Petroleum pipe microscopic fracture morphology identification method, system, equipment and medium

Abstract

The invention discloses a petroleum pipe microscopic fracture morphology identification method, system and device and a medium, and belongs to the technical field of failure analysis image process.The method comprises the steps that a deep learning classification network with a residual module is constructed and trained, and a deep learning classification network loaded with a network connection weight matrix is obtained; and finally, the deep learning classification network is utilized to efficiently and accurately identify the to-be-tested object region image. According to the method, massive data sets can be quickly processed and analyzed by using the high-speed processing capability of a modern computer, so that the efficiency is greatly improved. Compared with a traditional method, the automatic fracture classification method can reduce subjective interference of people and enhance reliability and consistency of results. And meanwhile, the identification method is simple, engineering application integration is facilitated, and good practicability and expansibility are achieved.

Description

Technical Field

[0001] The invention belongs to the technical field of failure analysis image processing, and in particular relates to a method, system, equipment and medium for identifying microscopic fracture morphology of petroleum pipes. Background Art

[0002] Pattern recognition of typical microscopic fractures of oil pipes is one of the important contents of intelligent diagnosis of oil pipe failure analysis. Each stage of the fracture process will leave corresponding traces, morphology and characteristics on the fracture. By studying the characteristics, properties, morphology and other issues of the metal fracture of oil pipes, we can better reveal the fracture mechanism, study the various factors of the fracture process and fracture morphology, so as to better guide production practice and provide a basis for the design and manufacturing of pipe products.

[0003] In the failure analysis of oil pipes, the typical micro-fracture morphology is currently mainly identified by manual visual recognition, which has problems of low efficiency and low precision. Regarding low efficiency: Manual recognition requires professionals, so it takes a lot of time and effort, especially for large-scale data sets, it takes a long time to complete the classification. This not only increases the cost, but also limits its application in actual production. Regarding low precision: Manual recognition is often subject to individual differences, fatigue, emotions and other factors, and is a relatively subjective task. Even experienced professionals find it difficult to guarantee the accuracy and consistency of the results, and rely heavily on expert experience. In addition, human visual perception and processing capabilities are also limited, and it is easy to miss detection or misjudge when faced with fractures with complex and changeable shapes. Summary of the invention

[0004] In order to overcome the shortcomings of the above-mentioned technology, the present invention provides a method, system, equipment and medium for identifying micro-fracture morphology of petroleum pipes, which can solve the technical problem that the existing technology uses manual identification of micro-fracture morphology, which is not only inefficient but also has poor accuracy.

[0005] In order to achieve the above object, the present invention adopts the following technical solution:

[0006] A method for identifying micro fracture morphology of petroleum pipes, comprising:

[0007] S1: Obtain the microscopic fracture electron microscope image of the oil pipe and perform preprocessing to obtain a single morphological target image;

[0008] S2: classify the single morphological target image obtained in S1 to obtain a sample image, and construct training data for completing deep learning classification network training based on the sample image;

[0009] S3: construct a deep learning classification network, and use the training data obtained in S2 to train the deep learning classification network to obtain a deep learning classification network with a network connection weight matrix loaded;

[0010] S4: Input the image of the area of ​​the object to be tested into the deep learning classification network that has been loaded with the network connection weight matrix, and output the micro-fracture morphology recognition result of the oil pipe.

[0011] Furthermore, in S1, the specific steps of preprocessing the microscopic fracture electron microscope image of the oil pipe are:

[0012] The microscopic fracture electron microscope image of the oil pipe is converted into a grayscale image, and the grayscale image is normalized, cropped and marked in sequence to obtain a single morphological target image.

[0013] Furthermore, in S2, the classification categories of a single morphological target image include cleavage fracture, dimple fracture, intergranular fracture and fatigue fracture.

[0014] Furthermore, in S2, the training data is divided into a training set, a test set and a validation set, specifically including cleavage fracture sample images and their corresponding type labels, dimple fracture sample images and their corresponding type labels, intergranular fracture sample images and their corresponding type labels and fatigue fracture sample images and their corresponding type labels.

[0015] Furthermore, in S3, the deep learning classification network adopts a deep learning network based on a residual structure, wherein the initial end of the residual structure is integrated with a SE attention mechanism module.

[0016] Furthermore, the specific structure of the deep learning classification network includes a network input part, a network intermediate convolution part and a network output part; wherein the network input part includes a convolution kernel with a size of 7×7 and a step size of 2, and a maximum pooling layer with a size of 3×3 and a step size of 2; the network intermediate convolution part includes four residual structures with SE attention mechanism modules integrated at the initial end; the network output part includes a spatial pyramid pooling layer and a fully connected layer.

[0017] Furthermore, in S3, the training data is input into the deep learning classification network, and the deep learning classification network is trained in turn using label smoothing, classification weights, and the Adam optimization algorithm combined with the learning rate piecewise constant decay strategy to obtain a deep learning classification network loaded with the network connection weight matrix.

[0018] A system for identifying microscopic fracture morphology of a petroleum tubular product, used to implement the steps of the above-mentioned method for identifying microscopic fracture morphology of a petroleum tubular product, comprising:

[0019] An acquisition and processing module is used to acquire the microscopic fracture electron microscope image of the oil pipe and perform preprocessing to obtain a single morphological target image;

[0020] The image classification module is used to classify a single morphological target image to obtain a sample image, and to construct training data for completing deep learning classification network training based on the sample image;

[0021] Constructing a network module, which is used to construct a deep learning classification network, and train the deep learning classification network using training data to obtain a deep learning classification network with a network connection weight matrix loaded;

[0022] The morphology recognition module is used to input the image of the area of ​​the object to be tested into the deep learning classification network that has been loaded with the network connection weight matrix, and output the micro-fracture morphology recognition result of the oil pipe.

[0023] A device comprising:

[0024] Memory for storing computer programs;

[0025] The processor is used to implement the steps of the above-mentioned method for identifying the micro-fracture morphology of petroleum tubular goods when executing the computer program.

[0026] A computer-readable storage medium stores a computer program, which is used to implement the steps of the above-mentioned method for identifying microscopic fracture morphology of petroleum pipes when executed by a processor.

[0027] Compared with the prior art, the present invention has the following beneficial effects:

[0028] The present invention provides a method for identifying micro-fracture morphology of petroleum pipes. The present invention constructs and trains a deep learning classification network with a residual module to obtain a deep learning classification network loaded with a network connection weight matrix, and finally uses the deep learning classification network to efficiently and accurately identify the regional image of the object to be tested; the method can quickly process and analyze massive data sets by using the high-speed processing capabilities of modern computers, greatly improving efficiency. Compared with traditional methods, the automated fracture classification method can reduce subjective interference of people and enhance the reliability and consistency of the results. At the same time, the recognition method is simple, convenient for engineering application integration, and has good practicality and expansibility.

[0029] The method of the present invention has a wide range of application objects, is simple and easy to apply, has high recognition efficiency, and model maintenance does not require staff to have knowledge and experience in professional feature extraction of computer image recognition. It has great application potential in fracture micro-morphology recognition.

[0030] The present invention has good accuracy and real-time performance in classifying the microscopic fracture morphology of oil pipes. The classification time for a 512×512 image to be classified is only 5ms. Deep learning has a more abstract feature extraction capability, thereby improving the classification accuracy of the typical fracture microscopic morphology of oil pipes.

[0031] Preferably, the specially constructed network structure of the present invention makes image detail feature extraction richer through the improved residual structure combined with spatial pyramid pooling, improves the classification effect, and further enhances the image recognition ability.

[0032] Preferably, the deep learning classification network of the present invention adopts a deep learning network based on a residual structure. In this way, the structure can learn channel feature weights according to global information, recalibrate the channel information to increase the weights of effective channel features, reduce the weights of invalid or less effective channel features, and improve the model feature extraction capability.

[0033] Preferably, the output part of the deep learning classification network of the present invention improves the global average pooling into spatial pyramid pooling to extract spatial feature information of different output sizes, thereby further improving the robustness of the model to spatial layout and object deformation.

[0034] Preferably, after the present invention inputs the training data into the deep learning classification network, it uses label smoothing to process it, so as to alleviate the problems of low information entropy in classification and overly arbitrary models, and enhance the generalization ability of the model.

[0035] Preferably, after the present invention inputs the training data into the deep learning classification network, it is processed using classification weights to alleviate the class imbalance problem. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 A flow chart of a method for identifying micro-fracture morphology of a petroleum pipe provided by an embodiment of the present invention;

[0037] Figure 2 A model structure diagram provided for an embodiment of the present invention, wherein Figure 2 (a) is the schematic diagram of SE (SE-Basic Block) module. Figure 2 (b) is the structure diagram of the deep learning classification network;

[0038] Figure 3 A schematic diagram of a microscopic fracture sample image provided by an embodiment of the present invention, wherein: Figure 3 (a) is the image of the cleavage fracture sample; Figure 3 (b) is the image of dimple fracture sample; Figure 3 (c) is the image of the sample with intergranular fracture; Figure 3 (d) is the image of fatigue fracture sample;

[0039] Figure 4 Provides training sample result graphs using three deep learning classification networks for embodiments of the present invention;

[0040] FIG5 is a comparison diagram provided by an embodiment of the present invention; wherein FIG5(a) is a comparison diagram of the accuracy curve of the verification set; and FIG5(b) is a comparison diagram of the loss curve of the verification set;

[0041] Figure 6 A flow chart of a method for identifying micro-fracture morphology of a petroleum pipe provided by the present invention;

[0042] Figure 7 The present invention provides a structural schematic diagram of a system for identifying micro-fracture morphology of petroleum pipes. DETAILED DESCRIPTION

[0043] The present invention provides a method for identifying microscopic fracture morphology of petroleum pipes. Figure 6 As shown, the following steps are included:

[0044] S1: Obtain the microscopic fracture electron microscope image of the oil pipe and perform preprocessing to obtain a single morphological target image;

[0045] S2: classify the single morphological target image obtained in S1 to obtain a sample image, and construct training data for completing deep learning classification network training based on the sample image;

[0046] S3: construct a deep learning classification network, and use the training data obtained in S2 to train the deep learning classification network to obtain a deep learning classification network with a network connection weight matrix loaded;

[0047] S4: Input the image of the area of ​​the object to be tested into the deep learning classification network that has been loaded with the network connection weight matrix, and output the micro-fracture morphology recognition result of the oil pipe.

[0048] Specifically, in S1, the specific steps of preprocessing the microscopic fracture electron microscope image of the oil pipe are: converting the microscopic fracture electron microscope image of the oil pipe into a grayscale image, and normalizing, cropping, and marking the grayscale image in sequence to obtain a single morphological target image.

[0049] In S2, the classification categories of a single morphological target image include cleavage fracture, dimple fracture, intergranular fracture, and fatigue fracture.

[0050] In S2, the training data is divided into a training set, a test set and a validation set, specifically including cleavage fracture sample images and their corresponding type labels, dimple fracture sample images and their corresponding type labels, intergranular fracture sample images and their corresponding type labels, and fatigue fracture sample images and their corresponding type labels.

[0051] In S3, the deep learning classification network adopts a deep learning network based on a residual structure, wherein the initial end of the residual structure is integrated with a SE attention mechanism module.

[0052] The specific structure of the above-mentioned deep learning classification network includes a network input part, a network intermediate convolution part and a network output part; wherein, the network input part includes a convolution kernel of size 7×7 and stride 2, and a maximum pooling layer of size 3×3 and stride 2; the network intermediate convolution part includes four residual structures with SE attention mechanism modules integrated at the initial end; the network output part includes a spatial pyramid pooling layer and a fully connected layer.

[0053] In S3, the training data is input into the deep learning classification network, and the deep learning classification network is trained in turn using label smoothing, classification weight, and Adam optimization algorithm combined with a learning rate piecewise constant decay strategy to obtain a deep learning classification network with a network connection weight matrix loaded.

[0054] like Figure 7 As shown, the present invention also provides a system for identifying micro-fracture morphology of oil pipes, including: an acquisition processing module, an image classification module, a network construction module and a morphology recognition module; the acquisition processing module is used to acquire the electron microscope image of the micro-fracture of the oil pipe and perform preprocessing to obtain a single morphology target image; the image classification module is used to classify the single morphology target image to obtain a sample image, and construct training data for completing deep learning classification network training based on the sample image; the network construction module is used to construct a deep learning classification network, and use the training data to train the deep learning classification network to obtain a deep learning classification network loaded with a network connection weight matrix; the morphology recognition module is used to input the image of the area of ​​the object to be tested into the deep learning classification network loaded with the network connection weight matrix, and output the micro-fracture morphology recognition result of the oil pipe.

[0055] The present invention also provides a device, comprising: a memory for storing a computer program; a processor for implementing the steps of the oil pipe micro-fracture morphology recognition method when executing the computer program.

[0056] When the processor executes the computer program, the above-mentioned steps of micro-fracture morphology recognition of the oil pipe are implemented, for example: obtaining an electron microscope image of the micro-fracture of the oil pipe and preprocessing it to obtain a single morphology target image; classifying the single morphology target image to obtain a sample image, and constructing training data for completing deep learning classification network training based on the sample image; constructing a deep learning classification network, and using the training data to train the deep learning classification network to obtain a deep learning classification network loaded with a network connection weight matrix; inputting the image of the area of ​​the object to be tested into the deep learning classification network loaded with the network connection weight matrix, and outputting the micro-fracture morphology recognition result of the oil pipe.

[0057] Alternatively, the processor implements the functions of each module in the above system when executing the computer program, for example: an acquisition processing module, used to acquire the electron microscope image of the microscopic fracture of the oil pipe and perform preprocessing to obtain a single morphological target image; an image classification module, used to classify the single morphological target image to obtain a sample image, and construct training data for completing the deep learning classification network training based on the sample image; a network construction module, used to construct a deep learning classification network, and use the training data to train the deep learning classification network to obtain a deep learning classification network loaded with a network connection weight matrix; a morphology recognition module, used to input the image of the area of ​​the object to be tested into the deep learning classification network loaded with a network connection weight matrix, and output the microscopic fracture morphology recognition result of the oil pipe.

[0058] Exemplarily, the computer program can be divided into one or more modules / units, and the one or more modules / units are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units can be a series of computer program instruction segments that can complete preset functions, and the instruction segments are used to describe the execution process of the computer program in the oil pipe micro-fracture morphology recognition device. For example, the computer program can be divided into an acquisition processing module, an image classification module, a network construction module and a morphology recognition module; the acquisition processing module is used to acquire the micro-fracture electron microscope image of the oil pipe and perform preprocessing to obtain a single morphology target image; the image classification module is used to classify the single morphology target image to obtain a sample image, and construct training data for completing the deep learning classification network training according to the sample image; the network construction module is used to construct a deep learning classification network, and use the training data to train the deep learning classification network to obtain a deep learning classification network that has been loaded with a network connection weight matrix; the morphology recognition module is used to input the image of the area of ​​the object to be tested into the deep learning classification network that has been loaded with a network connection weight matrix, and output the micro-fracture morphology recognition result of the oil pipe.

[0059] The oil pipe micro fracture morphology recognition device can be a computing device such as a desktop computer, a notebook, a PDA, and a cloud server. The oil pipe micro fracture morphology recognition device may include, but is not limited to, a processor and a memory. Those skilled in the art can understand that the above is an example of an oil pipe micro fracture morphology recognition device and does not constitute a limitation on the oil pipe micro fracture morphology recognition device. It may include more components than the above, or a combination of certain components, or different components. For example, the oil pipe micro fracture morphology recognition device may also include input and output devices, network access devices, buses, etc.

[0060] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may be any conventional processor, etc. The processor is the control center of the micro-fracture morphology recognition of the oil pipe material, and uses various interfaces and lines to connect various parts of the entire micro-fracture morphology recognition device of the oil pipe material.

[0061] The memory can be used to store the computer program and / or module, and the processor realizes various functions of the oil pipe micro-fracture morphology recognition device by running or executing the computer program and / or module stored in the memory and calling the data stored in the memory.

[0062] The memory may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), etc. In addition, the memory may include a high-speed random access memory, and may also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (SmartMediaCard, SMC), a secure digital (SecureDigital, SD) card, a flash card (FlashCard), at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0063] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method for identifying microscopic fracture morphology of petroleum pipes are implemented.

[0064] If the modules / units integrated in the oil pipe micro-fracture morphology recognition system are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium.

[0065] Based on such understanding, the present invention implements all or part of the processes in the above-mentioned oil pipe micro-fracture morphology recognition method, and can also be completed by instructing related hardware through a computer program, and the computer program can be stored in a computer-readable storage medium, and when the computer program is executed by a processor, the steps of the above-mentioned oil pipe micro-fracture morphology recognition method can be implemented. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or preset intermediate form, etc.

[0066] The computer-readable storage medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal and software distribution medium, etc.

[0067] It should be noted that the content contained in the computer-readable storage medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable storage media do not include electrical carrier signals and telecommunication signals.

[0068] Example

[0069] The present invention will be further described below in conjunction with the embodiments and drawings:

[0070] In order to solve the technical problem of how to improve the classification and recognition efficiency and recognition accuracy of typical micro-fracture morphology of oil pipes in the above-mentioned background technology, this embodiment provides a method for identifying micro-fracture morphology of oil pipes. By using the high-speed processing capabilities of modern computers, this method can quickly process and analyze massive data sets, greatly improving efficiency. Compared with traditional methods, the automated fracture classification method can reduce human subjective interference and enhance the reliability and consistency of the results. At the same time, the recognition method is simple, convenient for engineering application integration, and has good practicality and scalability; the specific steps include:

[0071] like Figure 1 As shown, the specific steps include:

[0072] Step (1): obtaining an electron microscope image of a microscopic fracture of a petroleum pipe, wherein the electron microscope image specifically refers to a scanning electron microscope image of a typical microscopic fracture morphology of a petroleum pipe, converting the obtained image into a grayscale image, normalizing and cropping the image, and extracting a single morphology target image;

[0073] Step (2): Divide all the images in step (1) into cleavage fracture images, dimple fracture images, intergranular fracture images, and fatigue fracture images, and construct training data, which is used to train a deep learning classification network. The training data specifically includes cleavage sample images and their corresponding type labels, dimple sample images and their corresponding type labels, intergranular sample images and their corresponding type labels, and fatigue sample images and their corresponding type labels.

[0074] Step (3): Design a deep learning classification network for typical micro-fracture pattern recognition of oil pipes, use the training data of step (2) for training, and after the training is completed, save the deep learning classification network with the network connection weight matrix H loaded for subsequent recognition;

[0075] Step (4): convert the image of the test sample to be tested into a grayscale image and then perform normalization and cropping processing, obtain the image area of ​​each test object and input it into the deep learning classification network obtained in step (3) in turn, complete the classification of typical micro-fracture morphology of oil pipes, that is, output the micro-fracture morphology recognition result of oil pipes.

[0076] Specifically, the typical microscopic fracture image collected in the above step (1) is converted into a grayscale image, and then the grayscale image is processed by normalization, cropping and other operations to obtain an image label to which it belongs, and the image category of the grayscale image is obtained based on the label.

[0077] The specific steps of step 2 are as follows:

[0078] 2.1, classify the image categories of step (1) into cleavage fracture, dimple fracture, intergranular fracture, and fatigue fracture, and the sample image is an image in which the classified object has a large range of a single fracture mode in the image area;

[0079] 2.2, construct a set of cleavage fracture sample images and their corresponding type labeled 0, construct a set of dimple fracture sample images and their corresponding type labeled 1, construct a set of intergranular fracture sample images and their corresponding type labeled 2, and construct a set of fatigue fracture sample images and their corresponding type labeled 3, thereby forming training data containing typical microscopic fracture patterns of the fracture surface.

[0080] The above-mentioned fracture modes are cleavage fracture mode, dimple fracture mode, intergranular fracture mode, and fatigue fracture mode.

[0081] The deep learning classification network designed in step (3) for typical micro-fracture pattern recognition of oil pipes is further described as follows:

[0082] 3.1, such as Figure 2 As shown in (b), a new deep learning network ResSNet based on residual structure is built; the residual structure incorporates SE (Squeeze-and-Excitation) attention mechanism, which can learn channel feature weights based on global information, recalibrate channel information to increase the weight of effective channel features, reduce the weight of invalid or less effective channel features, and improve the model feature extraction capability. Figure 2 As shown in (a), the SE module is added to the initial and end of the residual structure (the initial end has a better effect), followed by a 3×3 convolution, followed by batch normalization and ReLU function activation layer, and then another 3×3 convolution and normalization, and finally added to the input feature map (i.e., training set) and activated by ReLU function. The output part extracts spatial feature information of different output sizes by improving the global average pooling to spatial pyramid pooling (Spp), further improving the model's robustness to spatial layout and object deformation.

[0083] 3.2 The deep learning classification network is mainly composed of the network input part, the network intermediate convolution part and the network output part.

[0084] The network input consists of a large convolution kernel of size 7×7 and stride 2, and a maximum pooling layer of size 3×3 and stride 2. The middle convolution part of the network mainly extracts information through the stacking of 3×3 convolutions, which is mainly composed of four improved residual structures in 3.1 (i.e., the residual structure with SE attention mechanism module integrated at the initial end).

[0085] In addition, the network output part consists of a one-dimensional data conversion layer plus a fully connected layer (FC). The one-dimensional data conversion layer is composed of spatial pyramid pooling: 1×1, 2×2, 4×4 three maximum pooling methods, each input feature is pooled, and then the pooled features are spliced ​​to obtain the final one-dimensional feature vector. Then the output of the fully connected layer is connected, and the one-dimensional data conversion layer is connected to the output layer through the fully connected layer Softmax function classification layer. The number of output nodes is 4, which is consistent with the predicted category (cleavage, dimple, intergranular, fatigue).

[0086] 3.3. Input the training data of step (2) into the input layer of the deep learning classification network, and use label smoothing to alleviate the problem of insufficient information entropy in the classification and the overly arbitrary model, so as to enhance the generalization ability of the model; adopt the classification weight method to alleviate the class imbalance problem, and use the Adam optimization algorithm combined with the learning rate piecewise constant decay strategy to train the deep learning classification network until the error of the deep learning classification network reaches the minimum value and converges, and obtain the network connection weight matrix H composed of various parameters in the deep learning classification network after training, where the loss calculation adopts the multi-classification cross entropy loss function.

[0087] It should be noted that the deep learning classification network in step (4) is a deep learning classification network that has been loaded with the network connection weight matrix H.

[0088] Combine the following Figure 2 FIG5 further explains a method for identifying micro fracture morphology of a petroleum pipe provided in this embodiment, and the specific implementation process is as follows:

[0089] Step 1: Collect typical micro-fracture morphology scanning electron microscope images of oil pipes. The data set consists of valuable micro-fracture pattern images taken by scanning electron microscope and in books and papers. Convert them into grayscale images and crop them to obtain the key feature image area of ​​each object;

[0090] Step 2: Divide all image areas in step 1 into training sets and test sets, and construct training data for training the deep learning classification network, wherein the training data includes cleavage fracture sample images and their corresponding type labels, dimple fracture sample images and their corresponding type labels, intergranular fracture sample images and their corresponding type labels, and fatigue fracture sample images and their corresponding type labels; the specific process is as follows:

[0091] The key feature image categories of step 1 are divided into cleavage fracture, dimple fracture, intergranular fracture, and fatigue fracture. The sample image here is an image in which the identified object has a large range of a single fracture mode in the image area;

[0092] Construct a set of cleavage fracture sample images and their corresponding type labeled 0, construct a set of dimple fracture sample images and their corresponding type labeled 1, construct a set of intergranular fracture sample images and their corresponding type labeled 2, and construct a set of fatigue fracture sample images and their corresponding type labeled 3, thereby forming training data containing typical microscopic fracture modes of the fracture surface.

[0093] The above training data are randomly divided into batches, 50% of the training data are used as the training set, 20% of the training data are used as the validation set, and 30% of the training data are used as the test set.

[0094] Step 3: Design Figure 2 The deep learning classification network of the typical micro-fracture morphology shown in the figure is trained. After the training, the classification network connection weight matrix H is obtained for the subsequent identification and detection steps. The specific process is as follows:

[0095] First, a new deep learning network ResSNet based on residual structure is built. The residual structure incorporates the SE (Squeeze-and-Excitation) attention mechanism, which can learn channel feature weights based on global information, recalibrate channel information to increase the weight of effective channel features, reduce the weight of invalid or less effective channel features, and improve the model feature extraction capability. The SE module was added to the initial end and the end of the residual structure (the initial end has a better effect), and then a 3×3 convolution was performed, followed by batch normalization and ReLU function activation layer, and then 3×3 convolution and normalization were performed again, and finally it was added to the input feature map and activated by ReLU function. The output part uses spatial pyramid pooling (Spp) to extract spatial feature information of different output sizes, further improving the robustness of the model to spatial layout and object deformation.

[0096] The deep learning classification network mentioned above is mainly composed of a network input part, a network intermediate convolution part, and a network output part.

[0097] The network input consists of a large convolution kernel of size 7×7 and stride 2, and a maximum pooling layer of size 3×3 and stride 2. The middle convolution part of the network mainly extracts information through the stacking of 3×3 convolutions, which is mainly composed of four improved residual structures.

[0098] The network output consists of a one-dimensional data conversion layer plus a fully connected layer (FC). The one-dimensional data conversion layer consists of spatial pyramid pooling: 1×1, 2×2, 4×4 three maximum pooling methods, each input feature is pooled, and then the pooled features are concatenated to obtain the final one-dimensional feature vector. Then the output of the fully connected layer is connected, and the one-dimensional data conversion layer is connected to the output layer through the fully connected layer Softmax function classification layer. The number of output nodes is 4, which is consistent with the predicted category (cleavage, dimple, intergranular, fatigue).

[0099] Second, the training data is input into the input layer of the deep learning classification network. Label smoothing is used to alleviate the problem of insufficient information entropy and overly arbitrary model in classification, enhance the generalization ability of the model, and use classification weights to alleviate the class imbalance problem. The Adam optimization algorithm is combined with the learning rate piecewise constant decay strategy to train the deep learning classification network until the error of the deep learning classification network reaches the minimum value and converges. The network connection weight matrix H composed of various parameters in the deep learning classification network after training is obtained, and the loss calculation adopts the multi-classification cross entropy loss function;

[0100] Step 4: Collect the scanning electron microscope micro-fracture image to be tested, convert it into a grayscale image, normalize and crop it, and input it into the deep learning classification network of the connection weight matrix H loaded in step 3 to complete the recognition of the typical micro-fracture morphology of oil pipes.

[0101] This embodiment adopts Figure 3 The micro-fracture morphology shown is used as a training sample, where Figure 3 (a) is the image of the cleavage fracture sample; Figure 3 (b) is the image of dimple fracture sample; Figure 3 (c) is the image of the sample with intergranular fracture; Figure 3 (d) is the image of fatigue fracture sample; it can be seen from the figure that the cleavage fracture microstructure is composed of a series of small crack planes (cleavage planes of each grain); on each cleavage plane, some steps very close to the crack propagation direction can be seen, usually called cleavage steps. The dimple fracture microstructure is honeycomb-shaped, and the fracture surface is composed of some tiny pits. The pits are actually grown-up hollow cores, usually called dimples. The intergranular fracture microstructure is a polyhedral morphology of each grain, similar to the accumulation of rock candy. The fatigue fracture microstructure is a stripe pattern, with a certain spacing, perpendicular to the crack propagation direction, light and dark intersecting and parallel to each other.

[0102] like Figure 4As shown, the classification and identification of typical micro-fracture morphology are completed by this method, and the classification accuracy is compared by using the Resnet18 model, the Resnet18 optimization model and the ResSNet model, and the classification of typical micro-fracture morphology is completed. Among them, the ResSNet model is the basic architecture of the deep learning classification network in this embodiment, and Resnet18 is a model in which the network depth of the ResSNet model is designed to be 18 layers. The Resnet18 optimization model is obtained by re-adjusting the parameters of Resnet18 using optimization methods such as category weight, label smoothing, and learning rate decay. As shown in Figure 5, this embodiment is verified using a validation set, and the accuracy curve and loss curve of each model are plotted, see Figure 5 (a) and Figure 5 (b) for details. It can be seen from the figure that the classification model provided by this embodiment has good accuracy and practicality, and has great application potential in the diagnosis of typical micro-fractures of oil pipes.

[0103] This embodiment provides a method for identifying the micro-fracture morphology of oil pipes, which obtains the electron microscope image of the micro-fracture of the oil pipe to be identified, converts the grayscale image for normalization and obtains each image area of ​​the object to be identified; prepares cleavage fracture, dimple fracture, fatigue fracture, and intergranular fracture sample image data for training and testing; designs a deep learning classification network and completes the classification model training, and saves the trained neural network connection weight matrix; sends the image of the area of ​​the object to be tested to the trained deep learning classification network for image classification detection, and completes the classification detection of typical micro-fracture morphology. The present invention can accurately and effectively identify typical micro-fracture morphology, facilitate failure analysis and engineering applications, and has great value; compared with the existing technology, this method has the following advantages:

[0104] 1. The specially constructed network structure of this method makes the image detail feature extraction richer through the improved residual structure combined with spatial pyramid pooling, improves the classification effect, and further enhances the image recognition ability.

[0105] 2. This method has good accuracy and real-time performance for the classification of micro-fracture morphology of oil pipes. The classification time for a 512×512 image to be classified is only 5ms. Deep learning has more abstract feature extraction capabilities, thereby improving the classification accuracy of typical fracture micro-morphology of oil pipes.

[0106] 3. This method has a wide range of applications, is simple and easy to apply, has high recognition efficiency, and model maintenance does not require staff to have knowledge and experience in professional feature extraction of computer image recognition. It has great application potential in fracture micromorphology recognition.

[0107] The above embodiment is only one of the implementation methods that can realize the technical solution of the present invention. The scope of protection claimed by the present invention is not limited only to this embodiment, but also includes changes, replacements and other implementation methods that can be easily thought of by any technician familiar with the technical field within the technical scope disclosed by the present invention.

Claims

1. A method for identifying micro fracture morphology of petroleum pipes, characterized in that: include: S1: Obtain the microscopic fracture electron microscope image of the oil pipe and perform preprocessing to obtain a single morphological target image; S2: classify the single morphological target image obtained in S1 to obtain a sample image, and construct training data for completing deep learning classification network training based on the sample image; S3: construct a deep learning classification network, and use the training data obtained in S2 to train the deep learning classification network to obtain a deep learning classification network with a network connection weight matrix loaded; S4: Input the image of the area of ​​the object to be tested into the deep learning classification network that has been loaded with the network connection weight matrix, and output the micro-fracture morphology recognition result of the oil pipe.

2. A method for identifying micro fracture morphology of a petroleum pipe according to claim 1, characterized in that: In S1, the specific steps of preprocessing the microscopic fracture electron microscope image of the oil pipe are as follows: The microscopic fracture electron microscope image of the oil pipe is converted into a grayscale image, and the grayscale image is normalized, cropped and marked in sequence to obtain a single morphological target image.

3. The method for identifying micro fracture morphology of a petroleum pipe according to claim 1, characterized in that: In S2, the classification categories of a single morphological target image include cleavage fracture, dimple fracture, intergranular fracture, and fatigue fracture.

4. A method for identifying micro fracture morphology of a petroleum pipe according to claim 3, characterized in that: In S2, the training data is divided into a training set, a test set and a validation set, specifically including cleavage fracture sample images and their corresponding type labels, dimple fracture sample images and their corresponding type labels, intergranular fracture sample images and their corresponding type labels, and fatigue fracture sample images and their corresponding type labels.

5. The method for identifying micro fracture morphology of a petroleum pipe according to claim 1, characterized in that: In S3, the deep learning classification network adopts a deep learning network based on a residual structure, wherein the initial end of the residual structure is integrated with a SE attention mechanism module.

6. A method for identifying micro fracture morphology of a petroleum pipe according to claim 5, characterized in that: The specific structure of the deep learning classification network includes a network input part, a network intermediate convolution part and a network output part; wherein the network input part includes a convolution kernel with a size of 7×7 and a stride of 2, and a maximum pooling layer with a size of 3×3 and a stride of 2; the network intermediate convolution part includes four residual structures with SE attention mechanism modules integrated at the initial end; the network output part includes a spatial pyramid pooling layer and a fully connected layer.

7. A method for identifying micro fracture morphology of a petroleum pipe according to claim 1, characterized in that: In S3, the training data is input into the deep learning classification network, and the deep learning classification network is trained in turn using label smoothing, classification weights, and the Adam optimization algorithm combined with the learning rate piecewise constant decay strategy to obtain a deep learning classification network with a network connection weight matrix loaded.

8. A system for identifying microscopic fracture morphology of a petroleum tubular product, used to implement the steps of the method for identifying microscopic fracture morphology of a petroleum tubular product according to any one of claims 1 to 7, characterized in that: include: An acquisition and processing module is used to acquire the microscopic fracture electron microscope image of the oil pipe and perform preprocessing to obtain a single morphological target image; The image classification module is used to classify a single morphological target image to obtain a sample image, and to construct training data for completing deep learning classification network training based on the sample image; Constructing a network module, which is used to construct a deep learning classification network, and train the deep learning classification network using training data to obtain a deep learning classification network with a network connection weight matrix loaded; The morphology recognition module is used to input the image of the area of ​​the object to be tested into the deep learning classification network that has been loaded with the network connection weight matrix, and output the micro-fracture morphology recognition result of the oil pipe.

9. A device, characterized in that: include: Memory for storing computer programs; A processor is used to implement the steps of the method for identifying micro-fracture morphology of petroleum tubular products according to any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, it is used to implement the steps of the method for identifying micro-fracture morphology of petroleum tubular products according to any one of claims 1 to 7.

Images