduce the dimension of the feature map output by each convolution module. Then, the feature map after dimension reduction will be input into the next convolution module.
[0069] (5) The process of feature extraction is as described in (2) and (4), except that the number of convolution kernels contained in the convolution module used in each feature extraction operation is different. The number of convolution kernels contained in the convolution module used in the five feature extraction operations is 32, 64, 128, 256, and 512 respectively.
[0070] (6) The processing flow of the upsampling part is as described in (3). The upsampling part uses a transposed convolution with a kernel size of 2×2 and a step size of 2 to upsample the feature map, doubling the size of the feature map. The difference is that the number of transposed convolutions used in each upsampling operation is different. According to the upsampling order, the number of transposed convolutions used is 256, 128, 64, and 32 respectively.
[0071] In addition, the obtained new feature map is channel-joined with the feature map of the same size obtained by the feature extraction operation through skip connection to obtain a fused feature map;
[0072] The number of convolution kernels contained in the convolution module used for the feature map after channel splicing is different. The number of convolution kernels contained in the convolution module used in the four convolution operations are 256, 128, 64, and 32 respectively.
[0073] (7) Finally, the 32-channel features obtained by the first convolution module of the feature extraction part and the 32-channel features obtained by the last upsampling operation are spliced, and the spliced feature map is operated using a convolution module containing 32 kernel functions; the number of channels is adjusted through 1×1 convolution to obtain a single-channel feature for classification, and then this feature is processed through the sigmoid function to obtain a pixel-by-pixel classification result of the same size as the original input image, and a mask map is generated.
[0074] S3: Quantify the surface defects to obtain the monitoring signal Ap, i.e., the quantitative index of the surface defects;
[0075] After completing the surface defect segmentation, it is necessary to quantify the degree of surface defects and use it as a monitoring signal to input into the control system. The number of all segmented pixel points in the semantic segmentation mask image is counted, and the proportion of the number of defect points to the entire image is calculated to obtain the surface defect quantification index Ap, which is achieved through the following formula:
[0076]
[0077] Where R and C are the width and length of the original image, respectively, and II[·] is an indicator function used to determine which pixels’ predicted probability exceeds a predefined threshold α. If it exceeds the threshold α, then II(P(r,c)) = 1, otherwise it is 0.
[0078] S4: The image after laser scanning is input into a three-classification four-layer convolutional neural network for judging the melting state of the part surface to obtain a signal M; Figure 6 (a) Figure 6 (b) and Figure 6 As shown in (c), the three different melting states after laser scanning are under-melting state, normal melting state, and over-melting state. The three-classification four-layer convolutional neural network has the following characteristics:
[0079] (1) The image used for surface melting state recognition is a laser scanned image. In order to adapt to the trained convolutional neural network in this embodiment, the input laser scanned image needs to be a grayscale image with no size restriction.
[0080] (2) The surface melting state recognition of this embodiment adopts a four-layer convolutional neural network, whose input is the image taken after laser scanning, and undergoes four feature extraction operations. Among them, the input of the first feature extraction operation is the image taken after laser scanning, and the input of the remaining feature extraction operations are the output of the previous feature extraction operation. Each feature extraction operation includes convolution, batch normalization, leaky ReLU activation and maximum pooling operation. The parameters of the convolution operation are: convolution with a convolution kernel size of 3×3, a step size of 1, and a padding of 1; a maximum pooling operation with a sampling window size of 2×2 and a step size of 2; the number of convolution kernels used in the four feature extraction operations is 32, 64, 128 and 256 respectively. Finally, the extracted features are subjected to a global average pooling operation and three convolution layers with a kernel size of 1×1, and the probability value of each category (normal melting, over-melting, under-melting) is output through a softmax function.
[0081] like Figure 7 As shown in the figure, the size of each convolution kernel is 3×3, the stride is 1, and the padding is 1. Convolution operation, batch normalization, Leaky ReLu activation, and maximum pooling downsampling with a size of 2×2, a stride of 2, and a padding of 0 are performed in sequence. The number of convolution kernels used in these four convolution layers is 32, 64, 128, and 256 respectively. The extracted features are subjected to global average pooling sampling operation, and then three 1×1 convolution layers are used to adjust the number of channels. The probability value of the melting state is calculated by the SoftMax function, and the maximum probability value is selected as the final result to obtain the M value. The calculation formula is as follows:
[0082]
[0083] S5: Input the monitoring signal Ap and signal M into the proportional control system, calculate the process parameters required for the next layer, and adjust the process parameters layer by layer;
[0084] In this embodiment, for the layer-by-layer monitoring and control of part surface defects in LPBF, the controller adopts proportional control. The signal Ap detected for each layer is used to determine whether the area of the surface without defects has reached a preset value, and proportional control is performed based on the error between it and the preset value; the detected signal M is used to determine whether the part surface has melted. Part surface defects are usually caused by over-melting or under-melting, and the control directions for over-melting and under-melting are different. Therefore, it is possible to determine whether to increase or decrease the laser power for the subsequent layer based on M. During the processing, first, it is determined whether the surface of the metal part meets the desired requirements. If the requirements are met, that is, when Ap > E (expected value), the process parameters for the subsequent layer are kept unchanged. If Ap < E, the power for the next layer is updated according to the following formula:
[0085] P 下一层 = P 当前层 + M × Kp × Error - (1 - M)Kp × Error
[0086] Wherein, P 下一层 is the laser power required for the next layer, P 当前层 is the laser power of the current layer, M is a Boolean value used to represent the current melting state, that is, if the current melting state is over-melting, M = 1, otherwise M = 0, Kp is the proportionality coefficient, which can be adjusted according to the material and the LPBF equipment itself. Taking the Di-Metal 300 equipment produced by Guangzhou Leijia Additive Manufacturing Company and using 316L as the processing material as an example, Kp can be set to 180, and Error represents the error between the currently detected signal Ap and the expected value.
[0087] The above embodiments are the preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications made without departing from the spirit and principle of the present invention shall be equivalent replacement methods and are all included in the protection scope of the present invention.
Claims
1. A method for monitoring and controlling surface defect lines of parts in a laser powder bed melting process, characterized in that: The steps include: During the processing, the powder spreading image of the part surface and the image after laser scanning are collected layer by layer, and the powder spreading image of the part surface is enhanced to obtain the enhanced powder spreading image of the part surface; Construct and train a Unet semantic segmentation model for part surface defect segmentation, input the enhanced part surface powder spreading image into the trained Unet semantic segmentation model, segment the surface defect area, and generate a mask image; Quantify the surface defects and obtain the surface defect quantification index Ap; The image after laser scanning is input into a three-classification four-layer convolutional neural network for judging the melting state of the part surface to obtain a signal M; The surface defect quantitative index Ap and signal M are input into the proportional control system to calculate the process parameters required for the next layer, and the process parameters are adjusted layer by layer until the final forming.
2. A method for monitoring and controlling surface defect lines of parts in a laser powder bed melting process, characterized in that: The image enhancement process adopts limited contrast histogram equalization process.
3. The method for monitoring and controlling the surface defect line of a part during laser powder bed melting according to claim 1, characterized in that: The Unet semantic segmentation model includes a feature extraction part and an upsampling part; The feature extraction part extracts features from the powder spreading image of the part surface through the convolution operation of the convolution module, and uses maximum pooling downsampling to reduce the size of the feature map; The upsampling part uses a transposed convolution operation to upsample the feature map to enlarge the size of the feature map; Use skip connections to concatenate the feature maps obtained by the maximum pooling downsampling operation with the feature maps of the upsampling operation; The feature map obtained by the last upsampling and the feature map obtained by the first feature extraction are concatenated, the feature map is processed by the convolution operation of the convolution module, the number of channels is adjusted by 1x1 convolution, and a single-channel feature for classification is obtained. The sigmoid function is used to process the feature map to obtain a pixel-by-pixel classification result of the same size as the original input image, and the final mask map is generated to achieve semantic segmentation of the image.
4. The method for monitoring and controlling the surface defect lines of parts in a laser powder bed melting process according to claim 3, characterized in that: The convolution module includes two convolution layers, namely a first convolution layer with a kernel function size of 3×3, a stride of 1, and a padding of 1, and a second convolution layer with a kernel function size of 3×3, a stride of 1, and a padding of 1. Each convolution layer is followed by a batch normalization operation and a leaky ReLu activation. The input of the second convolution layer is the output of the first convolution layer, and the input of the first convolution layer is a feature map obtained by downsampling the output of the previous convolution module through maximum pooling.
5. The method for monitoring and controlling the surface defect line of a part during laser powder bed melting according to claim 3, characterized in that: The feature extraction part performs feature extraction on the part surface powder image through the convolution operation of the convolution module, and performs feature extraction on the part surface powder image through multiple feature extraction operations, and the number of convolution kernels of the convolution module corresponding to each feature extraction operation is different.
6. The method for monitoring and controlling the surface defect line of a part during laser powder bed melting according to claim 3, characterized in that: The upsampling part uses a transposed convolution with a kernel size of 2×2 and a step size of 2 to upsample the feature map. The number of transposed convolutions used in each upsampling operation is different. The new feature map obtained is channel-joined with the feature map of the same size obtained by the feature extraction part through a skip connection to obtain a fused feature map. The convolution module performs convolution operations on the fused feature maps, and the number of convolution modules used each time is different.
7. The method for monitoring and controlling the surface defect lines of parts in a laser powder bed melting process according to claim 1, characterized in that: The surface defects are quantified to obtain the quantitative index of surface defects, which is specifically expressed as: Where R and C are the width and length of the original image, respectively. II[·] is an indicator function used to determine whether the predicted probability of a pixel exceeds a predefined threshold α. If it exceeds the threshold α, II(P(r,c)) = 1, otherwise it is 0.
8. The method for monitoring and controlling the surface defect lines of parts in a laser powder bed melting process according to claim 1, characterized in that: The image after laser scanning is input into a three-classification four-layer convolutional neural network for judging the melting state of the part surface to obtain a signal M, which specifically includes: The image after laser scanning undergoes four feature extraction operations in sequence. Each feature extraction operation includes convolution, batch normalization, Leaky ReLU activation and maximum pooling downsampling operation. The extracted features undergo global average pooling operation and 3 convolution layers with a kernel size of 1×1, and the probability value of the melting state is output through the softmax function.
9. The method for monitoring and controlling the surface defect lines of parts in a laser powder bed melting process according to claim 7, characterized in that: The probability value of the melting state is output through the softmax function, and the maximum probability value is selected as the final result to obtain the M value. The calculation formula is expressed as: Among them, P 欠熔化 , P 正常 , P 过熔化 They represent the probability values of under-melting, normal melting, and over-melting respectively.
10. The method for monitoring and controlling the surface defect lines of parts in a laser powder bed melting process according to claim 1, characterized in that: The surface defect quantitative index Ap and signal M are input into the proportional control system to calculate the process parameters required for the next layer and adjust the process parameters layer by layer, including: P 下一层 =P 当前层 +M×Kp×Error-(1-M)Kp×Error Among them, P 下一层 is the laser power required for the next layer, P 当前层 is the laser power of the current layer, M is a Boolean value used to indicate the current melting state, that is, if the current melting state is over-melting, M=1, otherwise M=0, Kp is the proportional coefficient, and Error indicates the error between the current monitoring signal Ap and the expected value.
Citation Information
Patent Citations
3D Printing Process Fault Diagnosis Methods and Devices
CN109968671B
On-line monitoring system and on-line monitoring method for powder laying additive manufacturing defects
CN115582559A
Online monitoring method for selective laser melting defects
CN116740001A
Online monitoring method, device and equipment for melting defects of laser powder bed and medium
CN117390433A
Powder bed defect visual detection method based on image feature fusion
CN112001909A
Cited By
Laser powder bed melting defect monitoring method and system based on data fusion
CN120009282A
Topological structure additive manufacturing local deposition defect online monitoring and control method
CN120205841A
Turret head assembly of numerical control turret punch press provided with multi-specification die system
CN120363537A
Additive manufacturing powder laying defect recognition early warning method and system
CN120833315A
Molten pool image defect detection method and system based on artificial intelligence
CN121788487A
