Method for determining the state of corrosion of steel reinforcement in concrete

Abstract

The application discloses a kind of concrete internal steel bar corrosion state probability determination method, this scheme can establish sample library by literature research, utilize sample library to train machine learning model, and machine learning model after training can export the probability that the durability of reinforced concrete structure belongs to three structure durability stages according to the input electrochemical index;The beneficial technical effect of the application is: a kind of concrete internal steel bar corrosion state probability determination method is proposed, the scheme can train machine learning model using existing research results, realize the probability that structure is in each durability stage under the condition that part of electrochemical index is not collected.

Description

Technical Field

[0001] This invention relates to a structural durability condition assessment technology, and more particularly to a method for determining the probability of steel reinforcement corrosion inside concrete. Background Technology

[0002] Reinforced concrete is the most widely used man-made material in the construction industry worldwide due to its superior mechanical properties. The internal steel reinforcement significantly improves the mechanical properties of the structure without altering the cementitious matrix properties of the concrete. At the same time, the presence of concrete provides a protective barrier for the steel reinforcement, which is normally susceptible to corrosion.

[0003] Concrete structures are designed to resist atmospheric effects, chemical erosion, and other deteriorating factors throughout their expected service life, maintaining their safety, serviceability, and appearance without requiring significant maintenance costs during operation. However, due to the inherent characteristics of concrete materials and the environment in which they are used, concrete structures inevitably suffer from durability issues, meaning they may fail to meet these design requirements after a period of service.

[0004] For reinforced concrete in a chloride environment, the main process of structural durability degradation is considered to be: "chloride transfer within the concrete - chloride ion concentration on the steel surface reaches the critical corrosion concentration - steel corrosion within the concrete - formation of corrosion products leading to rust expansion - concrete cracking - continued cracking accelerating the transfer of harmful ions, water, and oxygen, further exacerbating steel corrosion - structural performance degradation." Existing theories summarize and simplify this degradation process into three main stages: steel passivation, initial rusting, and corrosion propagation.

[0005] The passivation stage, initial rusting stage, and corrosion propagation stage of reinforced concrete structures correspond to three states of durability damage: slight or no damage, moderate damage, and severe damage, respectively. From the perspective of reinforced concrete structure reinforcement and maintenance, compared to when a structure has already suffered moderate or severe durability damage, maintaining the structure when the damage is relatively minor can cost nearly 20 times more to achieve the same results. This would impose a heavy financial burden on the country, contradicting my country's basic national policies of sustainable development and environmental protection, and hindering the overall healthy and rapid development of the Chinese economy. Furthermore, this type of durability degradation can lead to structural collapse in severe cases, seriously threatening national safety. Therefore, it is of great significance to collect indicators to assess the durability status of a structure without compromising its overall integrity, and to use the assessment results to guide structural durability repair and reinforcement or to identify high-risk structures with potential safety hazards.

[0006] Several electrochemical methods have been developed to assess the degree of steel corrosion within reinforced concrete structures, i.e., the structural durability. These methods can test the condition of the steel reinforcement without damaging the overall concrete structure. These indicators and testing methods mainly include: half-cell potential (two-electrode half-cell potential method), concrete resistivity (four-electrode Wenner method), corrosion current density (Tafel method), and steel polarization resistance (linear polarization method).

[0007] Currently, the main guidelines for evaluating the corrosion state of reinforcing steel bars using electrochemical indicators are as follows:

[0008] 1. Zhang Jingquan, "Guidelines and Engineering Examples for Material Condition and Durability Testing and Evaluation of Old Concrete Bridges":

[0009] Criteria for determining the corrosion potential of steel reinforcement in structural concrete.

[0010]

[0011] 2. Technical Specification for Testing Reinforcing Steel in Concrete (JGJT 152-2008):

[0012] Criteria for evaluating the corrosion characteristics of reinforcing steel bars based on half-cell potential

[0013]

[0014] 3. Technical Standard for On-site Testing of Concrete Structures (GB / T 50784-2013):

[0015] Table 1. Corrosion Current and Corrosion Rate of Reinforcing Steel and Determination of Component Damage Age

[0016]

[0017] Table 2. Determination of Concrete Resistivity and Steel Corrosion Status

[0018] Serial Number Concrete resistivity (kΩcm) Determination of steel bar corrosion status 1 ＞100 The steel bars will not rust. 2 50～100 low corrosion rate 3 10～50 When steel bars are activated, medium to high corrosion rates can occur. 4 ＜10 Resistivity is not the controlling factor of corrosion.

[0019] 4. According to the American ASTM C876 standard, when the corrosion current density is in the range of 1-10 μA·cm-2, the corrosion rate of steel bars is low, and when it is in the range of 10-100 μA·cm-2, the corrosion rate of steel bars is high.

[0020] 5. Ba Hengjing et al. believe that when the polarization resistance of steel bars is less than 27 kΩ·cm2, the steel bars begin to depassivate and enter the initial rusting state. (Ba Hengjing, Zhao Weixuan. Using polarization resistance to test the critical chloride ion concentration of steel bars in simulated pore solution of concrete [J] Concrete, 2010(12): 1-4.)

[0021] The shortcomings of existing technology:

[0022] 1. Existing research findings and national standards for assessing structural durability using electrochemical indicators do not establish a mapping relationship between these indicators and the three specific stages of durability development: the passivation stage, the initial rusting stage, and the rust propagation stage. Instead, they focus on evaluating whether the steel reinforcement has rusted and the rate of rust corrosion. Except for the "Technical Standard for On-site Testing of Concrete Structures" (GB / T 50784-2013), which specifies a rust current density of less than 0.2 μA·cm⁻¹, this approach is not feasible. -2 If the reinforcing steel is in a passive state, such test results are detrimental to the final decision on reinforcement and repair.

[0023] 2. The test results of various electrochemical indicators will be affected by many factors, including structural material characteristics (water-cement ratio), structural characteristics (protective layer thickness, crack width), the water-filling rate of pores in the concrete, and the ambient temperature and humidity during testing. Even under conditions of consistent steel reinforcement corrosion and structural durability, multiple tests may yield significantly different results. Therefore, using a single electrochemical indicator to assess structural durability is inaccurate; a comprehensive evaluation using four durability indicators is necessary. However, obtaining some electrochemical indicators, particularly corrosion current density and polarization resistance, requires specialized instruments, considerable electrochemical testing experience, and a lengthy testing period. During on-site testing, it is difficult to simultaneously obtain all four electrochemical indicators and achieve a comprehensive evaluation of the structural durability stage.

[0024] 3. Considering that the test results of various electrochemical indicators will be affected by a series of factors including structural material properties and structural characteristics, the deterministic classification results may not be able to describe the influence of subjective and objective uncertainties.

[0025] 4. Existing methods cannot comprehensively utilize the four electrochemical indicators. Even if all four indicators are obtained, the final evaluation results cannot be unified. Summary of the Invention

[0026] To address the problems in the background technology, the inventors considered that existing scientific research results contain a large number of findings regarding different electrochemical indicators obtained from tests at different structural durability stages (passivation stage, initial rusting stage, and corrosion propagation stage). These findings can be used to develop machine learning models that take four electrochemical indicators as input and the probability of the structure being in different durability stages as output. Therefore, the inventors proposed the solution of this invention, the specific solution of which is as follows:

[0027] An innovative method for determining the probability of steel reinforcement corrosion state inside concrete is as follows: the method includes:

[0028] The durability of reinforced concrete structures includes three stages: passivation, initial rusting, and corrosion propagation.

[0029] Testing reinforced concrete structures yields four electrochemical indicators: the half-cell potential parameter E can be obtained from the two-electrode half-cell potential method. corr The concrete resistivity parameter ρ can be obtained using the four-electrode Wenner method, and the corrosion current density parameter i can be obtained using the Tafel method. corr The polarization resistance parameter R of the reinforcing steel can be obtained by the linear polarization method. ρ ;

[0030] 1) Through literature review, various electrochemical indicators under three structural durability stages were obtained. Theoretically, a certain corrosion state should correspond to four electrochemical indicators. In the actual literature review process, it may not be possible to obtain all four electrochemical indicators corresponding to a certain steel corrosion state. The obtained electrochemical indicators are marked as definite parameters, and the unobtained electrochemical indicators are marked as missing parameters.

[0031] 2) Then, organize the obtained parameters as follows: Two electrochemical indices under the same steel reinforcement corrosion state constitute a parameter pair, and multiple parameter pairs corresponding to the same two electrochemical indices constitute a parameter set: The first parameter set corresponds to E corr and i corr The parameter pairs formed, the second parameter set corresponds to E corr and R ρ The parameter pairs formed, the third parameter set corresponds to E corr The parameter pair formed by ρ, the fourth parameter set corresponds to the set of parameters formed by i. corr The parameter pair formed by ρ, the fifth parameter set corresponds to i corr and R ρ The parameter pairs formed, the sixth parameter set corresponds to R ρ The parameter pair formed by ρ;

[0032] 3) The EM algorithm is used to process the aforementioned six parameter sets one by one to establish the joint probability density function between the two electrochemical indices;

[0033] 4) Sample construction: A single sample consists of four electrochemical indicators and their corresponding structural durability stages; if an electrochemical indicator under a certain sample is missing, the sample is recorded as a parameter missing sample; a sample without missing electrochemical indicators is recorded as a complete sample.

[0034] 5) Repair samples with missing parameters. When repairing a single sample with missing parameters, follow these steps:

[0035] A) Duplicate the sample with missing parameters into n copies to obtain n duplicate samples;

[0036] B) Directly fill the determined parameters and corresponding structural durability stages into the corresponding positions in the replicated sample;

[0037] C) The four electrochemical indicators are each treated as four nodes. The node corresponding to the parameter is determined as the parent node, and the node corresponding to the missing parameter is determined as the child node. Then, a Bayesian network is established based on the rule that the parent node and child node are connected and the parent node points to the child node, as well as the corresponding joint probability density function. The distribution law of the electrochemical indicators corresponding to the child node is calculated by Bayesian network inference method.

[0038] Based on the distribution pattern of a certain child node, random sampling is performed n times according to the corresponding distribution pattern. The n parameters obtained by sampling are the repair parameters corresponding to the missing parameters. Then, these n repair parameters are filled into the positions corresponding to the missing parameters in the n replicated samples respectively. (Considering that Bayesian network inference is an existing theory, this paper introduces it in a relatively brief manner. If there are any omissions, those skilled in the art should refer to the existing theory for understanding.)

[0039] The repaired n duplicate samples and the complete sample together constitute the sample library;

[0040] 6) Train the machine learning model based on the sample library. During training, the electrochemical index is used as input, and the probability of the reinforced concrete structure's durability belonging to the three structural durability stages is used as output.

[0041] 7) Once the machine learning model is trained, it can be put into use. The measured electrochemical indicators are input into the machine learning model, and the machine learning model outputs the probability that the durability of the reinforced concrete structure belongs to one of the three structural durability stages. (In practice, as long as one electrochemical indicator can be obtained, this invention can output the final result. Of course, the more electrochemical indicators input, the higher the accuracy of the final result).

[0042] The aforementioned scheme involves the problem of repairing missing parameter samples. The specific reason is that, due to various reasons, most of the electrochemical parameters obtained in existing studies are incomplete, and some electrochemical indicators have not been tested. In order to make full use of existing research results, this invention uses relevant methods to repair missing parameter samples, which can also further increase the sample size and improve the training accuracy and evaluation accuracy.

[0043] The technical significance of this invention is mainly in the following aspects:

[0044] (1) The probability of the structure being in each durability stage can be determined by collecting and testing electrochemical indicators on site;

[0045] (2) It is possible to determine the probability of the structure being in each durability stage when some electrochemical indicators are not collected;

[0046] (3) A complete implementation method for determining the probability of a structure being in each durability stage using electrochemical indicators was proposed.

[0047] (4) The sample source makes full use of the existing electrochemical index testing results, without the need to conduct a large number of experiments to obtain samples, and the implementation difficulty and cost are relatively low.

[0048] While the aforementioned scheme can determine the probability of a structure being in each durability stage when only some electrochemical indicators are input, it is clear that the accuracy is best when all four electrochemical indicators are input. However, in practical engineering, due to limitations, it is often impossible to obtain all four electrochemical indicators. Therefore, this invention also proposes the following preferred scheme to supplement the measured data and improve the accuracy of the final result, as follows:

[0049] When there are four electrochemical indicators measured in step 7), the four electrochemical indicators are directly input into the machine learning model, and the machine learning model outputs the probability that the durability of the reinforced concrete structure belongs to the three structural durability stages.

[0050] When the electrochemical indicators measured in step 7) are one, two, or three, they should be processed as follows:

[0051] (1) Construct a prediction sample containing four electrochemical indicators, where the obtained electrochemical indicators are marked as known parameters and the unobtained electrochemical indicators are marked as unknown parameters;

[0052] (2) Copy the predicted sample into x copies to obtain x copied predicted samples;

[0053] (3) Fill the known parameters directly into the corresponding positions in the copied prediction samples;

[0054] (4) Complete the replication prediction sample: The four electrochemical indicators are respectively used as four nodes, the node corresponding to the known parameter is used as the parent node, and the node corresponding to the unknown parameter is used as the child node; then, a Bayesian network is established according to the rule that the parent node and the child node are connected and the parent node points to the child node and the corresponding joint probability density function. The distribution law of the electrochemical indicators corresponding to the child node is calculated by Bayesian network inference method.

[0055] Based on the distribution pattern of a certain child node, random sampling is performed x times according to the corresponding distribution pattern. The x parameters obtained by sampling are the completion parameters corresponding to the unknown parameters. Then, these x completion parameters are filled into the positions corresponding to the unknown parameters in x replicated prediction samples.

[0056] (5) Input the completed x replicated prediction samples into the machine learning model one by one, and average the x results output by the machine learning model to obtain the final result.

[0057] By adopting this preferred approach, as long as one electrochemical index can be obtained during actual testing, the remaining electrochemical indexes can be supplemented, ensuring that the machine learning model can always obtain four electrochemical index inputs, thereby improving the accuracy of the machine learning model's output.

[0058] The beneficial technical effect of this invention is that it proposes a method for determining the probability of steel reinforcement corrosion state inside concrete. This method can use existing research results to train a machine learning model and determine the probability of the structure being in each durability stage under the condition that some electrochemical indicators are not collected. Attached Figure Description

[0059] Figure 1 A potential detection result cloud map obtained by testing the potential of the reinforcing bars at the bottom plate position at the mid-span of a reinforced concrete beam;

[0060] Figure 2 , Figure 1 The statistical distribution of the half-cell potential parameters corresponding to the cloud map shown is illustrated.

[0061] Figure 3 A statistical distribution diagram of the corrosion current density obtained by testing the corrosion current density of the reinforcing bars at the bottom plate position at the mid-span of a reinforced concrete beam. Detailed Implementation

[0062] A method for probabilistically determining the corrosion state of reinforcing steel bars inside concrete, characterized in that: the method includes:

[0063] The durability of reinforced concrete structures includes three stages: passivation, initial rusting, and corrosion propagation.

[0064] Testing reinforced concrete structures yields four electrochemical indicators: the half-cell potential parameter E can be obtained from the two-electrode half-cell potential method. corr The concrete resistivity parameter ρ can be obtained using the four-electrode Wenner method, and the corrosion current density parameter i can be obtained using the Tafel method. corr The polarization resistance parameter R of the reinforcing steel can be obtained by the linear polarization method. ρ ;

[0065] 1) Through literature review, various electrochemical indicators under three structural durability stages were obtained. Theoretically, a certain corrosion state should correspond to four electrochemical indicators. In the actual literature review process, it may not be possible to obtain all four electrochemical indicators corresponding to a certain steel corrosion state. The obtained electrochemical indicators are marked as definite parameters, and the unobtained electrochemical indicators are marked as missing parameters.

[0066] 2) Then, organize the obtained parameters as follows: Two electrochemical indices under the same steel reinforcement corrosion state constitute a parameter pair, and multiple parameter pairs corresponding to the same two electrochemical indices constitute a parameter set: The first parameter set corresponds to E corr and i corr The parameter pairs formed, the second parameter set corresponds to E corr and R ρ The parameter pairs formed, the third parameter set corresponds to E corr The parameter pair formed by ρ, the fourth parameter set corresponds to the set of parameters formed by i. corr The parameter pair formed by ρ, the fifth parameter set corresponds to i corr and R ρ The parameter pairs formed, the sixth parameter set corresponds to R ρ The parameter pair formed by ρ;

[0067] The relationship between the aforementioned steel reinforcement corrosion state and structural durability stage is as follows: Based on common sense, steel reinforcement corrosion is a slow-changing process. A single structural durability stage refers to a corrosion process with a long time span, while the steel reinforcement corrosion state refers to the corrosion state of the steel reinforcement at a certain point in time. From a time domain perspective, a single structural durability stage includes multiple steel reinforcement corrosion states.

[0068] 3) The EM algorithm is used to process the aforementioned six parameter sets one by one to establish the joint probability density function between the two electrochemical indices;

[0069] 4) Sample construction: A single sample consists of four electrochemical indicators and their corresponding structural durability stages; if an electrochemical indicator under a certain sample is missing, the sample is recorded as a parameter missing sample; a sample without missing electrochemical indicators is recorded as a complete sample.

[0070] 5) Repair samples with missing parameters. When repairing a single sample with missing parameters, follow these steps:

[0071] A) Duplicate the sample with missing parameters into n copies to obtain n duplicate samples;

[0072] B) Directly fill the determined parameters and corresponding structural durability stages into the corresponding positions in the replicated sample;

[0073] C) Treat the four electrochemical indicators as four nodes, determine the node corresponding to the parameter as the parent node, and the node corresponding to the missing parameter as the child node; then establish a Bayesian network based on the rule that the parent node and child node are connected and the parent node points to the child node and the corresponding joint probability density function, and use the Bayesian network inference method to calculate the distribution law of the electrochemical indicators corresponding to the child node.

[0074] Based on the distribution pattern of a certain child node, random sampling is performed n times according to the corresponding distribution pattern. The n parameters obtained by sampling are the repair parameters corresponding to the missing parameters. Then, these n repair parameters are filled into the positions corresponding to the missing parameters in the n replicated samples.

[0075] The repaired n duplicate samples and the complete sample together constitute the sample library;

[0076] 6) Train the machine learning model based on the sample library. During training, the electrochemical index is used as input, and the probability of the reinforced concrete structure's durability belonging to the three structural durability stages is used as output.

[0077] In practice, to improve the accuracy of machine learning models, an ensemble machine learning model can be used. For example, before training, the training samples are bagged and resampled to obtain M sets of training sample subsets with low correlation. Then, the M sets of training sample subsets are used to train M / 3 sets of nearest neighbor node models, Naive Bayes and random forest models respectively. Then, the M sets of models are combined to form the final machine learning model required. After inputting electrochemical indicators, the M sets of models can all give the probability that different structures are in different durability stages. The average of the M results is the final result we need.

[0078] The specific model used can be selected from existing technologies. For example, the nearest neighbor model can be used to find the k points in the training samples that are closest to the mapping point of the sample to be predicted based on Euclidean distance. If the number of points representing the passivation stage, the initial rusting stage, and the corrosion expansion stage in the k points are β1, β2, and β3 respectively, then the probability of belonging to the passivation stage, the initial rusting stage, and the corrosion expansion stage in the prediction result is β1 / k, β2 / k, and β3 / k respectively. For another example, the inherent properties of the Naive Bayes model ensure that it can directly output the probability results of different classifications. For the random forest model, it is assumed that each model consists of l decision trees. If the number of decision trees that indicate that the structure is in the passivation stage, the initial rusting stage, and the corrosion expansion stage are α1, α2, and α3 respectively, then the probability of belonging to the passivation stage, the initial rusting stage, and the corrosion expansion stage in the prediction result is α1 / k, α2 / k, and α3 / k respectively.

[0079] 7) Once the machine learning model is trained, it can be put into use. The measured electrochemical indicators are input into the machine learning model, and the machine learning model outputs the probability that the durability of the reinforced concrete structure belongs to one of the three structural durability stages.

[0080] Furthermore, when there are four electrochemical indicators measured in step 7), the four electrochemical indicators are directly input into the machine learning model, and the machine learning model outputs the probability that the durability of the reinforced concrete structure belongs to one of the three structural durability stages.

[0081] When the electrochemical indicators measured in step 7) are one, two, or three, they should be processed as follows:

[0082] (1) Construct a prediction sample containing four electrochemical indicators, where the obtained electrochemical indicators are marked as known parameters and the unobtained electrochemical indicators are marked as unknown parameters;

[0083] (2) Copy the predicted sample into x copies to obtain x copied predicted samples;

[0084] (3) Fill the known parameters directly into the corresponding positions in the copied prediction samples;

[0085] (4) Complete the replication prediction sample: The four electrochemical indicators are respectively used as four nodes, the node corresponding to the known parameter is used as the parent node, and the node corresponding to the unknown parameter is used as the child node; then, a Bayesian network is established according to the rule that the parent node and the child node are connected and the parent node points to the child node and the corresponding joint probability density function. The distribution law of the electrochemical indicators corresponding to the child node is calculated by Bayesian network inference method.

[0086] Based on the distribution pattern of a certain child node, random sampling is performed x times according to the corresponding distribution pattern. The x parameters obtained by sampling are the completion parameters corresponding to the unknown parameters. Then, these x completion parameters are filled into the positions corresponding to the unknown parameters in x replicated prediction samples.

[0087] (5) Input the completed x replicated prediction samples into the machine learning model one by one, and average the x results output by the machine learning model to obtain the final result.

[0088] Example 1:

[0089] The potential of the reinforcing bars at the bottom slab location at mid-span of a reinforced concrete beam was tested, and the results were as follows: Figure 1 The potential detection result contour map shown (the contour map position corresponds to the location of the rebar); the statistical law of the half-cell potential parameters corresponding to the contour map is plotted as follows. Figure 2As shown, the mean value of the potential detection results within the section is -38mV, and the standard deviation is 43mV. The lower bound of the 95% confidence interval is calculated to be -121mV. Inputting the lower bound of the 95% confidence interval -121mV into the machine learning model of the present invention yields the following output:

[0090] The probability of passivation stage is 88.9%.

[0091] The probability of initial rust stage is 11.1%.

[0092] The probability of corrosion propagation stage is 0.0%.

[0093] Based on this, it can be determined that most of the reinforcing bars at this location in the reinforced concrete beam are in a passive state, and the structure still has a relatively long service life.

[0094] Example 2:

[0095] The corrosion current density of the reinforcing bars at the bottom slab location at mid-span of a reinforced concrete beam was tested, and the statistical law of the corrosion current density is as follows: Figure 3 As shown, the mean value of the corrosion current density test results within the section is 0.147 μA·cm. -2 The standard deviation is 0.040 μA·cm. -2 The calculated upper bound of the 95% confidence interval is 0.227 μA·cm. -2 The upper bound of the 95% confidence interval is set at 0.227 μA·cm. -2 Inputting the machine learning model of the present invention yields the following output:

[0096] The probability of passivation stage is 79.6%.

[0097] The probability of initial rusting is 20.4%.

[0098] The probability of corrosion propagation stage is 0.0%.

[0099] Based on this, it can be determined that most of the reinforcing bars at this location in the reinforced concrete beam are in a passive state, and the structure still has a relatively long service life.

[0100] Example 3:

[0101] By directly inputting the half-cell potential parameter -650mV into the model of this invention, we can obtain:

[0102] The probability of passivation stage is 0.1%.

[0103] The probability of initial rust stage is 32.6%.

[0104] The probability of the corrosion propagation stage is 67.3%.

[0105] At this point, it can be determined that the structure is in the corrosion expansion stage, posing a significant safety hazard. However, if the half-cell potential parameter of -650mV is used to determine the corrosion probability of the steel reinforcement according to the "Technical Specification for Testing Steel Reinforcement in Concrete" (JGJT 152-2008), it can only be determined that the steel reinforcement has a 95% probability of corrosion, but the severity of the structural durability failure cannot be further determined.

[0106] Example 4:

[0107] The corrosion current density parameter is directly 3.5 μA·cm. -2 By inputting the model of this invention, we can obtain:

[0108] The probability of passivation stage is 0.1%.

[0109] The probability of initial rust stage is 62.3%.

[0110] The probability of the corrosion propagation stage is 37.6%.

[0111] At this point, the structure can be determined to be in the initial rusting stage. However, according to the American ASTM C876 standard, the corrosion current density parameter is 3.5 μA·cm. -2 Making a judgment only provides information that the corrosion rate of the steel bars is relatively slow, making it difficult to further evaluate the durability of the structure.

[0112] Example 5:

[0113] By inputting the half-cell potential parameter -525mV into the model of this invention, we can obtain:

[0114] The probability of passivation stage is 0.1%.

[0115] The probability of initial rust stage is 79.1%.

[0116] The probability of the corrosion propagation stage is 20.8%.

[0117] At this point, the model judges the structure to be highly likely in the initial rust stage. However, we can set the quantile of the guarantee rate based on the importance of the structure. For example, for extremely important structures, a probability greater than 5% indicates the structure has entered the rust expansion stage; for ordinary structures, a probability greater than 45%–50% indicates the structure has entered the rust expansion stage. Based on the above calculation results, important structures are judged to be in the rust expansion stage; ordinary structures are judged to be in the initial rust stage.

[0118] Example 6: The half-cell potential parameter measured at the same location on the same structure was -121mV, and the corrosion current density parameter was 0.23μA·cm. -2 .

[0119] By inputting the half-cell potential parameter -121mV into the model of this invention alone, we obtain:

[0120] The probability of passivation stage is 88.9%.

[0121] The probability of initial rust stage is 11.1%.

[0122] The probability of corrosion propagation stage is 0.0%.

[0123] The corrosion current density parameter is 0.23 μA·cm. -2 Inputting the model of this invention alone yields:

[0124] The probability of passivation stage is 80.5%.

[0125] The probability of initial rust stage is 19.5%.

[0126] The probability of corrosion propagation stage is 0.0%.

[0127] The half-cell potential parameter is -121mV, and the corrosion current density parameter is 0.23μA·cm. -2 Simultaneously inputting the model of this invention, we obtain:

[0128] The probability of passivation stage is 80.9%.

[0129] The probability of initial rust stage is 19.1%.

[0130] The probability of corrosion propagation stage is 0.0%.

[0131] The model is able to evaluate the structural durability state with only a few electrochemical indicators, and the evaluation results are close to those obtained by combining various electrochemical indicators.

[0132] Example 7:

[0133] Using repaired samples, we established machine learning models with only one of the following as input: potential, corrosion current, polarization resistance, and concrete resistivity, and outputting the structural durability stage. Based on this, we selected 100 samples from existing research for evaluation, using accuracy as the metric. The accuracy was 97% when considering all four electrochemical indicators, 93% when considering only potential, 79% when considering only corrosion current, 80% when considering only polarization resistance, and 73% when considering only concrete resistivity.

[0134] In conclusion, using a combination of four electrochemical indicators is more effective than using a single electrochemical indicator to assess the structural durability stage.

[0135] Example 8:

[0136] Based on the research findings, four electrochemical indicators were used to classify the structural durability. C1, C2, C3, C4, and C5, from low to high, correspond to the structural durability from safe to dangerous states.

[0137]

[0138]

[0139]

[0140]

[0141] After a certain test, the potential detection result was -250mV and the corrosion current density was 0.8μA·cm. -2 Concrete resistivity is 120 kΩ·cm, polarization resistance is 27 kΩ·cm2. (Based on the correlation study between various electrochemical indicators, such results are considered reasonable).

[0142] Evaluating structural durability using various electrochemical indicators separately suggests the structure may be in stages C2, C3, C1, or C2 (or C3). This implies that, under current technological conditions, even if all four indicators are obtained, the final evaluation result cannot be unified. However, using the solution of this invention, inputting the four electrochemical indicators yields the following result:

[0143] The probability of passivation stage is 83.2%.

[0144] The probability of initial rust stage is 16.7%.

[0145] The probability of the corrosion propagation stage is 0.1%.

[0146] This invention demonstrates that it can provide a unified judgment result based on a comprehensive consideration of various electrochemical indicators.

Claims

1. A method for determining a probability of a state of corrosion of an internal steel reinforcement of concrete, characterized in that: The concrete internal steel bar corrosion state probability determination method comprises: The durability of reinforced concrete structure includes three structural durability stages of passivation stage, initial rust stage and corrosion expansion stage; Four electrochemical parameters can be obtained by testing reinforced concrete structures, they are: half-cell potential parameter obtained by two-electrode half-cell potential method E corr , concrete resistivity parameter obtained by four-electrode Wenner method ρ , corrosion current density parameter obtained by Tafel method i corr , and steel polarization resistance parameter obtained by linear polarization method R ρ ; 1) Obtain various electrochemical indexes under the conditions of the three structural durability stages through literature research. In theory, a certain corrosion state should correspond to four electrochemical indexes. In the actual literature research process, it may not be possible to obtain all four electrochemical indexes corresponding to a certain steel bar corrosion state. The obtained electrochemical indexes are marked as certain parameters, and the unobtained electrochemical indexes are marked as missing parameters; 2) then the obtained determination parameters are arranged as follows: two electrochemical indexes under the same steel bar corrosion state constitute a parameter pair, and multiple parameter pairs corresponding to the same two electrochemical indexes constitute a parameter set: the first parameter set corresponds to the parameter pair composed of E corr and i corr , the second parameter set corresponds to the parameter pair composed of E corr and R ρ , the third parameter set corresponds to the parameter pair composed of E corr and ρ , the fourth parameter set corresponds to the parameter pair composed of i corr and ρ , the fifth parameter set corresponds to the parameter pair composed of i corr and R ρ , and the sixth parameter set corresponds to the parameter pair composed of R ρ and ρ ; 3) The EM algorithm is used to process the six parameter sets one by one to establish the joint probability density function between the two corresponding electrochemical indexes; 4) Construct a sample: a single sample is four electrochemical indexes and the corresponding structural durability stage. If there is a missing electrochemical index in a certain sample, the sample is marked as a parameter missing sample. The sample without missing electrochemical index is marked as a complete sample; 5) Repair the parameter missing sample. When repairing a single parameter missing sample, the following steps are taken: A) Duplicate the parameter missing sample into n copies to obtain n replicated samples; B) Fill the certain parameters and the corresponding structural durability stage into the corresponding positions in the replicated samples; C) Take the four electrochemical indexes as four nodes, the node corresponding to the certain parameter as the parent node, and the node corresponding to the missing parameter as the child node. Then, according to the rule that the parent node and the child node are connected and the parent node points to the child node, and the corresponding joint probability density function, a Bayesian network is established. The distribution rule of the electrochemical index corresponding to the child node is calculated by using the Bayesian network inference method; On the basis of obtaining the distribution rule of a certain child node, according to the corresponding distribution rule, randomly sample n times. The n parameters obtained by sampling are the repair parameters corresponding to the missing parameters. Then, fill the n repair parameters into the positions corresponding to the missing parameters in the n replicated samples; The repaired n replicated samples and the complete sample together constitute a sample library; 6) Train the machine learning model according to the sample library. During training, the electrochemical indexes are used as input, and the probability that the reinforced concrete structure durability belongs to the three structural durability stages is used as output; 7) After the machine learning model is trained, it can be put into use. The measured electrochemical indexes are input into the machine learning model, and the machine learning model outputs the probability that the reinforced concrete structure durability belongs to the three structural durability stages.

2. The method of claim 1, wherein: When the measured electrochemical indexes in step 7) are four kinds, the four electrochemical indexes are directly input into the machine learning model, and the machine learning model outputs the probability that the reinforced concrete structure durability belongs to the three structural durability stages; When the measured electrochemical indexes in step 7) are one, two or three, the following processing is performed: (1) Construct a prediction sample containing four electrochemical indexes, wherein the obtained electrochemical indexes are marked as known parameters, and the unobtained electrochemical indexes are marked as unknown parameters; (2) copy the predicted samples as x parts, to obtain x copy-predicted samples; (3) Fill the known parameters into the corresponding positions in the replicated prediction samples; (4) Completing the copy prediction sample: taking four electrochemical indexes as four nodes, the node corresponding to the known parameter as the parent node, and the node corresponding to the unknown parameter as the child node; then establishing the Bayesian network according to the rule that the parent node and the child node are connected and the parent node points to the child node, and the corresponding joint probability density function, and calculating the distribution rule of the electrochemical index corresponding to the child node by using the Bayesian network inference method; On the basis of obtaining the distribution law of a certain sub-node, then according to the corresponding distribution law random sampling x , the obtained x parameters are the corresponding complete parameters of the unknown parameters, and then the x complete parameters are respectively filled into the positions corresponding to the unknown parameters in the x copy prediction samples; (5) the completed x copy prediction samples are input into the machine learning model one by one, and the results output by the machine learning model are averaged to obtain the final result. x copy prediction samples are input into the machine learning model one by one, and the results output by the machine learning model are averaged to obtain the final result.

Images