CN116522797A - A Method of Predicting the Failure Mode of RC Rectangular Columns Based on Artificial Neural Network - Google Patents

Abstract

The invention relates to the field of artificial neural networks, in particular to a method for predicting an RC rectangular column failure mode based on an artificial neural network. The method comprises the following steps: based on the prediction method of the artificial neural network, the failure mode of the rectangular column member is judged by combining Levenberg Marquardt back propagation training algorithm; firstly, setting an artificial neural network structure of a reinforced concrete rectangular column member in a damage mode; processing a data set of the rectangular column member, and performing sensitivity analysis on the hoop distribution rate, the longitudinal reinforcement rate, the effective height ratio of the stirrup spacing to the section, the shear span ratio, the axial compression ratio and the bending shear ratio in the damage mode of the rectangular column member to obtain a prediction result of the damage mode of the rectangular column member; and carrying out statistical analysis on the prediction result of the damage mode of the rectangular column member. The prediction accuracy in the aspect of judging the damage mode of the column member is improved by the method based on the artificial neural network.

Description

A Method of Predicting the Failure Mode of RC Rectangular Columns Based on Artificial Neural Network

technical field

The invention relates to the field of artificial neural networks, in particular to a method for predicting the failure mode of RC rectangular columns based on artificial neural networks.

Background technique

According to post-earthquake surveys and a large number of seismic tests of reinforced concrete columns, it can be found that different reinforcement characteristics, section geometric characteristics, and stress conditions will lead to different failure modes of column members. General engineering divides the failure modes into bending failure, bending-shear failure and shear failure. The failure mode of reinforced concrete rectangular column members has a direct impact on its ductility and seismic performance, so it is very important to predict the failure mode of reinforced concrete rectangular column members.

At present, the methods for predicting the failure mode of reinforced concrete column members mainly include: prediction methods based on shear-span ratio, prediction methods based on discriminant analysis, and prediction methods based on engineering simplified discriminant analysis. Among them, the prediction method based on shear-span ratio only considers the parameter of shear-span ratio; the prediction method based on engineering simplified discriminant analysis considers the two parameters of shear-span ratio and bending-shear ratio; and the prediction method based on discriminant analysis takes into account The three parameters of stirrup spacing and section effective height ratio, shear-span ratio and bending-shear ratio. The common methods for predicting the failure mode of reinforced concrete column members do not consider the influence of hoop ratio, axial compression ratio, and longitudinal reinforcement ratio, which will affect the accuracy of the prediction results of the failure mode of reinforced concrete column members, so The accuracy of prediction results of failure modes of reinforced concrete column members is low.

Contents of the invention

In order to solve the problem of low accuracy of the prediction results of the failure mode of reinforced concrete column members in the existing methods, the purpose of the present invention is to provide a method for predicting the failure mode of RC rectangular columns based on artificial neural network, and the adopted technical scheme details as follows:

The present invention provides a kind of method based on artificial neural network prediction RC rectangular column damage pattern, and this method comprises the following steps:

The prediction method based on the artificial neural network, combined with the Levenberg Marquardt backpropagation training algorithm to identify the failure mode of the rectangular column members;

Among them, the method of judging the failure mode of rectangular column members is as follows: setting the artificial neural network structure of the failure mode of reinforced concrete rectangular column members; Sensitivity analysis of reinforcement ratio, ratio of stirrup spacing to section effective height, shear-span ratio, axial compression ratio, and bending-shear ratio to obtain the prediction results of failure mode of rectangular column members; statistical analysis of the prediction results of failure mode of rectangular column members .

Preferably, the parameters of the artificial neural network structure of the failure mode of the reinforced concrete rectangular column member include: the number of input layer nodes, the number of hidden layer nodes, the number of output layer nodes, the activation function of the input layer, the activation function of the hidden layer and the activation function of the output layer.

Preferably, the input layer has 6 nodes, and the 6 nodes respectively represent the stirrup ratio, the longitudinal reinforcement ratio, the ratio of the stirrup spacing to the effective height of the section, the shear-span ratio, the axial compression ratio, and the bending-shear ratio.

Preferably, the expression of the input layer activation function is:

y=f(Wp+b)

Among them, y is the output vector of the activation function; W is the weight matrix; p is the initial input vector; b is the bias vector.

Preferably, the hidden layer includes two layers, the first layer has 100 nodes, and the second layer has 10 nodes; wherein, the activation function of the hidden layer is a sigmoid function.

Preferably, the number of nodes in the output layer is 3, and each node corresponds to a damage mode; the activation function of the output layer is a softmax function.

Preferably, the expression of the Levenberg Marquardt backpropagation training algorithm is:

△x＝-[J ^T J+μI] ^-1 J ^T ε

Among them, J is the Jacobian matrix; J ^T is the transpose matrix of the Jacobian matrix J; μ is a normal number; I is the identity matrix; ε is the network error.

Preferably, the bending-shear ratio and the shear-span ratio are related to the failure mode of the reinforced concrete rectangular column member.

Preferably, the accuracy rate of the training data of the trained artificial neural network reaches 100%, and the accuracy rate of the test data reaches 95.3%, wherein the minimum mean square error of the training data is 0.0018, and the minimum mean square error of the test data is 0.038.

The invention has the following beneficial effects: compared with the traditional prediction method, the method based on the artificial neural network of the invention performs better than the traditional discrimination prediction method in the aspect of discriminating the damage mode of the column member. With the continuous increase of column component data sets, the accuracy of artificial neural network models will be significantly improved in the future. Along with the integration and development of civil engineering and artificial neural network, it is foreseeable that not only the failure mode prediction will use artificial neural network, but even the deformation limit of components and the constitutive of concrete materials can also be directly based on artificial neural network (Artificial Neural Network). , ANN) combined with a large number of experimental data sets, the original experimental data will be rejuvenated with the rise of artificial neural networks.

Description of drawings

In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or in the prior art, the following will briefly introduce the drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description The drawings are only some embodiments of the present invention, and those skilled in the art can also obtain other drawings based on these drawings without creative work.

FIG. 1 is a schematic diagram of the ANN structure provided by an embodiment of the present invention;

Fig. 2 is a box-line diagram of the influence of the hoop ratio on the failure mode provided by an embodiment of the present invention;

Fig. 3 is the box diagram of the effect of the longitudinal reinforcement ratio on the failure mode provided by one embodiment of the present invention;

Fig. 4 is the box diagram of the effect of the ratio of the stirrup spacing and the section effective height on the failure mode provided by an embodiment of the present invention;

Fig. 5 is the box plot of the influence of shear-span ratio on failure mode provided by one embodiment of the present invention;

Fig. 6 is a box plot of the influence of the axial pressure ratio on the failure mode provided by an embodiment of the present invention;

Fig. 7 is the box diagram of the influence of the bending-shear ratio on the failure mode provided by one embodiment of the present invention;

Fig. 8 is the Pearson correlation coefficient matrix provided by one embodiment of the present invention;

Fig. 9 is a schematic diagram of data prediction accuracy provided by an embodiment of the present invention;

FIG. 10 is a schematic diagram of the minimum mean square error provided by an embodiment of the present invention;

FIG. 11 is a schematic diagram of a confusion matrix provided by an embodiment of the present invention;

Fig. 12 is a schematic diagram of failure mode prediction based on bending-shear ratio and shear-span ratio provided by an embodiment of the present invention;

Fig. 13 is a schematic diagram of a method for predicting the failure mode of an RC rectangular column based on an artificial neural network provided by an embodiment of the present invention.

Detailed ways

In order to further explain the technical means and effects of the present invention to achieve the intended purpose of the invention, the method for predicting the damage mode of RC rectangular columns based on artificial neural network proposed in the present invention will be carried out below in conjunction with the accompanying drawings and preferred embodiments. The details are as follows.

Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the technical field of the invention.

The embodiment of the present invention provides a specific implementation method of a method for predicting the failure mode of an RC rectangular column based on an artificial neural network, and the method is applicable to a scenario of predicting a failure mode of a reinforced concrete rectangular column member. In order to solve the problem of low accuracy of the prediction results of the failure mode of reinforced concrete column members in the existing methods. The invention uses a prediction method based on an artificial neural network and combines a Levenberg Marquardt backpropagation training algorithm to discriminate the failure mode of a rectangular column member.

A specific scheme of a method for predicting the failure mode of an RC rectangular column based on an artificial neural network provided by the present invention will be described below in conjunction with the accompanying drawings.

Please refer to Fig. 13, which shows that a method for predicting the damage mode of RC rectangular columns based on artificial neural network in this embodiment includes the following contents:

The prediction method based on artificial neural network is combined with the Levenberg Marquardt backpropagation training algorithm to identify the failure mode of rectangular column members. It should be noted that the Levenberg Marquardt backpropagation training algorithm is the Marquardt method, which is a well-known technology for those skilled in the art, and will not be repeated here.

Among them, the method of judging the failure mode of rectangular column members is as follows: setting the artificial neural network structure of the failure mode of reinforced concrete rectangular column members; Sensitivity analysis of reinforcement ratio, ratio of stirrup spacing to section effective height, shear-span ratio, axial compression ratio, and bending-shear ratio to obtain the prediction results of failure mode of rectangular column members; statistical analysis of the prediction results of failure mode of rectangular column members .

The parameters of the artificial neural network structure of the failure mode of reinforced concrete rectangular column members include: the number of input layer nodes, the number of hidden layer nodes, the number of output layer nodes, the activation function of the input layer, the activation function of the hidden layer and the activation function of the output layer;

Among them, there are 6 nodes in the input layer, and the 6 nodes represent the stirrup ratio, longitudinal reinforcement ratio, ratio of stirrup spacing to section effective height, shear-span ratio, axial compression ratio, and bending-shear ratio;

Among them, the expression of the input layer activation function is:

y=f(Wp+b)

Among them, y is the output vector of the activation function; W is the weight matrix; p is the initial input vector; b is the bias vector.

The hidden layer includes two layers, the first layer has 100 nodes, and the second layer has 10 nodes; wherein, the activation function of the hidden layer is a sigmoid function.

The expression of the sigmoid function is:

Among them, f(x) is the output value of the sigmoid function; x is the input variable; e is a natural constant.

The number of nodes in the output layer is 3, and each node corresponds to a damage mode; the activation function of the output layer is the softmax function.

The expression of the softmax function is:

Among them, softmax(x) is the output value of the softmax function; e is a natural constant; x _i is the i-th input variable; x _j is the j-th input variable; n is the number of input variables.

Please refer to FIG. 1 . FIG. 1 is a schematic diagram of an artificial neural network structure, that is, a schematic diagram of an ANN structure.

Among them, the expression of the Levenberg Marquardt backpropagation training algorithm is:

△x＝-[J ^T J+μI] ^-1 J ^T ε

Among them, J is the Jacobian matrix; J ^T is the transpose matrix of the Jacobian matrix J; μ is a normal number; I is the identity matrix; ε is the network error.

When μ is 0, the Levenberg-Marquardt backpropagation training algorithm is Newton's method; when the value of μ is larger, the Levenberg-Marquardt backpropagation training algorithm is close to the gradient descent method.

After testing the artificial neural network model, it was found that the model trained without normalization or normalization of the data produced better results. It may be that the input parameter scales are relatively close, and for a failure mode, the amount of data is relatively scarce. Therefore, some sensitivity to this corruption mode may be lost if the dataset is normalized.

Please refer to Figures 2 to 7. Figure 2 is the box-line diagram of the effect of the ratio of stirrups on the failure mode; Figure 3 is the box-line diagram of the effect of the ratio of longitudinal reinforcement on the failure mode; Figure 4 is the ratio of the spacing of the stirrups to the effective height of the section The boxplot of the influence on the failure mode; Figure 5 is the boxplot of the influence of the shear-span ratio on the failure mode; Figure 6 is the boxplot of the influence of the axial compression ratio on the failure mode; Figure 7 is the influence of the bending-shear ratio on the failure mode Box plots, Figures 2 to 7 are the qualitative analysis of the failure mode by the six input parameters of stirrup ratio, longitudinal reinforcement ratio, ratio of stirrup spacing to section effective height, shear-span ratio, axial compression ratio, and bending-shear ratio Influence. The horizontal axes of Figures 2 to 7 are failure modes, where 1 in the horizontal axes of Figures 2 to 7 represents the bending failure mode, 2 in the horizontal axis represents the bending-shear failure mode, and 3 in the horizontal axis represents the failure mode. Represents the shear failure mode; the vertical axis in Figure 2 is the stirrup ratio; the vertical axis in Figure 3 is the longitudinal reinforcement ratio; the vertical axis in Figure 4 is the ratio of the stirrup spacing to the section effective height; the vertical axis in Figure 5 is Shear-span ratio; the vertical axis in Figure 6 is the axial compression ratio; the vertical axis in Figure 7 is the bending-shear ratio.

Please refer to Figure 8, Figure 8 is the Pearson correlation coefficient matrix, and Figure 8 reflects the impact of each parameter on the failure mode in the quantitative analysis of the Pearson correlation coefficient matrix; in Figure 8, ρ" is the stirrup ratio; ρ _l is the longitudinal reinforcement ratio; s/d is the ratio of the stirrup spacing to the effective height of the section; a/d is the shear-span ratio; P/A _g *f _c ' is the axial compression ratio; V _p /V _n is the bending-shear ratio. From the data at the bottom of the figure, it can be seen that the bending-shear ratio (V _p /V _n ) and the shear-span ratio (a/d) are relatively related to the failure mode of reinforced concrete rectangular column members. The bending-shear ratio (V _p /V _n ) and shear-span ratio (a/d), stirrup ratio (ρ"), longitudinal reinforcement ratio (ρ _l ), axial compression ratio (P/A _g *f _c '), and stirrup spacing and section effective The height ratio (s/d) has no significant effect on the failure mode.

Among the 111 sets of data in Table 1, 89 sets are used as model training data, and the remaining 22 sets are used as test data. After the model is optimized, the accuracy of the training data and test data reaches 100% and 95.3% respectively, which means that the accuracy of the training data of the trained artificial neural network reaches 100%, and the accuracy of the test data reaches 95.3%. The minimum mean square error (MSE) of the training data is 0.0018, and the minimum mean square error (MSE) of the test data is 0.038. Please refer to Figure 9 and Figure 10, Figure 9 is a schematic diagram of data prediction accuracy, Figure 10 is a schematic diagram of the minimum mean square error, where the horizontal axis of Figure 9 and Figure 10 is the number of iterations.

As shown in Table 1, Table 1 shows the parameter data and failure mode of the rectangular column.

Table 1

Among them, FM represents the random variable earthquake damage mode.

Please refer to Figure 11. Figure 11 is a schematic diagram of the confusion matrix. According to the confusion matrix in Figure 11, among the 22 prediction results, there is 1 prediction error, and 1 "bending-shear failure" mode is wrongly identified For "shear failure" mode.

The prediction results are summarized in Table 2, and the details of Table 2 are as follows:

(1) Among the 13 sets of data where bending failure occurred, all the data were correctly predicted as bending failure.

(2) Among the 7 sets of data in which bending-shear failure occurred, 6 sets of data were correctly predicted as bending-shear failure, and 1 set of data was wrongly predicted as shear failure.

(3) In the 2 sets of data where shear failure occurred, all the data were correctly predicted as shear failure.

Table 2

The statistical analysis of the predicted results is listed in Table 3.

table 3

From the statistical analysis of the prediction results, it was found that 95.3% (95.3%=(100%+86%+100%)/3) of the column members were correctly predicted.

In the shear-span ratio-based method, the failure mode of the column member is only related to the shear-span ratio. When the shear-span ratio λ=a/d≤2, this failure mode is classified as shear failure; when the shear-span ratio λ=a/d≥4, this failure mode is classified as bending failure; when 2<a/d When d<4, this failure mode is classified as bending-shear failure. Among them, a represents the span length of the shear span, and d represents the effective height of the section.

When the shear-span ratio is λ=a/d≤2, 63% of the column members undergo shear failure; when the shear-span ratio is λ=a/d≥4", 51% of the column members undergo bending failure; when the shear-span ratio When the span ratio is 2<a/d<4, 92% of the column members suffer bending-shear failure.

The results showed that 68.7% (68.7%=(63%+51%+92%)/3) of the column members were correctly predicted. This also shows that the shear-span ratios of λ=a/d=2 and λ=a/d=4 are not enough to distinguish different failure modes of components. Therefore, it is inappropriate to divide the failure mode only by shear-span ratio.

Discriminant analysis includes parametric and non-parametric methods. The parametric method assumes that the observations of each class come from a multivariate normal distribution population, and the distribution mean center of each class can be different. The non-parametric method does not need to know the distribution of each class from the population. It uses a non-parametric method for each class and estimates the distribution density of each class, and then establishes a criterion based on this.

The discriminant analysis method is based on the Fisher criterion combined with the Bayes criterion to discriminate the failure modes of 111 reinforced concrete rectangular column members, and then three discriminant equations can be obtained, as shown in the following formula. The discriminant equation is a linear statistical equation that uses the ratio of stirrup spacing to section effective height (χ), shear-span ratio (λ) and bending-shear ratio (m) to calculate the values of three independent output variables: y ₁ , y ₂ and y ₃ . Based on the relative values of y ₁ , y ₂ and y ₃ , the failure mode of the reinforced concrete column is predicted. If _y1 is the largest, the failure of the reinforced concrete column is bending failure. If _y2 is the largest, the failure of the reinforced concrete column is bending-shear failure. If _y3 is the largest, the failure of the reinforced concrete column is shear failure. The coefficients of the discriminant equation represent weights. It can be seen that the coefficient of the bending-shear ratio is the largest, indicating that the discriminant equation is most sensitive to the value of the bending-shear ratio. This shows that the bending-shear ratio can reflect the failure mode of reinforced concrete columns to a certain extent. This is why the bending-shear ratio can be more accurate than the other parameters when analyzed using a single parameter. However, other parameters cannot be ignored.

y ₁ =-7.664χ+6.448λ+16.86m-19.455

y ₂ =-0.168χ+4.942λ+21.232m-20.815

y ₃ =-2.935χ+4.038λ+25.726m-23.226

Among them, y ₁ , y ₂ and y ₃ are independent output variable values; χ is the ratio of the stirrup spacing to the effective height of the section; λ is the shear-span ratio; m is the bending-shear ratio.

The results of the prediction method based on the discriminant equation are listed in Table 4. The detailed description of Table 4 is as follows:

(1) Among the 64 sets of data where bending failure occurred, 59 sets of data were correctly predicted as bending failure, and 5 sets of data were wrongly predicted as bending-shear failure.

(2) Among the 36 sets of data in which bending-shear failure occurred, 26 sets of data were correctly predicted as bending-shear failure, 5 sets of data were wrongly predicted as bending failure, and 5 sets of data were wrongly predicted as Shear failure.

(3) Among the 11 sets of data with shear failure, 10 sets of data were correctly predicted as shear failure, and 1 set of data was wrongly predicted as bending-shear failure.

Table 4

It was found that 85.1% (85.1%=(92.2%+72.2%+90.9%)/3) of the column members were correctly predicted by the discriminant analysis. Its accuracy is much higher than other forecasting methods that only use a single parameter.

Through the comprehensive analysis of the above discriminant analysis method, it can be seen that the ratio of stirrups, axial compression ratio, longitudinal reinforcement ratio, stirrup spacing and section effective height ratio have no significant impact on the failure mode of column members. The two parameters that have the greatest influence on the failure mode of column members are the bending-shear ratio and the shear-span ratio. Please refer to Figure 12. Figure 12 is a schematic diagram of failure mode prediction based on the bending-shear ratio and shear-span ratio. The failure modes of these 111 rectangular column members are divided according to the bending-shear ratio and shear-span ratio. Span ratio, the vertical axis is the reduced bending-shear ratio.

Based on Figure 12, a fast and simple method for manually predicting the failure mode of column members based on the bending-shear ratio and shear-span ratio can be developed—the engineering simplified discriminant analysis method. According to Fig. 12, the failure modes of RC columns can be predicted using the following criteria.

(1) For column members with shear-span ratio λ≤2, when the bending-shear ratio V _p /V _n ≥ 0.8, the shear failure is dominant; when the bending-shear ratio is less than V _p /V _n <0.8, the Bending-shear failure is dominant.

(2) For column members with a shear-span ratio of 2<λ<4, when the bending-shear ratio V _p /V _n <0.8, bending failure is dominant; when the bending-shear ratio is 0.8≤V _p /V _n ≤1.5, Bending-shear failure is the main form; when V _p /V _n >1.5, shear failure occurs in column members.

(3) For column members with a shear-span ratio λ≥4, when the bending-shear ratio V _p /V _n ≤ 1.2, bending failure is dominant; when the bending-shear ratio V _p /V _n > 1.2, due to the lack No conclusions can be drawn from the sample.

The failure mode prediction criteria for column members are summarized in Table 5.

table 5

The prediction results based on the prediction criteria in Table 5 are listed in Table 6. The detailed description of Table 6 is as follows:

(1) Among the 64 sets of data in which bending failure occurred, 49 sets of data were correctly predicted as bending failure, and 15 sets of data were wrongly predicted as bending-shear failure.

(2) Among the 36 sets of data in which bending-shear failure occurred, 31 sets of data were correctly predicted as bending-shear failure, 2 sets of data were wrongly predicted as bending failure, and 3 sets of data were wrongly predicted as Shear failure.

(3) Among the 11 sets of data with shear failure, 10 sets of data were correctly predicted as shear failure, and 1 set of data was wrongly predicted as bending-shear failure.

Table 6

It can be seen that the engineering simplified discriminant analysis method correctly predicts 84.5% (84.5% = (76.6% + 86.1% + 90.9%) / 3) column members, which is slightly lower than the accuracy rate of the discriminant analysis method.

The present invention compares with traditional prediction method, and ANN network prediction method has following advantages:

(1) The accuracy of the ANN network prediction method is higher than that of the traditional prediction method. Among the traditional prediction methods, the accuracy rate of the prediction method based on discriminant analysis is the highest, reaching 85.1%, while the accuracy rate of the ANN network prediction method is as high as 95.3%.

(2) The prediction method of ANN network considers more comprehensive factors. For example, the prediction method based on shear-span ratio only considers a single parameter, the prediction method based on engineering simplified discriminant analysis considers 2 parameters, the prediction method based on discriminant analysis considers 3 parameters, and the prediction method of ANN network considers 6 parameters.

(3) With the expansion of the database, the prediction ability of the model can be dynamically improved.

This embodiment is based on the artificial neural network prediction method, combined with the Levenberg Marquardt backpropagation training algorithm to identify the failure mode of the rectangular column member. First, the artificial neural network structure of the failure mode of reinforced concrete rectangular column members is set; the data set of rectangular column members is processed, which is effective for affecting the stirrup ratio, longitudinal reinforcement ratio, stirrup spacing and cross-section in the failure mode of rectangular column members Sensitivity analysis of height ratio, shear-span ratio, axial compression ratio, and bending-shear ratio is carried out to obtain the failure mode prediction results of rectangular column members; statistical analysis is carried out on the failure mode prediction results of rectangular column members.

It should be noted that: the above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the principles of the present invention shall be included in the present invention. within the scope of protection of the invention.

Claims (9)

1. A method for predicting RC rectangular column failure mode based on artificial neural network, is characterized in that, the method comprises the following steps: The prediction method based on the artificial neural network, combined with the Levenberg Marquardt backpropagation training algorithm to identify the failure mode of the rectangular column members; Among them, the method of judging the failure mode of rectangular column members is as follows: setting the artificial neural network structure of the failure mode of reinforced concrete rectangular column members; Sensitivity analysis of reinforcement ratio, ratio of stirrup spacing to section effective height, shear-span ratio, axial compression ratio, and bending-shear ratio to obtain the prediction results of failure mode of rectangular column members; statistical analysis of the prediction results of failure mode of rectangular column members . 2. the method for predicting RC rectangular column failure mode based on artificial neural network according to claim 1, is characterized in that, the parameter of the artificial neural network structure of reinforced concrete rectangular column member failure mode comprises: input layer node number, hidden layer node The number of nodes in the output layer, the activation function of the input layer, the activation function of the hidden layer and the activation function of the output layer. 3. the method for predicting RC rectangular column failure mode based on artificial neural network according to claim 2, it is characterized in that, input layer has 6 nodes, and 6 nodes represent respectively furnishing rate, longitudinal reinforcement ratio, stirrup spacing The effective height ratio of the section, the shear-span ratio, the axial compression ratio, and the bending-shear ratio. 4. the method for predicting RC rectangular column failure mode based on artificial neural network according to claim 2, is characterized in that, the expression of input layer activation function is: y=f(Wp+b) Among them, y is the output vector of the activation function; W is the weight matrix; p is the initial input vector; b is the bias vector. 5. the method for predicting RC rectangular column failure mode based on artificial neural network according to claim 2, is characterized in that, hidden layer comprises two layers, and the first layer has 100 nodes, and the second layer has 10 nodes; Wherein, The hidden layer activation function is the sigmoid function. 6. the method for predicting RC rectangular column failure mode based on artificial neural network according to claim 2, is characterized in that, output layer node number is 3, and wherein each node corresponds to a kind of failure mode; Wherein output layer activation function is softmax function. 7. the method for predicting RC rectangular column failure mode based on artificial neural network according to claim 1, is characterized in that, the expression of Levenberg Marquardt backpropagation training algorithm is: △x＝-[J ^T J+μI] ^-1 J ^T ε Among them, J is the Jacobian matrix; J ^T is the transpose matrix of the Jacobian matrix J; μ is a normal number; I is the identity matrix; ε is the network error. 8. The method for predicting the failure mode of RC rectangular columns based on artificial neural network according to claim 1, wherein the bending-shear ratio and the shear-span ratio are related to the failure mode of reinforced concrete rectangular column members. 9. the method for predicting RC rectangular column failure mode based on artificial neural network according to claim 1, is characterized in that, The accuracy rate of the training data of the trained artificial neural network reaches 100%, and the accuracy rate of the test data reaches 95.3%. The minimum mean square error of the training data is 0.0018, and the minimum mean square error of the test data is 0.038.