Casting internal defect detection method based on deep learning

Abstract

The invention belongs to the field of casting production full-process automation and intelligentization, and provides a casting internal defect detection method based on deep learning, which comprises the steps of to-be-detected image shooting, HDR image enhancement processing, one-key image enhancement processing and mask image fusion, industrial X-ray AI network detection, post-processing operation and data statistics. According to the method, the industrial X-ray AI network is constructed, so that the defects that conventional defect detection depends on personnel experience and is relatively poor in precision are overcome; through HDR image enhancement processing, one-key image enhancement processing and mask image fusion, the influence of the complex structure of a detection device, the defects of shooting equipment and noisy points is reduced.

Description

Technical Field

[0001] The present invention relates to the field of full-process automation and intelligence of casting production, and in particular to a method for detecting internal defects of castings based on deep learning. Background Art

[0002] The metal industry is one of the important basic industries in the manufacturing industry. Complex metal components are mostly formed as a whole using precision casting technology. During the casting process, defects such as shrinkage cavities, shrinkage, inclusions and cracks often appear inside the castings. Hot isostatic pressing and repair welding have limited ability to eliminate the above internal defects. Therefore, non-destructive testing of castings based on X-ray images has become a necessary process to ensure product quality.

[0003] However, the traditional X-ray casting internal defect detection or manual visual casting X-ray image defect detection is easily affected by the proficiency and physical and mental state of the detection personnel. The efficiency, accuracy, reliability and stability of the detection are difficult to guarantee, and the preservation and query of the detection information are quite time-consuming and labor-intensive, which seriously hinders the development of automation and intelligence of the whole process of casting production. Nondestructive testing based on X-ray images has now become a widely used detection technology. However, in traditional radiographic imaging technology, due to the different materials of the workpiece being detected, the complexity of the structure, the superposition of the structure of the workpiece itself and the scattered X-rays, the defects of the imaging equipment itself, and the influence of noise, the clarity of the obtained image is usually limited, and the contrast of the radiographic image is low, the key features are blurred, and even if the image contains rich information, it is difficult to be directly observed by the human eye. Traditional visible light imaging can form a massive digital image database, and its acquisition and collection methods are relatively convenient and fast. However, the acquisition of X-ray casting images must rely on inherent Xray detection equipment, and the types of castings are diverse, so it is difficult to collect and obtain them. Summary of the invention

[0004] In order to overcome the shortcomings of the prior art, the purpose of the present invention is to provide a method for detecting internal defects of castings based on deep learning. By constructing an industrial X-ray AI network, the defects of conventional defect detection that rely on human experience and have poor accuracy are avoided; through HDR image enhancement processing, one-click image enhancement processing and mask image fusion, the influence of the complex structure of the detection device, the defects of the shooting equipment itself and the noise are reduced.

[0005] To achieve the above object, the present invention provides the following solutions:

[0006] A method for detecting internal defects of castings based on deep learning, comprising:

[0007] Collect the image of the casting workpiece to be tested and obtain a 16-bit X-ray image to be tested;

[0008] The 16-bit X-ray image to be detected is converted into an 8-bit image by using HDR image enhancement processing to obtain a pre-processed image to be detected;

[0009] Perform one-key image enhancement processing on the pre-processed image to be trained and fuse it with a mask image generated according to the ROI detection area parameters set by the user to obtain the pre-processed data to be detected;

[0010] Inputting the pre-processed data to be detected into a pre-trained industrial X-ray AI network for detection to obtain detection data;

[0011] Performing post-processing operations on the detection data to obtain post-processed data; the post-processing operations include: original image defect display and image prediction;

[0012] Performing data statistics on the post-processing data to obtain an output result, and saving the defect data in the output result according to the file type;

[0013] The training process of the industrial X-ray AI network includes:

[0014] Collect existing casting workpiece images to obtain 16-bit X-ray images to be trained;

[0015] The 16-bit X-ray image to be trained is converted into an 8-bit image by using HDR image enhancement processing to obtain a pre-processed image to be trained;

[0016] Performing one-key image enhancement processing on the pre-processed image to be trained to obtain pre-processed data to be trained; the one-key image enhancement processing includes: brightness adjustment, contrast adjustment, filtering and sharpening;

[0017] Using an AI network generation algorithm based on a GAN network to generate images of the pre-processed data to be trained, to obtain generated data;

[0018] Combining the pre-processed data to be trained and the generated data and performing a super-resolution enhancement operation based on an SR algorithm to obtain SR data;

[0019] Performing labeled data sharing and combined pairing on the pre-processed data to be trained and the generated data respectively, and performing defect annotation on the SR data to obtain an original paired data set in a binary image format;

[0020] Performing data enhancement operations on the original paired data set to obtain a generalized data set; the data enhancement operations include: rotation, scaling, shearing, and noise addition;

[0021] Randomly divide the generalized data set into a validation data set and a training data set according to a preset matching ratio;

[0022] The pre-training data set obtained by combining the verification data set and the training data set is input into the W-Net network and the model training is performed using the GPU to obtain the trained industrial X-ray AI network.

[0023] Preferably, the identification objects of the industrial X-ray AI network include: pores, slag inclusions, porosity, cracks and shrinkage cavities.

[0024] Preferably, the parameter control method of the industrial X-ray AI network includes: regularization technology.

[0025] Preferably, it also includes:

[0026] When process parameters change, the defect data is updated and the industrial X-ray AI network is retrained using the updated defect data.

[0027] Preferably, it also includes:

[0028] When the performance indicator of the monitoring data exceeds the preset warning threshold, an early warning message is sent to the target object.

[0029] Preferably, it also includes:

[0030] Introduce anomaly detection mechanisms and feedback loop mechanisms;

[0031] Performing secondary analysis on the output result using the anomaly detection mechanism to obtain defect omission data;

[0032] The feedback loop mechanism is used to collect feedback data from target users, and the feedback data is used to optimize the industrial X-ray AI network.

[0033] The present invention discloses the following technical effects:

[0034] The present invention provides a method for detecting internal defects of castings based on deep learning. By constructing an industrial X-ray AI network, the problem of reliance on human experience and poor accuracy in conventional defect detection is solved, and the fine detection of the defect position of the target detection device is realized; through HDR image enhancement processing, one-key image enhancement processing and mask image fusion, the complex structure of the detection device, the defects of the shooting equipment itself and the influence of noise are solved, and the enhancement of the defective part and the enhancement of image clarity are realized. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0036] Figure 1 A schematic diagram of a casting internal defect detection process based on deep learning provided by an embodiment of the present invention;

[0037] Figure 2 A flowchart of AI model file training provided by an embodiment of the present invention;

[0038] Figure 3 A flowchart of AI model detection provided by an embodiment of the present invention;

[0039] Figure 4 A schematic diagram of a network structure provided in an embodiment of the present invention, Figure 4 (a) is the first sub-graph, Figure 4 (b) is the second sub-image. DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0041] The purpose of the present invention is to provide a method for detecting internal defects of castings based on deep learning. By constructing an industrial X-ray AI network, the defects of conventional defect detection that rely on human experience and have poor accuracy are avoided; through HDR image enhancement processing, one-key image enhancement processing and mask image fusion, the influence of the complex structure of the detection device, the defects of the shooting equipment itself and the noise are reduced.

[0042] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0043] Figure 1 A schematic diagram of a casting internal defect detection process based on deep learning provided in an embodiment of the present invention, Figure 2 The AI ​​model file training flow chart provided in the embodiment of the present invention is as follows: Figure 3 The AI ​​model detection flow chart provided in the embodiment of the present invention is as follows: Figure 1 , Figure 2 as well as Figure 3 As shown, the present invention provides a casting internal defect detection method based on deep learning, comprising:

[0044] Step 100: Acquire an image of a casting workpiece to be inspected to obtain a 16-bit X-ray image to be inspected;

[0045] Step 200: Convert the 16-bit X-ray image to be detected into an 8-bit image by using HDR image enhancement processing to obtain a pre-processed image to be detected;

[0046] Step 300: Perform one-key image enhancement processing on the pre-processed image to be trained and fuse the mask image generated according to the ROI detection area parameters set by the user to obtain the pre-processed data to be detected;

[0047] Step 400: input the pre-processed data to be detected into the pre-trained industrial X-ray AI network for detection to obtain detection data;

[0048] Step 500: performing post-processing operations on the detection data to obtain post-processed data; the post-processing operations include: original image defect display and image prediction;

[0049] Step 600: Perform data statistics on the post-processing data to obtain output results, and save the defect data in the output results according to the file type; the output results include: xml file, json file, excel file, yaml file, etc.;

[0050] The training process of the industrial X-ray AI network includes:

[0051] Collect existing casting workpiece images to obtain 16-bit X-ray images to be trained;

[0052] The 16-bit X-ray image to be trained is converted into an 8-bit image by using HDR image enhancement processing to obtain a pre-processed image to be trained;

[0053] Perform one-click image enhancement processing on the pre-processed image to be trained to obtain the pre-processed data to be trained; the one-click image enhancement processing includes: brightness adjustment, contrast adjustment, filtering and sharpening;

[0054] Use the AI ​​network generation algorithm based on the GAN network to generate images for the pre-training processed data to obtain generated data;

[0055] The pre-training processed data and the generated data are combined and subjected to super-resolution enhancement operation based on the SR algorithm to obtain SR data;

[0056] The pre-training processed data and generated data are labeled for data sharing and combined pairing respectively, and the SR data is labeled for defects to obtain the original paired data set in binary image format;

[0057] Perform data augmentation operations on the original paired data set to obtain a generalized data set; data augmentation operations include: rotation, scaling, shearing, and noise addition;

[0058] The generalization dataset is randomly divided into a validation dataset and a training dataset according to the preset matching ratio;

[0059] The pre-training data set obtained by combining the verification data set and the training data set is input into the W-Net network and the model is trained using GPU to obtain a trained industrial X-ray AI network.

[0060] Specifically, the identification objects of the industrial X-ray AI network include: pores, slag inclusions, looseness, cracks and shrinkage cavities.

[0061] Preferably, the parameter control method of the industrial X-ray AI network includes: regularization technology.

[0062] Optionally, it also includes:

[0063] When process parameters change, the defect data is updated and the industrial X-ray AI network is retrained using the updated defect data.

[0064] Preferably, it also includes:

[0065] When the performance indicators of the monitoring data exceed the preset warning threshold, an early warning message is sent to the target object.

[0066] Furthermore, it also includes:

[0067] Introduce anomaly detection mechanisms and feedback loop mechanisms;

[0068] Use the anomaly detection mechanism to conduct secondary analysis on the output results to obtain defect omission data;

[0069] A feedback loop mechanism is used to collect feedback data from target users, and the feedback data is used to optimize the industrial X-ray AI network.

[0070] Specifically, the process of the casting image data acquisition preprocessing method is as follows: first, the collected 16-bit X-ray imaging image is subjected to a high dynamic range enhancement preprocessing algorithm (HDR algorithm) to display the detailed information of the image defect position; secondly, in order to address the problem of image noise introduced during HDR algorithm enhancement, super-resolution image enhancement technology (SR) is used to effectively reduce image noise while enhancing image clarity; finally, the 16-bit image is converted into an 8-bit image through image normalization and saved for subsequent image processing operations and to improve image processing speed.

[0071] Furthermore, the process of establishing the training set and validation set of casting image data is as follows: first, the collected preprocessed casting workpiece images are manually generalized (due to the scarcity of industrial image data, in order to increase the number of defective images, data generalization operations are required). This operation is mainly achieved through the virtual generation algorithm constructed by the generative adversarial network in AI technology; then, the original preprocessed casting workpiece images and the image data generated by the virtual generation algorithm are merged; finally, the data set composed of the merged image data is annotated with image defects using ImageJ software. The image formed after annotation is a binary reference image. The annotated data set is divided into a training sample data set, a verification sample data set, and a test sample data set according to actual needs.

[0072] Preferably, the AI ​​industrial network model for detecting internal defects of casting products with different processes (the scope of casting identification includes: defects such as pores, slag inclusions, looseness, cracks, shrinkage holes, etc.) is designed with reference to its core network algorithm. Figure 4 (a) and Figure 4 (b). The designed AI industrial network model mainly uses the jump and residual connection method to deeply extract the features of casting images by cascading three U-type networks. At the same time, the channel attention and spatial attention mechanisms are integrated in the process of jump connection to solve the problem that small defective targets are easily missed, and multiple activation functions are combined to achieve more accurate casting X-ray image segmentation, increase the fitting ability of the network, and improve the robustness and accuracy of the network. The specific implementation process is as follows: First, the low-frequency feature information of the casting image is extracted through the first U-type network. This process mainly involves 4 downsampling operations and 4 upsampling operations; secondly, the second U-type network is used to capture the intermediate frequency information of the casting image. This process also involves 4 downsampling operations and 4 upsampling operations. At the same time, in order to avoid the problem of increasing network parameters due to continuous stacking, the bridge and residual connection methods are used to cascade the two networks before and after; finally, the third U-type network is used to mainly obtain fine image details, that is, high-frequency component information. This process also involves 4 downsampling and 4 upsampling. The three U-type networks can extract different features of the image at different levels of frequency, which can accelerate the convergence speed of the entire model network. In addition, channel attention and spatial attention mechanisms are integrated in the process of bridging and residual connection to improve the image segmentation accuracy of the entire network model.

[0073] Furthermore, the post-processing process of casting image data is as follows: first, in order to improve the detection rate of defect positions in casting images and reduce the false detection rate, a mask image for shielding areas is established (for example, to prevent the background from being treated as a defect point, a mask image template for shielding the background area of ​​the image can be established, or areas that are often misdetected can also be shielded by this method). Then, the binary image predicted by the AI ​​industrial network model is combined with the mask image to be annotated in the original image (generally in red or yellow). Finally, the defect area or size can be secondary processed according to the parameter settings of the xml or json file (the operation parameter value is generally determined by the user). In this way, the threshold parameters set can be used to finally determine which of the marked areas are OK and which are NG. Afterwards, the statistical results are written into the xml or json file, and the defective image with annotations is output.

[0074] Preferably, in the later stages of ADR algorithm optimization, when image collection becomes relatively difficult, improving the accuracy of ADR can be achieved through a variety of strategies and technical means:

[0075] (a) Data augmentation and synthesis:

[0076] Data augmentation: Generate new training samples by rotating, flipping, scaling, cropping, color changing, etc. existing images. This not only increases the diversity of the dataset, but also helps the model generalize better to different situations.

[0077] Synthetic data: Generate new, high-quality image data from a small number of real samples using techniques such as generative adversarial networks (GANs). This approach can significantly expand the dataset without increasing the actual collection cost.

[0078] (b) Transfer Learning:

[0079] Use pre-trained models: Use models pre-trained on large datasets (such as pre-trained models on ImageNet) as a starting point and fine-tune them to fit the dataset of a specific task. This can take advantage of the powerful feature representation capabilities that the pre-trained model has learned to improve the accuracy of new tasks.

[0080] (c) Semi-supervised and unsupervised learning:

[0081] Semi-supervised learning: Use partially labeled data and a large amount of unlabeled data for training. By combining supervised and unsupervised learning methods, useful information can be extracted from unlabeled data to assist in model training.

[0082] Unsupervised learning: Through unsupervised learning methods such as clustering and autoencoders, potential structures and features are mined from unlabeled data, and these features are then used to improve the performance of the model.

[0083] (d) Model compression and distillation: Model compression: By means of pruning, quantization, etc., the size of the model is reduced while maintaining or improving the performance of the model. This helps to better train the model on a limited dataset.

[0084] Knowledge distillation: Distill the knowledge of large, complex models into small, simple models. In this way, a small model with good performance can be trained on a smaller dataset.

[0085] (e) Regularization and optimization algorithms:

[0086] Regularization: Use L1, L2 regularization or other forms of regularization techniques to prevent the model from overfitting. This helps improve the generalization ability of the model on a limited dataset.

[0087] Optimization algorithm: Select appropriate optimization algorithm (such as Adam, RMSprop, etc.) and learning rate scheduling strategy to improve the training efficiency and convergence of the model.

[0088] Ensemble learning and model fusion:

[0089] (f) Ensemble learning: training multiple different models and then combining their prediction results to improve the overall accuracy.

[0090] Model fusion: Fuse the outputs of different models (such as CNN, RNN, etc.) to make full use of their respective advantages and improve the overall performance.

[0091] Therefore, in the later stages of ADR algorithm optimization, even if it is difficult to collect images, the accuracy of ADR can still be improved through a variety of strategies and technical means. The key is to make full use of existing resources, give full play to the advantages of deep learning models, and carry out targeted optimization and adjustments based on actual problems.

[0092] Furthermore, when process parameters change, the ADR algorithm may indeed face defects that have not appeared before or have rarely appeared, which may increase the probability of ADR missing judgments. In order to ensure the accuracy of ADR in this case, the following strategies can be adopted:

[0093] (a) Real-time data update and retraining:

[0094] When process parameters change, new defect data should be collected as soon as possible and the AI ​​model should be retrained. This ensures that the model can adapt to new environments and situations and reduce the chance of missed judgments.

[0095] In order to maintain the real-time and accuracy of the model, a continuous learning mechanism can be established to enable the model to continuously learn and evolve from new data.

[0096] (b) Enhanced model robustness:

[0097] During the model training process, we should focus on improving the robustness of the model so that it has stronger adaptability to changes in input data. The robustness of the model can be enhanced through technical means such as data enhancement, regularization, and model integration.

[0098] (c) Threshold adjustment and confidence analysis:

[0099] According to the new data distribution, the classification or detection threshold is adjusted to ensure that the model can more accurately identify new defects. For each prediction result, its confidence can be calculated, and the decision on whether to accept the prediction can be made based on the confidence level. For predictions with lower confidence, further manual inspection or analysis can be performed.

[0100] (d) Monitoring and early warning system:

[0101] Establish a monitoring and early warning system to monitor the performance and prediction results of the model in real time. When the model frequently misses or misjudges, an early warning should be issued in time and corresponding measures should be taken. The early warning function can be achieved by setting thresholds for performance indicators (such as accuracy, recall, etc.).

[0102] (e) Integration of expert knowledge and rules:

[0103] Integrating the knowledge and experience of experts into the model can be achieved through rule engines or hybrid models, etc. This can make up for the shortcomings of the model in dealing with new situations to a certain extent.

[0104] By working with experts, identification rules and strategies for new defects are developed and integrated into the ADR system.

[0105] (f) Anomaly detection and feedback loop:

[0106] Introduce anomaly detection mechanisms to discover data that differs significantly from normal patterns. When anomalies are detected, further investigation and analysis can be triggered to determine if new defects exist. Establish a feedback loop mechanism so that the system can continuously learn and improve from actual applications. By collecting user feedback and new data, the performance and accuracy of the model are continuously optimized.

[0107] Therefore, through real-time data updates, model robustness enhancement, threshold adjustment, monitoring and early warning, expert knowledge fusion, anomaly detection and feedback loops, etc., the probability of ADR missing judgment after process parameter changes can be effectively reduced to ensure the accuracy of ADR. At the same time, these methods also need to be flexibly adjusted and optimized in combination with actual application scenarios.

[0108] Optionally, when the ADR algorithm cannot accurately judge a situation, some strategies and technical means can still be adopted to minimize this risk and improve the accuracy of the AI ​​algorithm.

[0109] (a) In-depth analysis of known defects:

[0110] Carefully study the cases of major defects that the AI ​​algorithm failed to identify and understand their characteristics and causes. This helps identify possible blind spots and limitations of the algorithm. By analyzing these cases, useful information can be extracted to improve the algorithm or design new feature representation methods.

[0111] (b) Using unsupervised learning: Since there is less labeled data, we can consider using unsupervised learning methods to mine the underlying structure and features in the data. For example, clustering algorithms can be used to discover hidden patterns in the data that may be related to defects.

[0112] Unsupervised learning can help algorithms better understand the distribution and characteristics of data, thereby improving their ability to identify new defects.

[0113] (c) Combining expert knowledge and rules:

[0114] Work with domain experts to translate their knowledge and experience into rules and constraints in the algorithm. This helps to compensate for insufficient data and improve the algorithm’s ability to identify specific defects. By incorporating expert knowledge, more comprehensive guidance can be provided to the algorithm, enabling it to make more accurate judgments when faced with complex and rare defects.

[0115] (d) Using transfer learning and pre-trained models:

[0116] Transfer learning allows knowledge learned from other tasks or fields to be transferred to the current task. Consider using a model pre-trained on a related task as a starting point and fine-tune it to fit the current data and task. Pre-trained models usually have learned a large number of common feature representations, which helps improve performance in the current task.

[0117] (e) Anomaly detection and alarm mechanism:

[0118] Design an anomaly detection mechanism to monitor the output of the algorithm and detect possible anomalies or outliers. When the algorithm output does not match expectations, an alarm mechanism can be triggered so that potential defects can be discovered and handled in a timely manner.

[0119] By combining statistical methods and machine learning techniques, anomaly detection can be effectively achieved and the risk of missed detection can be reduced.

[0120] (f) Continuous monitoring and feedback loop:

[0121] Establish a mechanism for continuous monitoring and feedback loops so that new defect data can be collected and processed in a timely manner. Even if it cannot be used for training immediately, this data can be used to analyze and improve the algorithm. By continuously collecting feedback and updating the algorithm, its ability to identify rare and large defects can be gradually improved.

[0122] Since data collection is particularly difficult, these optimization strategies may not be able to completely eliminate the risk of missed judgments. Therefore, in practical applications, it is necessary to formulate appropriate solutions based on specific circumstances and needs, and comprehensively consider factors such as the accuracy, reliability and efficiency of the algorithm.

[0123] The beneficial effects of the present invention are as follows:

[0124] The present invention avoids the defects of relying on human experience and poor accuracy in conventional defect detection by constructing an industrial X-ray AI network; through HDR image enhancement processing, one-key image enhancement processing and mask image fusion, the influence of complex structure of detection devices, defects of shooting equipment itself and noise is reduced.

[0125] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0126] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for detecting internal defects of castings based on deep learning, characterized in that: include: Collect the image of the casting workpiece to be tested and obtain a 16-bit X-ray image to be tested; The 16-bit X-ray image to be detected is converted into an 8-bit image by using HDR image enhancement processing to obtain a pre-processed image to be detected; Perform one-key image enhancement processing on the pre-processed image to be trained and fuse it with a mask image generated according to the ROI detection area parameters set by the user to obtain the pre-processed data to be detected; Inputting the pre-processed data to be detected into a pre-trained industrial X-ray AI network for detection to obtain detection data; Performing post-processing operations on the detection data to obtain post-processed data; the post-processing operations include: original image defect display and image prediction; Performing data statistics on the post-processing data to obtain an output result, and saving the defect data in the output result according to the file type; The training process of the industrial X-ray AI network includes: Collect existing casting workpiece images to obtain 16-bit X-ray images to be trained; The 16-bit X-ray image to be trained is converted into an 8-bit image by using HDR image enhancement processing to obtain a pre-processed image to be trained; Performing one-key image enhancement processing on the pre-processed image to be trained to obtain pre-processed data to be trained; the one-key image enhancement processing includes: brightness adjustment, contrast adjustment, filtering and sharpening; Using an AI network generation algorithm based on a GAN network to generate images of the pre-processed data to be trained, to obtain generated data; Combining the pre-processed data to be trained and the generated data and performing a super-resolution enhancement operation based on an SR algorithm to obtain SR data; Performing labeled data sharing and combined pairing on the pre-processed data to be trained and the generated data respectively, and performing defect annotation on the SR data to obtain an original paired data set in a binary image format; Performing data enhancement operations on the original paired data set to obtain a generalized data set; the data enhancement operations include: rotation, scaling, shearing, and noise addition; Randomly divide the generalized data set into a validation data set and a training data set according to a preset matching ratio; The pre-training data set obtained by combining the verification data set and the training data set is input into the W-Net network and the model training is performed using the GPU to obtain the trained industrial X-ray AI network.

2. A casting internal defect detection method based on deep learning according to claim 1, characterized in that: The identification objects of the industrial X-ray AI network include: pores, slag inclusions, looseness, cracks and shrinkage cavities.

3. A casting internal defect detection method based on deep learning according to claim 1, characterized in that: The parameter control method of the industrial X-ray AI network includes: regularization technology.

4. A casting internal defect detection method based on deep learning according to claim 1, characterized in that: Also includes: When process parameters change, the defect data is updated and the industrial X-ray AI network is retrained using the updated defect data.

5. The method for detecting internal defects of castings based on deep learning according to claim 1, characterized in that: Also includes: When the performance indicator of the monitoring data exceeds the preset warning threshold, an early warning message is sent to the target object.

6. The method for detecting internal defects of castings based on deep learning according to claim 1, characterized in that: Also includes: Introduce anomaly detection mechanisms and feedback loop mechanisms; Performing secondary analysis on the output result using the anomaly detection mechanism to obtain defect omission data; The feedback loop mechanism is used to collect feedback data from target users, and the feedback data is used to optimize the industrial X-ray AI network.

Images