A pipeline damage assessment method and anti-corrosion method based on multidimensional data fusion

Abstract

The present invention relates to the field of chemical production technology, and discloses a pipeline damage assessment method and anti-corrosion method based on multidimensional data fusion. The pipeline damage assessment method includes: configuring a trained lightweight neural network model, inputting a pipeline damage image into the network model, and obtaining a probability distribution of damage types after feature extraction, feature enhancement, feature mapping and output; using a corrosion-related indicator data set as input, applying a correlation coefficient method to evaluate the degree of correlation between corrosion factors and corrosion rates, and obtaining an initial strongly correlated data set; expanding the initial strongly correlated data set to obtain a strongly correlated data set; using a strongly correlated data set as input, applying an Apriori association algorithm to evaluate the strong correlation rules between corrosion factors; and performing qualitative and quantitative assessments of chemical pipeline damage based on the probability distribution of damage types and strong correlation rules. The present invention can provide an in-depth understanding of the mechanism and influencing factors of pipeline corrosion, and provide a basis for formulating targeted corrosion protection strategies.

Description

Technical Field

[0001] The present invention belongs to the technical field of chemical production, and in particular relates to a pipeline damage assessment method and an anti-corrosion method based on multi-dimensional data fusion. Background Art

[0002] In the chemical production industry, equipment is frequently exposed to a range of complex operating conditions, including but not limited to corrosion, wear, thermal stress, and pressure cycling. The combined effects of these factors can lead to structural damage to metal materials, compromising equipment reliability and safety. Metal damage not only shortens the expected lifespan of equipment and increases the financial burden of repair and replacement, but in extreme cases, can also lead to serious safety incidents, posing a threat to operator safety and environmental health. Therefore, accurate identification and assessment of steel structural damage in chemical equipment is essential, not only to extend equipment lifespan and reduce maintenance costs, but also to prevent potential safety risks and ensure personnel safety and environmental protection.

[0003] There are many factors affecting pipeline corrosion, and their correlation analysis has been widely studied in data mining. However, in the field of data mining, the correlation analysis of corrosion influencing factors by domestic and foreign scholars is mainly limited to the analysis of individual corrosion factors of specific common pipelines, such as the corrosion analysis of pipelines transporting sour natural gas. However, there is less research on the correlation of corrosion influencing factors of process pipelines in industrial plants. The correlation of process pipeline corrosion factors is still unclear, making process pipeline corrosion prevention measures not targeted and the corrosion prevention efficiency low. In addition, the data sources of some current studies are not real pipeline corrosion data, but data obtained from simulation experiments or tests. These data cannot fully reflect the laws contained in real pipeline corrosion data and may deviate slightly from the actual laws. In addition, each study only studies the correlation analysis of various corrosion influencing factors and does not provide how to optimize existing pipeline corrosion prevention measures based on the current laws.

[0004] In summary, the existing technologies have shortcomings such as low pipeline damage detection accuracy, insufficient or uneven distribution of research data samples, and failure to propose effective anti-corrosion measures. Summary of the Invention

[0005] The purpose of the present invention is to address the deficiencies of the existing technology and propose a pipeline damage assessment method and an anti-corrosion method based on multidimensional data fusion.

[0006] To achieve the above objectives, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides a pipeline damage assessment method based on multidimensional data fusion, comprising the following steps:

[0008] Configure a trained lightweight neural network model and input the pipeline damage image into the lightweight neural network model. After feature extraction, feature enhancement, feature mapping and output, the probability distribution of damage types is obtained.

[0009] Taking the corrosion-related index dataset as input, the correlation coefficient method is applied to evaluate the correlation between corrosion factors and corrosion rates, and an initial strong correlation dataset is obtained. The initial strong correlation dataset is a small sample dataset.

[0010] Expanding the initial strongly correlated data set to obtain a strongly correlated data set;

[0011] Taking the strongly correlated dataset as input, the Apriori association algorithm is applied to evaluate the strong association rules between corrosion factors;

[0012] Qualitative and quantitative evaluation of chemical pipeline damage is performed based on the probability distribution of damage types and strong association rules.

[0013] As a possible implementation method, the backbone network of the lightweight neural network model is MobileNetV2, and the attention mechanism module is placed after MobileNetV2; feature extraction specifically includes:

[0014] S10. The pipeline damage image is subjected to up-convolution to expand its channels. Up-convolution is depth-wise convolution, meaning each channel is convolved separately.

[0015] S11. Apply activation functions to enhance the nonlinear characteristics of MobileNetV2 and constrain the output values ​​to be between [0, 6].

[0016] S12. Reduce the number of channels from the expanded number of channels to the original number of channels through dimensionality reduction convolution. Dimensionality reduction convolution is point-by-point convolution, which integrates information between channels.

[0017] After S10 to S12, the features of the pipeline damage image are extracted and output as a feature map Y0.

[0018] As a possible implementation method, the attention mechanism module includes channel attention and spatial attention, and feature enhancement specifically includes:

[0019] Channel attention generates channel attention weights through average pooling and maximum pooling to highlight important channel information;

[0020] The pooled results are processed by a fully connected network to obtain the channel attention weights;

[0021] Multiply the calculated channel attention weight by the feature map Y0 to obtain the channel attention enhanced feature map Y1.

[0022] As a possible implementation method, feature enhancement also includes:

[0023] Perform global average pooling and global maximum pooling on the feature map Y1 to obtain the pooling result at each position;

[0024] After concatenating the pooling at each position, the spatial attention weight is generated through convolution;

[0025] The spatial attention weight is multiplied with the feature map Y1 to highlight the important spatial regions in the feature map Y1 and obtain the spatial attention enhanced feature map Y2.

[0026] As a possible implementation method, feature mapping and output specifically include:

[0027] The feature map Y2 is flattened into a vector, and the flattened vector is input into the fully connected layer to obtain the probability distribution of the damage category.

[0028] As a possible implementation method, the initial strongly correlated dataset is obtained by the following method:

[0029] S20. Obtain an original data set and preprocess it to obtain a corrosion-related indicator data set, where the preprocessing includes at least data cleaning, removing outliers, and filling missing values. The corrosion-related indicator data set includes multiple corrosion factors X and corresponding corrosion rates Y.

[0030] S21. For normally distributed data, calculate the Pearson product-moment correlation coefficient τ between the corrosion factor X and the corrosion rate Y xy For non-normally distributed data, calculate the Spearman rank correlation coefficient τ between the corrosion factor X and the corrosion rate Y s ; For each corrosion factor X in the corrosion-related index data set, the corresponding correlation coefficient is obtained

[0031] S22. Configure the correlation coefficient threshold and set the correlation coefficient Corrosion factors X greater than or equal to the correlation coefficient threshold are retained, and the correlation coefficient is discarded The corrosion factors X that are smaller than the correlation coefficient threshold are composed of all retained corrosion factors X and their corresponding corrosion rates Y to form a strongly correlated data set.

[0032] As a possible implementation method, strong association rules are determined by the following method:

[0033] The Apriori association algorithm gradually scans the strongly correlated data set and generates frequent item sets using the properties of frequent item sets;

[0034] For X and Y in the frequent itemset, if And X∪Y is a frequent item set, then it can form an association rule Its confidence level is:

[0035]

[0036] As a possible implementation method, the chemical pipeline damage assessment method also includes: formulating an anti-corrosion mechanism based on strong association rules, specifically including:

[0037] S30. Obtaining a recommended value of the corrosion rate, and obtaining a recommended value of the corrosion influencing factor and a recommended value of the corrosion rate based on the strong association rule;

[0038] S31. Obtaining a monitoring value of the corrosion rate, monitoring the corrosion rate and corrosion factors, and obtaining a monitoring value of the corrosion rate;

[0039] S32 determines whether the monitoring value exceeds the recommended value. If so, execute S30. Otherwise, execute S31 to S32 in a loop.

[0040] S33. Alarm and decision support;

[0041] S34. Execute decision adjustment according to strong association rules;

[0042] S35. Determine whether the alarm is eliminated. If so, execute S32 to S35 in a loop. Otherwise, perform a process anti-corrosion inspection.

[0043] As a possible implementation approach, a generative adversarial network is applied to expand the initial strongly correlated dataset to obtain a strongly correlated dataset.

[0044] In a second aspect, the present invention provides a pipeline anti-corrosion method based on multidimensional data fusion, comprising the following steps:

[0045] The configured trained lightweight neural network model is used to assess the damage locations and types of chemical pipelines, and a dangerous point inspection log is established; damage types include mechanical damage and chemical corrosion damage;

[0046] Apply the mined strong association rules to improve the existing chemical pipeline anti-corrosion strategy.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] 1. The pipeline damage assessment method based on multidimensional data fusion provided by the present invention proposes the use of a lightweight neural network model to process pipeline damage images. This model uses MobileNetV2 as the backbone network, and places an attention mechanism module after MobileNetV2. The backbone network MobileNetV2 itself is based on depthwise separable convolution and has the characteristics of lightweight and efficient computing. After the introduction of the attention mechanism module, the features are dynamically weighted and optimized through the channel attention mechanism and the spatial attention mechanism, which significantly enhances the model's ability to focus on key features.

[0049] 2. The pipeline damage assessment method based on multidimensional data fusion, provided by this invention, uses a small sample size as its initial strongly correlated dataset. It then combines a generative adversarial network with the Apriori association rule algorithm to evaluate strong association rules between corrosion factors. This small sample size-based strong association rule approach addresses the issues of insufficient corrosion data and inefficient rule mining, enabling more accurate mining of association rules between corrosion-influencing factors and providing theoretical support for corrosion prediction and prevention.

[0050] 3. This invention combines the correlation coefficient method, the Generative Adversarial Network (GAN) adversarial neural network, and the Apriori association rule algorithm to implement a data mining algorithm for studying the correlation between various corrosion factors and corrosion rates. The correlation coefficient method is used to evaluate the degree of correlation between each factor and the corrosion rate in the original dataset, generating a small sample dataset. Highly correlated corrosion influencing factors are selected and this small sample dataset is expanded using the Generative Adversarial Network (GAN) adversarial neural network. Finally, the Apriori association rule mining algorithm is used to discover potential association rules. This ensures the accuracy of the correlation assessment between each corrosion factor and the corrosion rate, facilitates a deeper understanding of the mechanisms and influencing factors of pipeline corrosion, and provides a scientific basis for developing targeted corrosion protection strategies.

[0051] 4. This invention also proposes a pipeline corrosion prevention method based on multidimensional data fusion. By inspecting and assessing pipeline corrosion damage based on multidimensional data, this method is developed to optimize existing maintenance measures, ultimately maintaining a high level of process pipeline safety at a lower cost. These improvements will help improve pipeline safety and reliability, extend its service life, and reduce maintenance and repair costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] The accompanying drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the accompanying drawings:

[0053] Figure 1 A flow chart of a pipeline damage assessment method based on multidimensional data fusion is provided for an embodiment of the present invention;

[0054] Figure 2 A graph showing the iterative training process of the CBAM-MobileNet V2 model provided in an embodiment of the present invention;

[0055] Figure 3 A flowchart of the GAN-Apriori association rule pipeline corrosion factor data mining process provided by an embodiment of the present invention;

[0056] Figure 4 A flow chart of a pipeline damage assessment and anti-corrosion method based on multidimensional data fusion provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0057] To facilitate a clear description of the technical solutions of the embodiments of the present invention, the words "first" and "second" are used in the embodiments of the present invention to distinguish between identical or similar items with substantially the same functions and effects. For example, the first threshold and the second threshold are merely used to distinguish between different thresholds and do not limit their order. Those skilled in the art will understand that the words "first" and "second" do not limit the quantity or execution order, and the words "first" and "second" do not necessarily mean different.

[0058] It should be noted that, in the present invention, words such as "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in the present invention should not be construed as being preferred or advantageous over other embodiments or designs. Rather, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0059] In the present invention, "at least one" refers to one or more, and "more" refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent: the existence of A alone, the existence of A and B at the same time, and the existence of B alone, where A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. The following at least one item (item) or similar expressions thereof refer to any combination of these items, including any combination of single items (items) or plural items (items). For example, at least one item (item) of a, b or c can represent: a, b, c, the combination of a and b, the combination of a and c, the combination of b and c, or the combination of a, b and c, where a, b, c can be single or multiple.

[0060] The embodiments of the present invention aim to provide a pipeline damage assessment method and anti-corrosion method based on multidimensional data fusion, and propose a trained lightweight neural network model. In this model, an attention mechanism module is placed after the backbone network MobileNetV2, enabling the model to accurately extract features of pipeline damage images with almost no increase in computing resource consumption and running time.

[0061] In the first aspect, the present invention provides a chemical pipeline damage assessment method based on multi-dimensional data fusion, see Figure 1 , including the following steps:

[0062] Configure a trained lightweight neural network model and input the pipeline damage image into the lightweight neural network model. After feature extraction, feature enhancement, feature mapping and output, the probability distribution of damage types is obtained.

[0063] As a possible implementation, the lightweight neural network model uses MobileNetV2 as its backbone, followed by an attention mechanism module. While the MobileNetV2 backbone, built around depthwise separable convolutions, is lightweight and computationally efficient, its ability to model global feature dependencies is limited. The introduction of the attention mechanism module (CBAM) significantly enhances the model's ability to focus on key features by dynamically weighting and optimizing features through channel-wise and spatial-wise attention mechanisms.

[0064] As an example, the structure of the lightweight neural network model (CBAM-MobileNet) is shown in Table 1:

[0065] Table 1 Lightweight neural network model structure

[0066]

[0067]

[0068] As a possible implementation method, feature extraction specifically includes:

[0069] S10. The pipeline damage image is subjected to up-convolution to expand its channels. Up-convolution is depth-wise convolution, meaning each channel is convolved separately.

[0070] As an example, MobileNetV2 is a lightweight deep neural network whose core modules are inverted residual structure and depthwise separable convolution. Assuming that the pipeline damage image is X, the input image X first passes through the inverted residual module of MobileNetV2, which is composed of multiple depthwise separable convolutions and inverted residual structures. First, the input image X is convolved with the dimension-raising convolution W.expand Expand the number of channels. The convolution operation at this time belongs to depth convolution, that is, each channel is convolved separately.

[0071] S11. Apply activation functions to enhance the nonlinear characteristics of MobileNetV2 and constrain the output values ​​to be between [0, 6].

[0072] As an example, the ReLU6 activation function is applied to the result after dimensionality upgrade in step S10 to enhance the nonlinear characteristics of MobileNetV2 and limit the output value to [0, 6], thus avoiding the situation where the value is too large or too small.

[0073] S12. Reduce the number of channels from the expanded number of channels to the original number of channels through dimensionality reduction convolution. Dimensionality reduction convolution is point-by-point convolution, which integrates information between channels.

[0074] As an example, using the dimensionality reduction convolution W proj The number of channels is reduced from the expanded number to the original number. This operation is also a point-by-point convolution, i.e., a 1×1 convolution. Its purpose is to integrate information between channels through 1×1 convolution. After S10 to S12, the features of the pipeline damage image are extracted and output as feature map Y0 as follows:

[0075] Y0=W proj ReLU6(W expand ·X)

[0076] Among them, W expand is the dimension-raising weight, used to expand the number of channels, W proj is the dimensionality reduction weight used to reduce the number of channels, and ReLU6(z)=min(max(0,z),6) is the activation function used to limit the value range.

[0077] Next, the above feature map Y0 is input into the attention mechanism module, which includes channel attention and spatial attention. The feature representation is further enhanced by channel attention and spatial attention.

[0078] As a possible implementation method, feature enhancement specifically includes:

[0079] Channel attention generates channel attention weights through average pooling and maximum pooling to highlight important channel information;

[0080] The pooled results are processed by a fully connected network to obtain the channel attention weights, as shown in the following formula:

[0081] M c (Y0)=σ(MLP(AvgPool(Y0)+MLP(MaxPool(Y0))));

[0082] Among them, σ is the Sigmoid activation function, MLP is the fully connected layer, AvgPool is the global average pooling, and MaxPool is the global maximum pooling.

[0083] Multiply the calculated channel attention weight by the feature map Y0 to obtain the channel attention enhanced feature map Y1, as shown in the following formula:

[0084] Y1=M c (Y0)·Y0.

[0085] Next, spatial attention generates attention weights for spatial positions to highlight important areas in the image. Specifically, feature enhancement also includes:

[0086] Perform global average pooling and global maximum pooling on the feature map Y1 to obtain the pooling result at each position;

[0087] After concatenating the pooling at each position, the spatial attention weight is generated through convolution;

[0088] Multiply the spatial attention weight with the feature map Y1 to highlight the important spatial regions in the feature map Y1 and obtain the spatial attention enhanced feature map Y2, as shown in the following formula:

[0089] M s (Y1)=σ(Conv2D([AvgPool(Y1),MaxPool(Y1)])),

[0090] Y2=M s (Y1)·Y1.

[0091] Among them, Conv2D is a 7×7 convolution operation, AvgPool is global average pooling, MaxPool is global maximum pooling, and [·,·] is feature concatenation.

[0092] Next, the feature map Y2 is flattened into a vector and the flattened vector is input into the fully connected layer to obtain the probability distribution of the damage category, as shown in the following formula:

[0093] P=Softmax(W cls Flatten(Y2));

[0094] Among them, W cls is the weight matrix of the classification head, Flatten(Y2) is the vector of the flattened feature map, and Softmax converts the output into category probability.

[0095] See also Figure 2, which is a schematic diagram of the training process of the lightweight neural network model proposed in an embodiment of the present invention. It can be observed from the figure that the training and validation losses of the model drop rapidly in the early stage and then tend to be stable, which shows that the model effectively learns the data features during the training process and avoids overfitting to a certain extent. The training accuracy and validation accuracy rise rapidly in the early stage and then remain at a high level, showing that the model has good generalization ability. In addition, the average accuracy of training and validation also shows the same trend, further confirming the performance of the model in the steel defect classification task. Adam was used for training with an initial learning rate of 0.001 to help the model converge better. As can be seen from the figure, the model showed good performance during the training process. The reduction in loss function and the improvement in accuracy both indicate the effectiveness of model training.

[0096] Table 2 shows the performance indicators of the lightweight neural network model during training and verification:

[0097] Table 2 Performance indicators of lightweight neural network model

[0098]

[0099] As shown in Table 2, the model achieved 99% accuracy on the validation set, demonstrating its excellent classification capabilities and ability to accurately identify samples from different categories. During training, the model reached this optimal accuracy after 90 training epochs. The model has a relatively small number of parameters, 2.44M, while maintaining high accuracy, which helps reduce the model's storage and computational costs. Its computational complexity is 266.9MMac, which is acceptable in resource-constrained scenarios such as mobile devices. The total training time was 729.0776489 seconds, while the optimal validation time was 1.352496624 seconds, indicating that the model was able to quickly find the optimal solution during training. This reflects the model's demand for computing resources during training and provides valuable guidance for hardware configuration during actual deployment.

[0100] See also Figure 1 and Figure 3 ,Next, this embodiment takes the corrosion related index dataset as input, applies the ,correlation coefficient method to evaluate the degree of correlation between the ,corrosion factors and the corrosion rate, and obtains an initial ,strongly correlated dataset.

[0101] Corrosion is a widespread phenomenon in industrial equipment, materials, and the environment, and its influencing factors are complex and diverse, including material properties, environmental factors (such as humidity, temperature, and chemical composition), time, and human factors. Accurately mining the association rules between corrosion influencing factors is of great significance for preventing and controlling corrosion. However, corrosion data often has problems such as insufficient samples and uneven distribution, which makes it difficult for traditional association rule mining methods (such as the Apriori algorithm) to effectively extract high-quality rules. This embodiment solves the problems of insufficient corrosion data samples and low rule mining efficiency by combining the generative adversarial network (GAN) and the Apriori association rule algorithm, thereby more accurately mining the association rules between corrosion influencing factors and providing theoretical support for corrosion prediction and protection.

[0102] See also Figure 3 As a possible implementation method, the initial strong correlation dataset is obtained by preprocessing the original dataset, calculating the correlation coefficient, and analyzing the corrosion factors, as follows:

[0103] S20. Obtain an original data set and preprocess it to obtain a corrosion-related indicator data set, where the preprocessing includes at least data cleaning, removing outliers, and filling missing values. The corrosion-related indicator data set includes multiple corrosion factors X and corresponding corrosion rates Y.

[0104] As an example, assume that the original data set is D0. After data cleaning, outlier removal, and missing value filling, the preprocessed data set D is obtained. The data is analyzed by Pearson product-moment correlation coefficient and Spearman rank correlation coefficient is verified. The Pearson product-moment correlation coefficient is used to calculate the strength of the linear relationship between each factor and the corrosion rate. Assume that the sample data of the corrosion factor X and the corresponding corrosion rate Y are X={x1,x2,…,x n} and Y={y1,y2,…y n}.

[0105] S21. For normally distributed data, calculate the Pearson product-moment correlation coefficient τ between the corrosion factor X and the corrosion rate Y xy , as follows:

[0106]

[0107] in, is the sample mean of the corrosion factor X, is the sample mean of the corrosion rate Y.

[0108] For non-normally distributed data, calculate the Spearman rank correlation coefficient τ between the corrosion factor X and the corrosion rate Y s ; The details are as follows:

[0109]

[0110] Among them, d i is the rank difference between the corrosion factor X and the corrosion rate Y of the i-th sample, that is, d i =R(x i )―R(y i ), R(x i ) and R(y i ) are x i and y i The rank in the respective data, n is the number of samples.

[0111] For each corrosion factor X in the corrosion-related index data set, the corresponding correlation coefficient is obtained

[0112] S22. Configure the correlation coefficient threshold and set the correlation coefficient Corrosion factors X greater than or equal to the correlation coefficient threshold are retained, and the correlation coefficient is discarded The corrosion factors X that are smaller than the correlation coefficient threshold are composed of all retained corrosion factors X and their corresponding corrosion rates Y to form a strongly correlated data set.

[0113] As an example, the correlation coefficient threshold is configured as θ. For the calculated correlation coefficients between the various corrosion factors and the corrosion rate, if Then X i It can be used as an input feature, otherwise it is discarded. The initial strongly correlated data set F is composed of all retained corrosion factors X and their corresponding corrosion rates Y.

[0114] To solve the problem of insufficient samples, see Figure 3 As a possible implementation method, this embodiment applies a generative adversarial network to expand the initial strongly correlated dataset to obtain a strongly correlated dataset.

[0115] As an example, a generative adversarial network (GAN) consists of a generator and a discriminator. The goal of the generator is to generate synthetic data that is similar to real data. Let z be random noise and G(z) be the synthetic data generated by the generator. The logic of the generator can be expressed as:

[0116] G:z→G(z)

[0117] The parameters of the generator are updated by minimizing the probability of misclassification of the generated data by the discriminator, that is:

[0118]

[0119] Where D(G(z)) is the classification probability of the discriminator for the generated data, and p(z) is the random noise distribution.

[0120] The parameters of the discriminator are updated by maximizing the probability of correct classification of real data and generated data, that is:

[0121]

[0122] Where p(F) is the true data distribution.

[0123] See also Figure 3 , taking the strongly correlated dataset as input, the Apriori association algorithm is applied to evaluate the strong association rules between corrosion factors; as a possible implementation method, the strong association rules are determined by the following method:

[0124] The Apriori association algorithm gradually scans the strongly correlated data set and generates frequent item sets using the properties of frequent item sets;

[0125] As an example, the property of frequent itemsets is that if X is a frequent itemset, then all subsets of X are also frequent itemsets. Each element in the itemset I is 1 item. For the itemset If its support meets the following conditions, then X is a frequent itemset:

[0126]

[0127] For X and Y in the frequent itemset, if And X∪Y is a frequent item set, then it can form an association rule Its confidence level is:

[0128]

[0129] The confidence of the strong association rules selected in this embodiment is 100%, and they are sorted from high to low according to support. Generally speaking, the higher the support and confidence, the more credible the strong association rules are, and the more reasonable it is to use these strong association rules.

[0130] The embodiment of the present invention combines the correlation coefficient method, the GAN adversarial neural network and the Apriori association rule algorithm to realize a data mining algorithm for studying the correlation between each corrosion factor and the corrosion rate. The correlation coefficient method is used to evaluate the degree of correlation between each factor and the corrosion rate in the original data set, select the highly correlated corrosion influencing factors, and expand the data set through the GAN adversarial neural network. Finally, the Apriori association rule mining algorithm is used to discover potential association rules. In-depth understanding of the mechanism and influencing factors of pipeline corrosion is obtained, and a scientific basis is provided for the formulation of targeted corrosion protection strategies. In this embodiment, the corrosion sample data of 20# steel material is shown in Table 3, the GAN adversarial neural network model parameter settings are shown in Table 4, the data set expanded by GAN is encoded according to Table 5, and the association rules are obtained through the Apriori algorithm, as shown in Table 6.

[0131] Table 3 Corrosion sample data of 20# steel material

[0132] Serial number <![CDATA[CO2 partial pressure MPa]]> <![CDATA[Partial pressure of H2S, MPa]]> <![CDATA[Cl - Content mg / L]]> Temperature Pitting rate mm / a 1 0.01 0.0003 52500 50 0.0626 2 0.5 0.0003 52500 50 0.3076 3 0.01 0.05 52500 50 0.2034 4 0.5 0.05 52500 50 0.524 5 0.255 0.02515 5000 20 0.1121 6 0.255 0.02515 100000 20 0.1486 7 0.255 0.02515 5000 80 0.3206 8 0.255 0.02515 100000 80 0.4693 9 0.01 0.02515 52500 20 0.0913 10 0.5 0.02515 52500 20 0.2086 11 0.01 0.02515 52500 80 0.3206 12 0.5 0.02515 52500 80 0.5475 13 0.255 0.0003 5000 50 0.2738 14 0.255 0.05 5000 50 0.3076 15 0.255 0.0003 100000 50 0.3076 16 0.255 0.05 100000 50 0.3598 17 0.01 0.02515 5000 50 0.1981 18 0.5 0.02515 5000 50 0.4093 19 0.01 0.02515 100000 50 0.2815 20 0.5 0.02515 100000 50 0.4901 21 0.255 0.0003 52500 20 0.0938 22 0.255 0.05 52500 20 0.1772 23 0.255 0.0003 52500 80 0.3285 24 0.255 0.05 52500 80 0.4119 25 0.255 0.02515 52500 50 0.2476

[0133] Table 4 GAN model parameter settings

[0134]

[0135]

[0136] Table 5 Dataset encoding rules

[0137]

[0138] Table 6 Corrosion influencing factors association rules

[0139]

[0140]

[0141] By mining strong association rules, corrosion factors can be quantitatively controlled, thereby maintaining the corrosion rate at a low level. In practical applications, pipeline anti-corrosion mechanisms based on strong association rules should be used for anti-corrosion treatment according to the specific conditions of the factory.

[0142] In practical applications, chemical pipeline damage can also be qualitatively and quantitatively evaluated based on the probability distribution of damage types and strong association rules.

[0143] As a possible implementation, see Figure 4 ,The chemical pipeline damage assessment method also includes: ,the formulation of anti-corrosion mechanism based on strong ,association rules, specifically including:

[0144] S30. Obtaining a recommended value of the corrosion rate, and obtaining a recommended value of the corrosion influencing factor and a recommended value of the corrosion rate based on the strong association rule;

[0145] S31. Obtaining a monitoring value of the corrosion rate, monitoring the corrosion rate and corrosion factors, and obtaining a monitoring value of the corrosion rate;

[0146] S32 determines whether the monitoring value exceeds the recommended value. If so, execute S30. Otherwise, execute S31 to S32 in a loop.

[0147] S33. Alarm and decision support;

[0148] As an example, if the monitoring value exceeds the recommended value, the data monitoring system will issue an alarm to remind personnel to come and conduct corrosion condition analysis and corrosion condition log recording in time, or the system can record it automatically.

[0149] S34. Execute decision adjustment according to strong association rules;

[0150] As an example, the decision adjustment performed is to determine which values ​​exceed the range required by the strong association rule according to the strong association rule, and adopt a specific value adjustment strategy according to the type of the value.

[0151] S35. Determine whether the alarm is eliminated. If so, execute S32 to S35 in a loop. Otherwise, perform a process anti-corrosion inspection.

[0152] In the second aspect, the embodiment of the present invention provides a chemical pipeline anti-corrosion method based on multi-dimensional data fusion, see Figure 4 , including the following steps:

[0153] The configured trained lightweight neural network model is used to assess the damage locations and types of chemical pipelines, and a dangerous point inspection log is established; damage types include mechanical damage and chemical corrosion damage;

[0154] Apply the mined strong association rules to improve the existing chemical pipeline anti-corrosion strategy.

[0155] See also Figure 4 As an example, the CBAM-MobileNetV2 model was trained based on images of pipeline damage from historical factory inspection data to improve its accuracy. An inspection and patrol system was developed based on the equipment's operating status, including the determination of inspection locations, inspection cycles, and inspection strategies. Using edge devices deployed with CBAM-MobileNetV2, damaged pipelines were inspected, assessed, and a dangerous point inspection log was established.

[0156] On the other hand, based on the mined strong association rules, the existing corrosion prevention strategy can be improved by adjusting Cl - The risk of pipeline corrosion can be reduced by analyzing CO2 content, CO2 partial pressure, H2S partial pressure, and temperature. In practical applications, corrosion prevention strategies must also be refined based on corrosion factors specific to specific pipeline operating conditions. Based on basic process pipeline data and operating parameters, corrosion detection and monitoring of process pipeline operation are conducted, focusing on monitoring the various correlation factors and corrosion rates. Corrosion status logging and analysis are also performed. Finally, process corrosion prevention is implemented based on strong association rules, optimizing existing operations. Anti-corrosion effectiveness analysis is conducted to verify the conclusions drawn from existing strong association rules.

[0157] By inspecting and assessing pipeline corrosion damage based on multidimensional data, we can inform pipeline anti-corrosion maintenance decisions and optimize existing maintenance measures, ultimately maintaining a high level of process pipeline safety at a lower cost. These improvements will help improve pipeline safety and reliability, extend its service life, and reduce maintenance and repair costs.

[0158] Although the present invention is described herein in conjunction with various embodiments, in the process of implementing the claimed invention, those skilled in the art can understand and implement other variations of the disclosed embodiments by viewing the drawings, the disclosure, and the drawings, etc. In the specification, the word "comprising" does not exclude other components or steps, and "one" or "an" does not exclude multiple situations. A single processor or other unit can implement several functions listed in the specification. Certain measures are recorded in different embodiments, but this does not mean that these measures cannot be combined to produce good results.

[0159] Although the present invention has been described with reference to specific features and embodiments thereof, it will be apparent that various modifications and combinations thereof may be made without departing from the spirit and scope of the invention. Accordingly, this specification and drawings are merely illustrative of the present invention and are deemed to cover any and all modifications, variations, combinations or equivalents within the scope of the invention. It will be apparent that various modifications and variations of the present invention may be made by those skilled in the art without departing from the spirit and scope of the invention. Thus, the present invention is intended to include such modifications and variations as fall within the scope of the invention and its equivalents.

Claims

1. A pipeline damage assessment method based on multidimensional data fusion, characterized in that: The steps include: Configure a trained lightweight neural network model and input the pipeline damage image into the lightweight neural network model. After feature extraction, feature enhancement, feature mapping and output, the probability distribution of damage types is obtained. The backbone network of the lightweight neural network model is MobileNetV2, and an attention mechanism module is placed after MobileNetV2; the feature extraction specifically includes: S10. The pipeline damage image is subjected to a depthwise convolution to expand its channels. This convolution is performed on each channel independently. S11. Apply an activation function to enhance the nonlinear characteristics of MobileNetV2 and constrain the output value to the range [0, 6]. S12. Reducing the number of channels from the expanded number of channels to the original number of channels by performing a dimensionality reduction convolution, wherein the dimensionality reduction convolution is a point-by-point convolution that integrates information between channels. After S10 to S12, the features of the pipeline damage image are extracted and output as feature map Y0; Taking the corrosion-related index dataset as input, the correlation coefficient method is applied to evaluate the correlation between the corrosion factors and the corrosion rate, and an initial strong correlation dataset is obtained, wherein the initial strong correlation dataset is a small sample dataset; The initial strongly correlated data set is obtained by the following method: S20. Obtain the original data set, and obtain the corrosion-related indicator data set through preprocessing, wherein the preprocessing includes at least data cleaning, removal of outliers, and filling of missing values; the corrosion-related indicator data set has multiple corrosion factors And the corresponding corrosion rate ; S21. Calculate the corrosion factor for normally distributed data and corrosion rate Pearson product-moment correlation coefficient ; For non-normally distributed data, calculate the corrosion factor and corrosion rate The Spearman rank correlation coefficient between ; For each corrosion factor in the corrosion-related indicator dataset The corresponding correlation coefficients were obtained ; S22. Configure the correlation coefficient threshold and set the correlation coefficient Corrosion factors greater than or equal to the correlation coefficient threshold Keep and discard the correlation coefficient Corrosion factors less than the correlation coefficient threshold , by all the corrosion factors that are retained and its corresponding corrosion rate Form a strongly correlated data set; Expanding the initial strongly correlated data set to obtain a strongly correlated data set; Taking the strongly correlated dataset as input, the Apriori association algorithm is applied to evaluate the strong association rules between corrosion factors; Qualitative and quantitative assessment of chemical pipeline damage based on the probability distribution of damage types and strong association rules; The chemical pipeline damage assessment method further includes: formulating an anti-corrosion mechanism based on strong association rules, specifically including: S30 obtains the recommended value of the corrosion rate, and obtains the recommended value of the corrosion influencing factor and the recommended value of the corrosion rate based on the strong association rule; S31. Obtaining corrosion rate monitoring values, monitoring corrosion rates and corrosion factors, and obtaining corrosion rate monitoring values; S32 determines whether the monitoring value exceeds the recommended value, if so, execute S30, otherwise, loop execution S31 ~ S32; S33. Alarm and decision support; S34. Execute decision adjustment based on strong association rules; S35. Determine whether the alarm is cleared. If so, execute S32 to S35 in a loop. Otherwise, perform a process corrosion inspection.

2. The pipeline damage assessment method based on multidimensional data fusion according to claim 1 is characterized in that: The attention mechanism module includes channel attention and spatial attention, and the feature enhancement specifically includes: The channel attention generates channel attention weights through average pooling and maximum pooling to highlight important channel information; The pooled results are processed by a fully connected network to obtain the channel attention weights; Multiply the calculated channel attention weight by the feature map Y0 to obtain the channel attention enhanced feature map Y1.

3. The pipeline damage assessment method based on multidimensional data fusion according to claim 2 is characterized in that: The feature enhancements also include: Perform global average pooling and global maximum pooling on the feature map Y1 to obtain the pooling result at each position; After concatenating the pooling at each position, the spatial attention weight is generated through convolution; The spatial attention weight is multiplied with the feature map Y1 to highlight the important spatial regions in the feature map Y1 and obtain the spatial attention enhanced feature map Y2.

4. The pipeline damage assessment method based on multidimensional data fusion according to claim 3 is characterized in that: The feature mapping and output specifically include: The feature map Y2 is flattened into a vector, and the flattened vector is input into the fully connected layer to obtain the probability distribution of the damage category.

5. The pipeline damage assessment method based on multidimensional data fusion according to claim 1 is characterized in that: The strong association rule is determined by the following method: The Apriori association algorithm gradually scans the strongly correlated data set and generates frequent item sets using the properties of frequent item sets; For frequent itemsets and ,like and Is a frequent item set, then it can form an association rule , and its confidence is: 。 6. The pipeline damage assessment method based on multidimensional data fusion according to claim 1 is characterized in that: Generative adversarial networks are applied to expand the initial strongly correlated dataset to obtain a strongly correlated dataset.

7. A pipeline anti-corrosion method based on multidimensional data fusion, characterized in that: The steps include: The configured trained lightweight neural network model is used to assess the damage locations and types of chemical pipelines, and a dangerous point inspection record is established. The damage types include mechanical damage and chemical corrosion damage. The assessment of the damage locations and types of chemical pipelines specifically includes: configuring the trained lightweight neural network model, inputting pipeline damage images into the lightweight neural network model, and obtaining the probability distribution of damage types after feature extraction, feature enhancement, feature mapping and output; The backbone network of the lightweight neural network model is MobileNetV2, and an attention mechanism module is placed after MobileNetV2; the feature extraction specifically includes: S10. The pipeline damage image is subjected to a depthwise convolution to expand its channels. This convolution is performed on each channel independently. S11. Apply an activation function to enhance the nonlinear characteristics of MobileNetV2 and constrain the output value to the range [0, 6]. S12. Reducing the number of channels from the expanded number of channels to the original number of channels by performing a dimensionality reduction convolution, wherein the dimensionality reduction convolution is a point-by-point convolution that integrates information between channels. After S10 to S12, the features of the pipeline damage image are extracted and output as feature map Y0; The mined strong association rules are applied to improve the existing chemical pipeline anti-corrosion strategy. The strong association rules are mined by the following method: using a corrosion-related indicator dataset as input, the correlation coefficient method is used to evaluate the correlation between corrosion factors and corrosion rates, and an initial strong correlation dataset is obtained. The initial strong correlation dataset is a small sample dataset. The initial strongly correlated data set is obtained by the following method: S20. Obtain the original data set, and obtain the corrosion-related indicator data set through preprocessing, wherein the preprocessing includes at least data cleaning, removal of outliers, and filling of missing values; the corrosion-related indicator data set has multiple corrosion factors And the corresponding corrosion rate ; S21. Calculate the corrosion factor for normally distributed data and corrosion rate Pearson product-moment correlation coefficient ; For non-normally distributed data, calculate the corrosion factor and corrosion rate The Spearman rank correlation coefficient between ; For each corrosion factor in the corrosion-related indicator dataset The corresponding correlation coefficients were obtained ; S22. Configure the correlation coefficient threshold and set the correlation coefficient Corrosion factors greater than or equal to the correlation coefficient threshold Keep and discard the correlation coefficient Corrosion factors less than the correlation coefficient threshold , by all the corrosion factors that are retained and its corresponding corrosion rate Form a strongly correlated data set; Expanding the initial strongly correlated data set to obtain a strongly correlated data set; Taking the strongly correlated dataset as input, the Apriori association algorithm is applied to evaluate the strong association rules between corrosion factors.

Images