Online Monitoring Method for Porosity in Laser Metal Deposition Based on Deep Transfer Learning

Abstract

The present invention discloses an online monitoring method for the porosity of laser melting deposition based on deep transfer learning, comprising: step 1: obtaining a source domain sample set and a target domain sample set; step 2: extracting pore data of molten pool images in the source domain sample set and the target sample set respectively, and classifying the pores by label according to the pore area; step 3: establishing a source domain porosity online monitoring model by using ShuffleNetV2 with grouped convolution and channel shuffling; step 4: constructing a target domain porosity online monitoring model based on deep sub-domain adaptive network transfer learning with the help of the source domain porosity online monitoring model; the present invention first extracts pore features with the help of the lightweight characteristics of the ShuffleNetV2 network and the channel shuffling technology, establishes a porosity monitoring model in the source domain, and then realizes porosity monitoring in the target domain through the deep sub-domain adaptation method, thereby realizing porosity monitoring of different LMD process.

Description

Technical Field

[0001] The present invention belongs to the technical field of laser additive manufacturing, and in particular relates to an online monitoring method for the porosity of laser melting deposition based on deep transfer learning. Background Art

[0002] Laser melting deposition (LMD) is a process in which a laser beam acts on metal powder delivered by a powder sprayer, causing it to melt and deposit along a predetermined route. After solidification, a layer of deposited metal is formed. This reciprocating process ultimately completes the geometric shape of the part. It has a faster deposition rate and higher manufacturing freedom, and is advantageous in rapidly manufacturing large parts with good mechanical properties.

[0003] High part quality and repeatability remain key challenges for laser metal additive manufacturing (LMD) technology in high-end applications. In LMD manufacturing, the material's melting and solidification process can introduce various defects, such as cracks, balling, delamination, porosity, and poor melting. These defects can degrade the mechanical properties and quality of parts and hinder their use in critical physical applications. Porosity, in particular, is generally considered a key defect affecting part fatigue performance and crack growth characteristics. To reduce porosity in products, online porosity monitoring systems are essential. Combining sensor data with machine learning algorithms to monitor the porosity of printed parts has attracted widespread attention. However, machine learning algorithms rely on sufficient labeled data and the assumption that the data are independent and identically distributed. In laser melting deposition, changes in process parameters can significantly affect the distribution and correlation of experimental process data. Collecting sufficient labeled data from scratch for a new process is extremely time-consuming and expensive, and it is difficult to conduct experiments covering all process conditions to construct a sample library. Summary of the Invention

[0004] The purpose of the present invention is to provide an online porosity monitoring method for laser melting deposition based on deep transfer learning. This method first uses the lightweight characteristics of the ShuffleNetV2 network and the channel shuffling technology to extract pore features, establishes a porosity monitoring model in the source domain, and then realizes porosity monitoring in the target domain through a deep sub-domain adaptation method. It can effectively utilize the historical knowledge of relevant experimental data to realize porosity monitoring of different LMD process.

[0005] To achieve the above objectives, the present invention has the following beneficial effects:

[0006] A method for online monitoring of porosity in laser melting deposition based on deep transfer learning includes the following steps:

[0007] Step 1: Obtain source domain sample set and target domain sample set;

[0008] The source domain sample set refers to a melt pool image dataset during the historical deposition process;

[0009] The target domain sample set refers to a molten pool image dataset collected during the deposition process of new process parameters;

[0010] Step 2: Extract pore data from the melt pool images in the source domain sample set and the target sample set respectively, and classify the pores according to their labels based on their pore area;

[0011] Step 3: Establish a source domain porosity online monitoring model using ShuffleNetV2 with grouped convolution and channel shuffling, and use the source domain sample set and its label classification to train the source domain porosity online monitoring model;

[0012] Step 4: Using the source domain porosity online monitoring model, a target domain porosity online monitoring model is constructed based on deep sub-domain adaptive network transfer learning.

[0013] The trained source domain porosity online monitoring model is used as the initial target domain porosity online monitoring model. The source domain sample set and the target domain pore features are extracted by the source domain porosity online monitoring model and the initial target domain porosity online monitoring model respectively. Then, similar samples of the source domain pore features and the target domain pore features are projected into different sub-domains, and the distribution alignment is performed by calculating the LMMD distance between the sub-domains; the differences between label classifications are highlighted in the process of transfer learning.

[0014] Furthermore, the melt pool image dataset is preprocessed before step 2. The preprocessing refers to determining the size of the melt pool image according to the maximum pixel value and the connected domain and cropping it, then scaling it to a fixed size through a bilinear interpolation algorithm, and finally performing grayscale normalization.

[0015] Furthermore, in step 2, X-rays are used to scan the printed part along the deposition direction to extract a tomographic image, and then the pore data in the tomographic image is extracted through an image processing algorithm. The pore data includes shape information, position and area, and the pores are labeled and classified according to the pore area.

[0016] Furthermore, the loss function Loss of the porosity online monitoring model is:

[0017] Loss = L cross_entropy_target +λ1×L cross_entropy_source +λ2×L LMMD ;

[0018] Among them, L cross_entropy_target is the cross entropy loss function of the target domain, L cross_entropy_source is the cross entropy loss function of the source domain; λ1 is the loss ratio parameter of the source domain; LLMMD is the local maximum difference loss function, λ2 is the domain adaptation loss weight; Here, i represents the current training cycle and nepoch represents the total training cycle.

[0019] Compared with the prior art, the present invention has the following beneficial effects:

[0020] (1) In view of the structural characteristics and dynamic correlation of the samples, ShuffleNetv2 is used as the basic network to extract the features of the melt pools of different layers through group convolution, and channel shuffling is used to enhance the information exchange between layers, while maintaining low computational and memory requirements.

[0021] (2) The porosity monitoring method in the LAM process based on deep sub-domain adaptation realizes knowledge transfer by aligning the sub-domains of different porosity levels in the source and target domains, providing an effective solution for porosity monitoring in data-scarce and complex process conditions, significantly reducing the cost and time required for monitoring.

[0022] (3) An experimental dataset was constructed to train the proposed model and verify the method, and compared with other transfer learning methods and no-transfer models. The experimental results demonstrated the effectiveness and superiority of the proposed method.

[0023] (4) By learning cross-domain features or conversion mechanisms, the model can be adjusted according to the characteristics of the domain, and has better adaptability to the complex and changeable process of LMD. Moreover, adaptation within the subdomain is more targeted and efficient. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 This is a schematic structural diagram of the porosity online monitoring model of the present invention.

[0025] Figure 2 Flowchart of image processing in label classification; (a) physical image of the specimen; (b) melt pool image preprocessing; (c) CT image pore extraction process.

[0026] Figure 3 The pore distribution of the source domain and the target domain; (a) pore area distribution on the source domain, (b) pore area distribution on the target domain, and (c) pore category threshold division.

[0027] Figure 4 Figure 2 shows the porosity monitoring performance results of different base networks on the source domain.

[0028] Figure 5 This is the training process curve of ShuffleNetV2 under the optimal hyperparameters

[0029] Figure 6 Classification confusion matrix of different methods on the target domain; (a) no transfer, (b) fine-tuning, (c) freezing, (d) DAN, (e) DSAS.

[0030] Figure 7 TSNE visualization of models trained for different target domain sample sets; (a) 10%, (b) 20%, (c) 40%, (d) 60%, (e) 80%, (f) 100%.

[0031] Figure 8 This is a graph showing the impact of increased data volume on accuracy and training time. DETAILED DESCRIPTION

[0032] This embodiment provides a method for online monitoring of porosity of laser melting deposition based on deep transfer learning, including the following steps:

[0033] Step 1: Obtain source domain sample set and target domain sample set;

[0034] The source domain sample set refers to a molten pool image dataset in a historical deposition process; the historical deposition process refers to a deposition process with different process parameters or different deposition equipment.

[0035] The target domain sample set refers to a molten pool image dataset collected during the deposition process of new process parameters; the new process parameter deposition process refers to the deposition process of new process parameters or new deposition equipment.

[0036] This embodiment can effectively utilize historical knowledge of relevant experimental data to realize porosity monitoring of different LMD processes; in the case of data scarcity, it significantly improves the model development efficiency and reduces development costs, which is of great significance to the implementation and promotion of LAM quality monitoring.

[0037] In this embodiment, an industrial camera is coaxially mounted on the laser melting deposition (LMD) printhead to capture images of the melt pool during the laser melting deposition (LMD) process. Specifically, a CF8 / 5MX industrial camera from Kappa Optronics is used. During the deposition process, a high-energy 1030nm laser beam is emitted by the laser. This beam is reflected by a beam splitter covered with a 1030nm high-reflectivity film onto the substrate, where it interacts with the material, forming an interaction zone centered on the molten pool. Simultaneously, light emitted from the molten pool passes through the beam splitter and is reflected by a reflector back to the coaxially mounted industrial camera, enabling the camera to capture images of the interaction zone.

[0038] The melt pool images in the source domain sample set and the target domain sample set are preprocessed. The approximate location of the melt pool is determined by the maximum pixel value and the connected domain. A fixed-size ROI cropping window is created to include the melt pool in the window. To match the input size of the model and ensure the size consistency of the input image, the cropped image size is scaled to 224×224 using a bilinear interpolation algorithm. The image is then grayscale normalized so that the grayscale value of each process image is in the interval [0, 255].

[0039] Step 2: Extract pore data from the melt pool image and classify the pores according to the pore data;

[0040] The printed sample is cut and scanned along the stacking direction using X-rays to obtain a horizontal cross-sectional tomographic image. An image processing algorithm is then used to extract the pores in the tomographic image. The position and actual area of the pores are calibrated using the pore shape information and CT pixel resolution. Based on the size of the pore area, an appropriate pore threshold is set to classify the pores into small, medium, and large pores.

[0041] Comparing the source domain sample set and the target domain sample set, they have great similarity in the feature space. The same pore threshold division ensures the same label space. Due to the differences in process between the source domain sample set and the target domain sample set, the printed parts have different characteristics in interlayer bonding, heat transfer, defect formation, etc. The marginal probability distribution of the two is different, and the distribution state of pores and the conditional distribution of labels are also different.

[0042] Step 3: Establish a source domain porosity online monitoring model using ShuffleNetV2 with grouped convolution and channel shuffling;

[0043] ShuffleNetV2 is a lightweight and efficient deep learning model that follows the residual connection concept of ResNet and adopts depthwise separable convolution and channel shuffling mechanism in the residual block, achieving the highest possible classification or detection accuracy with the least possible computing resources and memory usage.

[0044] Grouped convolution and channel shuffling are well suited to the structural characteristics of LAM process samples. First, the samples contain dynamic melt pool images of multiple layers, and grouped convolution can precisely divide the melt pool images of different layers into multiple groups for feature extraction. Second, considering factors such as the dynamics of the melt pool and heat transfer, the melt pools of different layers interact and influence each other. Grouped convolution will cause the features of the melt pools of each layer to be independent of each other, making it difficult to extract global feature information. The channel shuffling mechanism will shuffle and rearrange the grouped feature maps. Channel shuffling can strengthen the information exchange between the melt pool images of different layers, helping the model learn richer and more accurate feature representations. Finally, the LAM process samples contain multiple layers of dynamic melt pool images. The lightweight nature of ShuffleNetV2 enables it to process such high-dimensional samples efficiently and quickly, reducing the demand for computing resources.

[0045] The source domain sample set and its pore data are used to train the source domain porosity online monitoring model.

[0046] Step 4: Construct a target domain porosity online monitoring model (DSAS) based on deep sub-domain adaptive network transfer learning; realize real-time monitoring of the pores in the LAM process through the target domain porosity online monitoring model.

[0047] like Figure 1 As shown in the figure, the trained source domain porosity online monitoring model is used as the initial target domain porosity online monitoring model. The source domain porosity online monitoring model and the target domain porosity online monitoring model are used to extract the source domain pore features and the target domain pore features respectively. The similar samples in the source domain pore features and the target domain pore features are projected into different sub-domains through the deep sub-domain adaptive network. The distribution alignment is performed by calculating the LMMD distance between the sub-domains; and the differences between categories are highlighted in the process of transfer learning.

[0048] The LMMD distance is used to align at different layers to achieve the effect of sub-domain alignment. The LMMD distance is:

[0049]

[0050] Where LMMD is the distribution distance, c is the label category, c∈{0,1,2}, C is the number of label categories, the label categories are small pores, medium pores, and large pores, so the number of label categories is 3. is the weight of the i-th sample in the source domain sample set belonging to category c, is the i-th sample in the source domain sample set; is the weight of the jth sample in the target domain sample set belonging to category c, is the jth sample in the target domain sample set.

[0051] and The calculation method is the same, specifically, the weight of the sample belonging to c is the weight of the i-th sample of category c in the domain, is the i-th category value of category c. The label vector is converted into one-hot encoding form, that is, only the current category is represented by 1, and the rest of the categories are represented by 0; this indicates that samples of the same category have the same weight value.

[0052] The source domain sample set and the target domain sample set both have corresponding label classifications. The DSAS method proposed in this embodiment uses a supervised approach to calculate the LMMD distance to reduce the error caused by pseudo-labels in the target domain. During backpropagation, the LMMD loss is not used alone as the network loss function, but is combined with the cross-entropy loss function to form the final total loss. The total network loss function is calculated as follows:

[0053] Loss = L cross_entropy_target +λ1×L cross_entropy_source +λ2×L LMMD ;

[0054] Among them, L cross_entropy_target is the cross entropy loss function of the target domain, L cross_entropy_source L is the cross entropy loss function of the source domain. Considering the difference in data distribution between the source domain and the target domain, the λ1 parameter is used to control the loss ratio of the source domain. LMMD is the local maximum difference loss, also known as the domain adaptation loss, and the parameter λ2 is the domain adaptation loss weight that changes with the training cycle.

[0055]

[0056] Here, i represents the current training cycle, and nepoch represents the total training cycle. As i increases, λ2 gradually increases from almost 0. In the early stages of training, the model may need to focus more on classification performance. Furthermore, because the model parameters have not yet converged to optimal values, the differences between the source and target domains cannot be well distinguished. Therefore, the weight of the domain adaptation loss is set relatively low. As performance improves, the model is able to better learn the features of the source domain, but lacks the ability to learn inter-domain features. Therefore, to improve the model's performance in target domain learning tasks, the proportion of the domain loss in the total loss is gradually increased, and the model parameters are then aligned with the target domain through backpropagation.

[0057] Experimental verification of the effectiveness of this embodiment

[0058] Experiments were conducted using 30CrNi2MoVA powder and a Q235 steel substrate. Laser power and scanning speed were used as process parameters to fabricate multiple thin-walled structure specimens. Melt pool images collected during deposition with fixed process parameters, as well as those obtained when process parameters were varied between specimens, served as the source domain sample set. To characterize a more complex process, melt pool images collected during deposition with a fixed laser scanning speed and varying laser power served as the target domain sample set. The source domain sample set was prepared using the process parameters listed in Table 1, and the target domain sample set was prepared using the process parameters listed in Table 2. The powder was first conditioned in a sealed chamber at 120°C for approximately 120 minutes to completely remove moisture from both the powder and the inter-particle spaces. Argon was then introduced to reduce the oxygen concentration to below 0.02%. The powder feed rate was set at 1.5 rpm (approximately 15 g / min), and the carrier gas flow rate was 6 L / min. During printing, the laser beam diameter was set to 3 mm, and the lift height per layer was 0.3 mm. All thin-walled components were 40 mm long, and 30 layers were printed in a single pass. The camera collects melt pool images with a pixel size of 920 × 488 in real time at a frame rate of 50 Hz. 1434 samples are collected in the source domain and 491 samples are collected in the target domain.

[0059] Table 1 Source domain sample set printing experiment parameters

[0060]

[0061] Table 2 Target domain sample set printing experiment parameters

[0062]

[0063]

[0064] After printing, the printed thin-walled parts are cut and inspected by X-ray using the nano Voxel4000 industrial CT system. Horizontal cross-section tomographic images are obtained along the stacking direction. Image segmentation algorithms are used to locate and quantitatively analyze micron-level pores in thin-walled parts. Figure 2 As shown in Figure 4, regional quality assessment of thin-walled parts is performed based on the spatial distribution of pores in the thin-walled parts.

[0065] The label distribution of the source domain sample set and the target domain sample set is statistically analyzed. Figure 3 (a) and (b) show the distribution of pore area in the source and target domain sample sets, respectively, while (c) illustrates the pore threshold classification. The distributions are significantly different. While the three types of pores are evenly distributed in the source domain sample set, the target domain sample set has the largest proportion of small pores, while the number of large pores is relatively small. Furthermore, the target domain sample set exhibits sample imbalance, further increasing the difficulty of porosity monitoring in the target domain.

[0066] (1) AlexNet, VGG19, ResNet50, DenseNet121, and ShuffleNetV2 are used as source domain pore online monitoring models. They are the most representative CNN models in the field of deep learning and have excellent performance and performance. The source domain sample set is divided into training set and test set in a ratio of 7:3. All models are trained and tested on four NVIDIA GeforceRTX4090 graphics cards and use the same hyperparameters. The classification accuracy, floating point operation amount, parameter amount and model inference speed of each model are shown in the figure below. Figure 4 As shown in the figure, it can be seen that deeper networks do not necessarily improve performance. In fact, the deepest network, DenseNet121, achieved the worst results. The porosity monitoring model based on ShuffleNetV2 achieved the highest accuracy. This is because the network characteristics of ShuffleNetV2 are more compatible with the high dimensionality and grouping nature of the samples used in this article. Furthermore, due to its lightweight nature, ShuffleNetV2 has the lowest FLOPs and parameter count, reducing computing resources and memory usage, making it applicable to terminal deployment in a wider range of industrial scenarios.

[0067] This example uses ShuffleNetV2 as the basic model for source domain porosity monitoring. To further optimize the performance of the model, adjustments are made to the learning rate, batch size, and optimizer weight decay parameters. A grid search method is used within a small range to find the optimal hyperparameter combination. When the initial learning rate is 0.001, the batch size is 32, and the SGD optimizer weight decay parameter is 5e-3, the model performance reaches the best with an accuracy of 90.5%. The loss value of the model on the training set and the accuracy change on the test set are shown in the following figure. Figure 5 shown.

[0068] (2) In order to demonstrate the superiority of the DSAS model, no-transfer and other popular transfer learning methods were used to train the model. To ensure fairness, the hardware configuration, environment and hyperparameters were kept the same. No-transfer means that the source domain-related model and parameters are not used on the target domain, and ShuffleNetV2 is directly used for training from scratch. Fine-tuning training is to load the pre-trained parameters on the source domain as the initial parameters, and use the target domain data for training to update the parameters. Frozen training means loading the pre-trained parameters on the source domain as the initial parameters, but freezing the underlying parameters of the model. The frozen parameters remain unchanged during the training process, and only the unfrozen network layers are adjusted. Different stages of ShuffleNetV2 are frozen separately, and finally the best results are obtained when the first two stages are frozen. DAN is a transfer learning network that performs domain adaptation on a global scale. In order to reflect the superiority of the DSAS model under fair conditions, the baseline network of DAN is set to be the same as DSAS, using the same source domain pre-trained weights. The specific performance indicators of porosity monitoring of different models on the target domain are shown in Table 3, and the classification confusion matrix of each model is shown in Figure 6 .

[0069] Table 3 Porosity monitoring performance of different methods in the target domain

[0070]

[0071] As can be seen in Table 3, when transfer learning is not used, the training accuracy is the lowest. This is because the target domain data is small and the model training is insufficient. The accuracy of several models trained using transfer learning methods is 3.8%, 5.4%, 3.1%, and 6.2% higher than that of the models without transfer learning, respectively. This shows that transfer learning is effective for porosity monitoring under variable process parameters. Although the manufacturing processes under fixed and variable process parameters differ significantly, the overall patterns of the two are similar and are not completely unrelated. Drawing on the historical knowledge of the model under fixed process parameters can achieve better results in porosity monitoring under variable process parameter data.

[0072] Comparing several transfer learning methods, freezing and fine-tuning are the most commonly used transfer learning methods, and they have also achieved considerable results in porosity monitoring with variable process parameters. The frozen model achieved an accuracy of 88.2% in the target domain. Furthermore, since some model parameters are not trained during frozen training, the entire model training process takes less time. The batch time in the table represents the average training time for each batch. The fine-tuned model achieved an accuracy of 89.8%. Fine-tuning training can adaptively adjust the model's weights, biases, and other parameters based on the data characteristics of the target domain, enabling the model to better adapt to new data. Both the DAN and DSAS methods are deep adaptation networks. The DAN method performs distributed global adaptation calculations between the source and target domains, while the DSAS method no longer performs global adaptation but instead adapts to related sub-domains separately. The distance between domains is also measured using the local maximum mean difference. In terms of training time, the deep adaptation method requires longer training time than freezing and fine-tuning. In terms of monitoring accuracy, the DAN method has the lowest accuracy among several transfer learning methods, only 87.5%, while the DSAS method achieves the best transfer effect, with a porosity classification accuracy of 90.6% for the target domain.

[0073] (3) In order to study the impact of the amount of data in the source and target domains on model performance, the model was trained using data of different proportions. First, the pre-training model was constructed using the entire source domain sample set, and the model was trained using DSAS using 10%, 20%, 40%, 60%, 80%, and 100% of the target domain sample set, respectively. Increasing the amount of target domain data can improve monitoring accuracy. When the amount of data is small, the model has poor recognition effect on the three categories of pores. However, when the amount of data exceeds 60%, the distribution differences of pores with different labels in the feature space are already quite obvious.

[0074] Secondly, all target domain data are used, and 10%, 20%, 40%, 60%, 80% and 100% of the source domain data are used to build pre-training models and DSAS training respectively, and their accuracy and training time are compared on the target domain test set, as shown in Figure 7 and Figure 8 As shown in Figure 2, it can be seen that increasing the amount of data in the source domain can improve the monitoring accuracy of the model in the target domain. This may be because the source domain data not only determines the quality of the pre-training model, but also participates in the adaptation training between domains.

[0075] When the combined source and target domain data exceeds 60%, the impact of increasing data volume on accuracy gradually decreases. This provides guidance for future research on the efficient use of small sample data. Another conclusion is that model training time increases with the amount of source domain data, while the amount of target domain data has little effect on training time. This is because the source domain data far exceeds the target domain data and occupies the majority of the training resources.

[0076] The deep transfer learning method proposed in this example effectively leverages historical knowledge from relevant experimental data to enable porosity monitoring of complex process parameters, even when data is scarce. In particular, in monitoring processes like LAM, which have high experimental and testing costs, using transfer learning to reduce the amount of data required and improve the generalizability of monitoring models can significantly improve model development efficiency and reduce monitoring costs, significantly impacting the implementation and promotion of LAM quality monitoring.

[0077] The above description is only a preferred embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any modification and replacement based on the technical solution and inventive concept provided by the present invention should be covered by the protection scope of the present invention.

Claims

1. A method for online monitoring of porosity in laser melting deposition based on deep transfer learning, characterized in that: The steps include: Step 1: Obtain source domain sample sets and target domain sample sets; The source domain sample set refers to a melt pool image dataset during the historical deposition process; The target domain sample set refers to a molten pool image dataset collected during the deposition process of new process parameters; Step 2: Extract pore data from the melt pool images in the source domain sample set and the target sample set respectively, and classify the pores according to their labels based on their pore area; Step 3: Establish a source domain porosity online monitoring model using ShuffleNetV2 with grouped convolution and channel shuffling, and use the source domain sample set and its label classification to train the source domain porosity online monitoring model; Step 4: Using the source domain porosity online monitoring model, a target domain porosity online monitoring model is constructed based on deep sub-domain adaptive network transfer learning. The trained source domain porosity online monitoring model is used as the initial target domain porosity online monitoring model. The source domain sample set and the target domain sample set are respectively extracted using the source domain porosity online monitoring model and the initial target domain porosity online monitoring model. Then, similar samples of the source domain pore characteristics and the target domain pore characteristics are projected into different sub-domains, and the distribution alignment is performed by calculating the LMMD distance between the sub-domains. The differences between the label classifications are highlighted in the process of transfer learning.

2. The method for online monitoring porosity of laser melting deposition based on deep transfer learning according to claim 1 is characterized in that: Before step 2, the melt pool image dataset is preprocessed. The preprocessing refers to determining the size of the melt pool image according to the maximum pixel value and the connected domain and cropping it, then scaling it to a fixed size through a bilinear interpolation algorithm, and finally performing grayscale normalization.

3. The online monitoring method for porosity of laser melting deposition based on deep transfer learning according to claim 1 is characterized in that: In step 2, X-rays are used to scan the printed part along the deposition direction to extract a tomographic image. Then, an image processing algorithm is used to extract pore data from the tomographic image. The pore data includes shape information, position, and area, and the pores are labeled and classified according to the pore area.

4. The online monitoring method for porosity of laser melting deposition based on deep transfer learning according to claim 1 is characterized in that: The loss function of the target domain porosity online monitoring model is for: ； in, is the cross entropy loss function of the initial target domain, is the cross entropy loss function of the source domain; is the loss ratio parameter of the source domain; is the local maximum difference loss function, is the domain adaptation loss weight; ; Where i represents the current training cycle and nepoch represents the total training cycle.

Images