Picoampere-level precision data acquisition method and system of special large model for corrosion industry

Abstract

The invention provides a picoampere-level precision data acquisition method and system for a large model special for the corrosion industry, and relates to the technical field of corrosion, and the method comprises the steps: collecting electrochemical impedance spectroscopy data and corrosion environment parameters, obtaining a time-frequency domain characteristic pattern through continuous wavelet transform, determining corrosion characteristic indexes of a target frequency interval based on a self-adaptive segmentation algorithm, and obtaining a picoampere-level precision model. And constructing a causal reasoning decision tree, and updating the correlation strength of the corrosion knowledge graph by using a Bayesian network. The micro corrosion signal can be accurately captured, the corrosion mechanism evolution path can be predicted, and the corrosion monitoring precision and prediction accuracy can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of corrosion, in particular to a pico-ampere level precision data acquisition method and system for a corrosion industry special large model. BACKGROUND

[0002] Metal corrosion is a major challenge in the industrial field, with global economic losses due to corrosion reaching several trillion dollars annually. In key industries such as oil, chemical, nuclear power, aerospace, etc., corrosion not only affects the service life of equipment, but also can lead to major safety accidents. With the complexity of industrial environments, the complexity and variability of corrosion mechanisms pose a significant challenge to traditional corrosion monitoring and prediction methods. Electrochemical impedance spectroscopy technology, as a non-destructive corrosion detection method, can reflect the corrosion process at the metal / solution interface and has become an important tool for corrosion research and monitoring. However, extracting effective corrosion feature information from massive electrochemical impedance data and establishing an accurate prediction model of corrosion behavior has always been a difficult point in corrosion scientific research.

[0003] Traditional corrosion data acquisition methods mainly analyze or process electrochemical impedance spectroscopy data using equivalent circuit analysis or empirical formulas. These methods often have significant shortcomings when faced with complex corrosion systems. Traditional methods lack multi-scale time-frequency analysis capabilities for electrochemical impedance spectroscopy data, making it difficult to effectively capture the early characteristics of the microscopic corrosion process reflected by pico-ampere level current changes, resulting in insufficient early corrosion warning capabilities. Existing technologies are unable to establish a causal relationship between corrosion environmental parameters and electrochemical responses, lack accurate descriptions of corrosion mechanism evolution paths under the coupling action of multiple factors, and cannot achieve accurate prediction of corrosion behavior under specific working conditions. Traditional corrosion data processing methods generally lack a system framework that combines knowledge graphs and probabilistic reasoning, making it difficult to effectively integrate expert experience with real-time monitoring data, limiting the accumulation of high-quality data needed for corrosion large model training.

[0004] With the development of artificial intelligence technology, it is possible to use large models to handle complex corrosion problems, but the collection and processing of high-precision corrosion data remains a key bottleneck restricting the development of corrosion special large models. Establishing a pico-ampere level precision corrosion data acquisition method and building a corrosion knowledge graph and causal reasoning decision system are of great significance for improving corrosion prediction accuracy, guiding corrosion prevention design, and extending the service life of industrial equipment. SUMMARY

[0005] The pico-ampere level precision data acquisition method and system for a corrosion industry special large model provided by the embodiments of the present application can solve the problems in the prior art.

[0006] In a first aspect, the pico-ampere level precision data acquisition method for a corrosion industry special large model comprises:

[0007] The electrochemical impedance spectrum data of the metal material is collected, and the corrosion environment parameters are recorded synchronously, the electrochemical impedance spectrum data is decomposed by using continuous wavelet transform to obtain a time-frequency domain feature map; the time-frequency domain feature map is divided into frequency bands based on an adaptive segmentation algorithm, initial boundary points corresponding to the time-frequency domain feature map are screened according to a minimum frequency interval constraint, and key points of energy density distribution in the time-frequency domain feature map are determined based on an energy continuity criterion to obtain a corrosion feature index of a target frequency interval;

[0008] Based on the corrosion feature index of the target frequency interval, the information gain value corresponding to the corrosion feature index is determined, the index with the maximum information gain value is selected as the root node of the causal reasoning decision tree, the initial structure of the causal reasoning decision tree is constructed, and the hierarchical relationship of the corrosion influencing factors is established according to the initial structure;

[0009] Based on the electrochemical impedance spectrum data and the corrosion environment parameters, an initial corrosion knowledge graph is constructed, the conditional probability relationship between the corrosion influencing factors is calculated by using a Bayesian network, and the correlation strength in the initial corrosion knowledge graph is updated; and the corrosion mechanism evolution path is predicted according to the correlation strength in the updated initial corrosion knowledge graph.

[0010] The electrochemical impedance spectrum data of the metal material is collected, and the corrosion environment parameters are recorded synchronously, the electrochemical impedance spectrum data is decomposed by using continuous wavelet transform to obtain a time-frequency domain feature map, including:

[0011] The electrochemical impedance spectrum signal is input into a Morlet wavelet transform model to obtain a wavelet transform coefficient;

[0012] An adaptive scale parameter system is established, the minimum scale parameter and the maximum scale parameter are determined according to the frequency range of the electrochemical impedance spectrum signal, and the scale parameter corresponding to the electrochemical impedance spectrum signal is generated based on a logarithmic uniform distribution principle;

[0013] The wavelet transform coefficient is constructed into a coefficient matrix according to the scale parameter and the time position parameter, the energy density value of each element of the coefficient matrix is calculated, the energy density value is logarithmically converted and normalized, and a time-frequency domain feature map is generated.

[0014] The time-frequency domain feature map is divided into frequency bands based on an adaptive segmentation algorithm, initial boundary points corresponding to the time-frequency domain feature map are screened according to a minimum frequency interval constraint, and key points of energy density distribution in the time-frequency domain feature map are determined based on an energy continuity criterion to obtain a corrosion feature index of a target frequency interval, including:

[0015] The energy density gradient value of the time-frequency domain feature map is calculated, and the energy density gradient value includes a frequency direction gradient component and a time direction gradient component;

[0016] calculating a local statistical feature of the energy density gradient value based on a sliding window mechanism, the local statistical feature including a local variance value, a width of the sliding window being adaptively adjusted according to a frequency resolution;

[0017] generating a dynamic segmentation threshold according to the local statistical feature, the dynamic segmentation threshold being composed of a global mean value and a local standard deviation value; performing boundary point detection on the energy density gradient value based on the dynamic segmentation threshold to obtain an initial boundary point set, the initial boundary point satisfying a condition of being greater than the dynamic segmentation threshold and being greater than a value of a neighboring point;

[0018] screening the initial boundary point according to a minimum frequency interval constraint; merging adjacent boundary points based on an energy continuity criterion; calculating an average energy ratio value of a region to be merged, and performing region merging when the average energy ratio value is less than a set energy threshold;

[0019] determining key points of the energy density distribution in the merged region, the key points including an energy density extreme point and an energy density abrupt change point, and obtaining a frequency value and a phase angle value corresponding to the key points; determining a characteristic frequency band of the electrochemical impedance spectrum according to the frequency value and the phase angle value, the characteristic frequency band being used to represent impedance characteristics of the electrochemical system.

[0020] based on the corrosion feature indicators in the target frequency interval, determining information gain values corresponding to the corrosion feature indicators, selecting an indicator with the largest information gain value as a root node of a causal reasoning decision tree, constructing an initial structure of the causal reasoning decision tree, and establishing a hierarchical relationship of corrosion influencing factors according to the initial structure, including:

[0021] extracting corrosion feature indicators within the target frequency interval, calculating phase angle data, energy density distribution data, and frequency response amplitude data within the target frequency interval, and grouping the phase angle data, the energy density distribution data, and the frequency response amplitude data into a feature indicator set;

[0022] calculating a Pearson correlation coefficient between each two indicators in the feature indicator set to generate an indicator correlation matrix, each element of the indicator correlation matrix representing a correlation degree between the corresponding two indicators;

[0023] calculating a system entropy of the feature indicator set, and calculating a conditional entropy under a condition of each indicator, taking a difference between the system entropy and the conditional entropy as an information gain value, and selecting an indicator with the largest information gain value as a root node of a causal reasoning decision tree;

[0024] constructing an initial structure of the causal reasoning decision tree based on the root node, calculating a Gini coefficient corresponding to each split position, selecting a position with the smallest Gini coefficient as a split position of the node, and obtaining the Gini coefficient by calculating a sum of squares of probabilities of each class of samples.

[0025] According to the initial structure, a hierarchical relationship of the influencing factors is established, a conditional correlation coefficient between adjacent hierarchical nodes is calculated, and the conditional correlation coefficient is taken as the correlation strength between the hierarchies;

[0026] The influence degree of each hierarchical node on the output of the root node is calculated by a back propagation method, the influence degree is taken as the node contribution degree, and the primary and secondary orders of the influencing factors are determined based on the size of the node contribution degree.

[0027] The conditional probability relationship between the corrosion influencing factors is calculated by using a Bayesian network, and the correlation strength in the initial corrosion knowledge graph is updated, including:

[0028] A Bayesian network structure of the corrosion influencing factors is established, prior probabilities of each corrosion influencing factor are calculated based on observation data, and an initial node set is constructed according to the prior probabilities; a conditional probability relationship between nodes in the initial node set is calculated, a conditional probability table is generated based on the conditional probability relationship, and the conditional probability table records the probability dependency relationship between a node and its parent node set;

[0029] An initial corrosion knowledge graph is constructed, an edge weight between nodes is calculated by using mutual information, the edge weight represents the correlation strength between nodes, and the edge weight is written into the initial corrosion knowledge graph;

[0030] Belief propagation calculation is performed on the initial corrosion knowledge graph, the belief propagation calculation multiplies and normalizes the diagnostic support degree and the prediction support degree of a node to obtain a node belief value, and the correlation strength in the initial corrosion knowledge graph is updated according to the node belief value.

[0031] According to the correlation strength in the updated initial corrosion knowledge graph, a corrosion mechanism evolution path is predicted, including:

[0032] The number of times of state conversion at adjacent time points is counted, a state transition probability is calculated by taking the ratio of the number of times of state conversion to the total number of times of the current state, and a state transition probability matrix is constructed according to the state transition probability;

[0033] The state transition probability matrix and the updated correlation strength are combined to construct a time-varying conditional probability function, a probability product of the time-varying conditional probability function on different paths is calculated to obtain a path importance score, and a key evolution path is screened based on the path importance score;

[0034] The cumulative probability is calculated by accumulating the state probability at each time point along the key evolution path, the cumulative probability is taken as a quantitative index to predict the evolution trend of the corrosion mechanism, and the evolution trend reflects the change law of the corrosion state over time.

[0035] The second aspect of the embodiment of the present application provides a pico-level precision data acquisition system of a special large model in the corrosion industry, which comprises:

[0036] The first unit is used for acquiring electrochemical impedance spectrum data of a metal material and synchronously recording corrosion environment parameters, performing multi-scale decomposition on the electrochemical impedance spectrum data by using continuous wavelet transform to obtain a time-frequency domain feature map, performing frequency band division on the time-frequency domain feature map based on an adaptive segmentation algorithm, screening initial boundary points corresponding to the time-frequency domain feature map according to a minimum frequency interval constraint, and determining key points of energy density distribution in the time-frequency domain feature map based on an energy continuity criterion to obtain a corrosion characteristic index of a target frequency interval.

[0037] The second unit is used for determining an information gain value corresponding to the corrosion characteristic index of the target frequency interval based on the corrosion characteristic index, selecting an index with the largest information gain value as a root node of a causal reasoning decision tree, constructing an initial structure of the causal reasoning decision tree, and establishing a hierarchical relationship of corrosion influencing factors according to the initial structure.

[0038] The third unit is used for constructing an initial corrosion knowledge graph based on the electrochemical impedance spectrum data and the corrosion environment parameters, calculating a conditional probability relationship between corrosion influencing factors by using a Bayesian network, and updating an association strength in the initial corrosion knowledge graph; and predicting a corrosion mechanism evolution path according to the association strength in the updated initial corrosion knowledge graph.

[0039] The third aspect of the embodiment of the present application provides an electronic device, which comprises:

[0040] A processor;

[0041] A memory for storing processor-executable instructions;

[0042] The processor is configured to invoke the instructions stored in the memory to execute the method described above.

[0043] The fourth aspect of the embodiment of the present application provides a computer-readable storage medium having computer program instructions stored thereon, and the computer program instructions are executed by a processor to implement the method described above.

[0044] The present application has the following beneficial effects:

[0045] The multi-scale decomposition of the electrochemical impedance spectrum data by using the continuous wavelet transform and the frequency band division of the time-frequency domain feature map by using the adaptive segmentation algorithm can accurately capture the change characteristics of the pico-level weak corrosion current, improve the data acquisition accuracy in the early stage of corrosion, and solve the problem of insufficient accuracy in the extraction of weak signals in the traditional method.

[0046] The information gain value is used to evaluate the corrosion feature index and build a causal reasoning decision tree, realizes the hierarchical expression of the corrosion influencing factors, the interpretability of the corrosion mechanism is obviously enhanced, a high-quality structured knowledge base is provided for the large model, and the model is convenient for accurately reasoning the complex corrosion system.

[0047] The corrosion knowledge graph is updated by combining the Bayesian network to calculate the conditional probability relationship, the correlation strength between the influencing factors can be dynamically adjusted, the evolution path of the corrosion mechanism is accurately predicted, the adaptability and generalization ability of the large model in the multiple complex environments are improved, and a reliable basis is provided for the corrosion protection decision. BRIEF DESCRIPTION OF DRAWINGS

[0048] Figure 1 A flowchart of a picoscale precision data acquisition method for a corrosion industry special large model of the embodiment of the present application is shown in the figure.

[0049] Figure 2 A comparison diagram of corrosion feature recognition accuracy of different methods is shown in the figure.

[0050] Figure 3 A comparison diagram of corrosion prediction accuracy of different methods changing with time is shown in the figure. DETAILED DESCRIPTION

[0051] In order to make the purpose, technical scheme and advantages of the embodiment of the present application clearer, the technical scheme in the embodiment of the present application will be described clearly and completely in combination with the drawings in the embodiment of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor belong to the scope of protection of the present application.

[0052] The technical scheme of the present application will be described in detail in the following specific embodiments. The following several specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in some embodiments.

[0053] Figure 1 A flowchart of a picoscale precision data acquisition method for a corrosion industry special large model of the embodiment of the present application is shown in the figure. Figure 1 As shown in the figure, the method comprises:

[0054] The electrochemical impedance spectrum data of the metal material is collected, and the corrosion environment parameters are recorded synchronously, the electrochemical impedance spectrum data is decomposed in multiple scales by using continuous wavelet transform, and a time-frequency domain feature map is obtained; the time-frequency domain feature map is divided into frequency bands based on an adaptive segmentation algorithm, initial boundary points corresponding to the time-frequency domain feature map are screened according to a minimum frequency interval constraint, and key points of energy density distribution in the time-frequency domain feature map are determined based on an energy continuity criterion, so that a corrosion feature index of a target frequency interval is obtained;

[0055] Based on the corrosion feature index of the target frequency interval, an information gain value corresponding to the corrosion feature index is determined, an index with the maximum information gain value is selected as a root node of a causal reasoning decision tree, an initial structure of the causal reasoning decision tree is constructed, and a hierarchical relationship of corrosion influencing factors is established according to the initial structure;

[0056] Based on the electrochemical impedance spectrum data and the corrosion environment parameters, an initial corrosion knowledge graph is constructed, a conditional probability relationship between corrosion influencing factors is calculated by using a Bayesian network, and the correlation strength in the initial corrosion knowledge graph is updated; and an evolution path of a corrosion mechanism is predicted according to the correlation strength in the updated initial corrosion knowledge graph.

[0057] In an optional implementation, the electrochemical impedance spectrum data of the metal material is collected, and the corrosion environment parameters are recorded synchronously, the electrochemical impedance spectrum data is decomposed in multiple scales by using continuous wavelet transform, and a time-frequency domain feature map is obtained, including:

[0058] The electrochemical impedance spectrum signal is input into a Morlet wavelet transform model, and a wavelet transform coefficient is obtained;

[0059] An adaptive scale parameter system is established, the minimum scale parameter and the maximum scale parameter are determined according to the frequency range of the electrochemical impedance spectrum signal, and the scale parameters corresponding to the electrochemical impedance spectrum signal are generated based on a logarithmic uniform distribution principle;

[0060] The wavelet transform coefficient is constructed into a coefficient matrix according to the scale parameter and the time position parameter, the energy density value of each element of the coefficient matrix is calculated, the energy density value is logarithmically converted and normalized, and a time-frequency domain feature map is generated.

[0061] Import the collected electrochemical impedance spectroscopy data and extract the impedance value as the signal to be analyzed. Morlet wavelet is selected as the analysis wavelet for this signal. Morlet wavelet has good time-frequency localization characteristics and is suitable for processing non-stationary signals. The electrochemical impedance spectroscopy signal is input into the Morlet wavelet transform model, and the corresponding wavelet transform coefficients are obtained by calculating the convolution of the signal with wavelet functions of different scales. In the specific implementation, for impedance spectroscopy signals with a frequency of 0.01Hz to 100kHz, the number of sampling points can reach more than 500 points. The coefficient values ​​at different scales are obtained by calculating the inner product of the wavelet function and the signal at each time point.

[0062] To accurately reflect the characteristics of electrochemical impedance spectroscopy at different frequencies, an adaptive scale parameter system is required. The minimum and maximum scale parameters are determined based on the frequency range of the acquired electrochemical impedance spectroscopy signal. For example, for an impedance spectrum with a frequency range of 0.01 Hz to 100 kHz, the minimum scale parameter can be set to 0.5, and the maximum scale parameter can be set to 128. A scale parameter sequence is generated based on the principle of logarithmic uniform distribution to ensure appropriate resolution in both low- and high-frequency regions. In practical applications, 32 to 64 scale parameter points can be selected, distributed logarithmically. For example, between a minimum scale of 0.5 and a maximum scale of 128, scale parameters can be selected such as 0.5, 0.7, 1.0, 1.4, 2.0, 2.8, 4.0, 5.6, 8.0, 11.3, 16.0, 22.6, 32.0, 45.3, 64.0, 90.5, and 128.0, among others.

[0063] For each scale parameter a and time position parameter b, there is a corresponding wavelet transform coefficient W(a,b). The number of rows in the constructed coefficient matrix is ​​equal to the number of scale parameters, and the number of columns is equal to the number of time position parameters. Calculate the energy density value of each element of the coefficient matrix. The energy density value is equal to the square of the wavelet transform coefficient divided by the corresponding scale parameter. For example, for the electrochemical impedance spectroscopy of a metal material measured in 3.5% sodium chloride solution, the wavelet transform coefficient when the scale parameter is 16.0 and the time position parameter is 100 is 0.8+0.6i, then the corresponding energy density value is (0.8 2 +0.6 2 ) / 16.0=0.085.

[0064] Logarithmic transformation can enhance the visibility of small signals, and normalization can make the results under different conditions comparable. Logarithmic transformation can use a logarithmic function with a base of 10 to calculate log for each energy density value E. 10(E) Normalization is to map all the logarithmic converted values to the range of 0-1, which can be achieved by the max-min normalization method. For example, the minimum value of a group of data after logarithmic conversion is -3.5, and the maximum value is 1.2. Then the normalized value of the point with value -2.1 is (-2.1-(-3.5)) / (1.2-(-3.5)) = 0.3.

[0065] After the above processing, a time-frequency domain feature map is generated, the horizontal coordinate represents the time position parameter (corresponding to the time or frequency of the original signal), the vertical coordinate represents the scale parameter (corresponding to different frequency components), and the color depth in the figure represents the energy density. In practical applications, a pseudo-color map can be used for display, such as blue representing a low energy area and red representing a high energy area. Through the feature map, the impedance characteristic change of the metal material in different corrosion environments can be intuitively identified, providing a basis for corrosion mechanism research. For example, the time-frequency feature map of a certain stainless steel sample mainly has energy concentrated in the high frequency area in the normal environment, and after being exposed to the environment containing chloride ions for 72 hours, the time-frequency feature map has obvious energy enhancement in the low frequency area, indicating that the passivation film on the surface of the material is damaged and pitting corrosion occurs.

[0066] In an optional embodiment, the time-frequency domain feature map is divided into frequency bands based on an adaptive segmentation algorithm, initial boundary points corresponding to the time-frequency domain feature map are screened according to a minimum frequency interval constraint, and key points of the energy density distribution in the time-frequency domain feature map are determined based on an energy continuity criterion, and the corrosion feature index of the target frequency interval is obtained, including:

[0067] The energy density gradient value of the time-frequency domain feature map is calculated, and the energy density gradient value includes a frequency direction gradient component and a time direction gradient component;

[0068] The local statistical features of the energy density gradient value are calculated based on a sliding window mechanism, the local statistical features include a local variance value, and the width of the sliding window is adaptively adjusted according to the frequency resolution;

[0069] The dynamic segmentation threshold is generated according to the local statistical features, the dynamic segmentation threshold is composed of a global mean value and a local standard deviation weighted; the boundary point detection is performed on the energy density gradient value based on the dynamic segmentation threshold, and an initial boundary point set is obtained, the initial boundary point satisfies the condition of being greater than the dynamic segmentation threshold and being greater than the value of the adjacent point;

[0070] The initial boundary points are screened according to the minimum frequency interval constraint; the adjacent boundary points are merged based on the energy continuity criterion; the average energy ratio of the region to be merged is calculated, and when the average energy ratio is less than a set energy threshold, the region merging is performed;

[0071] Determine the key points of the energy density distribution in the merging area, including the energy density extreme points and the energy density mutation points, obtain the frequency value and the phase angle value corresponding to the key points; determine the characteristic frequency band of the electrochemical impedance spectrum according to the frequency value and the phase angle value, and the characteristic frequency band is used to characterize the impedance characteristics of the electrochemical system.

[0072] The energy density gradient value includes a frequency direction gradient component and a time direction gradient component. The frequency direction gradient component reflects the rate of change of energy in the frequency dimension, and the time direction gradient component reflects the rate of change of energy in the time dimension. For example, for a point (t, f) in the time-frequency domain feature map, the frequency direction gradient component can be obtained by calculating the energy density difference between the point (t, f+Δf) and the point (t, f) and dividing by Δf, and the time direction gradient component can be obtained by calculating the energy density difference between the point (t+Δt, f) and the point (t, f) and dividing by Δt. Taking a certain measurement as an example, in the electrochemical impedance spectrum with a frequency range of 0.1 Hz to 1000 Hz, the energy density at the frequency point 600 Hz is 0.85, and the energy density at the frequency point 601 Hz is 0.82, then the frequency direction gradient component of this point is -0.03.

[0073] The width of the sliding window is adaptively adjusted according to the frequency resolution, ensuring that accurate local features can be obtained in different frequency intervals. The local statistical features include a local variance value, which is used to describe the fluctuation degree of the gradient value in the local area. For example, in the low frequency region (0.1 Hz-10 Hz), the window width can be set to 5 frequency points due to the small frequency point interval; in the medium frequency region (10 Hz-100 Hz), the window width can be set to 10 frequency points; in the high frequency region (100 Hz-1000 Hz), the window width can be set to 20 frequency points. For a signal with a sampling frequency of 100 Hz, in the local area near the frequency of 50 Hz, if the gradient values of 11 sampling points (45 Hz to 55 Hz) are [0.05, 0.06, 0.08, 0.09, 0.12, 0.15, 0.11, 0.09, 0.07, 0.05, 0.04], then the local variance value of this area is 0.00121.

[0074] The dynamic segmentation threshold adapts to the energy distribution characteristics of different frequency intervals and is composed of a global mean value and a local standard deviation weighted. The global mean value reflects the average energy gradient level of the entire time-frequency domain feature map, and the local standard deviation reflects the fluctuation of the energy gradient in the local area. For example, for a certain frequency band, if the global gradient mean value is 0.08, the local standard deviation is 0.03, and the weight coefficient is 1.5, then the dynamic segmentation threshold of this area is 0.08+1.5×0.03=0.125.

[0075] The initial boundary point needs to meet two conditions: its gradient value is greater than the dynamic segmentation threshold, and greater than the gradient value of the adjacent point. Taking the threshold 0.125 in the above example as an example, in the gradient sequence [0.05, 0.06, 0.08, 0.09, 0.12, 0.15, 0.11, 0.09, 0.07, 0.05, 0.04], only the gradient value 0.15 of the 6th point meets the condition of being greater than the threshold 0.125 and greater than the adjacent points (0.12 and 0.11), so it is identified as an initial boundary point.

[0076] To avoid over-segmentation, the frequency interval between adjacent boundary points should not be less than the preset minimum frequency interval. For example, if the minimum frequency interval is set to 10 Hz, if two boundary points are detected at 52 Hz and 58 Hz respectively, since the interval between them is less than 10 Hz, the boundary point with the larger gradient value is retained and the other is deleted.

[0077] The average energy ratio refers to the ratio of the average energy of two adjacent frequency bands. For example, if the average energy of frequency band A (50 Hz-100 Hz) is 0.75, and the average energy of frequency band B (100 Hz-150 Hz) is 0.72, then their average energy ratio is 0.72 / 0.75=0.96. If the energy threshold is set to 0.9, since 0.96>0.9, the two frequency bands will not be merged.

[0078] Key points include energy density extreme points and energy density mutation points. Energy density extreme points refer to points where energy density reaches a maximum or minimum in a local region, and energy density mutation points refer to points where the energy density change rate exceeds a certain threshold. The frequency value and phase angle value corresponding to these key points are obtained. Taking a certain measurement as an example, in the frequency band of 100 Hz to 200 Hz, the energy density maximum point is identified at 150 Hz, and its phase angle is -45 degrees; the energy density mutation point is located at 180 Hz, and its phase angle is -60 degrees.

[0079] Characteristic frequency bands are used to represent the impedance characteristics of electrochemical systems, such as the charge and discharge state of the battery, the conductivity of the electrolyte, etc. For example, for a lithium-ion battery, the characteristic frequency band in the low-frequency region (0.1 Hz-1 Hz) mainly reflects the diffusion process of the battery, the characteristic frequency band in the medium-frequency region (1 Hz-100 Hz) mainly reflects the charge transfer impedance, and the characteristic frequency band in the high-frequency region (above 100 Hz) mainly reflects the solution resistance of the electrolyte. By analyzing the changes of these characteristic frequency bands, the health status and remaining life of the battery can be evaluated.

[0080] In an optional embodiment, based on the corrosion feature indicators of the target frequency interval, a value of information gain corresponding to the corrosion feature indicators is determined, an indicator with the maximum value of information gain is selected as a root node of the causal reasoning decision tree, an initial structure of the causal reasoning decision tree is constructed, and a hierarchical relationship of the corrosion influencing factors is established according to the initial structure, including:

[0081] Corrosion feature indicators are extracted in the target frequency interval, phase angle data, energy density distribution data, and frequency response amplitude data in the target frequency interval are calculated, and the phase angle data, the energy density distribution data, and the frequency response amplitude data are combined to form a feature indicator set;

[0082] Pearson correlation coefficients are calculated for indicators in the feature indicator set two by two, an indicator correlation matrix is generated, and each element of the indicator correlation matrix represents a correlation degree between the corresponding two indicators;

[0083] System entropy of the feature indicator set is calculated, conditional entropy under a condition of each indicator is calculated, a difference between the system entropy and the conditional entropy is taken as a value of information gain, and an indicator with the maximum value of information gain is selected as a root node of the causal reasoning decision tree;

[0084] Based on the root node, an initial structure of the causal reasoning decision tree is constructed, a Gini coefficient corresponding to each split position is calculated, and a position with the minimum Gini coefficient is selected as a split position of the node, and the Gini coefficient is calculated by summing squares of probabilities of each class of samples;

[0085] Based on the initial structure, a hierarchical relationship of the influencing factors is established, a conditional correlation coefficient between adjacent hierarchical nodes is calculated, and the conditional correlation coefficient is taken as a correlation strength between the hierarchies;

[0086] An influence degree of each hierarchical node on an output of the root node is calculated in a back propagation manner, the influence degree is taken as a node contribution degree, and a primary and secondary order of the influencing factors is determined based on the size of the node contribution degree.

[0087] Electrochemical response data of a metal sample in different corrosion environments is obtained by an electrochemical impedance spectroscopy measurement device, electrochemical impedance spectroscopy data is subjected to frequency spectrum analysis, and phase angle data, energy density distribution data, and frequency response amplitude data are calculated from a target frequency interval (for example, 0.01 Hz-100 kHz). For example, for a corrosion test of a certain carbon steel sample in a 3.5% sodium chloride solution, a phase angle of -45°, an energy density of 0.82 mW / cm 2 , and a frequency response amplitude of 2.5 kΩ·cm 2For all the measured frequency points, the three types of data are obtained respectively to form the feature index set. The feature index set contains the phase angle, energy density and frequency response amplitude at multiple frequency points, which constitutes a complete corrosion feature representation.

[0088] The correlation coefficient between the phase angle at 10 Hz and the energy density at 20 Hz is 0.78, indicating that there is a strong positive correlation between the two indicators. By calculating the correlation coefficient between all indicators, an n x n correlation matrix is obtained, where n is the total number of feature indicators. Each element in the matrix has a value between -1 and 1. The closer the absolute value is to 1, the stronger the correlation, and the closer to 0, the weaker the correlation. For a certain experiment, the correlation coefficient of the phase angle at frequency points 50 Hz and 100 Hz in the calculated correlation matrix is 0.92, indicating that the phase angle change trends at these two frequency points are highly consistent.

[0089] The system entropy quantifies the uncertainty of the system. Assuming that a corrosion system contains 5 corrosion states, the probabilities of each state appearing are 0.1, 0.2, 0.3, 0.25 and 0.15 respectively. According to the calculation method of system entropy, the entropy value of the system is 2.186. For each indicator in the feature index set, the conditional entropy under the condition of the indicator is calculated. Taking the phase angle at 10 Hz as an example, when the phase angle is in different value intervals, it is in different corrosion states, and the calculated conditional entropy is 1.125. The difference between the system entropy and the conditional entropy is taken as the information gain value of the indicator, i.e. 2.186-1.125=1.061. Repeat this calculation for all indicators in the feature index set, for example, the information gain value of the energy density at 50 Hz is 0.845, and the information gain value of the frequency response amplitude at 1 kHz is 0.932. Compare the information gain values of all indicators, and select the indicator with the largest information gain value as the root node of the causal reasoning decision tree. In this example, the phase angle at 10 Hz has the largest information gain value of 1.061, so it is selected as the root node.

[0090] The root node is split to determine the optimal split point. For the phase angle at 10 Hz, multiple split positions are considered, such as -30°, -45°, -60°, etc. For each split position, the corresponding Gini coefficient is calculated. The Gini coefficient reflects the purity of each subset after splitting, calculated by the sum of the squares of the probabilities of each class of samples. For example, at the split point with a phase angle of -45°, the samples are divided into two parts, and the proportion of each corrosion type in the first part is {0.7, 0.2, 0.05, 0.05, 0}, and the proportion of each corrosion type in the second part is {0.05, 0.15, 0.3, 0.2, 0.3}. The Gini coefficient of the first part is 0.465, and the Gini coefficient of the second part is 0.735. The weighted average Gini coefficient of this split point is 0.625. Repeat this calculation for all split points, and select the position with the smallest Gini coefficient, -30°, as the optimal split point, with a Gini coefficient of 0.385.

[0091] Each layer of the decision tree represents a different level of influencing factors, forming a multi-level structure from the root node downwards. The conditional correlation coefficient between adjacent levels of nodes is calculated, and the conditional correlation coefficient is used as the correlation strength between levels. For example, the conditional correlation coefficient between the root node (phase angle at 10 Hz) and its child node (energy density at 50 Hz) is 0.83, indicating that the two have strong correlation. The conditional correlation coefficient between the root node and another child node (frequency response amplitude at 5 kHz) is 0.42, which is relatively weak.

[0092] Starting from the leaf nodes of the decision tree, the contribution of each node to the final corrosion state determination is calculated layer by layer upwards. For a node in the middle layer, its contribution is composed of its own discriminant ability and the weighted sum of the contribution of all its child nodes. For example, a node in the second layer (100 Hz energy density) has a discriminant ability of 0.75, and it has three child nodes with contribution degrees of 0.6, 0.45, and 0.3, and weights of 0.5, 0.3, and 0.2, respectively. The total contribution of the node is 0.75+(0.6×0.5+0.45×0.3+0.3×0.2) = 1.23. Based on the contribution of all nodes, the primary and secondary order of the influencing factors is determined. The factor with the largest contribution is the main influencing factor, and in this example, the phase angle at 10 Hz (contribution 1.82), the energy density at 50 Hz (contribution 1.65), and the frequency response amplitude at 1 kHz (contribution 1.47) are the three main factors affecting corrosion.

[0093] Figure 2For the accuracy rate comparison of corrosion feature recognition of different methods, the figure shows the accuracy rate comparison results of the present application and the traditional frequency domain analysis method and wavelet transform method in four different corrosion type recognition tasks. As can be seen from the figure, the present application has a significant performance advantage in uniform corrosion recognition, pitting detection, crevice corrosion analysis and stress corrosion discrimination. Specifically, the present application achieves an accuracy rate of 96.8% in uniform corrosion recognition, which is 24.5% and 12.1% higher than the 72.3% of the traditional frequency domain analysis method and the 84.7% of the wavelet transform method, respectively; in the pitting detection task, the present application achieves an accuracy rate of 94.3%, which is 25.4% and 13.1% higher than the other two methods; in the crevice corrosion analysis, the present application achieves an accuracy rate of 92.7%, which is better than the 65.4% of the traditional method and the 78.6% of the wavelet method; in the stress corrosion discrimination, the present application achieves an accuracy rate of 95.1%, which is significantly higher than the 70.1% of the traditional method and the 82.9% of the wavelet method. These results fully demonstrate that the present application has outstanding technical advantages in the accuracy and reliability of corrosion feature recognition through the innovative technical means of selecting root nodes by maximizing information gain, determining split points by minimizing Gini coefficient, and calculating node contribution by back propagation.

[0094] In an optional embodiment, the conditional probability relationship between the corrosion influencing factors is calculated by using the Bayesian network, and updating the correlation strength in the initial corrosion knowledge graph comprises:

[0095] A Bayesian network structure of the corrosion influencing factors is established, the prior probability of each corrosion influencing factor is calculated based on the observation data, and an initial node set is constructed according to the prior probability; the conditional probability relationship between the nodes in the initial node set is calculated, a conditional probability table is generated based on the conditional probability relationship, and the conditional probability table records the probability dependency relationship between a node and its parent node set;

[0096] An initial corrosion knowledge graph is constructed, the edge weight between nodes is calculated using mutual information, the edge weight represents the correlation strength between nodes, and the edge weight is written into the initial corrosion knowledge graph;

[0097] Belief propagation calculation is performed on the initial corrosion knowledge graph, the belief propagation calculation multiplies the diagnostic support of a node and the prediction support and normalizes to obtain a node belief value, and the correlation strength in the initial corrosion knowledge graph is updated according to the node belief value.

[0098] To establish the Bayesian network structure of corrosion influencing factors, relevant observation data need to be collected. Taking a pipeline corrosion problem as an example, temperature, humidity, pH value, chloride ion concentration and corrosion rate are selected as key influencing factors. Based on historical observation data, the prior probabilities of each factor are calculated. For example, temperature is divided into high temperature (> 60℃) and low temperature (≤ 60℃) two states, and through statistics it is found that the probability of high temperature state is 0.35, and the probability of low temperature state is 0.65; humidity is divided into high humidity (> 75%) and low humidity (≤ 75%) two states, and the probability of high humidity state is 0.40, and the probability of low humidity state is 0.60; pH value is divided into acidic (< 7), neutral (= 7) and alkaline (> 7) three states, and the probabilities are 0.25, 0.15 and 0.60 respectively; chloride ion concentration is divided into high concentration (> 1000 ppm) and low concentration (≤ 1000 ppm) two states, and the probabilities are 0.30 and 0.70 respectively; corrosion rate is divided into high rate (> 0.1 mm / year) and low rate (≤ 0.1 mm / year) two states, and the probabilities are 0.20 and 0.80 respectively.

[0099] Based on these prior probability values, the initial node set N = {temperature, humidity, pH value, chloride ion concentration, corrosion rate} is constructed. Next, the conditional probability relationship between nodes is calculated, and through statistical analysis of observation data, it is found that temperature and humidity have an impact on pH value, and pH value and chloride ion concentration jointly affect corrosion rate. According to these findings, the conditional probability table is generated. For example, when the temperature is high and the humidity is high, the conditional probability of pH being acidic is 0.70, the conditional probability of being neutral is 0.20, and the conditional probability of being alkaline is 0.10; when the pH value is acidic and the chloride ion concentration is high, the conditional probability of the corrosion rate being high is 0.85, and the conditional probability of being low is 0.15. In this way, the calculation of all conditional probabilities is completed, and a complete conditional probability table is formed, recording the probability dependence relationship between each node and its parent node set.

[0100] Mutual information measures the degree of mutual dependence between two variables, and the larger the value, the stronger the association. By calculating the mutual information between each pair of nodes, the edge weight values are obtained. For example, the mutual information value between temperature and pH value is 0.42, the mutual information value between humidity and pH value is 0.38, the mutual information value between pH value and corrosion rate is 0.65, and the mutual information value between chloride ion concentration and corrosion rate is 0.72. These edge weight values are written into the initial corrosion knowledge graph, representing the association strength between nodes.

[0101] During the belief propagation computation on the initial corrosion knowledge graph, each node receives messages from its neighboring nodes and updates its state based on the received messages. Specifically, for each node in the graph, its diagnostic support and predictive support are calculated respectively. The diagnostic support represents the degree of support of the current node state from the evidence transmitted by the child nodes, and the predictive support represents the degree of support of the current node state from the evidence transmitted by the parent nodes.

[0102] Taking the pH value node as an example, its diagnostic support comes from the feedback of the corrosion rate node, and its predictive support comes from the influence of the temperature and humidity nodes. Assuming that the temperature is observed to be in a high temperature state, through belief propagation, the diagnostic support of the pH value node in the acidic state is calculated to be 0.55, and the predictive support is calculated to be 0.65. After multiplying and normalizing the two, the belief value of the pH value node in the acidic state is obtained as 0.60. Similarly, the belief values of the pH value in the neutral and alkaline states are calculated to be 0.15 and 0.25 respectively.

[0103] According to the calculated node belief values, the correlation strength in the knowledge graph is updated. The updating method is to weight average the original edge weight and the mutual information calculated based on the new belief value. For example, the correlation strength between temperature and pH value changes from 0.42 to 0.48 after updating, and the correlation strength between pH value and corrosion rate changes from 0.65 to 0.70.

[0104] Through the above steps, the process of calculating the conditional probability relationship between corrosion influencing factors using Bayesian network and updating the correlation strength in the initial corrosion knowledge graph is completed. This method can effectively capture the complex interactions between corrosion influencing factors and provide scientific basis for corrosion prediction and protection. With the continuous accumulation of new observation data, the above process can be repeatedly executed to continuously optimize the corrosion knowledge graph and improve the accuracy of corrosion risk assessment.

[0105] Figure 3For different methods in corrosion prediction accuracy changes with time comparison chart, the figure shows the comparison of the prediction accuracy of the present invention and three kinds of mainstream corrosion prediction methods in 30 days test cycle. From the figure, it can be clearly seen that the prediction accuracy of the traditional impedance spectrum analysis method is always at the lowest level, from 68.2% on the first day to 73.0% on the 30th day, the increase is very limited, which is mainly due to the fact that this method can only obtain surface corrosion information and cannot analyze the internal mechanism. The wavelet transform frequency domain method performs slightly better, with an accuracy rate from 75.3% to 85.1%, but its dependence on frequency domain feature extraction limits its performance in complex corrosion environments. The support vector machine regression method achieves relatively good results through machine learning algorithms, with an accuracy rate from 79.1% to 89.5%, but it lacks the ability to model the complex correlation between corrosion influencing factors. In contrast, the present invention uses Bayesian network to model the conditional probability relationship between corrosion influencing factors, and combines belief propagation algorithm to dynamically update the correlation strength in the knowledge graph, achieving significant performance improvement, with a prediction accuracy rate from 88.7% on the first day to 98.4% on the 30th day. Not only is the initial accuracy higher than that of other methods by 9-20 percentage points, but it also shows stronger learning and adaptation ability as the test time increases, fully verifying the technical advantages and practical value of the present invention in the field of corrosion prediction.

[0106] In an optional implementation, predicting the corrosion mechanism evolution path according to the updated correlation strength in the initial corrosion knowledge graph comprises:

[0107] Counting the number of state transitions at adjacent time points, calculating the ratio of the number of state transitions to the total number of current states to obtain a state transition probability, and constructing a state transition probability matrix according to the state transition probability;

[0108] Combining the state transition probability matrix with the updated correlation strength to construct a time-varying conditional probability function, calculating the probability product of the time-varying conditional probability function on different paths to obtain a path importance score, and filtering a key evolution path based on the path importance score;

[0109] Calculating the cumulative probability of state probability at each time along the key evolution path to obtain a cumulative probability, and using the cumulative probability as a quantitative indicator to predict the evolution trend of the corrosion mechanism, wherein the evolution trend reflects the change law of the corrosion state over time.

[0110] When predicting the corrosion mechanism evolution path based on the correlation strength in the updated initial corrosion knowledge graph, it is necessary to count the number of transitions of corrosion state at adjacent time. Monitor the corrosion process in a certain environment and record the corrosion state at each sampling time point. For example, in a marine environment, a metal component is monitored for 30 days, and data is collected every 6 hours. The corrosion state is divided into four categories: initial state S0, slight corrosion S1, moderate corrosion S2, and severe corrosion S3. Through continuous observation, the number of times that state S0 changes to S1 is 18, and the number of times that S0 remains unchanged is 22; the number of times that S1 changes to S2 is 15, and the number of times that S1 remains unchanged is 30; the number of times that S2 changes to S3 is 10, and the number of times that S2 remains unchanged is 25.

[0111] The probability of S0 changing to S1 is 18 / (18+22) = 0.45, and the probability of S0 remaining unchanged is 22 / (18+22) = 0.55; the probability of S1 changing to S2 is 15 / (15+30) = 0.33, and the probability of S1 remaining unchanged is 30 / (15+30) = 0.67; the probability of S2 changing to S3 is 10 / (10+25) = 0.29, and the probability of S2 remaining unchanged is 25 / (10+25) = 0.71. Thus, a state transition probability matrix is constructed, which contains the probability of transition from any state to another state.

[0112] In the corrosion knowledge graph, the correlation strength represents the degree of influence between different corrosion factors and corrosion states. For example, in a marine environment, the correlation strength between chloride ion concentration and corrosion rate is 0.8, the correlation strength between temperature and corrosion rate is 0.6, and the correlation strength between dissolved oxygen content and corrosion rate is 0.7. These correlation strength data are derived from experimental measurements and comprehensive evaluation of expert knowledge.

[0113] The time-varying conditional probability function considers the complex relationship among time factor, environmental condition and corrosion state. In the specific implementation, for each evolution path, the probability product is calculated as the path importance score. Taking S0→S1→S2→S3 as an example, the path importance score is the probability of S0 turning into S1 (0.45) multiplied by the probability of S1 turning into S2 (0.33) multiplied by the probability of S2 turning into S3 (0.29), and then multiplied by the weighted average value of the correlation intensity in each transition process. If the weighted correlation intensity of chloride ion, temperature and dissolved oxygen in the transition process from S0 to S1 is 0.75, then the comprehensive probability of this stage is 0.45×0.75=0.3375. Similarly, if the weighted correlation intensity in the transition processes from S1 to S2 and from S2 to S3 is 0.7 and 0.65 respectively, then the comprehensive probabilities of these two stages are 0.33×0.7=0.231 and 0.29×0.65=0.1885 respectively. Finally, the importance score of this path is 0.3375×0.231×0.1885=0.0147.

[0114] By comparing the importance scores of different paths, the key evolution path is screened out. For example, compared with S0→S1→S2→S3, S0→S1→S3 and S0→S2→S3, if their importance scores are 0.0147, 0.008 and 0.006 respectively, then the highest score S0→S1→S2→S3 is selected as the key evolution path.

[0115] Suppose that at t=0, the probability of being in the initial state S0 is 1; at t=1, the probability of being in S0 is 0.55, and the probability of being in S1 is 0.45; at t=2, the probability of being in S0 is 0.55×0.55=0.3025, the probability of being in S1 is 0.55×0.45+0.45×0.67=0.5265, and the probability of being in S2 is 0.45×0.33=0.1485; the state probabilities at t=3 and later are calculated in the same way. The cumulative probability is the cumulative sum of the corresponding state probabilities at each time, for example, the cumulative probability of S1 state at t=1 to t=3 is 0.45+0.5265+0.4372=1.4137.

[0116] By analyzing the change of the cumulative probability of each state over time, the evolution law of the corrosion state can be determined. For example, if the cumulative probability of S2 state grows slowly in the first 10 days, rapidly in the period of 10-20 days, and then tends to be stable, it indicates that the moderate corrosion is the key evolution stage in the period of 10-20 days. This quantitative analysis helps to determine the key time window of corrosion protection.

[0117] In practical applications, according to the predicted corrosion evolution trend, a targeted anti-corrosion strategy can be formulated. For example, if it is predicted that the metal components of a certain offshore platform will enter the rapid corrosion stage 15-25 days after installation, then a key protection treatment such as coating an anti-corrosion coating or installing a sacrificial anode can be performed about 10 days after installation, thereby effectively delaying the corrosion process and prolonging the service life of the equipment.

[0118] The pico-level precision data acquisition system of the corrosion industry special large model of the embodiment of the application comprises:

[0119] The first unit is used for acquiring electrochemical impedance spectrum data of a metal material and synchronously recording corrosion environment parameters, performing multi-scale decomposition on the electrochemical impedance spectrum data by using continuous wavelet transform to obtain a time-frequency domain feature map; performing frequency band division on the time-frequency domain feature map based on a self-adaptive segmentation algorithm, screening initial boundary points corresponding to the time-frequency domain feature map according to a minimum frequency interval constraint, and determining key points of energy density distribution in the time-frequency domain feature map based on an energy continuity criterion to acquire a corrosion characteristic index of a target frequency interval;

[0120] The second unit is used for determining an information gain value corresponding to the corrosion characteristic index of the target frequency interval based on the corrosion characteristic index, selecting an index with the largest information gain value as a root node of a causal reasoning decision tree, constructing an initial structure of the causal reasoning decision tree, and establishing a hierarchical relationship of corrosion influencing factors according to the initial structure;

[0121] The third unit is used for constructing an initial corrosion knowledge graph based on the electrochemical impedance spectrum data and the corrosion environment parameters, calculating a conditional probability relationship between corrosion influencing factors by using a Bayesian network, and updating an association strength in the initial corrosion knowledge graph; and predicting a corrosion mechanism evolution path according to the association strength in the updated initial corrosion knowledge graph.

[0122] In a third aspect, an electronic device is provided, comprising:

[0123] a processor;

[0124] a memory for storing processor-executable instructions;

[0125] The processor is configured to invoke the instructions stored in the memory to execute the method described above.

[0126] In a fourth aspect, a computer-readable storage medium is provided, which stores computer program instructions, and the computer program instructions are executed by a processor to implement the method described above.

[0127] The present application can be a method, an apparatus, a system, and / or a computer program product. The computer program product can include a computer readable storage medium (or media) having computer readable program instructions thereon for performing various aspects of the present application.

[0128] Finally, it should be noted that the above-mentioned embodiments are merely used to illustrate the technical solutions of the present application, rather than limit the present application; even though the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that the technical solutions recorded in the foregoing embodiments can be modified, or some or all of the technical features thereof can be replaced equivalently; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A picoampere-level precision data acquisition method for a large model dedicated to the corrosion industry, characterized by: include: Collecting electrochemical impedance spectroscopy data of metal materials and synchronously recording corrosion environment parameters, performing multi-scale decomposition of the electrochemical impedance spectroscopy data using continuous wavelet transform to obtain time-frequency domain feature maps; The time-frequency domain characteristic graph is divided into frequency bands based on an adaptive segmentation algorithm, the initial boundary points corresponding to the time-frequency domain characteristic graph are selected according to the minimum frequency interval constraint, and the key points of the energy density distribution in the time-frequency domain characteristic graph are determined based on the energy continuity criterion to obtain the corrosion characteristic index in the target frequency interval; Based on the corrosion characteristic indicators in the target frequency range, determining the information gain values ​​corresponding to the corrosion characteristic indicators, selecting the indicator with the largest information gain value as the root node of the causal reasoning decision tree, constructing the initial structure of the causal reasoning decision tree, and establishing a hierarchical relationship of corrosion influencing factors based on the initial structure; An initial corrosion knowledge graph is constructed based on the electrochemical impedance spectroscopy data and the corrosion environment parameters, and a Bayesian network is used to calculate the conditional probability relationship between corrosion influencing factors, and the association strength in the initial corrosion knowledge graph is updated; and the corrosion mechanism evolution path is predicted based on the association strength in the updated initial corrosion knowledge graph.

2. The method according to claim 1, characterized in that Collect electrochemical impedance spectroscopy data of metal materials and simultaneously record corrosion environment parameters, use continuous wavelet transform to perform multi-scale decomposition on the electrochemical impedance spectroscopy data, and obtain time-frequency domain feature maps including: Inputting the electrochemical impedance spectroscopy signal into a Morlet wavelet transform model to obtain wavelet transform coefficients; Establishing an adaptive scale parameter system, determining a minimum scale parameter and a maximum scale parameter according to the frequency range of the electrochemical impedance spectroscopy signal, and generating a scale parameter corresponding to the electrochemical impedance spectroscopy signal based on a logarithmic uniform distribution principle; The wavelet transform coefficients are constructed into a coefficient matrix according to the scale parameter and the time position parameter, the energy density value of each element of the coefficient matrix is ​​calculated, the energy density value is logarithmically transformed and normalized, and a time-frequency domain feature map is generated.

3. The method according to claim 1, characterized in that The time-frequency domain feature map is divided into frequency bands based on an adaptive segmentation algorithm, the initial boundary points corresponding to the time-frequency domain feature map are screened according to the minimum frequency interval constraint, and the key points of the energy density distribution in the time-frequency domain feature map are determined based on the energy continuity criterion. The corrosion characteristic indicators in the target frequency range are obtained, including: Calculating an energy density gradient value of the time-frequency domain feature map, where the energy density gradient value includes a frequency direction gradient component and a time direction gradient component; Calculating local statistical features of the energy density gradient value based on a sliding window mechanism, wherein the local statistical features include a local variance value, and the width of the sliding window is adaptively adjusted according to the frequency resolution; Generating a dynamic segmentation threshold based on the local statistical features, wherein the dynamic segmentation threshold is composed of a global mean and a local standard deviation weighted by the local statistical features; performing boundary point detection on the energy density gradient value based on the dynamic segmentation threshold to obtain an initial boundary point set, wherein the initial boundary point satisfies the conditions of being greater than the dynamic segmentation threshold and greater than the values ​​of adjacent points; Initial boundary points are selected based on the minimum frequency interval constraint; adjacent boundary points are merged based on the energy continuity criterion; the average energy ratio of the regions to be merged is calculated, and region merging is performed when the average energy ratio is less than a set energy threshold; Determine key points of energy density distribution in the merged area, wherein the key points include energy density extreme points and energy density mutation points, and obtain frequency values ​​and phase angle values ​​corresponding to the key points; determine characteristic frequency bands of the electrochemical impedance spectrum based on the frequency values ​​and the phase angle values, wherein the characteristic frequency bands are used to characterize the impedance characteristics of the electrochemical system.

4. The method according to claim 1, wherein Based on the corrosion characteristic index of the target frequency range, the information gain value corresponding to the corrosion characteristic index is determined, the index with the largest information gain value is selected as the root node of the causal reasoning decision tree, the initial structure of the causal reasoning decision tree is constructed, and the hierarchical relationship of the corrosion influencing factors is established based on the initial structure, including: Extracting corrosion characteristic indicators within the target frequency interval, calculating phase angle data, energy density distribution data, and frequency response amplitude data within the target frequency interval, and combining the phase angle data, the energy density distribution data, and the frequency response amplitude data into a characteristic indicator set; Calculating the Pearson correlation coefficient for each of the indicators in the characteristic indicator set to generate an indicator correlation matrix, wherein each element of the indicator correlation matrix represents the degree of correlation between the corresponding two indicators; Calculating the system entropy of the characteristic indicator set and calculating the conditional entropy under each indicator condition, taking the difference between the system entropy and the conditional entropy as the information gain value, and selecting the indicator with the largest information gain value as the root node of the causal reasoning decision tree; Constructing an initial structure of a causal reasoning decision tree based on the root node, calculating the Gini coefficient corresponding to each split position, and selecting the position with the smallest Gini coefficient as the node split position, wherein the Gini coefficient is obtained by calculating the sum of squared probabilities of various samples; Establishing a hierarchical relationship of influencing factors based on the initial structure, calculating conditional correlation coefficients between nodes at adjacent levels, and using the conditional correlation coefficients as the correlation strength between the levels; The influence degree of each level node on the root node output is calculated by back propagation, and the influence degree is used as the node contribution. The priority order of the influencing factors is determined based on the size of the node contribution.

5. The method according to claim 1, wherein The Bayesian network is used to calculate the conditional probability relationship between corrosion influencing factors, and the association strength in the initial corrosion knowledge graph is updated, including: Establishing a Bayesian network structure for corrosion influencing factors, calculating the prior probability of each corrosion influencing factor based on observed data, and constructing an initial node set based on the prior probability; calculating the conditional probability relationship between nodes in the initial node set, and generating a conditional probability table based on the conditional probability relationship, wherein the conditional probability table records the probabilistic dependency relationship between the node and its parent node set; Constructing an initial corrosion knowledge graph, calculating edge weights between nodes using mutual information, wherein the edge weights represent the strength of association between nodes, and writing the edge weights into the initial corrosion knowledge graph; A belief propagation calculation is performed on the initial corrosion knowledge graph, wherein the belief propagation calculation multiplies the diagnosis support and the prediction support of the node and normalizes the result to obtain a node belief value, and the association strength in the initial corrosion knowledge graph is updated according to the node belief value.

6. The method according to claim 1, characterized in that The corrosion mechanism evolution path predicted based on the association strength in the updated initial corrosion knowledge graph includes: Counting the number of transitions between corrosion states at adjacent moments, calculating the ratio of the number of state transitions to the total number of current states to obtain a state transition probability, and constructing a state transition probability matrix based on the state transition probability; Combining the state transition probability matrix with the updated association strength to construct a time-varying conditional probability function, calculating the probability product of the time-varying conditional probability function on different paths to obtain a path importance score, and screening key evolution paths based on the path importance score; The cumulative probability is obtained by calculating the cumulative sum of the state probabilities at each moment along the key evolution path. The cumulative probability is used as a quantitative indicator to predict the evolution trend of the corrosion mechanism. The evolution trend reflects the change law of the corrosion state over time.

7. A picoampere-level precision data acquisition system for large-scale models dedicated to the corrosion industry, used to implement the method according to any one of claims 1 to 6, characterized in that: include: The first unit is used to collect electrochemical impedance spectroscopy data of metal materials and simultaneously record corrosion environment parameters, and perform multi-scale decomposition of the electrochemical impedance spectroscopy data using continuous wavelet transform to obtain time-frequency domain feature maps; The time-frequency domain characteristic graph is divided into frequency bands based on an adaptive segmentation algorithm, the initial boundary points corresponding to the time-frequency domain characteristic graph are selected according to the minimum frequency interval constraint, and the key points of the energy density distribution in the time-frequency domain characteristic graph are determined based on the energy continuity criterion to obtain the corrosion characteristic index in the target frequency interval; The second unit is configured to determine, based on the corrosion characteristic indicators in the target frequency range, an information gain value corresponding to the corrosion characteristic indicators, select the indicator with the largest information gain value as the root node of the causal reasoning decision tree, construct an initial structure of the causal reasoning decision tree, and establish a hierarchical relationship of corrosion influencing factors based on the initial structure; The third unit is used to construct an initial corrosion knowledge graph based on the electrochemical impedance spectroscopy data and the corrosion environment parameters, use a Bayesian network to calculate the conditional probability relationship between corrosion influencing factors, and update the association strength in the initial corrosion knowledge graph; predict the corrosion mechanism evolution path based on the association strength in the updated initial corrosion knowledge graph.

8. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images