Bolt tensile damage real-time monitoring method based on acoustic emission technology

Abstract

The invention belongs to the technical field of structure health monitoring and the field of machine learning, and discloses a bolt tensile damage real-time monitoring method based on an acoustic emission technology, which comprises the following steps: carrying out a bolt tensile damage experiment; drawing a bolt tensile damage experiment load loading curve; performing acoustic emission signal data processing; establishing a bolt tensile damage prediction model; and performing bolt tensile damage stage prediction. According to the bolt tensile damage real-time monitoring method, the real-time performance of a monitoring signal is ensured by using an acoustic emission technology, and the damage condition of the bolt can be checked even if large equipment is in a working state; the method comprises the following steps: carrying out multiple tensile experiments on similar bolt materials, extracting experimental parameters in different tensile stages, carrying out grouping, normalization and secondary grouping on the experimental parameters to obtain a mean value, and calculating the mean value, standard deviation, skewness and kurtosis of secondary grouping to obtain a trend of a monitoring signal changing along with time in a period of time; whether the bolt has the risk of breaking and falling off or not is predicted through the trend, and engineering practical value is achieved.

Description

Technical Field

[0001] The present invention belongs to the technical field of structural health monitoring and the field of machine learning, and specifically relates to a real-time monitoring method for bolt tensile damage based on acoustic emission technology. Background Art

[0002] The traditional method of monitoring bolt tensile damage is to regularly inspect the equipment, manually check the damage of the bolts, and replace the damaged bolts. The manual inspection method has the following problems: first, large equipment is usually in a closed environment when working, and it will work continuously for a long time. Maintenance personnel cannot enter the large equipment to check the damage of the bolts. The inspection cycle is long and real-time monitoring is impossible; second, it is difficult to identify early damage of the bolts with the naked eye. If the equipment is in working condition for a long time, there is still a high risk of bolts breaking and falling off in the working condition.

[0003] At present, there is an urgent need to develop a real-time monitoring method for bolt tensile damage based on acoustic emission technology. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a real-time monitoring method for bolt tensile damage based on acoustic emission technology, and to establish a gradient ascent decision tree model to perform real-time non-destructive monitoring of bolts in a working state.

[0005] The real-time monitoring method for bolt tensile damage based on acoustic emission technology of the present invention comprises the following steps: S10. Conduct bolt tensile damage test; Select bolts with the same material and diameter as the bolts to be tested on large equipment, conduct tensile damage experiments, and obtain acoustic emission signal data and tensile data during the bolt tensile process; The acoustic emission signal emitted during the bolt fracture process is collected by an acoustic emission acquisition device, and the acquisition device automatically calculates the amplitude, energy, effective voltage, average level, duration, rise time, rise count, and ring count of the acoustic emission signal of each tensile damage experiment as signal characteristics; S20. Draw the load curve of bolt tensile damage experiment; The loading curve of the bolt tensile damage experiment reflecting the change of tension over time is drawn, and the tensile stage is divided into three stages: early stage, middle stage and late stage; the dividing point between the early stage and the middle stage is the turning point where the slope of the curve decreases, and the dividing point between the middle stage and the late stage is the maximum value of the tension; S30. Performing acoustic emission signal data processing; S31. Obtaining the average value curve of the acoustic emission signal characteristic value of the bolt at different deformation stages; According to the load curve of the bolt tensile damage experiment, the stages are divided, and the collected discrete acoustic emission signals are distinguished by different colors in each stage, and the average value curve of the characteristic value of the acoustic emission signal of the bolt at different deformation stages is drawn; S32. Filtering the acoustic emission signal based on the signal characteristic distribution; The acoustic emission test parameters have the following changing rules in different deformation stages: rise time, rise count, duration and ring count remain stable in the early, middle and late stages; the characteristic values ​​are filtered out by rise time, rise count, duration and ring count to reduce redundant data; The filtering rules for characteristic values ​​are: 1200<rise time<1600; 30<rise count<60; 2500<duration<3000; 60<ring count<120; The filtered acoustic emission signal characteristic value data has the following data characteristics: Early stage: Amplitude, energy, effective voltage, and average level gradually increase over time, with an obvious upper bound but no obvious lower bound; rise count and ring count increase first and then decrease over time, with obvious upper and lower bounds; Mid-term: Amplitude, energy, effective voltage, and average level increase slowly over time, and the degree of dispersion gradually increases over time; rising counts and ringing counts increase slowly over time, and the degree of dispersion does not change significantly; Late stage: Amplitude, energy, effective voltage, and average level decrease with time, and the degree of dispersion continues to increase; there is no obvious change in rising count compared with the mid-stage; and ringing count decreases with time; S33. Perform secondary grouping on the acoustic emission signal characteristic values ​​to obtain the mean value, standard deviation, skewness and kurtosis characteristics of each group; In order to describe the changing trend and discreteness of the eigenvalues, the filtered acoustic emission signal data are grouped once. The rule for the first grouping is 600 acoustic emission signals per group. The next group of signals consists of 550 acoustic emission signals from the previous group and the next 50 acoustic emission signals. After the acoustic emission signals of each group are normalized, they are grouped again. The secondary grouping is divided into 30 groups. The mean of each group is calculated, and a new group is formed by the mean of each group. The mean, standard deviation, peak value and skewness of the new group are calculated. Normalized calculation formula: ; The purpose of normalization is to remove the numerical differences caused by grouping and focus on the change trend and distribution of the acoustic emission signals in each group; is the experimental parameter within the group; is the minimum value in the group; is the maximum value in the group; is the normalized experimental parameter within the group; Mean calculation formula: ; After normalization, the mean is calculated to show the overall tendency of the new group of acoustic emission signal data. The closer the mean is to 1, the higher the acoustic emission signal data of this group tends to be, and the closer the mean is to 0, the lower the acoustic emission signal data of this group tends to be. is the mean of the characteristics within the group; are the eigenvalues ​​within the group; is the total amount of eigenvalues ​​within the group; Skewness calculation formula: ; in, is the within-group characteristic standard deviation, is the characteristic skewness within the group; if the skewness is greater than 0, the data distribution of the new group of acoustic emission signals is right-skewed, that is, positively skewed, and the extreme values ​​on the left end are more than the extreme values ​​on the right end; if the skewness is less than 0, the data distribution is left-skewed, that is, negatively skewed, and the extreme values ​​on the right end are more than the extreme values ​​on the left end; if the skewness is close to 0, the data distribution of the new group of acoustic emission signals is close to symmetric; Kurtosis calculation formula: ; in, is the characteristic kurtosis within the group; kurtosis is a statistic used to describe the sharpness or flatness of the acoustic emission signal data distribution within the group; if the kurtosis is greater than 3, the new group acoustic emission signal data distribution is sharper than the normal distribution; if the kurtosis is less than 3, the new group acoustic emission signal data distribution is flatter than the normal distribution; S40. Establish a bolt tensile damage prediction model; By extracting the time domain features of the acoustic emission signal, the prediction of the bolt damage stage is converted into a multi-classification problem. For the multi-classification problem, multiple decision trees are trained for classification prediction. The 24 intra-group eigenvalues ​​calculated after the secondary grouping are used as eigenvalues, and the different stages distinguished by tension are used as categories. The information gain formula is used as the basic formula for decision tree fitting to train the decision tree. Information gain formula: ; in, The experimental parameters For the dataset The information gain value is larger, the experimental parameters The better the classification effect; is the training data set; Indicates experimental parameters A subset of the data; Represents the total number of samples in the training data set; The number of samples representing the data subset; is the information entropy calculation formula. The higher the information entropy, the greater the degree of confusion of the data set. Information entropy calculation formula: ; in, is a subset of data, is a category in the subset, For Category Probability of occurrence; Each time a decision tree is trained, it is added to the prediction model, and the residual between the predicted value and the fitting target is calculated using the cross entropy loss function, and the residual is used as the fitting target for the next decision tree. Cross entropy loss function: ; Among them, L is the cross entropy loss value of the training set. The larger the L value, the greater the difference between the true distribution and the predicted distribution. is the sample size; is the number of categories; is a symbolic function, taking the value 0 or 1. If the true category of the current sample i is equal to the category , Take 1, otherwise take 0; The current sample i belongs to the category The predicted probability of S50. Predict the bolt tensile damage stage; Use the prediction model to make stage predictions, and issue an alarm when the prediction result is mid-term or late.

[0006] The real-time monitoring method for bolt tensile damage based on acoustic emission technology of the present invention uses acoustic emission technology to ensure the real-time nature of the monitoring signal, and can check the damage of the bolt even when large equipment is in working condition; multiple tensile tests are carried out on the same bolt material, and the monitoring signal characteristics at different tensile stages are extracted. The monitoring signal is grouped, normalized, and averaged through secondary grouping, and the mean, standard deviation, skewness and kurtosis of the secondary grouping are calculated to obtain the trend of the monitoring signal changing with time within a period of time, and predict whether the bolt has the risk of breaking and falling off through the trend.

[0007] In short, the real-time monitoring method for bolt tensile damage based on acoustic emission technology of the present invention obtains the acoustic emission signal characteristics of the bolt to be monitored in the entire tensile process through a tensile experiment, and writes an online recognition algorithm according to the acoustic emission characteristics of the bolt in different states during the tensile process, so as to perform real-time non-destructive monitoring of the bolt in the working state and predict whether the bolt has the risk of breaking and falling off. It has strong anti-interference ability, low probability of false alarm, and has practical engineering value. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 It is a flow chart of the real-time monitoring method of bolt tensile damage based on acoustic emission technology of the present invention; Figure 2 This is the loading curve of the bolt tensile damage experiment; Figure 3a is the average value curve (amplitude) of the characteristic value of the acoustic emission signal of the bolt at different deformation stages; Figure 3b is the average value curve (energy) of the acoustic emission signal characteristic value of the bolt at different deformation stages; Figure 3c It is the average value curve (effective voltage) of the acoustic emission signal characteristic value of the bolt at different deformation stages; Figure 3d It is the average value curve (average level) of the characteristic value of the acoustic emission signal of the bolt at different deformation stages; Figure 3e It is the average value curve (duration) of the characteristic value of the acoustic emission signal of the bolt at different deformation stages; Figure 3f It is the average value curve (rise time) of the characteristic value of the acoustic emission signal of the bolt at different deformation stages; Figure 3g It is the average value curve (rising count) of the characteristic value of the acoustic emission signal of the bolt at different deformation stages; Figure 3h It is the average value curve (ring count) of the acoustic emission signal characteristic value of the bolt at different deformation stages.

[0009] In the figure, blue dots represent the early stage, green dots represent the middle stage, and gray dots represent the late stage. DETAILED DESCRIPTION

[0010] The present invention is described in detail below with reference to the accompanying drawings and embodiments.

[0011] Example: Figure 1 As shown, the real-time monitoring method for bolt tensile damage based on acoustic emission technology of this embodiment includes the following steps: S10. Conduct bolt tensile damage test; Select bolts with the same material and diameter as the bolts to be tested on large equipment, conduct tensile damage experiments, and obtain acoustic emission signal data and tensile data during the bolt tensile process; The acoustic emission signal emitted during the bolt fracture process is collected by an acoustic emission acquisition device, and the acquisition device automatically calculates the amplitude, energy, effective voltage, average level, duration, rise time, rise count, and ring count of the acoustic emission signal of each tensile damage experiment as signal characteristics; Among them, the amplitude reflects the strength of the energy released by the acoustic emission source; the energy reflects the amount of energy released by the acoustic emission source, which is used to indicate the degree of internal damage or structural change of the material; the effective voltage, that is, the root mean square value of the acoustic emission signal level, reflects the average energy of the acoustic emission signal; the average level, that is, the mean of the acoustic emission signal level, reflects the average energy level of the acoustic emission signal; the duration, that is, the physical time length of the acoustic emission signal; the rise time, reflects the speed of change of the acoustic emission signal; the rise count, reflects the volatility of the acoustic emission signal during the rise process; the ringing count, reflects the frequency characteristics of the acoustic emission signal; S20. Draw the load curve of bolt tensile damage experiment; Draw as Figure 2 The load curve of the bolt tensile damage experiment reflecting the change of tension with time is shown, and the tensile stage is divided into three stages: early stage, middle stage and late stage; the dividing point between the early stage and the middle stage is the turning point where the slope of the curve decreases, and the dividing point between the middle stage and the late stage is the maximum value of the tension; The three stages are shown on the pressure curve as follows: the early tensile force curve is close to the rising slope, which represents the plastic deformation stage in the tensile deformation stage of the bolt metal material; the mid-term tensile force curve is also close to the rising slope, but the rising slope decreases and gradually approaches 0, which represents the non-plastic deformation stage in the tensile deformation of the bolt metal material; the late tensile force curve is close to the descending slope, and the descending slope gradually changes from 0 to negative, which represents the destruction stage in the tensile deformation of the metal; S30. Performing acoustic emission signal data processing; S31. Obtaining the average value curve of the acoustic emission signal characteristic value of the bolt at different deformation stages; According to the loading curve of the bolt tensile damage experiment, the discrete acoustic emission signals collected are distinguished by different colors in each stage, and the following diagram is drawn: Figure 3a~Figure 3h The average value curve of the acoustic emission signal characteristic value of the bolt at different deformation stages is shown; S32. Filtering the acoustic emission signal based on the signal characteristic distribution; The acoustic emission test parameters have the following changing rules in different deformation stages: rise time, rise count, duration and ring count remain stable in the early, middle and late stages; the characteristic values ​​are filtered out by rise time, rise count, duration and ring count to reduce redundant data; The filtering rules for characteristic values ​​are: 1200<rise time<1600; 30<rise count<60; 2500<duration<3000; 60<ring count<120; The filtered acoustic emission signal characteristic value data has the following data characteristics: Early stage: Amplitude, energy, effective voltage, and average level gradually increase over time, with an obvious upper bound but no obvious lower bound; rise count and ring count increase first and then decrease over time, with obvious upper and lower bounds; Mid-term: Amplitude, energy, effective voltage, and average level increase slowly over time, and the degree of dispersion gradually increases over time; rising counts and ringing counts increase slowly over time, and the degree of dispersion does not change significantly; Late stage: Amplitude, energy, effective voltage, and average level decrease with time, and the degree of dispersion continues to increase; there is no obvious change in rising count compared with the mid-stage; and ringing count decreases with time; S33. Perform secondary grouping on the acoustic emission signal characteristic values ​​to obtain the mean value, standard deviation, skewness and kurtosis characteristics of each group; In order to describe the changing trend and discreteness of the eigenvalues, the filtered acoustic emission signal data are grouped once. The rule for the first grouping is 600 acoustic emission signals per group. The next group of signals consists of 550 acoustic emission signals from the previous group and the next 50 acoustic emission signals. After the acoustic emission signals of each group are normalized, they are grouped again. The secondary grouping is divided into 30 groups. The mean of each group is calculated, and a new group is formed by the mean of each group. The mean, standard deviation, peak value and skewness of the new group are calculated. Normalized calculation formula: ; The purpose of normalization is to remove the numerical differences caused by grouping and focus on the change trend and distribution of the acoustic emission signals in each group; is the experimental parameter within the group; is the minimum value in the group; is the maximum value in the group; is the normalized experimental parameter within the group; Mean calculation formula: ; After normalization, the mean is calculated to show the overall tendency of the new group of acoustic emission signal data. The closer the mean is to 1, the higher the acoustic emission signal data of this group tends to be, and the closer the mean is to 0, the lower the acoustic emission signal data of this group tends to be. is the mean of the characteristics within the group; are the eigenvalues ​​within the group; is the total amount of eigenvalues ​​within the group; Skewness calculation formula: ; in, is the within-group characteristic standard deviation, is the characteristic skewness within the group; if the skewness is greater than 0, the data distribution of the new group of acoustic emission signals is right-skewed, that is, positively skewed, and the extreme values ​​on the left end are more than the extreme values ​​on the right end; if the skewness is less than 0, the data distribution is left-skewed, that is, negatively skewed, and the extreme values ​​on the right end are more than the extreme values ​​on the left end; if the skewness is close to 0, the data distribution of the new group of acoustic emission signals is close to symmetric; Kurtosis calculation formula: ; in, is the characteristic kurtosis within the group; kurtosis is a statistic used to describe the sharpness or flatness of the acoustic emission signal data distribution within the group; if the kurtosis is greater than 3, the new group acoustic emission signal data distribution is sharper than the normal distribution; if the kurtosis is less than 3, the new group acoustic emission signal data distribution is flatter than the normal distribution; S40. Establish a bolt tensile damage prediction model; By extracting the time domain features of the acoustic emission signal, the prediction of the bolt damage stage is converted into a multi-classification problem. In order to achieve better prediction results, multiple decision trees are trained for classification prediction for multi-classification problems. The 24 intra-group eigenvalues ​​calculated after secondary grouping are used as eigenvalues, and the different stages distinguished by tension are used as categories. The information gain formula is used as the basic formula for decision tree fitting to train the decision tree. Information gain formula: ; in, The experimental parameters For the dataset The information gain value is larger, the experimental parameters The better the classification effect; is the training data set; Indicates experimental parameters A subset of the data; Represents the total number of samples in the training data set; The number of samples representing the data subset; is the information entropy calculation formula. The higher the information entropy, the greater the degree of confusion of the data set. Information entropy calculation formula: ; in, is a subset of data, is a category in the subset, For Category Probability of occurrence; Each time a decision tree is trained, it is added to the prediction model, and the residual between the predicted value and the fitting target is calculated using the cross entropy loss function, and the residual is used as the fitting target for the next decision tree. Cross entropy loss function: ; Among them, L is the cross entropy loss value of the training set. The larger the L value, the greater the difference between the true distribution and the predicted distribution. is the sample size; is the number of categories; is a symbolic function, taking the value 0 or 1. If the true category of the current sample i is equal to the category , Take 1, otherwise take 0; The current sample i belongs to the category The predicted probability of Adjust the maximum depth of the decision tree. A larger depth generally leads to better prediction results, but may cause the prediction model to be too complex and have overfitting problems. Adjust the minimum number of samples of child nodes to help control the complexity of the tree and prevent overfitting. Divide the maximum number of samples of child nodes to help ensure that each leaf node has enough samples to improve the stability of the prediction model. Adjust the number of decision trees. Increasing the number of decision trees generally leads to better prediction results, but may lead to overfitting. Adjust the update weights of the decision trees. Small weights will reduce the update speed of the prediction model, but will make the prediction model more stable, but the training time will also increase. Evaluate the prediction model. If overfitting occurs, it is necessary to reduce some parameters and train the model again based on existing expert experience. Compare the evaluation results of each prediction model and select the prediction model that meets the requirements; S50. Predict the bolt tensile damage stage; Use the prediction model to make stage predictions, and issue an alarm when the prediction result is mid-term or late.

[0012] Although the embodiments of the present invention have been disclosed as above, they are not limited to the applications listed in the specification and implementation modes. For those familiar with the art, all features disclosed in the present invention, or steps in all methods or processes disclosed, except for mutually exclusive features and / or steps, can be combined in any way without departing from the principles of the present invention. The present invention is not limited to the specific details and illustrations shown and described herein.

Claims

1. A real-time monitoring method for bolt tensile damage based on acoustic emission technology, characterized in that: The following steps are involved: S10. Conduct bolt tensile damage test; S20. Draw the load curve of bolt tensile damage experiment; S30. Performing acoustic emission signal data processing; S40. Establish a bolt tensile damage prediction model; S50. Predict the tensile damage stage of bolts.

2. The real-time monitoring method for bolt tensile damage based on acoustic emission technology according to claim 1 is characterized in that: The S10 is used for bolt tensile damage test. The process includes: Select bolts with the same material and diameter as the bolts to be tested on large equipment, conduct tensile damage experiments, and obtain acoustic emission signal data and tensile data during the bolt tensile process; The acoustic emission signals emitted during the bolt fracture process are collected by the acoustic emission acquisition equipment. The acquisition equipment automatically calculates the amplitude, energy, effective voltage, average level, duration, rise time, rise count, and ringing count of the acoustic emission signals of each tensile damage experiment as signal characteristics.

3. The real-time monitoring method for bolt tensile damage based on acoustic emission technology according to claim 2 is characterized in that: The S20 drawing bolt tensile damage test load curve, The process includes: A bolt tensile damage test load curve reflecting the change of tension over time is drawn, and the tensile stage is divided into three stages: early, middle and late stages; the dividing point between the early and middle stages is the turning point where the slope of the curve decreases, and the dividing point between the middle and late stages is the maximum tension.

4. The real-time monitoring method for bolt tensile damage based on acoustic emission technology according to claim 3 is characterized in that: The acoustic emission signal data processing in S30 includes the following steps: S31. Obtaining the average value curve of the acoustic emission signal characteristic value of the bolt at different deformation stages; According to the load curve of the bolt tensile damage experiment, the stages are divided, and the collected discrete acoustic emission signals are distinguished by different colors in each stage, and the average value curve of the characteristic value of the acoustic emission signal of the bolt at different deformation stages is drawn; S32. Filtering the acoustic emission signal based on the signal characteristic distribution; The acoustic emission test parameters have the following changing rules in different deformation stages: rise time, rise count, duration and ring count remain stable in the early, middle and late stages; the characteristic values ​​are filtered out by rise time, rise count, duration and ring count to reduce redundant data; The filtering rules for characteristic values ​​are: 1200<rise time<1600; 30<rise count<60; 2500<duration<3000; 60<ring count<120; The filtered acoustic emission signal characteristic value data has the following data characteristics: Early stage: Amplitude, energy, effective voltage, and average level gradually increase over time, with an obvious upper bound but no obvious lower bound; rise count and ring count increase first and then decrease over time, with obvious upper and lower bounds; Mid-term: Amplitude, energy, effective voltage, and average level increase slowly over time, and the degree of dispersion gradually increases over time; rising counts and ringing counts increase slowly over time, and the degree of dispersion does not change significantly; Late stage: Amplitude, energy, effective voltage, and average level decrease with time, and the degree of dispersion continues to increase; there is no obvious change in rising count compared with the mid-stage; and ringing count decreases with time; S33. Perform secondary grouping on the acoustic emission signal characteristic values ​​to obtain the mean value, standard deviation, skewness and kurtosis characteristics of each group; In order to describe the changing trend and discreteness of the eigenvalues, the filtered acoustic emission signal data are grouped once. The rule for the first grouping is 600 acoustic emission signals per group. The next group of signals consists of 550 acoustic emission signals from the previous group and the next 50 acoustic emission signals. After the acoustic emission signals of each group are normalized, they are grouped again. The secondary grouping is divided into 30 groups. The mean of each group is calculated, and a new group is formed by the mean of each group. The mean, standard deviation, peak value and skewness of the new group are calculated. Normalized calculation formula: ； The purpose of normalization is to remove the numerical differences caused by grouping and focus on the change trend and distribution of the acoustic emission signals in each group; is the experimental parameter within the group; is the minimum value in the group; is the maximum value in the group; is the normalized experimental parameter within the group; Mean calculation formula: ； After normalization, the mean is calculated to show the overall tendency of the new group of acoustic emission signal data. The closer the mean is to 1, the higher the acoustic emission signal data of this group tends to be, and the closer the mean is to 0, the lower the acoustic emission signal data of this group tends to be. is the mean of the characteristics within the group; are the eigenvalues ​​within the group; is the total amount of eigenvalues ​​within the group; Skewness calculation formula: ； in, is the within-group characteristic standard deviation, is the characteristic skewness within the group; if the skewness is greater than 0, the data distribution of the new group of acoustic emission signals is right-skewed, that is, positively skewed, and the extreme values ​​on the left end are more than the extreme values ​​on the right end; if the skewness is less than 0, the data distribution is left-skewed, that is, negatively skewed, and the extreme values ​​on the right end are more than the extreme values ​​on the left end; if the skewness is close to 0, the data distribution of the new group of acoustic emission signals is close to symmetric; Kurtosis calculation formula: ； in, is the characteristic kurtosis within the group; kurtosis is a statistic used to describe the sharpness or flatness of the acoustic emission signal data distribution within the group; if the kurtosis is greater than 3, the new group acoustic emission signal data distribution is sharper than the normal distribution; if the kurtosis is less than 3, the new group acoustic emission signal data distribution is flatter than the normal distribution.

5. The real-time monitoring method for bolt tensile damage based on acoustic emission technology according to claim 4 is characterized in that: The S40 bolt tensile damage prediction model includes the following steps: By extracting the time domain features of acoustic emission signals, the prediction of bolt damage stages is converted into a multi-classification problem. For multi-classification problems, multiple decision trees are trained for classification prediction; The 24 eigenvalues ​​calculated after the secondary grouping are used as eigenvalues, and the different stages distinguished by the pulling force are used as categories. The decision tree is trained based on the information gain formula as the basic formula for decision tree fitting. Information gain formula: ； in, The experimental parameters For the dataset The information gain value is larger, the experimental parameters The better the classification effect; is the training data set; Indicates experimental parameters A subset of the data; Represents the total number of samples in the training data set; The number of samples representing the data subset; is the information entropy calculation formula. The higher the information entropy, the greater the degree of confusion of the data set. Information entropy calculation formula: ； in, is a subset of data, is a category in the subset, For Category Probability of occurrence; Each time a decision tree is trained, it is added to the prediction model, and the residual between the predicted value and the fitting target is calculated using the cross entropy loss function, and the residual is used as the fitting target for the next decision tree. Cross entropy loss function: ； Among them, L is the cross entropy loss value of the training set. The larger the L value, the greater the difference between the true distribution and the predicted distribution. is the sample size; is the number of categories; is a symbolic function, taking the value 0 or 1. If the true category of the current sample i is equal to the category , Take 1, otherwise take 0; The current sample i belongs to the category The predicted probability of Compare the evaluation results of each prediction model and select the prediction model that meets the requirements.

6. The real-time monitoring method for bolt tension damage based on acoustic emission technology according to claim 5 is characterized in that: The S50 is used to predict the tensile damage stage of bolts. The process includes: Use the prediction model to make stage predictions, and issue an alarm when the prediction result is mid-term or late.

Images