CN114739894B - Neural network prediction method and device for titanium alloy corrosion rate in fluoride ion environment - Google Patents

Neural network prediction method and device for titanium alloy corrosion rate in fluoride ion environment Download PDF

Info

Publication number
CN114739894B
CN114739894B CN202210295759.1A CN202210295759A CN114739894B CN 114739894 B CN114739894 B CN 114739894B CN 202210295759 A CN202210295759 A CN 202210295759A CN 114739894 B CN114739894 B CN 114739894B
Authority
CN
China
Prior art keywords
data
neural network
titanium alloy
corrosion
current density
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202210295759.1A
Other languages
Chinese (zh)
Other versions
CN114739894A (en
Inventor
张卫冬
崔鹏
谢正荣
艾轶博
张琬滢
张媛媛
王校源
马宁
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
University of Science and Technology Beijing USTB
Original Assignee
University of Science and Technology Beijing USTB
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by University of Science and Technology Beijing USTB filed Critical University of Science and Technology Beijing USTB
Priority to CN202210295759.1A priority Critical patent/CN114739894B/en
Publication of CN114739894A publication Critical patent/CN114739894A/en
Application granted granted Critical
Publication of CN114739894B publication Critical patent/CN114739894B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N17/00—Investigating resistance of materials to the weather, to corrosion, or to light
    • G01N17/006—Investigating resistance of materials to the weather, to corrosion, or to light of metals
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N17/00—Investigating resistance of materials to the weather, to corrosion, or to light
    • G01N17/02—Electrochemical measuring systems for weathering, corrosion or corrosion-protection measurement
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
    • Y02P90/30—Computing systems specially adapted for manufacturing

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Theoretical Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Biophysics (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Biodiversity & Conservation Biology (AREA)
  • Ecology (AREA)
  • Environmental & Geological Engineering (AREA)
  • Environmental Sciences (AREA)
  • Chemical & Material Sciences (AREA)
  • Analytical Chemistry (AREA)
  • Biochemistry (AREA)
  • Immunology (AREA)
  • Pathology (AREA)
  • Testing Resistance To Weather, Investigating Materials By Mechanical Methods (AREA)

Abstract

The invention provides a neural network prediction method and device for titanium alloy corrosion rate in a fluoride ion environment, and relates to the technical field of metal material corrosion rate prediction. The method comprises the steps of selecting a corrosion current density square value as an intermediate variable based on dimension analysis, fitting a time-corrosion current density square value relation, carrying out numerical integration on the corrosion current density square value along time by using a complex trapezoid product formula, fitting a time-yield strength degradation amount relation to obtain a degradation curve of the yield strength of the titanium alloy along time for engineering service reference, and scanning the curve by using a differential matrix on the basis of the obtained yield strength degradation curve to obtain a relative instantaneous corrosion rate v m. The prediction model provided by the invention has good performance and excellent fitting effect, can be used as a monitoring method for service safety of the titanium alloy material for the underwater vehicle under the condition of fluoride ion corrosion in the deep open sea environment, and supplements the prediction method for corrosion rate of the titanium alloy material for the conventional underwater vehicle.

Description

Neural network prediction method and device for titanium alloy corrosion rate in fluoride ion environment
Technical Field
The invention relates to the technical field of metal material corrosion rate prediction, in particular to a neural network prediction method and device for titanium alloy corrosion rate in a fluoride ion environment.
Background
Cybenko G in 1989, it was demonstrated that the general approximation theorem of neural networks, i.e., single hidden layer feedforward neural networks with a sufficient number of neurons, can approximate a function arbitrarily. The rise of deep learning indicates that compared with a single hidden layer neural network, the method can require a huge number of neurons for high-precision approximation of an objective function, and the approximation capability of the neural network can be effectively improved by increasing the number of hidden layers of the neural network, namely deepening the depth of the neural network, and using relatively fewer neurons. Unlike a multi-layer feed forward network, the radial basis function network has only one hidden layer, and the net input of the hidden layer transfer function is the Euclidean distance between the input vector and the weight matrix, rather than the weighted sum of the weight matrix to the input vector in the feed forward network. A generalized recurrent neural network is a special radial basis network whose hidden layer neurons count is consistent with the total number of input samples, and whose linear output layer has no bias values.
In the prior art, the corrosion experimental data of the titanium alloy material for the submarine aircraft under the condition of fluoride ions in the deep-open sea environment are less, the existing simulation corrosion experimental period is shorter, and an effective fitting and long-term predictable corrosion rate model is lacking for experimental results.
Disclosure of Invention
Aiming at the problems of less corrosion experimental data and shorter period of the existing simulated corrosion experiment in the prior art, and the lack of an effective fitting and long-term predictable corrosion rate model for experimental results, the invention provides a neural network prediction method and device for the corrosion rate of titanium alloy in a fluoride ion environment.
In order to solve the technical problems, the invention provides the following technical scheme:
In one aspect, a neural network prediction method for corrosion rate of a titanium alloy in a fluoride ion environment is provided, and the method is applied to electronic equipment and comprises the following steps:
s1, collecting experimental data of titanium alloy corrosion, and preprocessing the experimental data;
s2, performing segment fitting on the preprocessed data by combining multiple neural networks;
S3, finishing a fitting result of data in the neural network to obtain a titanium alloy yield strength degradation curve;
And S4, scanning the yield strength degradation curve by using a differential matrix on the basis of the yield strength degradation curve to obtain a relative instantaneous corrosion rate v m, and finishing neural network prediction of the corrosion rate of the titanium alloy in the fluoride ion environment.
Optionally, in step S1, experimental data of corrosion of the titanium alloy is collected, and the experimental data is preprocessed, including:
S11, dividing titanium alloy samples into two groups, respectively placing the titanium alloy samples into soaking solutions with different fluoride ion concentrations for soaking corrosion, and dividing each group into seven time gradients;
s12, performing electrochemical parameter measurement on the titanium alloy sample after each gradient soaking corrosion is completed;
S13, calculating a numerical average value of the data of the parallel experimental group, calculating the corrosion current density and further calculating the square value of the corrosion current density by combining the size of the test piece if the corrosion current data are calculated, and cleaning the data of the results obtained by the operation.
Optionally, in S2, the pre-processed data is fitted in segments by using multiple neural networks in combination, and training the neural networks includes:
s21, inputting experimental data of each time gradient after pretreatment by using a plurality of neural networks in a combined way, training the neural networks, and carrying out piecewise fitting on time-corrosion current density square values;
s22, predicting the expected time length of the user Discrete intoInputting the corrosion current into a trained neural network, and outputting a corresponding corrosion current density square value sequenceThe corrosion current density square value sequence is integrated to obtain a corrosion current density square integral value sequence
S23, training the neural network by combining a plurality of neural networks, performing sectional fitting on the data, namely the corrosion current density square integral and the yield strength degradation, and performing pre-fitting by adopting a traditional mathematical method in a time period with insufficient data so as to realize data enhancement and then using the data enhancement for training the neural network.
Optionally, the data segmentation comprises segmenting titanium alloy sample data with different fluoride ion concentrations, wherein the first segment is 0-21 days, and the second segment is 21 days later.
Optionally, in step S21, the experimental data of each time gradient after the preprocessing is input by using a plurality of neural networks in combination, and training the neural networks includes:
for the first segment of data of two groups of concentrations, BP neural network is selected, the maximum training period number epochs =8000, the iteration termination condition is that the maximum absolute value error MAE of the training sample set is less than 10 -15, Wherein isFor the output value of the target it is,K is the sample number and is the actual output value;
selecting a feedforward neural network for the second segment data of the first group, wherein a preset algorithm is a Levenberg-Marquardt algorithm;
And selecting a generalized regression neural network for the second segment data of the second group.
Optionally, the data enhancement includes data enhancement of the second segment of data, i.e., supplementing the data by a pre-fit formula to a time of 2 times the actual experimental days.
Optionally, in step S21, the length of time the user desires to predict is setDiscrete intoInputting the corrosion current into a trained neural network, and outputting a corresponding corrosion current density square value sequenceComprising the following steps:
s211, discrete time sequence Input into trained BP network to obtain outputOnly the first 21 output results are reserved
S212, discrete time sequenceInput into trained generalized regression neural network net_Time_ Jcor2_F8 to obtain output
S213, combining the two output sequences to obtain a corrosion current density square value sequence of a first group of concentration;
and S214, repeating the steps S211-S213 to obtain a second group of corrosion current density square value sequences.
Optionally, in S21, numerically integrating the sequence of square values of the corrosion current density includes numerically integrating the square values of the corrosion current density over time by a complex trapezoidal integration formula.
Optionally, in S4, scanning the yield strength degradation curve by using a differential matrix on the basis of the yield strength degradation curve to obtain a relative instantaneous corrosion rate v m, and completing neural network prediction of the corrosion rate of the titanium alloy in the fluoride ion environment, including:
S41, taking the single-day degradation quantity-delta sigma m of the yield strength of the titanium alloy on any day as the relative instantaneous corrosion rate v m of the day, wherein, δT=1;
S42, using the differential matrix to multiply the degradation curve data sequence obtained in the step S3Obtaining a sequence of relative instantaneous corrosion ratesI.e.And (5) finishing neural network prediction of the corrosion rate of the titanium alloy in the fluoride ion environment.
In one aspect, a neural network prediction apparatus for corrosion rate of titanium alloy in a fluoride ion environment is provided, the apparatus being applied to an electronic device, the apparatus comprising:
The data preprocessing module is used for collecting experimental data of titanium alloy corrosion and preprocessing the experimental data;
the data fitting module is used for carrying out segment fitting on the preprocessed data by combining multiple neural networks;
the curve integrating module is used for finishing fitting results of data in the neural network to obtain a titanium alloy yield strength degradation curve;
And the corrosion rate prediction module scans the yield strength degradation curve by using a differential matrix on the basis of the yield strength degradation curve to obtain a relative instantaneous corrosion rate v m and complete the neural network prediction of the corrosion rate of the titanium alloy in the fluoride ion environment.
In one aspect, an electronic device is provided that includes a processor and a memory having at least one instruction stored therein that is loaded and executed by the processor to implement the neural network prediction method of titanium alloy corrosion rate in a fluoride ion environment described above.
In one aspect, a computer readable storage medium having stored therein at least one instruction loaded and executed by a processor to implement the neural network prediction method of titanium alloy corrosion rate in a fluoride ion environment described above is provided.
The technical scheme provided by the embodiment of the invention has at least the following beneficial effects:
In the scheme, the prediction model provided by the invention has good performance and excellent fitting effect, can be used as a monitoring method for the service safety of the titanium alloy material for the underwater vehicle under the condition of fluoride ion corrosion in the deep-open sea environment, and supplements the prediction method for the corrosion rate of the titanium alloy material for the conventional underwater vehicle. Based on dimensional analysis, selecting a corrosion current density square value as an intermediate variable, fitting a time-corrosion current density square value relationship, performing numerical integration on the corrosion current density square value along time by using a complex trapezoid product formula, fitting a corrosion current density square integral value-yield strength degradation amount relationship, obtaining a degradation curve of the yield strength of the titanium alloy along time for engineering service reference, and scanning the curve by using a differential matrix on the basis of the obtained yield strength degradation curve to finally obtain the relative instantaneous corrosion rate of the titanium alloy in the fluoride ion environment.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly described below, and it is apparent that the drawings in the following description are only some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of a neural network prediction method for titanium alloy corrosion rate in a fluoride ion environment according to an embodiment of the present invention;
FIG. 2 is a flowchart of a neural network prediction method for corrosion rate of titanium alloy in a fluoride ion environment according to an embodiment of the present invention;
FIG. 3 is a graph of experimental sample sizes of a neural network prediction method for titanium alloy corrosion rate in a fluoride ion environment according to an embodiment of the present invention;
FIG. 4 is a graph showing the fitting effect of a model of time-corrosion current density square value under the condition of 4mmol/L fluoride ions in a neural network prediction method of titanium alloy corrosion rate in a fluoride ion environment, which is provided by the embodiment of the invention;
FIG. 5 is a graph showing the fitting effect of a model of the corrosion current density square integral value-yield strength under the condition of 4mmol/L fluoride ion for 0 to 21 days in the neural network prediction method of the titanium alloy corrosion rate in the fluoride ion environment;
FIG. 6 is a graph showing the fitting effect of a model of 'corrosion current density square integral value-yield strength' under the condition of 4mmol/L fluorine ions for 21 to 600 days in a neural network prediction method of titanium alloy corrosion rate in a fluorine ion environment;
FIG. 7 is a graph of the fitting effect of a model of time-corrosion current density square value under 8mmol/L fluoride ion condition in a neural network prediction method of titanium alloy corrosion rate in a fluoride ion environment, which is provided by the embodiment of the invention;
FIG. 8 is a graph showing the fitting effect of a model of the corrosion current density square integral value-yield strength under the condition of 8mmol/L fluoride ion for 0 to 28 days in the neural network prediction method of titanium alloy corrosion rate in a fluoride ion environment, which is provided by the embodiment of the invention;
FIG. 9 is a graph showing the fitting effect of a model of the corrosion current density square integral value-yield strength under the condition of 8mmol/L fluoride ion for 28 to 600 days in a neural network prediction method of titanium alloy corrosion rate in a fluoride ion environment, which is provided by the embodiment of the invention;
FIG. 10 is a graph of relative instantaneous corrosion rates from 21 to 200 days under 4mmol/L fluoride ion conditions for a neural network prediction method of titanium alloy corrosion rates in a fluoride ion environment provided by an embodiment of the present invention;
FIG. 11 is a block diagram of a neural network prediction device for corrosion rate of titanium alloy in a fluoride ion environment according to an embodiment of the present invention;
fig. 12 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
In order to make the technical problems, technical solutions and advantages to be solved more apparent, the following detailed description will be given with reference to the accompanying drawings and specific embodiments.
The embodiment of the invention provides a neural network prediction method for titanium alloy corrosion rate in a fluoride ion environment, which can be realized by electronic equipment, wherein the electronic equipment can be a terminal or a server. A flow chart of a neural network prediction method for corrosion rate of titanium alloy in a fluoride ion environment as shown in fig. 1, the method comprising:
s101, collecting experimental data of titanium alloy corrosion, and preprocessing the experimental data;
s102, performing segment fitting on the preprocessed data by combining multiple neural networks;
s103, finishing a fitting result of data in a neural network to obtain a titanium alloy yield strength degradation curve;
And S104, scanning the yield strength degradation curve by using a differential matrix on the basis of the yield strength degradation curve to obtain a relative instantaneous corrosion rate v m, and finishing neural network prediction of the corrosion rate of the titanium alloy in the fluoride ion environment.
Optionally, in step S101, experimental data of corrosion of the titanium alloy is collected, and the experimental data is preprocessed, including:
s111, dividing titanium alloy samples into two groups, respectively placing the titanium alloy samples into soaking solutions with different fluoride ion concentrations for soaking corrosion, and dividing each group into seven time gradients;
S112, performing electrochemical parameter measurement on the titanium alloy sample after each gradient soaking corrosion is completed;
S113, calculating a numerical average value of the data of the parallel experimental group, calculating the corrosion current density and further calculating the square value of the corrosion current density by combining the size of the test piece if the corrosion current data are calculated, and cleaning the data of the results obtained by the operation.
Optionally, in step S102, the preprocessed data is fitted in segments by using multiple kinds of neural networks in combination, and training the neural networks includes:
S121, inputting experimental data of each time gradient after pretreatment by using a plurality of neural networks in a combined way, training the neural networks, and carrying out piecewise fitting on time-corrosion current density square values;
S122, predicting the expected time length of the user Discrete intoInputting the corrosion current into a trained neural network, and outputting a corresponding corrosion current density square value sequenceThe corrosion current density square value sequence is integrated to obtain a corrosion current density square integral value sequence
S123, training the neural network by combining multiple neural networks, performing sectional fitting on the data, namely the corrosion current density square integral and the yield strength degradation, and performing pre-fitting by adopting a traditional mathematical method in a time period with insufficient data so as to realize data enhancement, and then using the data enhancement for training the neural network.
Optionally, the data segmentation comprises segmenting titanium alloy sample data with different fluoride ion concentrations, wherein the first segment is 0-21 days, and the second segment is 21 days later.
Optionally, in step S121, by using a plurality of neural networks in combination, experimental data of each time gradient after preprocessing is input, the neural networks are trained, including:
for the first segment of data of two groups of concentrations, BP neural network is selected, the maximum training period number epochs =8000, the iteration termination condition is that the maximum absolute value error MAE of the training sample set is less than 10 -15, Wherein isFor the output value of the target it is,K is the sample number and is the actual output value;
selecting a feedforward neural network for the second segment data of the first group, wherein a preset algorithm is a Levenberg-Marquardt algorithm;
And selecting a generalized regression neural network for the second segment data of the second group.
Optionally, the data enhancement includes data enhancement of the second segment of data, i.e., supplementing the data by a pre-fit formula to a time of 2 times the actual experimental days.
Optionally, in step S121, the length of time the user desires to predict is setDiscrete intoInputting the corrosion current into a trained neural network, and outputting a corresponding corrosion current density square value sequenceComprising the following steps:
s1211 discrete time sequence Input into trained BP network to obtain outputOnly the first 21 output results are reserved
S1212, discrete time sequenceInput into trained generalized regression neural network net_Time_ Jcor2_F8 to obtain output
S1213, combining the two output sequences to obtain a corrosion current density square value sequence of a first group of concentration;
s1214, repeating the steps S1211-S1213 to obtain a second set of corrosion current density square value sequences.
Optionally, in S121, numerically integrating the sequence of square values of the corrosion current density includes numerically integrating the square values of the corrosion current density over time by a complex trapezoidal integration formula.
Optionally, in step S104, based on the yield strength degradation curve, scanning the yield strength degradation curve by using a differential matrix to obtain a relative instantaneous corrosion rate v m, and completing neural network prediction of the corrosion rate of the titanium alloy in the fluoride ion environment, including:
s141, taking the single-day degradation quantity-delta sigma m of the yield strength of the titanium alloy on any day as the relative instantaneous corrosion rate v m of the day, wherein, δT=1;
S142, using the degradation curve obtained in the differential matrix multiplication step S103 and the data sequenceObtaining a sequence of relative instantaneous corrosion ratesI.e.And (5) finishing neural network prediction of the corrosion rate of the titanium alloy in the fluoride ion environment.
The prediction model provided by the invention has good performance and excellent fitting effect, can be used as a monitoring method for service safety of the titanium alloy material for the underwater vehicle under the condition of fluoride ion corrosion in the deep open sea environment, and supplements the prediction method for corrosion rate of the titanium alloy material for the conventional underwater vehicle.
The embodiment of the invention provides a neural network prediction method for titanium alloy corrosion rate in a fluoride ion environment, which can be realized by electronic equipment, wherein the electronic equipment can be a terminal or a server. A flow chart of a neural network prediction method for corrosion rate of titanium alloy in a fluoride ion environment as shown in fig. 2, the method comprising:
s201, dividing titanium alloy samples into two groups, respectively placing the titanium alloy samples into soaking solutions with different fluoride ion concentrations for soaking corrosion, and dividing each group into seven time gradients;
S202, performing electrochemical parameter measurement on the titanium alloy sample after each gradient soaking corrosion is completed;
In a feasible implementation mode, the experimental method for titanium alloy corrosion provided by the invention comprises the steps of dividing titanium alloy samples into two groups, namely, 4 mmol.L -1 and 8 mmol.L -1 fluoride ion concentration soaking corrosion, wherein each group comprises 7 time gradient soaking corrosion, the beginning is 0 week in 1 week, one fifth of fresh solution is added into an original electrolyte solution every 40 hours, the soaking corrosion is heated in a water bath to keep the constant temperature at 65 ℃, and data acquisition is carried out, namely, electrochemical parameter measurement is carried out after each gradient soaking corrosion is completed. FIG. 3 is a graph showing the dimensions of the experimental samples.
In a possible implementation mode, titanium alloy samples are divided into two groups, namely 4 mmol.L -1 and 8 mmol.L -1 fluoride ion concentration soaking corrosion, and each group is 7 time gradient soaking corrosion, and the beginning is 0 weeks in 1 week;
Adding one fifth of fresh solution (namely 0.02L of 3.5% NaCl+4mmol.L -1 or 0.02L of 3.5% NaCl+8mmol.L -1) into the original electrolyte solution every 40 hours, soaking in a corrosive water bath, heating at a constant temperature of 65 ℃;
And after the locking time is reached, taking out the sample, firstly flushing the soaking solution on the surface of the sample by deionized water, then drying by a blower, and measuring electrochemical parameters, namely self-corrosion current, self-corrosion potential and open circuit by maintaining the constant temperature of 65 ℃ by using a water bath pot, and recording.
S203, calculating the number average value of the data of the parallel experimental group, if the corrosion current data are calculated, calculating the corrosion current density by combining the test piece size and further calculating the square value of the corrosion current density, and cleaning the data of the results obtained by the operation.
In a feasible implementation mode, the data preprocessing method comprises the steps of calculating a number average value of data of a parallel experiment group, calculating corrosion current density according to the size of a test piece if the corrosion current data is the corrosion current data, further calculating a square value of the corrosion current density, removing abnormal data which does not accord with common sense according to the result obtained by the operation, and simultaneously amplifying or reducing the cleaned data by the same proper multiple to remove the order of magnitude of the data.
In a practical implementation, the data of day 21 measured under the condition of 8 mmol.L -1 of fluoride ion concentration is only 3 groups of parallel samples (4 groups of parallel samples are all measured in other time periods), so that the data of day 21 corresponding to the condition of 4,8 mmol.L -1 cannot be uniformly divided by 3 when calculating an arithmetic average value;
The titanium alloy sample used in the invention is a disk with holes, the diameter D of the disk is 15mm, the diameter D of the holes is 1mm, and the surface area is A=pi/4 (D 2-d2);
The test data corresponding to the two-grade fluoride ion concentrations of 4 mmol.L -1 and 8 mmol.L -1 are all discarded from the 35 th day measurement results (including corrosion current density and yield strength);
The square value of the corrosion current density corresponding to the concentration of 4 mmol.L -1 fluoride ions is multiplied by 10 4 to remove the order of magnitude of the original data, and the square value of the corrosion current density corresponding to the concentration of 8 mmol.L -1 fluoride ions is multiplied by 10 5;
The yield strength data corresponding to each of the two fluoride ion concentrations are multiplied by 10 -2 to remove the order of magnitude.
S204, inputting the preprocessed experimental data of each time gradient by using a plurality of neural networks, training the neural networks, fitting time in a subsection manner, namely the corrosion current density square value, and performing pre-fitting by adopting a traditional mathematical method in a time period with insufficient data so as to realize data enhancement and then using the data enhancement for training the neural networks.
In a possible implementation mode, the method disclosed by the invention is used for combining multiple neural networks, and is used for fitting a time-corrosion current density square value in a segmented mode, wherein time is taken as an input quantity, and the corrosion current density square value is taken as an output quantity, wherein for 4 mmol.L -1 fluoride ions, data on days 0, 7, 14 and 21 are taken as a first segment, 4 layers (hidden layers are multiplied by 2) of custom BP networks are adopted for fitting, data on days 21, 28 and 42 are taken as a second segment, and the following negative exponential functions are adopted for generating supplementary data after the 42 th day: The supplementary data was substituted into the above formula at 2 times the actual experimental days, i.e., taking t=43, 44. At this time, data from day 21 to day 84 was simulated as the second segment using a 4-layer (hidden layer×2) feed-forward neural network (FeedforwardNet).
For 8 mmol.L -1 fluoride ions, the data on days 0, 7, 14 and 21 are used as a first segment, a custom BP neural network containing 4 layers (hidden layer multiplied by 2) is adopted for fitting, the data on days 21, 28 and 42 are used as a second segment, and interpolation and extrapolation values are carried out by adopting the following Gaussian function: Supplementary data were taken at 2 times the actual experimental days, i.e., t=21, 22,..42, 43,..84 was substituted into the above formula. At this time, data from day 21 to day 84 were simulated as the second segment using a generalized recurrent neural network (general regression neural network, GRNN) containing 64 radial basis neurons.
In a possible embodiment, the square value of the corrosion current density gradually degrades over time to a certain fixed non-negative value. To ensure that the neural network learns this trend, training data needs to be supplemented after the 42 th day, otherwise the neural network may predict the change rule after the 42 th day as continuously decreasing (even accelerating to decrease) to below 0 value, and cannot learn the real slow-down rule. By conventional mathematical methods, using negative exponential functions ae -bx +c or rational functionsExtrapolation is performed after day 42 to supplement the training data so that the training set covers the global behavior of the function to be fitted.
In a feasible implementation manner, if the neural network is trained by using only 4 pieces of actual measurement data of '0 th day, 7 th day, 14 th day and 21 th day', even if the fitting degree is higher at the actual measurement data points, the shape of the whole fitting curve is not unique, and due to the lack of actual experimental data as a reference, the fitting curve cannot be judged to be better, and the training of the neural network cannot be guided, so that the data enhancement is realized by using a traditional mathematical fitting method and is used as a reference value of the neural network.
S205, predicting the expected time length of the userDiscrete intoInputting the corrosion current into a trained neural network, and outputting a corresponding corrosion current density square value sequenceThe corrosion current density square value sequence is integrated to obtain a corrosion current density square integral value sequence
In a possible implementation, in step S205, the length of time that the user desires to predict is determinedDiscrete intoInputting the corrosion current into a trained neural network, and outputting a corresponding corrosion current density square value sequenceComprising the following steps:
s251, discrete time sequence Input into trained BP network to obtain outputOnly the first 21 output results are reserved
S252, discrete time sequenceInput into trained generalized regression neural network net_Time_ Jcor2_F8 to obtain output
S253, combining the two output sequences to obtain a corrosion current density square value sequence of a first group of concentration;
and S254, repeating the steps S251-S253 to obtain a second group of corrosion current density square value sequence.
S206, training the neural network by combining a plurality of neural networks, performing sectional fitting on the data, namely the corrosion current density square integral and the yield strength degradation, and performing pre-fitting by adopting a traditional mathematical method in a time period with insufficient data so as to realize data enhancement and then using the data enhancement for training the neural network.
In one possible implementation, for the first data of two sets of concentrations, BP neural network is selected, maximum training cycle number epochs =8000, iteration termination condition is training sample set maximum absolute value error MAE <10 -15,Wherein isFor the output value of the target it is,K is the sample number and is the actual output value;
The self-defined standard full-connection BP network has a network structure of 1-5-3-1. Hidden layer neuron transfer function hyperbolic tangent function Output layer neuron transfer function, linear function y=x, maximum training cycle number, epochs =8000. The training algorithm adopts the steepest descent method, and the initial learning rate is eta=0.01.
The learning rate self-adaptive regulation rule of the invention is as follows:
where n is the training period ordinal number.
In a possible implementation manner, a feedforward neural network is selected for the second segment data of the first group, and the preset algorithm is a Levenberg-Marquardt algorithm;
The feedforward neural network (FeedforwardNet) is named as net_Time_ Jcor _F4, the network structure is 1-5-3-1, and the training algorithm is the Levenberg-Marquardt algorithm:
net_Time_Jcor2_F4.trainParam.epochs=8000
net_Time_Jcor2_F4.trainParam.max_fail=20
net_Time_Jcor2_F4.trainParam.min_grad=10-20
net_Time_Jcor2_F4.trainParam.goal=0
net_Time_Jcor2_F4.trainParam.lr=0.01
net_Time_Jcor2_F4.trainParam.time=inf
In one possible implementation, a generalized regression neural network is selected for the second segment of the second set of data. The generalized regression neural network is expressed as net_Time_ Jcor _F8, the network structure is input-radial basis neuron layer-special linear output layer 1-64-1 radial basis neuron layer with deviation value, the special linear output layer has no deviation value, the number of radial basis neuron layer neurons is consistent with the number of samples, 64 radial basis neurons are obtained, and the radial basis neuron layer transfer function is Gaussian function (one radial basis function) Output layer transfer function, linear function y=x.
In a possible embodiment, the present invention uses a plurality of neural networks in combination, and fits the "corrosion current density square integral-yield strength degradation amount" in segments, so that the obtained corrosion current density square integral value I k is used as an input amount, the yield strength degradation amount- Δσ k is used as an output amount (emphasis is made that the degradation amount "- Δσ24" in the yield strength degradation curve is always referenced to the non-corroded titanium alloy yield strength σ 0 (σ 0 =860 MPa in the present invention), and is different from the single day degradation amount "- δσ" in the relative instantaneous corrosion rate):
For 4 mmol.L -1 fluoride ion, the data from day 0 to day 21 are the first segment, and the yield strength data from day 0, 7, 14 and 21 are interpolated by cubic smooth spline Yield strength degradation amount from day 1 to 21Δσ k＝σ0-σk in the sequence of the integrated value of the square of the corrosion current density obtained in step S205 under the condition that the concentration of fluorine ions is 4 mmol.L -1 For input toTo output reference values, a Radial Basis Function (RBF) network (including 21 radial basis neurons) is selected for approximation, data from day 21 to day 600 are used as the second segment, yield strength data from days 21, 28, 42 are Gaussian fitted and extrapolated to day 600 to obtainThe gaussian function selected is as follows:
Yield strength degradation amount from day 21 to 600 Δσ k＝σ0-σk in the sequence of the integrated value of the square of the corrosion current density obtained in step S205 under the condition that the concentration of fluorine ions is 4 mmol.L -1 For input toTo output the reference values, a 5-layer (hidden layer×3) feedforward neural network is trained.
In one possible implementation, the radial basis function (Radial basis function network, RBF) network is created by creating a network one neuron at a time, i.e., iteratively determining the network structure. And selecting the input vector which is most effective in reducing the network output error to generate a radial basis neuron when selecting the generation, then checking the error of the newly generated network, ending the network creation process if the expected error is not exceeded, otherwise, continuing to add the new neuron until the mean square error of the network reaches a set error target or the network reaches the maximum number of neurons.
Network structure of input-radial basis neuron layer (hidden layer) -linear output layer 1-21-1, radial basis neuron layer and linear output layer with offset value, radial basis neuron layer transfer function of Gaussian function (one of radial basis functions)Output layer transfer function, linear function y=x;
Feedforward neural network, here, the built neural network is noted as net_ Jcor _Sigma_F4, the network structure is 1-10-8-5-1, the training algorithm is the training parameters of the Levenberg-Marquardt algorithm as follows:
net_Jcor2_Sigma_F4.trainParam.epochs=8000
net_Jcor2_Sigma_F4.trainParam.max_fail=20
net_Jcor2_Sigma_F4.trainParam.min_grad=10-20
net_Jcor2_Sigma_F4.trainParam.goal=0
net_Jcor2_Sigma_F4.trainParam.lr=0.01
net_Jcor2_Sigma_F4.trainParam.time=inf
For 8 mmol.L -1 fluoride ion, the data from day 0 to day 28 are the first segment, and the yield strength data from day 0, day 7, day 14, day 21, day 28 are interpolated by cubic smooth spline Yield strength degradation amount from day 1 to 28Δσ k＝σ0-σk by the sequence of the square integral values of the corrosion current density obtained in step S205 under the condition that the concentration of fluorine ions is 8 mmol.L -1 As input, as obtained in step S5212To output reference value, radial Basis Function (RBF) network containing 28 radial basis neurons is selected for approximation, the data from 28 th day to 600 th day is the second segment, and the yield strength data from 28 th day and 42 th day is subjected to the action of regulating factorFitting and extrapolating to the gaussian function of 600 days to obtainSelected to carry a regulatory factorThe gaussian function of (c) is as follows:
wherein k=0.015.
In a possible implementation, the network creation mode of the radial basis function neural network is to create the network by adding one neuron at a time, namely, iteratively determining the network structure. And selecting the input vector which is most effective in reducing the network output error to generate a radial basis neuron when selecting the generation, then checking the error of the newly generated network, ending the network creation process if the expected error is not exceeded, otherwise, continuing to add the new neuron until the mean square error of the network reaches a set error target or the network reaches the maximum number of neurons. Network structure of input-radial basis neuron layer (hidden layer) -linear output layer 1-28-1, radial basis neuron layer and linear output layer with offset values, radial basis neuron layer transfer function of Gaussian function (one of radial basis functions)Output layer transfer function, linear function y=x.
Yield strength degradation amount from day 28 to 600Δσ k＝σ0-σk by the sequence of the square integral values of the corrosion current density obtained in step S205 under the condition that the concentration of fluorine ions is 8 mmol.L -1 For input toFor outputting the reference values, a 5-layer (hidden layer×3) feed-forward neural network (FeedforwardNet) is trained.
In one possible implementation, the feedforward neural network, here denoted as net_ Jcor2_Sigma_F4, has a network structure of 1-10-8-5-1, and the training algorithm, levenberg-Marquardt algorithm, has training parameters as follows:
net_Jcor2_Sigma_F4.trainParam.epochs=8000
net_Jcor2_Sigma_F4.trainParam.max_fail=20
net_Jcor2_Sigma_F4.trainParam.min_grad=10-20
net_Jcor2_Sigma_F4.trainParam.goal=0
net_Jcor2_Sigma_F4.trainParam.lr=0.01
net_Jcor2_Sigma_F4.trainParam.time=inf
In one possible embodiment, the data segmentation includes segmenting titanium alloy sample data of different fluoride ion concentrations, wherein the first segment is 0-21 days and the second segment is 21 days later.
In a possible embodiment, the data enhancement includes data enhancement of the second segment of data, i.e., supplementing the data by a pre-fit formula to a time of 2 times the actual experimental days.
In one possible embodiment, numerically integrating the sequence of square values of the corrosion current density includes numerically integrating the square values of the corrosion current density over time by a complex trapezoidal integration formula.
In the embodiment of the invention, the I cor and the sigma cor do not necessarily have a certain mapping relation in one-to-one correspondence, and even if the approximate relation is fitted from the perspective of pure data, the mapping relation does not necessarily conform to the real physical causal relation. The corrosion current characterizes the transfer of the charge in the electrochemical reaction, which in turn characterizes the increase or decrease in the amount of reactant product species, and the transition of the species tends to cause a change in its intrinsic properties, such as a change in mechanical properties. The present invention therefore contemplates fitting process quantities rather than state quantities, i.e., using the square integral of the corrosion currentAnd the degradation of yield strength in the same time period- Δσ i＝σi-1-σi. For analysis operation such as integration, for realizing by a computer, numerical integration processing is needed, and the invention selects a common trapezoidal numerical integration formulaInstead of
In a possible implementation, the gaussian function has an inflection point (where the second derivative function is 0) and gradually tapers towards 0 over time. The training set should cover the full behavior of the approximated function (similar to the probability statistical interpretation, the data distribution of the training set needs to be close to the pattern true probability distribution), and therefore the training set needs to include inflection points as well as the mitigation situation therein. The inflection point isTo include the situation where the function values in the training set are slowed down over time, suggesting that the selected interval isSelected as [21,600] in the invention
In a possible embodiment, the experimental data for the condition of 8mmol/L of fluoride ion shows that after 21 days there is a short rising process of the square value of the corrosion current density, which gradually decreases after 21 days, unlike the experimental data for the condition of 4mmol/L of fluoride ion, in which the square value of the corrosion current density gradually deteriorates to a certain fixed non-negative value with time evolution. To ensure that the neural network learns this trend of ascending followed by descending, interpolation on days 21 to 42 and extrapolation after day 42 is needed to supplement the training data. If the training data is not interpolated within 21-42 days, although the neural network still can learn the trend of rising and falling, the shape of the obtained fitting curve is more likely to be unstable, and because no reference value exists, the fitting curve cannot be evaluated and selected, so that the training of the neural network is better guided by the data is necessary to be interpolated within 21-42 days, and if the training data is not supplemented by the extrapolation value after 42 days, the neural network can predict the change rule after 42 days as continuously decreasing (even accelerating and decreasing) to be below 0 value, and the real slow-down rule cannot be learned. By a conventional mathematical method, using Gaussian functionsInterpolation (days 21-42) and extrapolation (after day 42) are performed after day 21 to supplement the training data so that the training set covers the global behavior of the function to be fitted.
In a possible implementation manner, if the neural network is trained by using only 5 pieces of measured data of '0 th, 7 th, 14 th, 21 th and 28 th days', even if the fitting degree is higher at the measured data points, the shape of the whole fitting curve is not unique, and due to the lack of actual experimental data as a reference, which fitting curve is better cannot be judged, and the training of the neural network cannot be guided, so that the data enhancement is realized by using a traditional mathematical fitting method and is used as a reference value of the neural network.
In a possible implementation, the gaussian function has an inflection point (where the second derivative function is 0) and gradually tapers off over time. The training set should cover the full behavior of the approximated function (similar to the probability statistical interpretation, the data distribution of the training set needs to be close to the pattern true probability distribution), and therefore the training set needs to include inflection points as well as the mitigation situation therein. The inflection point isTo include the situation where the function values in the training set are slowed down over time, suggesting that the selected interval isSelected as [28,600]
And S207, finishing the fitting result of the data in the neural network to obtain a titanium alloy yield strength degradation curve.
In a possible embodiment, the discrete time sequences are integratedAnd yield strength degradation amount sequenceThe degradation of the yield strength of the titanium alloy- Δσ m can be obtained from the beginning of corrosion (t 0 =0) to any time of day t m. The residual yield strength of the titanium alloy after corrosion in any day can be obtained by using the calculation formula 'sigma m＝σ0-(-Δσm', namely, the degradation curve of the yield strength of the titanium alloy along with time can be drawn
And S208, scanning the yield strength degradation curve by using a differential matrix on the basis of the yield strength degradation curve to obtain a relative instantaneous corrosion rate v m, and completing the neural network prediction of the corrosion rate of the titanium alloy in the fluoride ion environment.
In one possible embodiment, the single day degradation of the yield strength of the titanium alloy on any one day, delta sigma m, is taken as the relative instantaneous corrosion rate v m for that day, wherein,δT=1;
Degradation curve obtained by using differential matrix multiplication and data sequenceObtaining a sequence of relative instantaneous corrosion ratesI.e.
And (5) finishing neural network prediction of the corrosion rate of the titanium alloy in the fluoride ion environment. The fitted effect graph of the present invention, as well as the yield strength degradation curve and the relative instantaneous corrosion rate graph, are shown in fig. 4-10.
In a possible embodiment, "the process of degradation of yield strength of the fitted titanium alloy with time" of the present invention "selects" the square value of corrosion current density "as an intermediate variable is described as follows:
The dimensional analysis is as follows [ sigma SL ] = [ W ] = [ E ] = [ i 2 RT ], wherein sigma is stress, S is area, L is length, W is work, E is energy, i is current, R is resistance, T is time. Both ends are the dimensions of work or energy. Is known to be Where η is resistivity, i=j·s, where J is current density, then there is I.e., [ sigma ] = [ eta ] [ J 2][T]＝[η][J2 T ], so the present invention decides to use the "square value of corrosion current density"As an intermediate variable (corresponding to [ J 2 ] in the dimensional analysis), and integrating the square value of the corrosion current density along the time value (corresponding to [ J 2 T ] in the dimensional analysis), and fitting with the yield strength sigma cor after corrosion, namelyRather than fitting directly to J cor→σcor.
In a feasible implementation mode, the method aims to solve the technical problems that under the condition of fluoride ion corrosion, the corrosion current density and the change rule of the yield strength of the titanium alloy along with time (the minimum time scale delta T is set to be 1 day) are obtained through fitting according to experimental data, and the yield strength degradation amount per unit time (1 day) is used as a predicted relative instantaneous corrosion rate.
According to the method, based on dimension analysis, a corrosion current density square value is selected as an intermediate variable, a time-corrosion current density square value relation is fitted, a complex trapezoid product formula is used for carrying out numerical integration on the corrosion current density square value along time, a corrosion current density square integral value-yield strength degradation amount relation is fitted, and a degradation curve of titanium alloy yield strength along time is finally obtained for engineering service reference.
When the function relation is fitted, the invention fully examines the distribution characteristics of the original experimental data and decides to adopt the piecewise fitting;
Comprehensively considering the requirement on the minimum time scale (1 day), the actual experimental sampling period (7 days) and the trend of current and yield strength data inspired or constrained by priori knowledge of material science over time (the measured data should gradually trend to a very small fixed value along with the infinite extension of corrosion time), the invention firstly uses the traditional mathematical fitting method to realize the data enhancement and supplement necessary training data to cover all the behaviors of the function to be fitted so as to guide and constrain the next neural network training, and adopts different neural network models for different segments of data to ensure the enough accurate fitting degree when the invention combines the multi-layer feedforward neural network, the radial basis neural network and the generalized regression neural network.
In one possible embodiment, considering the actual service use of the material, the time required for the comprehensive material to generate obvious measurable corrosion behavior, and the actual experimental conditions and specific operations of experimental implementation, similar to the existence of the minimum energy unit in quantum mechanics, the corrosion process of the titanium alloy is considered to be reasonable and necessary with "day" as the minimum time scale, i.e. the variation of the mechanical properties of the titanium alloy in the time scale of seconds, milliseconds, microseconds, etc. is not necessarily fitted, as the evidence indicates that the service period of the titanium alloy is in time units of "year", so that if the daily mechanical properties of the titanium alloy can be obtained, the degradation rate of the mechanical properties of the material obtained in time units of "day" can be enough to meet engineering requirements, namely, the service rate is known as the relative instantaneous corrosion rate in the invention, compared with the service process in time units of "year" per se.
FIG. 11 is a block diagram illustrating a neural network prediction device for titanium alloy corrosion rate in a fluoride ion environment, in accordance with an exemplary embodiment. Referring to fig. 11, the apparatus 300 includes:
the data preprocessing module 310 is used for collecting experimental data of titanium alloy corrosion and preprocessing the experimental data;
a data fitting module 320, configured to perform segment fitting on the preprocessed data by using multiple neural networks in combination;
the curve integrating module 330 is used for finishing the fitting result of the data in the neural network to obtain a titanium alloy yield strength degradation curve;
The corrosion rate prediction module 340 scans the yield strength degradation curve by using a differential matrix on the basis of the yield strength degradation curve to obtain a relative instantaneous corrosion rate v k, and completes the neural network prediction of the corrosion rate of the titanium alloy in the fluoride ion environment.
Optionally, the data preprocessing module 310 is further configured to divide the titanium alloy samples into two groups, and respectively put the two groups into soaking solutions with different fluoride ion concentrations for soaking and etching, wherein each group is divided into seven time gradients;
Electrochemical parameter measurement is carried out on the titanium alloy sample after each gradient soaking corrosion is completed;
the method comprises the steps of calculating a numerical average value of data of a parallel experiment group, calculating corrosion current density and further calculating a square value of the corrosion current density by combining the size of a test piece if the corrosion current data are calculated, and cleaning the data of the results obtained by the operation.
Optionally, the data fitting module 320 is further configured to input the preprocessed experimental data of each time gradient by using multiple kinds of neural networks in combination, train the neural networks, and fit time-corrosion current density square values in segments;
The length of time that the user expects to predict Discrete intoInputting the corrosion current into a trained neural network, and outputting a corresponding corrosion current density square value sequenceThe corrosion current density square value sequence is integrated to obtain a corrosion current density square integral value sequence
The method comprises the steps of training a neural network by combining a plurality of neural networks, performing sectional fitting on data, namely the corrosion current density square integral and the yield strength degradation, and performing pre-fitting by adopting a traditional mathematical method in a time period with insufficient data so as to realize data enhancement, and then using the data enhancement in the training of the neural network.
Optionally, the data segmentation comprises segmenting titanium alloy sample data with different fluoride ion concentrations, wherein the first segment is 0-21 days, and the second segment is 21 days later.
Optionally, the data fitting module 320 is further configured to train the neural network by using a plurality of neural networks in combination, and inputting the preprocessed experimental data of each time gradient, including:
for the first segment of data of two groups of concentrations, BP neural network is selected, the maximum training period number epochs =8000, the iteration termination condition is that the maximum absolute value error MAE of the training sample set is less than 10 -15, Wherein isFor the output value of the target it is,K is the sample number and is the actual output value;
selecting a feedforward neural network for the second segment data of the first group, wherein a preset algorithm is a Levenberg-Marquardt algorithm;
And selecting a generalized regression neural network for the second segment data of the second group.
Optionally, the data enhancement includes data enhancement of the second segment of data, i.e., supplementing the data by a pre-fit formula to a time of 2 times the actual experimental days.
Optionally, the data fitting module 320 is further configured to fit a discrete time sequence to the dataInput into trained BP network to obtain outputOnly the first 21 output results are reserved
Will be discrete time seriesInput into trained generalized regression neural network net_Time_ Jcor2_F8 to obtain output
Combining the two output sequences to obtain a corrosion current density square value sequence of a first group of concentration;
repeating the steps to obtain a second group of corrosion current density square value sequences.
Alternatively, numerically integrating the sequence of square values of the corrosion current density includes numerically integrating the square values of the corrosion current density over time by a complex trapezoidal integration formula.
Optionally, the corrosion rate prediction module 340 is further configured to take the single day degradation of the yield strength of the titanium alloy, δσ m, for any day as the relative instantaneous corrosion rate for that day, v m, where,δT=1;
Degradation curve obtained by using differential matrix multiplication and data sequenceObtaining a sequence of relative instantaneous corrosion ratesI.e.
And (5) finishing neural network prediction of the corrosion rate of the titanium alloy in the fluoride ion environment.
In the embodiment of the invention, based on dimensional analysis, the square value of the corrosion current density is selected as an intermediate variable, the relation of time and the square value of the corrosion current density is fitted, the complex trapezoidal product formula is used for carrying out numerical integration on the square value of the corrosion current density along time, then the relation of the square integral value of the corrosion current density and the degradation quantity of yield strength is fitted, a degradation curve of the yield strength of the titanium alloy along time is obtained for engineering service reference, and on the basis of the obtained degradation curve of the yield strength, a differential matrix is used for scanning the curve to finally obtain the relative instantaneous corrosion rate of the titanium alloy in the fluoride ion environment.
Fig. 12 is a schematic structural diagram of an electronic device 400 according to an embodiment of the present invention, where the electronic device 400 may have a relatively large difference due to different configurations or performances, and may include one or more processors (central processing units, CPU) 401 and one or more memories 402, where at least one instruction is stored in the memories 402, and the at least one instruction is loaded and executed by the processors 401 to implement the following steps of a neural network prediction method for a titanium alloy corrosion rate in a fluoride ion environment:
s1, collecting experimental data of titanium alloy corrosion, and preprocessing the experimental data;
s2, performing segment fitting on the preprocessed data by combining multiple neural networks;
S3, finishing a fitting result of data in the neural network to obtain a titanium alloy yield strength degradation curve;
And S4, scanning the yield strength degradation curve by using a differential matrix on the basis of the yield strength degradation curve to obtain a relative instantaneous corrosion rate v m, and finishing neural network prediction of the corrosion rate of the titanium alloy in the fluoride ion environment.
In an exemplary embodiment, a computer readable storage medium, such as a memory, comprising instructions executable by a processor in a terminal to perform the neural network prediction method of titanium alloy corrosion rate in a fluoride ion environment described above is also provided. For example, the computer readable storage medium may be ROM, random Access Memory (RAM), CD-ROM, magnetic tape, floppy disk, optical data storage device, etc.
It will be understood by those skilled in the art that all or part of the steps for implementing the above embodiments may be implemented by hardware, or may be implemented by a program for instructing relevant hardware, where the program may be stored in a computer readable storage medium, and the storage medium may be a read-only memory, a magnetic disk or an optical disk, etc.
The foregoing description of the preferred embodiments of the invention is not intended to limit the invention to the precise form disclosed, and any such modifications, equivalents, and alternatives falling within the spirit and scope of the invention are intended to be included within the scope of the invention.

Claims (8)

1.一种氟离子环境中钛合金腐蚀速率的神经网络预测方法，其特征在于，包括：1. A neural network prediction method for the corrosion rate of titanium alloy in a fluoride ion environment, characterized by comprising: S1：采集钛合金腐蚀实验数据；对所述实验数据进行预处理；S1: Collecting titanium alloy corrosion experimental data; preprocessing the experimental data; 所述步骤S1中，采集钛合金腐蚀实验数据，对所述实验数据进行预处理，包括：In the step S1, titanium alloy corrosion experimental data is collected and preprocessed, including: S11：将钛合金试样分成两组，分别置于不同氟离子浓度的浸泡液中进行浸泡腐蚀，将每组分为七个时间梯度；两组钛合金试样的同一时间梯度，为平行实验组；S11: The titanium alloy specimens were divided into two groups, and were placed in immersion solutions with different fluoride ion concentrations for immersion corrosion. Each group was divided into seven time gradients. The same time gradient of the two groups of titanium alloy specimens was a parallel experimental group. S12：对每个梯度浸泡腐蚀完成后的所述钛合金试样进行电化学参数测量；S12: measuring electrochemical parameters of the titanium alloy sample after each gradient immersion corrosion; S13：对平行实验组的数据求算数平均值；若计算腐蚀电流数据则还需结合试件尺寸求出腐蚀电流密度并进一步求其平方值；对于上述操作所得结果，进行数据清洗；S13: Calculate the arithmetic mean of the data of the parallel experimental groups; if the corrosion current data is calculated, it is also necessary to calculate the corrosion current density in combination with the specimen size and further calculate its square value; for the results obtained from the above operations, perform data cleaning; S2：通过组合使用多种神经网络，对预处理后的数据进行分段拟合；S2: By combining multiple neural networks, the preprocessed data is piecewise fitted; 所述S2中，通过组合使用多种神经网络，对预处理后的数据进行分段拟合，包括：In S2, a plurality of neural networks are combined to perform segmented fitting on the preprocessed data, including: S21:通过组合使用多种神经网络，输入预处理后的各时间梯度的实验数据，对神经网络进行训练，分段拟合时间—腐蚀电流密度平方值；数据不足的时间段内，采用传统数学方法作预拟合以实现数据增强，再用于神经网络训练；S21: By combining and using a variety of neural networks, inputting the pre-processed experimental data of each time gradient, training the neural network, and fitting the time-corrosion current density square value in segments; in the time period with insufficient data, using traditional mathematical methods for pre-fitting to achieve data enhancement, and then using it for neural network training; S22: 将用户期望预测的时间长度，离散为，将其输入已训练好的神经网络，输出相应的腐蚀电流密度平方值序列；对腐蚀电流密度平方值序列作数值积分，获得腐蚀电流密度平方积分值序列；S22: The length of time that the user expects to be predicted , discretized into , input it into the trained neural network, and output the corresponding corrosion current density square value sequence ; Perform numerical integration on the square value sequence of corrosion current density to obtain the square integral value sequence of corrosion current density ; S23: 通过组合使用多种神经网络，对神经网络进行训练；分段拟合腐蚀电流密度平方积分—屈服强度退化量；数据不足的时间段内，采用传统数学方法作预拟合以实现数据增强，再用于神经网络训练；S23: Train the neural network by combining multiple neural networks; segmentally fit the square integral of corrosion current density-yield strength degradation; use traditional mathematical methods for pre-fitting to achieve data enhancement during the period of insufficient data, and then use it for neural network training; S3：整理神经网络中数据的拟合结果，得到钛合金屈服强度退化曲线；S3: Arrange the fitting results of the data in the neural network to obtain the yield strength degradation curve of the titanium alloy; S4：在所述屈服强度退化曲线的基础上，使用差分矩阵扫描所述屈服强度退化曲线，获得相对瞬时腐蚀速率，完成氟离子环境中钛合金腐蚀速率的神经网络预测。S4: Based on the yield strength degradation curve, the yield strength degradation curve is scanned using a difference matrix to obtain a relative instantaneous corrosion rate , completed the neural network prediction of titanium alloy corrosion rate in fluoride ion environment. 2.根据权利要求1所述的方法，其特征在于，数据分段包括：将不同氟离子浓度的钛合金试样数据进行分段，其中，0-21天为第一段，21天以后为第二段。2. The method according to claim 1 is characterized in that data segmentation includes: segmenting the titanium alloy sample data with different fluoride ion concentrations, wherein 0-21 days is the first segment and after 21 days is the second segment. 3.根据权利要求2所述的方法，其特征在于，所述步骤S21中，通过组合使用多种神经网络，输入预处理后的各时间梯度的实验数据，对神经网络进行训练，包括：3. The method according to claim 2, characterized in that in the step S21, a plurality of neural networks are used in combination, and the pre-processed experimental data of each time gradient are input to train the neural network, comprising: 对两组浓度的第一段数据，选用BP神经网络，最大训练周期数，迭代终止条件为训练样本集最大绝对值误差，，其中为为目标输出值，为实际输出值，为样本序号；For the first segment of the two concentration data, BP neural network was selected with a maximum number of training cycles. The iteration termination condition is the maximum absolute error of the training sample set , , where is the target output value, is the actual output value, is the sample serial number; 对第一组的第二段数据，选用前馈神经网络，预设算法为列文伯格-马夸特算法；For the second segment of data in the first group, a feedforward neural network was selected, and the default algorithm was the Levenberg-Marquardt algorithm; 对第二组的第二段数据，选用广义回归神经网络。For the second segment of data in the second group, a generalized regression neural network was selected. 4.根据权利要求3所述的方法，其特征在于，数据增强包括：对第二段数据进行数据增强，即通过预拟合公式补充数据至实际实验天数的2倍时间处。4. The method according to claim 3 is characterized in that data enhancement includes: performing data enhancement on the second segment of data, that is, supplementing the data to twice the actual experimental days through a pre-fitting formula. 5.根据权利要求4所述的方法，其特征在于，所述步骤S21中，将用户期望预测的时间长度，离散为，将其输入已训练好的神经网络，输出相应的腐蚀电流密度平方值序列，包括：5. The method according to claim 4, characterized in that in step S21, the time length that the user expects to be predicted is , discretized into , input it into the trained neural network, and output the corresponding corrosion current density square value sequence ,include: S211：将离散时间序列输入进已训练完成的BP网络中获得输出，只保留前21条输出结果即；S211: Discrete time series Input into the trained BP network to get the output , only keep the first 21 output results, that is ; S212：将离散时间序列输入进已训练完成的广义回归神经网络net_Time_Jcor2_F8中获得输出；S212: Discrete time series Input into the trained generalized regression neural network net_Time_Jcor2_F8 to get the output ; S213：合并上述两个输出序列，得到第一组浓度的腐蚀电流密度平方值序列；S213: merging the above two output sequences to obtain a first group of concentration corrosion current density square value sequences; S214：重复上述步骤S211- S213，得到第二组浓度的腐蚀电流密度平方值序列。S214: Repeat the above steps S211-S213 to obtain a second set of concentrations of corrosion current density square value sequences. 6.根据权利要求5所述的方法，其特征在于，所述S21中，对腐蚀电流密度平方值序列作数值积分，包括：通过复化梯形求积公式对腐蚀电流密度平方值沿时间作数值积分。6. The method according to claim 5 is characterized in that, in said S21, numerically integrating the sequence of square values of corrosion current density comprises: numerically integrating the square values of corrosion current density along time by using a complex trapezoidal quadrature formula. 7.根据权利要求6所述的方法，其特征在于，所述S4中，在所述屈服强度退化曲线的基础上，使用差分矩阵扫描所述屈服强度退化曲线，获得相对瞬时腐蚀速率，完成氟离子环境中钛合金腐蚀速率的神经网络预测，包括：7. The method according to claim 6, characterized in that in S4, based on the yield strength degradation curve, the yield strength degradation curve is scanned using a difference matrix to obtain a relative instantaneous corrosion rate , complete the neural network prediction of titanium alloy corrosion rate in fluoride ion environment, including: S41：以任意一天钛合金屈服强度的单日退化量作为该天的相对瞬时腐蚀速率，其中，，；S41: Single-day degradation of the yield strength of titanium alloy on any given day As the relative instantaneous corrosion rate of the day ,in, , ; S42：使用差分矩阵左乘步骤S3中所得的退化曲线数据序列，得到相对瞬时腐蚀速率序列，即完成氟离子环境中钛合金腐蚀速率的神经网络预测。S42: Use the difference matrix to multiply the degradation curve data sequence obtained in step S3 , the relative instantaneous corrosion rate series is obtained ,Right now Complete the neural network prediction of titanium alloy corrosion rate in fluoride ion environment. 8.一种氟离子环境中钛合金腐蚀速率的神经网络预测装置，其特征在于，所述装置适用于所述权利要求1-7中任一项的氟离子环境中钛合金腐蚀速率的神经网络预测方法，所述装置包括：8. A neural network prediction device for the corrosion rate of titanium alloy in a fluoride ion environment, characterized in that the device is applicable to the neural network prediction method for the corrosion rate of titanium alloy in a fluoride ion environment according to any one of claims 1 to 7, and the device comprises: 数据预处理模块，用于采集钛合金腐蚀实验数据；对所述实验数据进行预处理；A data preprocessing module is used to collect titanium alloy corrosion experimental data and preprocess the experimental data; 数据拟合模块，用于通过组合使用多种神经网络，对预处理后的数据进行分段拟合；A data fitting module is used to perform piecewise fitting on the preprocessed data by combining multiple neural networks; 曲线整合模块，整理神经网络中数据的拟合结果，得到钛合金屈服强度退化曲线；The curve integration module organizes the fitting results of the data in the neural network to obtain the yield strength degradation curve of the titanium alloy; 腐蚀速率预测模块，在所述屈服强度退化曲线的基础上，使用差分矩阵扫描所述屈服强度退化曲线，获得相对瞬时腐蚀速率，完成氟离子环境中钛合金腐蚀速率的神经网络预测。The corrosion rate prediction module uses a differential matrix to scan the yield strength degradation curve based on the yield strength degradation curve to obtain a relative instantaneous corrosion rate. , completed the neural network prediction of titanium alloy corrosion rate in fluoride ion environment.
CN202210295759.1A 2022-03-24 2022-03-24 Neural network prediction method and device for titanium alloy corrosion rate in fluoride ion environment Active CN114739894B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210295759.1A CN114739894B (en) 2022-03-24 2022-03-24 Neural network prediction method and device for titanium alloy corrosion rate in fluoride ion environment

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210295759.1A CN114739894B (en) 2022-03-24 2022-03-24 Neural network prediction method and device for titanium alloy corrosion rate in fluoride ion environment

Publications (2)

Publication Number Publication Date
CN114739894A CN114739894A (en) 2022-07-12
CN114739894B true CN114739894B (en) 2025-04-08

Family

ID=82277957

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210295759.1A Active CN114739894B (en) 2022-03-24 2022-03-24 Neural network prediction method and device for titanium alloy corrosion rate in fluoride ion environment

Country Status (1)

Country Link
CN (1) CN114739894B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115656013B (en) * 2022-09-14 2025-11-21 首钢集团有限公司 Method for detecting preferential corrosion phase in high-aluminum zinc aluminum magnesium coating

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103933611A (en) * 2014-03-27 2014-07-23 同济大学 Preparation method of hydroxyapatite/polylactic acid composite coating on surface of medical magnesium alloy
CN104239659A (en) * 2013-06-07 2014-12-24 中国石油天然气集团公司 Carbon steel corrosion rate prediction method of back propagation (BP) neural network

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP6264311B2 (en) * 2015-03-13 2018-01-24 Ｊｆｅスチール株式会社 Method for predicting corrosion products
CN110782096A (en) * 2019-10-29 2020-02-11 山东科技大学 A foreign exchange time series forecasting method

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104239659A (en) * 2013-06-07 2014-12-24 中国石油天然气集团公司 Carbon steel corrosion rate prediction method of back propagation (BP) neural network
CN103933611A (en) * 2014-03-27 2014-07-23 同济大学 Preparation method of hydroxyapatite/polylactic acid composite coating on surface of medical magnesium alloy

Also Published As

Publication number Publication date
CN114739894A (en) 2022-07-12

Similar Documents

Publication Publication Date Title
CN111260030B (en) A-TCN-based power load prediction method and device, computer equipment and storage medium
CN110083951B (en) Solid insulation life prediction method based on relevant operation data of transformer
CN109033513B (en) Power Transformer Fault Diagnosis Method and Power Transformer Fault Diagnosis Device
CN112288147B (en) A Method of BP-Adaboost Strong Predictor to Predict the Insulation State of Generator Stator
CN103778482A (en) Aquaculture dissolved oxygen short-term prediction method based on multi-scale analysis
CN116680639B (en) A deep learning-based anomaly detection method for deep-sea submersible sensor data
CN117330963A (en) Energy storage power station fault detection methods, systems and equipment
CN114739894B (en) Neural network prediction method and device for titanium alloy corrosion rate in fluoride ion environment
CN115329492A (en) Multi-objective optimization method for stainless steel turning technological parameters based on MLP and variant PSO
CN114154762B (en) Metal corrosion rate prediction method, device, computer equipment and storage medium
CN113486569B (en) A method for predicting the remaining life of centrifugal pumps
CN116303786A (en) Block chain financial big data management system based on multidimensional data fusion algorithm
CN115394380A (en) A Prediction Method of Material-related Parameters Based on Stochastic Degradation Process
CN112032578B (en) Parameter measurement method and equipment for corrosion point in pipeline
CN109614742A (en) An Algorithm for Continuous Prediction of Sea Surface Height
CN114676630B (en) Neural network prediction method and device for residual life of titanium alloy in fluoride ion environment
CN115541641B (en) XRF element quantitative analysis method based on AOG-BP neural network
CN114943360B (en) Neural network prediction photovoltaic short-term power generation method based on improved particle swarm
CN116861256A (en) Furnace temperature prediction method, system, equipment and medium for solid waste incineration process
CN116454880A (en) New energy prediction method based on meta-learning in small sample scene
CN115236149A (en) A method and system for water quality detection based on electrochemical sensor
CN117610397B (en) Intelligent analysis method and device for using effect based on sewage treatment equipment
Akhtari-Goshayeshi et al. Prediction of Corrosion Rate for Carbon Steel in Soil Environment by Artificial Neural Network and Genetic Algorithm
CN120724863B (en) A Probabilistic Fatigue Life Prediction Method for Complex Equipment Based on Physical Information Neural Networks
CN119830717B (en) Life prediction method of centrifugal impeller pump based on LSTM model

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant