L-shaped plate hook hoisting strength prediction method based on GA-BP neural network

Abstract

The invention relates to the technical field of building steel structure hoisting, in particular to an L-shaped plate hook hoisting strength prediction method based on a GA-BP neural network. Compared with the prior art, the method has the advantages that the strength of the L-shaped plate hook for hoisting the latticed column under various working conditions is calculated under different working conditions by utilizing a finite element method, and the data is normalized, so that a sample database is established, and on the basis, the strength of the L-shaped plate hook for hoisting the latticed column under various working conditions is calculated; the significance of each feature factor influencing the strength of the L-shaped plate hook is further quantified, and main feature factors are extracted from the feature factors to be used for constructing a neural network model; in order to optimize the performance of a neural network model, a BP neural network is adjusted and optimized by adopting a genetic algorithm (GA), and an L-shaped plate hook strength prediction model is generated; and then, verifying and testing the L-shaped plate hook strength prediction model by using the test set, so that the L-shaped plate hook strength prediction model can accurately calculate the maximum stress of the L-shaped plate hook.

Description

Technical Field

[0001] The invention relates to the technical field of building steel structure hoisting, in particular to a method for predicting the hoisting strength of an L-shaped plate hook based on a GA-BP neural network. Background Art

[0002] In the processing and manufacturing process of large composite lattice columns, steel columns need to undergo multiple hoisting and transportation. The commonly used hoisting tools are mainly L-shaped plate hooks, which are mostly made by steel structure processing plants and cut from plates. There are two main ways to calculate the strength of these self-cut L-shaped plate hooks: one is to use the simplified model in the traditional mechanical manual and the basic formula of material mechanics for calculation; the other is to use solid finite element analysis to perform detailed calculation and analysis on specific plate hooks. However, in actual engineering applications, the simplified force model provided by the traditional mechanical manual has over-simplified calculation conditions and external loads, which makes it difficult to accurately reflect the actual stress state of the L-shaped plate hook in actual application, and the calculation accuracy is low; in contrast, the finite element solid simulation can rely on the specific information of the external working conditions and provide high-precision calculation results, but if the parameters of the L-shaped plate hook change, it must be recalculated. In addition, the iterative method used in the finite element calculation itself has a high calculation cost, which greatly limits its practicality and flexibility in actual applications. Summary of the invention

[0003] The technical problem to be solved by the present invention is that the existing L-shaped plate hook strength calculation accuracy is low and there are calculation problems.

[0004] The present invention is to solve the above problems, and its specific scheme is:

[0005] A method for predicting the lifting strength of an L-type plate hook based on a GA-BP neural network comprises the following steps:

[0006] Step 1: Use solid finite element method to simulate and calculate the strength of the L-shaped plate hook for hoisting lattice columns under various working conditions, and obtain the maximum stress of the L-shaped plate hook under each working condition; analyze the characteristic factors affecting the maximum stress, and normalize the data of the maximum stress and characteristic factors;

[0007] Step 2: Perform significance analysis on the normalized data and select the main characteristic factors based on the significance analysis;

[0008] Step 3: randomly divide the maximum stress value data in step 1 into a training set and a test set, wherein the training set is used to train the neural network model, and the test set is used to verify the neural network results;

[0009] Step 4: Use the training set as the input of the neural network model, the main characteristic factors as the input layer of the neural network model, and the predicted strength of the L-type hook as the output layer; set the maximum number of iterations, learning rate and target error, and preliminarily determine the number of neurons in the hidden layer. Then, use the training set to train the neural network model and adjust the number of neurons in the hidden layer.

[0010] Step 5: Optimize the neural network model through genetic algorithm to generate the L-type plate hook strength prediction model;

[0011] Step 6: Use the L-type hook strength prediction model to predict the test set, generate prediction results, denormalize the prediction results back to the original scale, calculate the mean square error, and use the mean square error as an evaluation indicator. If the mean square error is greater than the expected value, the neural network training is completed, otherwise return to step 4 for retraining;

[0012] Step 7. Input the main characteristic factors of the L-type hook to be tested into the L-type hook strength prediction model to generate the maximum stress prediction value of the L-type hook to be tested; compare the maximum stress prediction value of the L-type hook to be tested with the yield strength of the L-type hook. If the maximum stress prediction value is less than the yield strength, it is determined that the working condition is safe, otherwise, it is determined that the working condition is unsafe.

[0013] Compared with the prior art, the present invention using the above technical solution has the following beneficial effects:

[0014] Compared with the prior art, the present invention firstly uses the finite element method to calculate the strength of the L-shaped plate hook of the hoisting lattice column under various working conditions, and normalizes these data to establish a sample database. On this basis, the present invention further quantifies the significance of each characteristic factor affecting the strength of the L-shaped plate hook, and extracts the main characteristic factors therefrom for constructing a neural network model; in order to optimize the performance of the neural network model, the present invention uses a genetic algorithm (GA) to tune the BP neural network and generate an L-shaped plate hook strength prediction model; subsequently, the L-shaped plate hook strength prediction model is verified and tested using a test set, so that the L-shaped plate hook strength prediction model can accurately calculate the maximum stress of the L-shaped plate hook.

[0015] Preferably, a further technical solution of the present invention is:

[0016] The characteristic factors include the handle length, thickness, net width and weight to be lifted of the L-shaped plate hook.

[0017] The normalization process of the data of the maximum stress and characteristic factors is specifically as follows:

[0018]

[0019]

[0020] in, is the unprocessed value and normalized processed value of the handle length of the L-type plate hook; is the net width of the L-type hook to be processed and the normalized processed value; is the thickness of the L-shaped hook to be processed and the normalized processed value; The waiting value and normalized value of the weight to be lifted of the L-type plate hook; δ limit It is the maximum stress to be processed value, normalized processed value and the yield limit value specified for the material that occurs in the L-type plate hook.

[0021] The adjustment of the number of neurons in the hidden layer is specifically as follows:

[0022]

[0023] Where m is the number of neurons in the hidden layer; l is the number of neurons in the input layer; b is the number of neurons in the output layer; and a is an integer constant in the interval [1,10].

[0024] The neural network model is optimized by genetic algorithm, specifically: the population size, iteration times, crossover probability and mutation probability of the genetic algorithm are set, the genetic algorithm and the neural network model are combined, and the L-type plate hook strength prediction model is established. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 is a flow chart of a prediction method according to an embodiment of the present invention;

[0026] Figure 2 is a schematic diagram of the structure of a neural network model according to an embodiment of the present invention; DETAILED DESCRIPTION

[0027] The present invention will be further described below in conjunction with embodiments, the purpose of which is only to provide a better understanding of the content of the present invention. Therefore, the examples given do not limit the protection scope of the present invention.

[0028] See attached Figure 1-2 The embodiment of the present invention discloses a method for predicting the lifting strength of an L-type plate hook based on a GA-BP neural network, comprising the following steps:

[0029] Step 1: Design an orthogonal test and simulate the strength of the L-shaped plate hook for hoisting lattice columns under various working conditions by using the solid finite element technology to obtain the maximum stress of the L-shaped plate hook under each working condition, such as the handle length L, thickness t, net width w and weight to be hoisted G of the L-shaped plate hook;

[0030] In this example, a four-factor five-level orthogonal experiment is designed as shown in Table 1 and Table 2:

[0031] Table 1 Factors and levels of orthogonal test

[0032]

[0033] Table 2 Factors and levels of orthogonal test

[0034]

[0035] The normalized orthogonal test scheme is shown in Table 3 below:

[0036] Table 3 Orthogonal test scheme

[0037] Sample L / mm t / mm w / mm G / t MPa(stress) 1 L1 t1 w1 G5 L1t1w1G5 2 L1 t2 w3 G2 L1t2w3G2 3 L1 t3 w5 G4 L1t3w5G4 4 L1 t4 w2 G1 L1t4w2G1 5 L1 t5 w4 G3 L1t5w4G3 6 L2 t1 w5 G2 L2t1w5G2 7 L2 t2 w2 G4 L2t2w2G4 8 L2 t3 w4 G1 L2t3w4G1 9 L2 t4 w1 G3 L2t4w1G3 10 L2 t5 w3 G5 L2t5w3G5 11 L3 t1 w4 G4 L3t1w4G4 12 L3 t2 w1 G1 L3t2w1G1 13 L3 t3 w3 G3 L3t3w3G3 14 L3 t4 w5 G5 L3t4w5G5 15 L3 t5 w2 G2 L3t5w2G2 16 L4 t1 w3 G1 L4t1w3G1 17 L4 t2 w5 G3 L4t2w5G3 18 L4 t3 w2 G5 L4t3w2G5 19 L4 t4 w4 G2 L4t4w4G2 20 L4 t5 w1 G4 L4t5w1G4 21 A5 t1 w2 G3 A5t1w2G3 22 A5 t2 w4 G5 A5t2w4G5 23 A5 t3 w1 G2 A5t3w1G2 24 A5 t4 w3 G4 A5t4w3G4 25 A5 t5 w5 G1 A5t5w5G1

[0038] Table 4 Orthogonal test scheme and simulation results

[0039]

[0040]

[0041] Analyze the characteristic factors that affect the maximum stress, and normalize the data of the maximum stress and characteristic factors, specifically:

[0042]

[0043] in, is the unprocessed value and normalized processed value of the handle length of the L-type plate hook; is the net width of the L-type hook to be processed and the normalized processed value; is the thickness of the L-shaped hook to be processed and the normalized processed value; The waiting value and normalized value of the weight to be lifted of the L-type plate hook; δ limit It is the maximum stress to be processed value, normalized processed value and the yield limit value specified for the material that occurs in the L-type plate hook.

[0044] Step 2: Perform significance analysis on the normalized data and select the main characteristic factors based on the significance analysis;

[0045] The significance analysis specifically uses the range analysis method to establish an analysis table, as shown below:

[0046] Table 5 Range Analysis Table

[0047]

[0048] In the table:

[0049]

[0050] Where:

[0051] n and i are natural positive integers, ranging from 0 to 5;

[0052] S is the number of levels, which is 4 in this example;

[0053] M is the combination of L, t, w, and G.

[0054] Substituting the data in this embodiment into the following table:

[0055] Table 6 Range analysis results

[0056]

[0057]

[0058] It can be found from the above table that in this example, the order of the characteristic factors in terms of their influence on the results from large to small is W>L>G>T, the optimal level of handle length is L4, the optimal level of thickness is t4, the optimal level of net width is w2, and the optimal level of weight to be hoisted is G5. Therefore, the optimal level result L4t4w2G5 can be obtained in this orthogonal test.

[0059] Step 3: Aggregate the maximum stress value data in step 1 into a data set, and randomly divide the data set into a training set and a test set, wherein the proportion of the training set is set to be no less than 70% of the total data in the data set, the training set is used to train the neural network model, and the test set is used to verify the neural network results;

[0060] Step 4: construct a neural network model, use the training set as the input of the neural network model, the main characteristic factors as the input layer of the neural network model, and the predicted strength of the L-type plate hook as the output layer; set the maximum number of iterations, learning rate and target error, and preliminarily determine the number of neurons in the hidden layer, then adjust the number of neurons in the hidden layer according to the number of neurons in the input layer and the output layer, and use the training set to train the neural network model. Preferably, MATLAB, PYTHON or FORTRAN software is used for model training, and the training algorithm uses the Levenberg-Marquard algorithm to simulate the nonlinear relationship between the main characteristic factors and the maximum stress of the L-type plate hook;

[0061] Specifically, the formula for adjusting the number of neurons in the hidden layer is:

[0062]

[0063] Where m is the number of neurons in the hidden layer; l is the number of neurons in the input layer; b is the number of neurons in the output layer; and a is an integer constant in the interval [1,10].

[0064] Step 5: Optimize the neural network model through genetic algorithm (GA). First, set the parameters of the genetic algorithm, such as population size, number of iterations, selection method, crossover probability, and mutation probability. Then, combine the genetic algorithm with the neural network model to establish an L-type plate hook strength prediction model to improve the prediction accuracy and generalization ability. The specific genetic algorithm parameters can be implemented by calling the gaoptimset function.

[0065] Step 6: Use the L-type hook strength prediction model to predict the test set, generate prediction results, denormalize the prediction results back to the original scale, calculate the mean square error, and use the mean square error as an evaluation indicator. If the mean square error is less than the expected value, the neural network training is completed, otherwise return to step 4 for retraining;

[0066] The mean square error is calculated using the following formula:

[0067]

[0068] Where:

[0069] n is the number of samples tested;

[0070] y i is the finite element data of the test set;

[0071] Predict the values ​​for the test set for the neural network.

[0072] The expected value is a positive number between 0 and 1. If the mean square error is less than the expected value, it means that the model fits well, the neural network training is qualified, and the training can be completed.

[0073] Step 7. Input the main characteristic factors of the L-type hook to be tested into the L-type hook strength prediction model to generate the maximum stress prediction value of the L-type hook to be tested; compare the maximum stress prediction value of the L-type hook to be tested with the yield strength of the L-type hook. If the maximum stress prediction value is less than the yield strength, it is determined that the working condition is safe, otherwise, it is determined that the working condition is unsafe.

[0074] In this embodiment, preferably, the strength of 25 sets of L-shaped plate hooks for hoisting lattice columns is simulated and calculated, and the 25 sets of data sets are divided into 20 training sets and 5 validation sets;

[0075] The data of the 36 training sets are as follows:

[0076]

[0077]

[0078] The data of the 6 validation sets are as follows:

[0079] Working conditions Mises stress (MPa) Working conditions Mises stress (MPa) CS-01 161.65 CS-04 256.2 CS-02 240.95 CS-05 298.9 CS-03 280.6 CS-06 384.3

[0080] The data of the training set and the validation set are normalized, where the limit of stress normalization is set to 295 MPa. The training set data after normalization is as follows:

[0081]

[0082]

[0083] The validation set data after normalization is as follows:

[0084] Working conditions Handle length L Thickness Net width W Weight to be hoisted G stress CS-01 1.000 0.500 0.625 0.600 1.130 CS-02 1.000 0.625 0.875 1.000 0.650 CS-03 1.000 0.750 0.500 0.400 1.018 CS-04 1.000 0.875 0.750 0.800 0.772 CS-05 1.000 1.000 1.000 0.200 0.341

[0085] In this embodiment, the neural network model is optimized by a genetic algorithm, the population size is set to 50, the maximum number of evolutions is set to 100, the crossover probability is set to 0.4, and the mutation probability is set to 0.1.

[0086] The data of the five validation sets were input into the L-type plate hook strength prediction model, and the validation results are shown in the following table:

[0087] Working conditions Finite element calculation value Predicted value Relative error Prediction / Calculation Com1 1.130 1.276 12.89％ 112.89％ Com2 0.650 0.581 10.66％ 89.34％ Com3 1.018 1.131 11.12％ 111.12％ Com4 0.772 0.739 4.33％ 95.67％ Com5 0.341 0.349 2.21％ 102.21％

[0088] In these five sets of validation data, the relative error between the maximum stress predicted by the L-type plate hook strength prediction model and the maximum stress calculated by the finite element method remained within 15%, with a minimum error of 2.21%, a maximum error of 12.89%, and an average relative error of 8.24%.

[0089]

[0090] 0 <MSE=0.147<1

[0091] The mean square error is less than 1, which shows that the prediction method in the present invention has high accuracy and the neural network result has good fitting.

[0092] Compared with the prior art, the present invention firstly uses the finite element method to calculate the strength of the L-shaped plate hook of the hoisting lattice column under various working conditions, and normalizes these data to establish a sample database. On this basis, the present invention further quantifies the significance of each characteristic factor affecting the strength of the L-shaped plate hook, and extracts the main characteristic factors therefrom for constructing a neural network model; in order to optimize the performance of the neural network model, the present invention uses a genetic algorithm (GA) to tune the BP neural network and generate an L-shaped plate hook strength prediction model; subsequently, the L-shaped plate hook strength prediction model is verified and tested using a test set, so that the L-shaped plate hook strength prediction model can accurately calculate the maximum stress of the L-shaped plate hook.

[0093] The above description is only a preferred feasible embodiment of the present invention, and does not limit the scope of rights of the present invention. All equivalent changes made using the contents of the present specification and its drawings are included in the scope of rights of the present invention.

Claims

1. A method for predicting the lifting strength of an L-type plate hook based on a GA-BP neural network, characterized in that: The following steps are involved: Step 1: Use solid finite element method to simulate and calculate the strength of the L-shaped plate hook for hoisting lattice columns under various working conditions, and obtain the maximum stress of the L-shaped plate hook under each working condition; analyze the characteristic factors affecting the maximum stress, and normalize the data of the maximum stress and characteristic factors; Step 2: Perform significance analysis on the normalized data and select the main characteristic factors based on the significance analysis; Step 3: randomly divide the maximum stress value data in step 1 into a training set and a test set, wherein the training set is used to train the neural network model, and the test set is used to verify the neural network results; Step 4: Use the training set as the input of the neural network model, the main characteristic factors as the input layer of the neural network model, and the predicted strength of the L-type hook as the output layer; set the maximum number of iterations, learning rate and target error, and preliminarily determine the number of neurons in the hidden layer. Then, use the training set to train the neural network model and adjust the number of neurons in the hidden layer. Step 5: Optimize the neural network model through genetic algorithm to generate the L-type plate hook strength prediction model; Step 6: Use the L-type hook strength prediction model to predict the test set, generate prediction results, denormalize the prediction results back to the original scale, calculate the mean square error, and use the mean square error as an evaluation indicator. If the mean square error is greater than the expected value, the neural network training is completed, otherwise return to step 4 for retraining; Step 7. Input the main characteristic factors of the L-type hook to be tested into the L-type hook strength prediction model to generate the maximum stress prediction value of the L-type hook to be tested; compare the maximum stress prediction value of the L-type hook to be tested with the yield strength of the L-type hook. If the maximum stress prediction value is less than the yield strength, it is determined that the working condition is safe, otherwise, it is determined that the working condition is unsafe.

2. The method for predicting the lifting strength of an L-type plate hook based on a GA-BP neural network according to claim 1 is characterized in that: The characteristic factors include the handle length, thickness, net width and weight to be lifted of the L-shaped plate hook.

3. The method for predicting the lifting strength of an L-type plate hook based on a GA-BP neural network according to claim 1 is characterized in that: The data of maximum stress and characteristic factors are normalized as follows: in, is the unprocessed value and normalized processed value of the handle length of the L-type plate hook; is the net width of the L-type hook to be processed and the normalized processed value; is the thickness of the L-shaped hook to be processed and the normalized processed value; The waiting value and normalized value of the weight to be lifted of the L-type plate hook; δ limit It is the maximum stress to be processed value, normalized processed value and the yield limit value specified for the material that occurs in the L-type plate hook.

4. The method for predicting the lifting strength of an L-type plate hook based on a GA-BP neural network according to claim 1 is characterized in that: The adjustment of the number of neurons in the hidden layer is specifically as follows: Where m is the number of neurons in the hidden layer; l is the number of neurons in the input layer; b is the number of neurons in the output layer; and a is an integer constant in the interval [1,10].

5. The method for predicting the lifting strength of an L-type plate hook based on a GA-BP neural network according to claim 1 is characterized in that: The neural network model is optimized by genetic algorithm, specifically: the population size, iteration times, crossover probability and mutation probability of the genetic algorithm are set, the genetic algorithm and the neural network model are combined, and the L-type plate hook strength prediction model is established.

Images