阅读说明：本技术 一种基于bp神经网络的振动时效工艺参数优化方法 (BP neural network-based vibration aging process parameter optimization method ) 是由 花文杰 顾邦平 肖光年 胡雄 霍志鹏 李帅振 薛文喆 季雨 王军硕 于 2021-06-18 设计创作，主要内容包括：一种基于BP神经网络的振动时效工艺参数优化方法,其特征在于：包括以下步骤：采用正交试验设计法制定振动时效实验方案,开展振动时效实验,获取实验数据；将所获得的数据样本集按照不同参数类别分别进行量纲一化处理,并按照一定的比例将该数据集分为训练样本与测试样本两部分；采用量纲一化后的数据对BP神经网络进行训练和测试；根据测试和实验得到的残余应力值的误差分别对网络各层权值和阈值进行修正,直到预测值与实际值的相对误差均值小于设定值为止,得到最优的工艺参数组合,最终形成基于BP神经网络的振动时效工艺参数优化方法。本发明具有简化振动时效工艺参数调整过程和得到最优的工艺参数的优点。(A vibration aging process parameter optimization method based on a BP neural network is characterized by comprising the following steps: the method comprises the following steps: an orthogonal test design method is adopted to formulate a vibration aging test scheme, a vibration aging test is carried out, and test data are obtained; respectively carrying out dimension normalization processing on the obtained data sample set according to different parameter types, and dividing the data set into a training sample and a test sample according to a certain proportion; adopting the data after dimension normalization to train and test the BP neural network; and respectively correcting the weight and the threshold value of each layer of the network according to the errors of the residual stress values obtained by testing and experiments until the average value of the relative errors of the predicted value and the actual value is less than a set value to obtain an optimal process parameter combination, and finally forming the BP neural network-based vibration aging process parameter optimization method. The invention has the advantages of simplifying the adjustment process of the vibration aging process parameters and obtaining the optimal process parameters.)

1. A vibration aging process parameter optimization method based on a BP neural network is characterized by comprising the following steps: the method comprises the following steps:

(1) an orthogonal test design method is adopted to formulate a vibration aging experimental scheme: analyzing technological parameters influencing the vibratory stress relief effect, namely main factors influencing the vibratory stress relief effect, selecting the same number of levels for each technological parameter, namely selecting the same number of levels for each factor, and selecting an orthogonal test table according to the number of the factors and the factor levels to formulate a vibratory stress relief experimental scheme; the vibration aging process parameters comprise vibration frequency, vibration amplitude and vibration time;

(2) developing a vibration aging experiment to obtain experiment data: carrying out a vibration aging experiment according to the vibration aging experiment scheme formulated in the step (1) to obtain a vibration aging effect under different process parameter combinations;

(3) determining a data sample set: determining a data sample set according to an experimental result, and dividing the data sample set into a training sample and a test sample according to a certain proportion;

(4) dimension normalization treatment: respectively carrying out dimension normalization processing on the data sample set obtained in the step (3) according to different parameter types;

(5) training a BP neural network: training a BP neural network by adopting data in a training sample, namely taking dimension-normalized vibration aging process parameters as input and residual stress reduction rate as output, and establishing a BP neural network model between the dimension-normalized vibration aging process parameters and the residual stress reduction rate;

(6) testing the BP neural network: testing the BP neural network model between the dimensional normalized vibration aging process parameters and the residual stress reduction rate established in the step (5) by adopting the test data in the test sample, namely inputting the dimensional normalized vibration aging process parameters in the test sample into the BP neural network model established in the step (5) to obtain the predicted value of the residual stress reduction rate corresponding to the vibration aging process parameters after dimensional normalization treatment in the test sample;

(7) judging whether the training process is finished according to the predicted value of the residual stress reduction rate obtained in the step (6) and the relative error mean value of the actual residual stress reduction rate obtained through the experimental test in the test sample: setting a set value of a relative error mean value between a predicted value of the residual stress reduction rate and an actual residual stress reduction rate obtained through an experimental test as beta, and finishing the training process if the relative error mean value between the predicted value of the residual stress reduction rate and the residual stress reduction rate obtained through the experimental test is smaller than the set value beta; if the mean value of the relative errors between the predicted value of the residual stress reduction rate and the actual residual stress reduction rate obtained through the experimental test is larger than the set value beta, respectively correcting the weight and the threshold value of each layer of the network according to the errors to realize the reverse propagation of the errors until the mean value of the relative errors between the predicted value of the residual stress reduction rate and the actual residual stress reduction rate obtained through the experimental test is smaller than the set value beta, and ending the training process;

(8) and obtaining the optimal process parameter combination.

2. The vibration aging process parameter optimization method based on the BP neural network as claimed in claim 1, wherein: the vibration aging effect is characterized by adopting a residual stress reduction rate which isWhere δ is the residual stress reduction ratio, σ_InitialFor aging testInitial residual stress, σ, of samples not subjected to vibratory ageing_{After vibration}The residual stress of the aged sample after vibration aging treatment is shown.

3. The vibration aging process parameter optimization method based on the BP neural network as claimed in claim 1, wherein: testing at two points of the test sample respectively, wherein a test point 1 is used for testing the residual stress of the test sample before vibratory stress relief, and a test point 2 is used for testing the residual stress of the test sample after vibratory stress relief; in order to improve the reliability of the experimental result, the average residual stress of three groups of samples is used for representing the residual stress before and after the vibration aging.

4. The vibration aging process parameter optimization method based on the BP neural network as claimed in claim 1, wherein: the software for predicting and optimizing the process parameters by the BP neural network is MATLAB.

5. The vibration aging process parameter optimization method based on the BP neural network as claimed in claim 1, wherein: the residual stress testing method is a small hole method.

6. The vibration aging process parameter optimization method based on the BP neural network as claimed in claim 1, wherein: the optimal process parameter combination in the step (8) is determined under the condition that the relative error between a predicted value of the residual stress reduction rate obtained by inputting the optimal process parameters into the established BP neural network model and an actual residual stress reduction rate obtained by an experimental test corresponding to the optimal process parameters is smaller than a set critical value alpha, and the predicted value of the residual stress reduction rate is larger than a set critical value lambda; and if a plurality of groups of process parameter combinations meet the conditions, selecting the corresponding process parameter combination as the optimal process parameter combination when the predicted value of the residual stress reduction rate is the maximum value.

Technical Field

The invention relates to the technical field of vibration aging, in particular to a vibration aging process parameter optimization method based on a BP neural network.

Background

In the conventional residual stress eliminating method, the vibration aging technology is favored by various enterprises by virtue of the advantages of good treatment effect, short treatment time, energy conservation, environmental protection, easy field operation and the like, and belongs to an efficient, energy-saving, green and environment-friendly aging treatment technology. The method is characterized in that the residual stress generated in the process of processing and manufacturing the component is eliminated by adopting the vibration aging technology, firstly, the process parameters of the vibration aging are determined, the ideal vibration aging effect can be achieved only by reasonable process parameters, and the optimization of the process parameters of the vibration aging is also one of the key research contents in the technical field of the vibration aging. At present, in the vibration aging process, a more ideal aging effect is obtained mainly by continuously changing process parameters, so that the parameter adjustment process is complex and the aging treatment efficiency is low, which is one of the main problems in the current vibration aging technical field. Therefore, further research needs to be carried out on the vibration aging technology to obtain an optimization method of the vibration aging process parameters, and technical support is provided for popularization and application of the vibration aging technology.

The BP neural network is a multilayer feedforward neural network trained according to an error back propagation algorithm, and is one of the most widely applied neural network models at present. The BP neural network can continuously reduce errors through repeated iteration to obtain more reasonable process parameters, so that a more ideal aging treatment effect is obtained. In view of the advantages of the BP neural network in the aspect of process parameter optimization, the invention provides the method for optimizing the vibration aging process parameters by adopting the BP neural network, and solves the problems of complex adjustment process of the vibration aging process parameters and low aging treatment efficiency in the prior art.

Aiming at the problems that the adjustment process of the existing vibration aging process parameters is complex and the aging treatment efficiency is low, the invention provides a vibration aging process parameter optimization method based on a BP neural network. The method provided by the invention is adopted to optimize the vibration aging process parameters to obtain more reasonable process parameters, and the ideal aging effect of the sample after vibration aging treatment can be ensured.

Disclosure of Invention

In order to solve the problems that the adjusting process of the existing vibration aging process parameters is complex and the aging treatment efficiency is low, the invention provides a vibration aging process parameter optimization method based on a BP neural network.

A vibration aging process parameter optimization method based on a BP neural network is characterized by comprising the following steps: the method comprises the following steps:

(1) an orthogonal test design method is adopted to formulate a vibration aging experimental scheme: analyzing technological parameters influencing the vibratory stress relief effect, namely main factors influencing the vibratory stress relief effect, selecting the same number of levels for each technological parameter, namely selecting the same number of levels for each factor, and selecting an orthogonal test table according to the number of the factors and the factor levels to formulate a vibratory stress relief experimental scheme; the vibration aging process parameters comprise vibration frequency, vibration amplitude and vibration time;

(2) developing a vibration aging experiment to obtain experiment data: carrying out a vibration aging experiment according to the vibration aging experiment scheme formulated in the step (1) to obtain a vibration aging effect under different process parameter combinations;

(3) determining a data sample set: determining a data sample set according to an experimental result, and dividing the data sample set into a training sample and a test sample according to a certain proportion;

(4) dimension normalization treatment: respectively carrying out dimension normalization processing on the data sample set obtained in the step (3) according to different parameter types;

(5) training a BP neural network: training a BP neural network by adopting data in a training sample, namely taking dimension-normalized vibration aging process parameters as input and residual stress reduction rate as output, and establishing a BP neural network model between the dimension-normalized vibration aging process parameters and the residual stress reduction rate;

(6) testing the BP neural network: testing the BP neural network model between the dimensional normalized vibration aging process parameters and the residual stress reduction rate established in the step (5) by adopting the test data in the test sample, namely inputting the dimensional normalized vibration aging process parameters in the test sample into the BP neural network model established in the step (5) to obtain the predicted value of the residual stress reduction rate corresponding to the vibration aging process parameters after dimensional normalization treatment in the test sample;

(7) judging whether the training process is finished according to the predicted value of the residual stress reduction rate obtained in the step (6) and the relative error mean value of the actual residual stress reduction rate obtained through the experimental test in the test sample: setting a set value of a relative error mean value between a predicted value of the residual stress reduction rate and an actual residual stress reduction rate obtained through an experimental test as beta, and finishing the training process if the relative error mean value between the predicted value of the residual stress reduction rate and the residual stress reduction rate obtained through the experimental test is smaller than the set value beta; if the mean value of the relative errors between the predicted value of the residual stress reduction rate and the actual residual stress reduction rate obtained through the experimental test is larger than the set value beta, respectively correcting the weight and the threshold value of each layer of the network according to the errors to realize the reverse propagation of the errors until the mean value of the relative errors between the predicted value of the residual stress reduction rate and the actual residual stress reduction rate obtained through the experimental test is smaller than the set value beta, and ending the training process;

(8) and obtaining the optimal process parameter combination.

Further, the vibration aging process parameter optimization method based on the BP neural network is characterized by comprising the following steps: the vibration aging effect is characterized by adopting a residual stress reduction rate which isWhere δ is the residual stress reduction ratio, σ_InitialInitial residual stress, σ, for aged specimens without vibratory aging_{After vibration}The residual stress of the aged sample after vibration aging treatment is shown.

Further, the vibration aging process parameter optimization method based on the BP neural network is characterized by comprising the following steps: testing at two points of the test sample respectively, wherein a test point 1 is used for testing the residual stress of the test sample before vibratory stress relief, and a test point 2 is used for testing the residual stress of the test sample after vibratory stress relief; in order to improve the reliability of the experimental result, the average residual stress of three groups of samples is used for representing the residual stress before and after the vibration aging.

Further, the vibration aging process parameter optimization method based on the BP neural network is characterized by comprising the following steps: the software for predicting and optimizing the process parameters by the BP neural network is MATLAB software.

Further, the vibration aging process parameter optimization method based on the BP neural network is characterized by comprising the following steps: the residual stress testing method is a small hole method.

Further, the vibration aging process parameter optimization method based on the BP neural network is characterized by comprising the following steps: the optimal process parameter combination in the step (8) is determined under the condition that the relative error between a predicted value of the residual stress reduction rate obtained by inputting the optimal process parameters into the established BP neural network model and an actual residual stress reduction rate obtained by an experimental test corresponding to the optimal process parameters is smaller than a set critical value alpha, and the predicted value of the residual stress reduction rate is larger than a set critical value lambda; and if a plurality of groups of process parameter combinations meet the conditions, selecting the corresponding process parameter combination as the optimal process parameter combination when the predicted value of the residual stress reduction rate is the maximum value. The critical value alpha is mainly determined according to the prediction precision requirement of the BP neural network model. The critical value lambda is mainly determined according to the requirement on residual stress relief effect.

The technical conception of the invention is as follows: firstly selecting a vibration aging process parameter range and formulating an orthogonal experiment scheme, then developing a vibration aging experiment to obtain a data sample set, dividing the data set into training data and testing data, then inputting the training sample into a BP neural network for training to obtain a complex nonlinear mapping relation between the residual stress reduction rate and the vibration aging process parameter, then inputting the testing sample into a neural network model for prediction until the relative error mean value of a predicted value and an actual value is less than a set value beta, and finishing training to obtain an optimal process parameter combination.

The invention has the beneficial effects that:

(1) the BP neural network is used for optimizing the process parameters of the vibration aging, has high mapping capability, can realize any nonlinear mapping from input to output, and establishes the nonlinear relation among the vibration frequency, the vibration amplitude, the vibration time and the residual stress reduction rate to solve the parameter optimization problem influenced by complex factors.

(2) The invention simplifies the adjustment process of the vibration aging process parameters, the adjustment of the process parameters is flexible, the error of the neural network is reduced by continuously adjusting the weight and the threshold, then the BP neural network is trained and tested, finally, a better process parameter combination can be obtained, and the efficiency and the effect of the vibration aging are greatly improved.

Drawings

FIG. 1 is a schematic flow diagram of a vibration aging process parameter optimization method based on a BP neural network.

FIG. 2 shows a residual stress prediction model network structure.

FIG. 3 is a schematic diagram of a residual stress test point.

Detailed Description

The invention is further illustrated with reference to the accompanying drawings:

a vibration aging process parameter optimization method based on a BP neural network is characterized by comprising the following steps: the method comprises the following steps:

(1) an orthogonal test design method is adopted to formulate a vibration aging experimental scheme: analyzing technological parameters influencing the vibratory stress relief effect, namely main factors influencing the vibratory stress relief effect, selecting the same number of levels for each technological parameter, namely selecting the same number of levels for each factor, and selecting an orthogonal test table according to the number of the factors and the factor levels to formulate a vibratory stress relief experimental scheme; the vibration aging process parameters comprise vibration frequency, vibration amplitude and vibration time;

(2) developing a vibration aging experiment to obtain experiment data: carrying out a vibration aging experiment according to the vibration aging experiment scheme formulated in the step (1) to obtain the residual stress reduction rate under different process parameter combinations;

(3) determining a data sample set: determining a data sample set according to an experimental result, and dividing 16 groups of data sample sets into 10 groups of training samples and 6 groups of testing samples according to a ratio of 5: 3;

the specific implementation details are as follows:

table 1 shows the process parameter data and the effectiveness in the training samples, wherein F represents the actual residual stress reduction value obtained by the experimental test.

TABLE 1 Process parameter data and Effect in training samples

Table 2 shows the process parameters and the effectiveness in the test samples, wherein F' represents the predicted value of the residual stress reduction rate,the relative error between the predicted value of the residual stress reduction rate and the actual residual stress reduction rate obtained by experimental test is adopted.

Table 2 testing of process parameters and time effectiveness in samples

(4) Dimension normalization treatment: respectively carrying out dimension normalization processing on the data sample set obtained in the step (3) according to different parameter types;

the unit of the vibration frequency f is kHz, the unit of the vibration amplitude A is mum, the unit of the vibration time t is min, and in order to eliminate the influence of different dimensions, the following three dimensionless variables are defined: dimensionless vibration frequencyDimensionless vibration amplitudeDimensionless vibration timeWherein f is₀Is 1kHz, A₀Is 1 μm, t₀Is 1 min;

(5) training a BP neural network: training a BP neural network by adopting data in a training sample, namely taking dimension-normalized vibration aging process parameters as input and residual stress reduction rate as output, and establishing a BP neural network model between the dimension-normalized vibration aging process parameters and the residual stress reduction rate;

the specific implementation details are as follows:

the structure of the vibration aging residual stress prediction model network is shown in FIG. 2. As can be seen from fig. 2, the process parameters affecting the residual stress elimination effect of vibro-aging mainly include vibration frequency F, vibration amplitude a and vibration time t, the three process parameters are input ends of the neural network model, the residual stress elimination effect F of vibro-aging is an output end of the neural network model, and the number of layers in the middle hidden layer is set to 8.

(6) Testing the BP neural network: testing the BP neural network model between the dimension-normalized vibration aging process parameters and the residual stress reduction rate, which are established in the step (5), by adopting the test data after dimension normalization in the test sample, namely inputting the dimension-normalized vibration aging process parameters in the test sample into the BP neural network model established in the step (5) to obtain a predicted value of the residual stress reduction rate corresponding to the dimension-normalized vibration aging process parameters in the test sample;

(7) judging whether the training process is finished according to the predicted value of the residual stress reduction rate obtained in the step (6) and the relative error mean value of the actual residual stress reduction rate obtained through the experimental test in the test sample: setting a set value of a relative error mean value between a predicted value of the residual stress reduction rate and an actual residual stress reduction rate obtained through experimental test as beta, and finishing the training process if the relative error mean value between the predicted value of the residual stress reduction rate and the actual residual stress reduction rate obtained through the experimental test is smaller than the set value beta; if the mean value of the relative errors between the predicted value of the residual stress reduction rate and the actual residual stress reduction rate obtained through the experimental test is larger than the set value beta, respectively correcting the weight and the threshold value of each layer of the network according to the errors to realize the reverse propagation of the errors until the mean value of the relative errors between the predicted value of the residual stress reduction rate and the actual residual stress reduction rate obtained through the experimental test is smaller than the set value beta, and ending the training process;

the specific implementation details are as follows:

the predicted value of the residual stress reduction rate can be calculated and obtained based on the BP neural network model for optimizing the vibration aging process parameters. The technological parameters in the 6 groups of test samples are input into the BP neural network model established by the invention, so that the predicted value of the residual stress reduction rate, such as the predicted value F' of the residual stress reduction rate in the table 2, can be solved, and the actual residual stress reduction rate F obtained by the experimental test is also recorded in the table 2. The set value beta is mainly determined according to the prediction precision requirement of the BP neural network model. Based on Table 2, it can be seen that the relative error between the predicted residual stress reduction rate and the experimentally measured actual residual stress reduction rate of the 6 test samplesIs 4.87%, and assuming that the set value beta is 5%, the relative error isIf the mean value of (b) is less than the set value beta, the training is finished.

(8) And obtaining the optimal process parameter combination.

Further, the vibration aging process parameter optimization method based on the BP neural network is characterized by comprising the following steps: the vibration aging effect is represented by the residual stress reduction rate which isWhere δ is the residual stress reduction ratio, σ_InitialInitial residual stress, σ, for aged specimens without vibratory aging_{After vibration}The residual stress of the aged sample after vibration aging treatment is shown.

Further, the vibration aging process parameter optimization method based on the BP neural network is characterized by comprising the following steps: and testing residual stress at two points of the test sample respectively, wherein a test point 1 is used for testing the residual stress of the test sample before vibration aging, a test point 2 is used for testing the residual stress of the test sample after vibration aging, and in order to improve the reliability of the experimental result, the residual stress before and after vibration aging is represented by the average residual stress of three groups of test samples.

The specific implementation details are as follows:

as shown in fig. 3, in order to evaluate the residual stress of the test piece before and after the vibration aging treatment, two test points were disposed on the test piece, wherein the test point 1 was used for testing the residual stress of the test piece before the vibration aging treatment, and the test point 2 was used for testing the residual stress of the test piece after the vibration aging treatment.

Further, the relative error of Table 2 in step (6)

Further, the process of correcting the weight and the threshold of each layer of the network according to the error and implementing the reverse propagation of the error in the step (7) is as follows: and (4) assigning the weight value and the threshold value which are adjusted according to the error magnitude to the BP neural network again, and then circulating the steps (5) and (6) until the relative error mean value of the residual stress predicted value of the test data and the residual stress actual value measured by the experiment is smaller than the set value beta.

Further, the vibration aging process parameter optimization method based on the BP neural network is characterized by comprising the following steps: the residual stress testing method is a small hole method.

Further, the vibration aging process parameter optimization method based on the BP neural network is characterized by comprising the following steps: the software for predicting and optimizing the process parameters by the BP neural network is MATLAB software.

Further, the vibration aging process parameter optimization method based on the BP neural network is characterized by comprising the following steps: the optimal process parameter combination in the step (8) is determined under the condition that the relative error between a predicted value of the residual stress reduction rate obtained by inputting the optimal process parameters into the established BP neural network model and an actual residual stress reduction rate obtained by an experimental test corresponding to the optimal process parameters is smaller than a set critical value alpha, and the predicted value of the residual stress reduction rate is larger than a set critical value lambda; and if a plurality of groups of process parameter combinations meet the conditions, selecting the corresponding process parameter combination as the optimal process parameter combination when the predicted value of the residual stress reduction rate is the maximum value. The critical value alpha is mainly determined according to the prediction precision requirement of the BP neural network model. The critical value lambda is mainly determined according to the requirement on residual stress relief effect.

The specific implementation details are as follows:

assuming that the critical value λ is determined to be 70% and the critical value α is determined to be 6%, we can find from the results of table 2 that the group 5 process parameter combinations are the optimal process parameter combinations.

It is found from table 2 that the residual stress reduction rate can be obtained without a vibration aging test, that is, the predicted value of the residual stress reduction rate can be directly obtained through the established BP neural network model, so that the vibration aging process parameter adjustment process can be simplified, the vibration aging treatment efficiency can be improved, unnecessary test processes can be reduced, and the cost and energy consumption can be reduced.

The predicted value of the residual stress reduction rate can be obtained by utilizing the BP neural network model, and the relative error between the predicted data and the actual data in the table 2 can find that the predicted value and the actual value obtained by the established BP neural network model have better goodness of fit, which shows that the optimization of the process parameters by adopting the established BP neural network model is reasonable and feasible, and the obtained data is credible.

The embodiments described in this specification are merely illustrative of implementations of the inventive concept and the scope of the present invention should not be considered limited to the specific forms set forth in the embodiments but rather by the equivalents thereof as may occur to those skilled in the art upon consideration of the present inventive concept.