Neural Network Prediction Method and Device for Residual Life of Titanium Alloy in Fluoride Ion Environment

Abstract

The present invention discloses a neural network prediction method and device for the remaining life of titanium alloy in a fluoride ion environment, which relates to the technical field of predicting the remaining life of metal materials. The method includes: obtaining the time length to be predicted and discretizing the time length; inputting the discretized time length into the constructed time-corrosion current density squared value model to obtain a sequence of corrosion current density squared values for the time length to be predicted; performing a composite trapezoidal numerical integration on the sequence of corrosion current density squared values to obtain a sequence of corrosion current density squared integral values; inputting the sequence of corrosion current density squared integral values into the constructed corrosion current density squared integral-yield strength degradation amount model to obtain a sequence of yield strength degradation amounts; and further obtaining the remaining life of the titanium alloy for the time length to be predicted. The present invention has good performance and excellent fitting effect, can be used as a monitoring method for the service safety of titanium alloy materials under fluoride ion corrosion conditions, and supplements the existing prediction methods for the remaining life of titanium alloy materials.

Description

Technical Field

[0001] The present invention relates to the technical field of predicting the remaining life of metal materials, and particularly to a neural network prediction method and device for the remaining life of titanium alloy in a fluoride ion environment. Background Art

[0002] At present, there is little corrosion experimental data on titanium alloy materials for submersibles under fluoride ion conditions in the deep sea environment, and the existing simulated corrosion experiment period is short. There is a lack of an effective fitting and long-term predictable remaining life model for the experimental results.

[0003] Cybenko G proved the universal approximation theorem of neural networks in 1989, that is, a single-hidden-layer feedforward neural network with a sufficient number of neurons can approximate any function arbitrarily. The rise of deep learning indicates that, compared with a single-hidden-layer neural network, a relatively large number of neurons may be required to approximate the target function with high precision. By increasing the number of hidden layers of the neural network, that is, deepening the depth of the neural network, a relatively smaller number of neurons can be used to effectively improve the approximation ability of the neural network. Different from the multi-layer feedforward network, the radial basis neural network has only one hidden layer, and the net input of the transfer function of the hidden layer 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 feedforward network. The generalized regression neural network is a special radial basis network, the number of neurons in its hidden layer is the same as the total number of input samples, and its linear output layer does not have a bias value.

[0004] Therefore, how to design a long-term predictable remaining life model for titanium alloy materials for submersibles under fluoride ion conditions in the deep sea environment is an urgent problem to be solved. Summary of the Invention

[0005] In view of the problem in the prior art of how to design a long-term predictable remaining life model for titanium alloy materials for submersibles under fluoride ion conditions in the deep sea environment, the present invention is proposed.

[0006] To solve the above technical problems, the present invention provides the following technical solutions:

[0007] On the one hand, the present invention provides a neural network prediction method for the remaining life of titanium alloy in a fluoride ion environment. This method is implemented by an electronic device, and the method includes:

[0008] S1. Obtain the time length to be predicted Discretize the time length into

[0009] S2. Input the discretized time length into the constructed time-corrosion current density squared value model to obtain a sequence of corrosion current density squared values for the time length to be predicted.

[0010] S3. Perform composite trapezoidal numerical integration on the sequence of squared corrosion current density values to obtain a sequence of integrated squared corrosion current density values.

[0011] S4. Input the sequence of integrated squared corrosion current density values into the constructed integrated squared corrosion current density - yield strength degradation model to obtain a sequence of yield strength degradation amounts.

[0012] S5. Based on the sequence of integrated squared corrosion current density values and the sequence of yield strength degradation amounts, obtain the remaining life of the titanium alloy for the length of time to be predicted.

[0013] Optionally, obtaining the remaining life of the titanium alloy for the length of time to be predicted according to the sequence of integrated squared corrosion current density values and the sequence of yield strength degradation amounts in S5 includes:

[0014] S51. Based on the sequence of integrated squared corrosion current density values and the sequence of yield strength degradation amounts, obtain the degradation curve of the yield strength of the titanium alloy.

[0015] S52. According to the degradation curve, piecewise construct an approximate calculation formula for the distribution of the yield strength of the titanium alloy with respect to the fluoride ion concentration.

[0016] S53. According to the approximate calculation formula, obtain the degradation of the yield strength of the titanium alloy over time.

[0017] S54. Based on the degradation curve and the qualified line required for the yield strength by the service safety of the material, obtain the remaining life of the titanium alloy for the length of time to be predicted.

[0018] Optionally, the construction process of the time - squared corrosion current density model in S2 includes:

[0019] S21. Obtain the experimental data of the titanium alloy corrosion.

[0020] S22. Preprocess the experimental data.

[0021] S23. Based on the preprocessed experimental data, construct a time - squared corrosion current density model.

[0022] Optionally, the time - squared corrosion current density model in S2 includes: a first - concentration first - stage squared - value model, a first - concentration second - stage squared - value model, a second - concentration first - stage squared - value model, and a second - concentration second - stage squared - value model.

[0023] The first - concentration first - stage squared - value model is a first - squared - value backpropagation BP neural network model.

[0024] The first - concentration second - stage squared - value model is a feed - forward neural network model FNN of squared values.

[0025] The square value second concentration first stage model is the second square value backpropagation BP neural network model.

[0026] The square value second concentration second stage model is the generalized regression neural network model GRNN.

[0027] Optionally, the preprocessing of the experimental data in S22 includes:

[0028] Taking the average value of the experimental data, calculating the square value, removing the order of magnitude, and discarding abnormal data.

[0029] Optionally, the corrosion current density square integral - yield strength degradation amount model in S4 includes: the degradation amount first concentration first stage model, the degradation amount first concentration second stage model, the degradation amount second concentration first stage model, and the degradation amount second concentration second stage model.

[0030] The degradation amount first concentration first stage model is the first degradation amount radial basis function neural network model RBF.

[0031] The degradation amount first concentration second stage model is the first degradation amount feedforward neural network model FNN.

[0032] The degradation amount second concentration first stage model is the second degradation amount radial basis function neural network model RBF.

[0033] The degradation amount second concentration second stage model is the second degradation amount feedforward neural network model FNN.

[0034] Optionally, the degradation curve of the yield strength of the titanium alloy in S51 is as shown in the following formula (1):

[0035]

[0036] where σ(t m ) is the Gaussian function of the yield strength of the titanium alloy on the m - th day; σ0 is the yield strength of the uncorroded titanium alloy; -Δσ m is the sequence of the yield strength degradation amount of the titanium alloy.

[0037] Optionally, the approximate calculation formula for the distribution of the yield strength of the titanium alloy with the fluoride ion concentration in S52 is as shown in the following formula (2):

[0038]

[0039] where, is the fluoride ion concentration; is the yield strength of the titanium alloy at time t after the start of corrosion under the condition that the fluoride ion concentration is m ; σ(ρ ref ; tm ) is the yield strength of the titanium alloy at time t after the start of corrosion under the condition of the reference fluoride ion concentration; t m is the mth day; ρ m is the reference fluoride ion concentration, which is 4 mmol·L ref or 8 mmol·L -1 ; σ -1 ; σ(ρ m ref ) is a shorthand notation for σ(ρ ref ; t m ); is the average power-law factor, and the values are as follows under different reference fluoride ion concentration conditions and different time periods:

[0040] 0 - 7 days: ρ ref = 4 mmol·L -1 .

[0041] 7 - 14 days: ρ ref = 4 mmol·L -1 .

[0042] 14 - 21 days: ρ ref = 4 mmol·L -1 .

[0043] 21 - 28 days: ρ ref = 8 mmol·L -1 .

[0044] After 28 days: ρ ref = 4 mmol·L -1 .

[0045] Optionally, obtaining the remaining life of the titanium alloy for the length of the time to be predicted according to the degradation curve and the qualified line required for the service safety of the material for the yield strength includes:

[0046] Determining the qualified line for the service safety of the titanium alloy based on the yield strength.

[0047] According to the time corresponding to the intersection point of the qualified line and the degradation curve, obtaining the maximum allowable service period T s of the titanium alloy.

[0048] According to the maximum allowable service period T s of the titanium alloy and the length of the time to be predicted T cor , obtaining the remaining life T r of the titanium alloy.

[0049] ​On the other hand, the present invention provides a neural network prediction device for the remaining life of a titanium alloy in a fluoride ion environment, which is applied to implement the neural network prediction method for the remaining life of a titanium alloy in a fluoride ion environment. The device includes:

[0050] An acquisition module, configured to acquire the time length to be predicted Discretize the time length into

[0051] A square value sequence module, configured to input the discretized time length into the constructed time-corrosion current density square value model to obtain a sequence of corrosion current density square values for the time length to be predicted.

[0052] A square integral value sequence module, configured to perform composite trapezoidal numerical integration on the sequence of corrosion current density square values to obtain a sequence of corrosion current density square integral values.

[0053] A degradation amount sequence module, configured to input the sequence of corrosion current density square integral values into the constructed corrosion current density square integral-yield strength degradation amount model to obtain a sequence of yield strength degradation amounts.

[0054] An output module, configured to obtain the remaining life of the titanium alloy for the time length to be predicted according to the sequence of corrosion current density square integral values and the sequence of yield strength degradation amounts.

[0055] Optionally, the output module is further configured to:

[0056] S51. Obtain a degradation curve of the yield strength of the titanium alloy according to the sequence of corrosion current density square integral values and the sequence of yield strength degradation amounts.

[0057] S52. Piecewise construct an approximate calculation formula for the distribution of the yield strength of the titanium alloy with respect to the fluoride ion concentration according to the degradation curve.

[0058] S53. Obtain the degradation of the yield strength of the titanium alloy with respect to time according to the approximate calculation formula.

[0059] S54. Obtain the remaining life of the titanium alloy for the time length to be predicted according to the degradation curve and the qualified line required for the yield strength by the service safety of the material.

[0060] Optionally, the square value sequence module is further configured to:

[0061] S21. Obtain the experimental data of the corrosion of the titanium alloy.

[0062] S22. Preprocess the experimental data.

[0063] S23. Construct a time-corrosion current density square value model according to the preprocessed experimental data.

[0064] Optionally, the time-corrosion current density squared value model includes: a first-concentration first-stage squared value model, a first-concentration second-stage squared value model, a second-concentration first-stage squared value model, and a second-concentration second-stage squared value model.

[0065] The first-concentration first-stage squared value model is a first squared value backpropagation BP neural network model.

[0066] The first-concentration second-stage squared value model is a feedforward neural network model FNN for the squared value.

[0067] The second-concentration first-stage squared value model is a second squared value backpropagation BP neural network model.

[0068] The second-concentration second-stage squared value model is a generalized regression neural network model GRNN.

[0069] Optionally, the squared value sequence module is further configured to:

[0070] Take the average of the experimental data, calculate the squared value, remove the order of magnitude, and discard abnormal data.

[0071] Optionally, the corrosion current density squared integral-yield strength degradation model includes: a first-concentration first-stage degradation model, a first-concentration second-stage degradation model, a second-concentration first-stage degradation model, and a second-concentration second-stage degradation model.

[0072] The first-concentration first-stage degradation model is a first degradation radial basis function neural network model RBF.

[0073] The first-concentration second-stage degradation model is a first degradation feedforward neural network model FNN.

[0074] The second-concentration first-stage degradation model is a second degradation radial basis function neural network model RBF.

[0075] The second-concentration second-stage degradation model is a second degradation feedforward neural network model FNN.

[0076] Optionally, the degradation curve of the yield strength of the titanium alloy is as shown in the following formula (1):

[0077]

[0078] where σ(t m ) is the Gaussian function of the yield strength of the titanium alloy on the mth day; σ0 is the yield strength of the uncorroded titanium alloy; -Δσ m is the sequence of the yield strength degradation amount of the titanium alloy.

[0079] Optionally, the approximate calculation formula for the yield strength of the titanium alloy with respect to the fluoride ion concentration distribution is shown in the following formula (2):

[0080]

[0081] Wherein, is the fluoride ion concentration; is the yield strength of the titanium alloy at time t after corrosion starts under the condition that the fluoride ion concentration is ; σ(ρ m ; t ref ) is the yield strength of the titanium alloy at time t after corrosion starts under the condition of the reference fluoride ion concentration; t m m m is the m-th day; ρ ref is the reference fluoride ion concentration, which is 4 mmol·L -1 or 8 mmol·L -1 ; σ m (ρ ref ) is a shorthand notation for σ(ρ ref ; t m ); is the average power-law factor, and its values are as follows under different reference fluoride ion concentration conditions and different time periods:

[0082] 0 - 7 days: ρ ref = 4 mmol·L -1 .

[0083] 7 - 14 days: ρ ref = 4 mmol·L -1 .

[0084] 14 - 21 days: ρ ref = 4 mmol·L -1 .

[0085] 21 - 28 days: ρ ref = 8 mmol·L -1 .

[0086] After 28 days: ρ ref = 4 mmol·L -1 .

[0087] Optionally, the output module is further configured to:

[0088]

[0089] Determine the pass line for the service safety of the titanium alloy based on the yield strength.

[0089] According to the time corresponding to the intersection point of the qualified line and the degradation curve, the maximum allowable service period T of the titanium alloy is obtained. s .

[0090] According to the maximum allowable service period T of the titanium alloy s and the length of the time to be predicted T cor , the remaining life T of the titanium alloy is obtained. r .

[0091] On the one hand, an electronic device is provided. The electronic device includes a processor and a memory. At least one instruction is stored in the memory, and the at least one instruction is loaded and executed by the processor to implement the above-mentioned neural network prediction method for the remaining life of titanium alloy in a fluoride ion environment.

[0092] On the one hand, a computer-readable storage medium is provided. At least one instruction is stored in the storage medium, and the at least one instruction is loaded and executed by a processor to implement the above-mentioned neural network prediction method for the remaining life of titanium alloy in a fluoride ion environment.

[0093] The beneficial effects brought by the technical solutions provided in the embodiments of the present invention at least include:

[0094] In the above solution, based on dimensional analysis, the square value of the corrosion current density is selected as an intermediate variable. First, the relationship between "time - square value of the corrosion current density" is fitted, then the composite trapezoidal quadrature formula is used to numerically integrate the square value of the corrosion current density along time, and then the relationship between "integral value of the square of the corrosion current density - degradation amount of yield strength" is fitted. Finally, the degradation curve of the yield strength of the titanium alloy with time is obtained for engineering service reference.

[0095] The prediction model designed by the present invention has good performance and excellent fitting effect. It can be used as a monitoring method for the service safety of titanium alloy materials used in submersibles under fluoride ion corrosion conditions in the deep and far sea environment, provide a reference value for the maximum allowable service period, and supplement the existing prediction methods for the remaining life of titanium alloy materials used in submersibles.

[0096] Under the condition of fluoride ion corrosion, the present invention fits the change laws of the corrosion current density and yield strength of the titanium alloy with time (the minimum time scale δT is set to 1 day) based on experimental data, so as to realize the prediction of the remaining life of the titanium alloy; in addition, the present invention further estimates the influence of different fluoride ion concentrations on the degradation degree of the yield strength of the titanium alloy. Description of the Drawings

[0097] To more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the accompanying drawings required for the description of the embodiments. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can be obtained based on these drawings.

[0098] Figure 1 It is a schematic flowchart of the neural network prediction method for the remaining life of titanium alloy in a fluoride ion environment provided by an embodiment of the present invention;

[0099] Figure 2 It is a drawing of the experimental specimen size provided by an embodiment of the present invention;

[0100] Figure 3 It is the fitting effect diagram of the square value model from 0 to 21 days under the condition of 4 mmol / L fluoride ions provided by an embodiment of the present invention;

[0101] Figure 4 It is the fitting effect diagram of the square value model from 21 to 84 days under the condition of 4 mmol / L fluoride ions provided by an embodiment of the present invention;

[0102] Figure 5 It is the fitting effect diagram of the square value model from 0 to 21 days under the condition of 8 mmol / L fluoride ions provided by an embodiment of the present invention;

[0103] Figure 6 It is the fitting effect diagram of the square value model from 21 to 84 days under the condition of 8 mmol / L fluoride ions provided by an embodiment of the present invention;

[0104] Figure 7 It is the fitting effect diagram of the degradation amount model from 0 to 21 days under the condition of 4 mmol / L fluoride ions provided by an embodiment of the present invention;

[0105] Figure 8 It is the fitting effect diagram of the degradation amount model from 21 to 84 days under the condition of 4 mmol / L fluoride ions provided by an embodiment of the present invention;

[0106] Figure 9 It is the fitting effect diagram of the degradation amount model from 0 to 21 days under the condition of 8 mmol / L fluoride ions provided by an embodiment of the present invention;

[0107] Figure 10 It is the fitting effect diagram of the degradation amount model from 21 to 84 days under the condition of 8 mmol / L fluoride ions provided by an embodiment of the present invention;

[0108] Figure 11 It is the degradation curve of the yield strength from 0 to 21 days under the condition of 4 mmol / L provided by an embodiment of the present invention;

[0109] Figure 12It is the degradation curve of the yield strength under the condition of 4 mmol / L provided by the embodiment of the present invention for 21 - 600 days;

[0110] Figure 13 It is the degradation curve of the yield strength under the condition of 8 mmol / L provided by the embodiment of the present invention for 0 - 28 days;

[0111] Figure 14 It is the degradation curve of the yield strength under the condition of 8 mmol / L provided by the embodiment of the present invention for 28 - 600 days;

[0112] Figure 15 It is the prediction diagram of the remaining life of the titanium alloy in the 65°C 4 mmol / L fluoride ion corrosion environment provided by the embodiment of the present invention;

[0113] Figure 16 It is the schematic diagram for predicting the influence of different fluoride ion concentrations on the yield strength of the titanium alloy provided by the embodiment of the present invention;

[0114] Figure 17 It is the block diagram of the neural network prediction device for the remaining life of the titanium alloy in the fluoride ion environment provided by the embodiment of the present invention;

[0115] Figure 18 It is the structural schematic diagram of an electronic device provided by the embodiment of the present invention. Detailed implementation manners

[0116] To make the technical problems, technical solutions and advantages to be solved by the present invention clearer, the following will be described in detail with reference to the accompanying drawings and specific embodiments.

[0117] As Figure 1 shown, the embodiment of the present invention provides a neural network prediction method for the remaining life of a titanium alloy in a fluoride ion environment. This method can be implemented by a blockchain management node, and the blockchain management node can be a terminal or a server. As Figure 1 shown in the flowchart of the neural network prediction method for the remaining life of a titanium alloy in a fluoride ion environment, the processing flow of this method can include the following steps:

[0118] S1. Obtain the time length to be predicted Discretize the time length into

[0119] S2. Input the discretized time length into the constructed time - corrosion current density squared value model to obtain the corrosion current density squared value sequence for the time length to be predicted.

[0120] In a feasible implementation manner, the time length (unit: days) that the user expects to predict is discretized into and then input into the trained neural network (note that the time length When it is relatively large, it may be necessary to input it in segments into the corresponding neural network model), and output the corresponding "square value of corrosion current density" sequence

[0121] Optionally, the construction process of the time-corrosion current density square value model in S2 includes:

[0122] S21. Obtain experimental data on the corrosion of titanium alloy.

[0123] For example, obtaining experimental data on the corrosion of titanium alloy can be to divide the titanium alloy specimens into two groups, which are 4 mmol·L -1 ) and 8 mmol·L -1 immersion corrosion with fluoride ion concentration, and each group is divided into 7 time gradients for immersion corrosion, with 1 week as the unit, starting from 0 weeks;

[0124] Add one-fifth of the fresh solution to the original electrolyte solution every 40 hours, that is: 0.02 L of 3.5% NaCl + 4 mmol·L -1 or 0.02 L of 3.5% NaCl + 8 mmol·L -1 , and the immersion corrosion is heated in a water bath to a constant temperature of 65 degrees Celsius.

[0125] After reaching the locking time, take out the specimens, first rinse the immersion solution on the surface of the specimens with deionized water, then dry them with a hair dryer, and keep the constant temperature of 65 degrees Celsius with a water bath to measure the electrochemical parameters: self-corrosion current, self-corrosion potential, and open-circuit potential, and make records.

[0126] S22. Preprocess the experimental data.

[0127] Optionally, the preprocessing of the experimental data in S22 includes:

[0128] Perform operations such as taking the average value, calculating the square value, removing the order of magnitude, and discarding abnormal data on the experimental data.

[0129] In a feasible implementation, the data preprocessing can be to calculate the arithmetic mean of the data in the parallel experimental groups. If it is corrosion current data, the corrosion current density needs to be calculated in combination with the specimen size and then its square value is further calculated; for the results obtained from the above operations, remove the abnormal data that does not conform to common sense; simultaneously magnify or reduce the cleaned data by the same appropriate multiple to remove the order of magnitude of the data.

[0130] For example, the preprocessing of the experimental data can be that the data measured on the 21st day under the condition of fluoride ion concentration of 8 mmol·L -1 only has 3 parallel samples (the data measured at other time periods all have 4 parallel samples), so when calculating the arithmetic mean, it cannot be uniformly divided by 4, 8 mmol·L-1 The data on the 21st day corresponding to the condition should be divided by 3.

[0131] As Figure 2 shown, the titanium alloy specimen used in the present invention is in the shape of a perforated disc, the disc diameter D = 15 mm, the hole diameter d = 1 mm, and the surface area taken is

[0132] 4 mmol·L -1 and 8 mmol·L -1 For the experimental data corresponding to the two levels of fluoride ion concentration, the measurement results on the 35th day (including corrosion current density and yield strength) are discarded.

[0133] 4 mmol·L -1 The squared value of the corrosion current density corresponding to the fluoride ion concentration needs to be multiplied by 10 4 to remove the original data order of magnitude, while for the squared value of the corrosion current density corresponding to the 8 mmol·L -1 fluoride ion concentration, it needs to be multiplied by 10 5 .

[0134] The yield strength data corresponding to the two levels of fluoride ion concentration are both multiplied by 10 -2 to remove the order of magnitude.

[0135] S23. According to the preprocessed experimental data, construct a time-corrosion current density squared value model.

[0136] Optionally, the time-corrosion current density squared value model in S2 includes: a first-concentration first-stage squared value model, a first-concentration second-stage squared value model, a second-concentration first-stage squared value model, and a second-concentration second-stage squared value model.

[0137] The first-concentration first-stage squared value model is a first squared value backpropagation BP neural network model.

[0138] The first-concentration second-stage squared value model is a feedforward neural network model FNN for the squared value.

[0139] The second-concentration first-stage squared value model is a second squared value backpropagation BP neural network model.

[0140] The second-concentration second-stage squared value model is a generalized regression neural network model GRNN.

[0141] In a feasible implementation manner, multiple neural networks are combined and used to fit the "time-corrosion current density squared value" in segments. In the time period with insufficient data, traditional mathematical methods are used for "pre-fitting" to achieve data enhancement, and then it is used for neural network training. Using time as the input quantity and the squared value of the corrosion current density as the output quantity, it may specifically include:

[0142] 1. 4 mmol·L -1 Fluoride ion:

[0143] 11. The data on the 0th, 7th, 14th, and 21st days are the first segment, and a custom BP (BackPropagation) network with 4 layers (hidden layer × 2) is used for fitting.

[0144] 12. The data on the 21st, 28th, and 42nd days are the second segment, and supplementary data are generated using a negative exponential function after the 42nd day, as shown in the following formula (1):

[0145]

[0146] The supplementary data are extended to twice the actual experimental days, that is, t = 43, 44,..., 84 are substituted into the above formula.

[0147] At this time, the data from the 21st day to the 84th day are used as the second segment, and a 4-layer (hidden layer × 2) feedforward neural network is used for simulation.

[0148] Furthermore, the square value of the corrosion current density gradually degrades to a certain fixed non-negative value over time. To ensure that the neural network learns this changing trend, supplementary training data are needed after the 42nd day. Otherwise, it may cause the neural network to predict that the changing pattern after the 42nd day continuously decreases (even accelerates to decrease) below 0, and fails to learn the true slowing-down pattern. Using traditional mathematical methods, a negative exponential function ae -bx + c or a rational function p / q + x is used for extrapolation after the 42nd day to supplement the training data, so that the training set covers the global behavior of the function to be fitted.

[0149] 2. 8 mmol·L -1 Fluoride ion:

[0150] 21. The data on the 0th, 7th, 14th, and 21st days are the first segment, and a custom BP neural network with 4 layers (hidden layer × 2) is used for fitting.

[0151] 22. The data on the 21st, 28th, and 42nd days are the second segment, and Gaussian functions are used for interpolation and extrapolation, as shown in the following formula (2):

[0152]

[0153] The supplementary data are extended to twice the actual experimental days, that is, t = 21, 22,..., 42, 43,..., 84 are substituted into the above formula.

[0154] At this time, the data from the 21st day to the 84th day is taken as the second segment and simulated using a GRNN (General Regression Neural Network) with 64 radial basis neurons.

[0155] Furthermore, if the neural network is trained only using the 4 measured data points of "the 0th, 7th, 14th, and 21st days", even if the fitting degree is high at the measured data points, the shape of the entire fitting curve is not unique. Due to the lack of actual experimental data for reference, it is impossible to judge which fitting curve is better, and thus impossible to guide the training of the neural network. Therefore, by means of traditional mathematical fitting methods, data augmentation is achieved as a reference value for the neural network.

[0156] Furthermore, the following explanation is given for selecting the "square value of the corrosion current density" as an intermediate variable for "fitting the degradation process of the yield strength of titanium alloy over time in the present invention":

[0157] The dimensional analysis is as follows:

[0158] [σSL] = [W] = [E] = [i 2 RT]

[0159] where σ: stress; S: area; L: length; W: work; E: energy; i: current; R: resistance; T: time. Both ends are in the dimension of work or energy.

[0160] It is known that where η is the resistivity; i = J·S, where J is the current density.

[0161] So there is That is, [σ] = [η][J 2 [T] = [η][J 2 T].

[0162] Therefore, the present invention decides to use the "square value of the corrosion current density " as the intermediate variable (corresponding to [J 2 in the dimensional analysis), and after integrating the square value of the corrosion current density along the time value (corresponding to [J 2 T] in the dimensional analysis), it is then fitted with the yield strength after corrosion, that is instead of directly fitting J cor →σ cor .

[0163] For example, for the piecewise fitting of "time - square value of the corrosion current density", the specific settings of the structural parameters and training parameters of the neural network used in each segment are as follows:

[0164] 1. The first - stage model of the square - value first - concentration is the first square - value back - propagation BP neural network model.

[0165] As Figure 3 shown, fitting [F - = 4 mmol·L -1 under the condition of the "time - corrosion current density square - value" data from 0 to 21 days:

[0166] The present invention customizes a standard fully - connected BP network.

[0167] Network structure: 1 - 5 - 3 - 1.

[0168] Transfer function of hidden - layer neurons: hyperbolic tangent function

[0169] Transfer function of output - layer neurons: linear function y = x.

[0170] Maximum number of training epochs: epochs = 8000.

[0171] Iteration termination condition: maximum absolute - value error MAE of the training sample set < 10 -15 .

[0172]

[0173] where is the target output value, the actual output value, and k is the sample serial number.

[0174] Training algorithm: steepest - descent method.

[0175] Initial learning rate: η = 0.01.

[0176] Learning - rate self - adaptive adjustment rule:

[0177] η (n+1) = η (n) ·0.85, if MAE (n) > MAE (n-1)

[0178] η (n+1) = η (n) ·1.005, if MAE (n) ≤MAE (n-1)

[0179] where n is the training - epoch ordinal number.

[0180] 2. The second - stage model of the square - value first - concentration is a square - value FNN (Feedforward Neural Network, feed - forward neural network model).

[0181] As shown Figure 4 in the figure, the data of "time - square value of corrosion current density" from 21 to 84 days under the fitting condition of - [F -1 = 4 mmol·L

[0182] Feed - forward neural network. The neural network established here is denoted as net_Time_Jcor2_F4:

[0183] Network structure: 1 - 5 - 3 - 1.

[0184] Training algorithm: Levenberg - Marquardt algorithm.

[0185] net_Time_Jcor2_F4.trainParam.epochs = 8000

[0186] net_Time_Jcor2_F4.trainParam.max_fail = 20

[0187] net_Time_Jcor2_F4.trainParam.min_grad = 10 -20

[0188] net_Time_Jcor2_F4.trainParam.goal = 0

[0189] net_Time_Jcor2_F4.trainParam.lr = 0.01

[0190] net_Time_Jcor2_F4.trainParam.time = inf

[0191] 3. The first - stage model of the second concentration of the square value is the second - square - value BP (Backpropagation) neural network model.

[0192] As Figure 5 shown in the figure, the data of "time - square value of corrosion current density" from 0 to 21 days under the fitting condition of - [F -1 = 8 mmol·L

[0193] The present invention customizes a standard fully - connected BP network.

[0194] Network structure: 1 - 5 - 3 - 1.

[0195] Activation function of neurons in the hidden layer: hyperbolic tangent function

[0196] Activation function of neurons in the output layer: linear function y = x.

[0197] Maximum number of training epochs: epochs = 8000.

[0198] Iteration termination condition: maximum absolute error of the training sample set MAE < 10 -15 .

[0199]

[0200] Among them, is the target output value, is the actual output value, and k is the sample serial number.

[0201] Training algorithm: steepest descent method.

[0202] Initial learning rate: η = 0.01.

[0203] Learning rate adaptive adjustment rule:

[0204]

[0205] Among them, n is the training epoch ordinal number.

[0206] 4. The second concentration second stage model of the squared value is GRNN (General Regression Neural Network).

[0207] As Figure 6 shown, fitting [F - = 8 mmol·L -1 under the condition of the "time - squared corrosion current density" data from 21 to 84 days:

[0208] General regression neural network, the neural network built here is denoted as net_Time_Jcor2_F8:

[0209] Network structure: input - radial basis neuron layer - special linear output layer 1 - 64 - 1.

[0210] The radial basis neuron layer has a bias value, and the special linear output layer has no bias value.

[0211] The number of neurons in the radial basis neuron layer is the same as the number of samples, which is 64 radial basis neurons here.

[0212] Transfer function of the radial basis neuron layer: Gaussian function (a type of radial basis function)

[0213] Transfer function of the output layer: linear function y = x.

[0214] Under the condition of a fluoride ion concentration of 8 mmol / L, obtain the sequence of "square values of corrosion current density from day 0 to day 28" The specific operation is as follows:

[0215] First, the discrete time series needs to be input into the custom BP network of the present invention that has been trained to obtain the output Then, only the first 21 output results are retained, that is

[0216] Secondly, the discrete time series is input into the trained generalized regression neural network net_Time_Jcor2_F8 to obtain the output

[0217] Finally, the above two output sequences are combined.

[0218] S3. Perform complex trapezoidal numerical integration on the sequence of square values of corrosion current density to obtain the sequence of integrated values of square of corrosion current density.

[0219] In a feasible implementation manner, the sequence of "square values of corrosion current density" output in step S2 is subjected to complex trapezoidal numerical integration to obtain the sequence of integrated values of square of corrosion current density

[0220] The numerical integration formula is as shown in formula (3) below:

[0221]

[0222] where δT = 1

[0223] S4. Input the sequence of integrated values of square of corrosion current density into the constructed model of integrated value of square of corrosion current density - yield strength degradation amount to obtain the sequence of yield strength degradation amount.

[0224] Optionally, the model of integrated value of square of corrosion current density - yield strength degradation amount in S4 includes: the first-stage model of degradation amount at the first concentration, the second-stage model of degradation amount at the first concentration, the first-stage model of degradation amount at the second concentration, and the second-stage model of degradation amount at the second concentration.

[0225] The first-stage model of degradation amount at the first concentration is the first degradation amount radial basis neural network model RBF.

[0226] The second-stage model of degradation amount at the first concentration is the first degradation amount feedforward neural network model FNN.

[0227] The first-stage model of degradation amount at the second concentration is the second degradation amount radial basis neural network model RBF.

[0228] The second concentration second-stage model of the degradation amount is the second degradation amount feedforward neural network model FNN.

[0229] In a feasible implementation, multiple neural networks are used in combination to piecewise fit the "square integral of corrosion current density - yield strength degradation amount", and the obtained square integral value I of the corrosion current density k is used as the input quantity, and the yield strength degradation amount -Δσ k is used as the output quantity (emphasis, the degradation amount "-Δσ" in the yield strength degradation curve is always referenced to the yield strength σ0 of the uncorroded titanium alloy (in this invention, σ0 = 860 MPa is taken), which is different from the single-day degradation amount "-δσ" in the relative instantaneous corrosion rate). Specifically as follows:

[0230] S41. 4 mmol·L -1 Fluoride ion:

[0231] S411. The data from day 0 to day 21 is the first segment.

[0232] S4111. Cubic smoothing spline interpolation is performed on the yield strength data of days 0, 7, 14, and 21 to obtain

[0233] Furthermore, the Gaussian function has an inflection point (where the second derivative function is 0), and gradually approaches 0 with the evolution of time. The training set should cover all the behaviors of the function to be approximated (similar to the explanation in probability statistics, the data distribution of the training set needs to be close to the true probability distribution of the pattern). Therefore, the training set needs to include the inflection point and the trend of slowing down. The inflection point is To include the trend of the function value slowing down with time in the training set,

[0234] It is recommended to select the interval as In this invention, it is selected as [21, 600]

[0235] S4112. The yield strength degradation amount from day 1 to day 21

[0236] S4113. Using the sequence of square integral values of the corrosion current density obtained under the condition of a fluoride ion concentration of 4 mmol·L -1 as the input, and using the obtained in step S4112 as the output reference value, a radial basis neural network (RBF) containing 21 radial basis neurons is selected for approximation.

[0237] S412. The data from day 21 to day 600 is the second segment.

[0238] S4121. Gaussian function fitting is performed on the yield strength data of days 21, 28, and 42 and extrapolated to day 600 to obtain The selected Gaussian function is shown in the following formula (4):

[0239]

[0240] Furthermore, different from the experimental data of "fluoride ion concentration of 4 mmol / L", the experimental data under the condition of fluoride ion concentration of 8 mmol / L show that after 21 days, there is still a short rising process in the square value of the corrosion current density, and then it gradually decreases; during the decreasing process, the square value of the corrosion current density gradually degrades to a certain fixed non-negative value over time. To ensure that the neural network learns this trend of rising first and then falling, it is necessary to interpolate within the 21st to 42nd days and extrapolate after the 42nd day to supplement the training data. If the training data is not interpolated within 21 - 42 days, although the neural network can still learn the trend of rising first and then falling, there are many possible shapes of the obtained fitting curve and it is not fixed. Without a reference value, it is impossible to evaluate and select, so it is necessary to perform interpolation within 21 - 42 days to enhance the data and better guide the training of the neural network; if the training data is not extrapolated after the 42nd day, it may cause the neural network to predict the change law after the 42nd day as continuously decreasing (even accelerating to decrease) below 0, and fail to learn the true slowdown law. Using traditional mathematical methods, with the Gaussian function Perform interpolation (21 - 42 days) and extrapolation (after 42 days) after the 21st day to supplement the training data, so that the training set covers the global behavior of the function to be fitted.

[0241] S4122. Yield strength degradation from the 21st to 600th day

[0242] S4123. Using the sequence of the integrated square value of the corrosion current density obtained under the condition of fluoride ion concentration of 4 mmol·L -1 as the input, and using the obtained in step S4122 as the output reference value to train a 5 - layer (hidden layer × 3) feed - forward neural network.

[0243] S42. 8 mmol·L -1 fluoride ion:

[0244] S421. The data from the 0th to 28th day is the first segment.

[0245] S4211. Perform cubic smoothing spline interpolation on the yield strength data of the 0th, 7th, 14th, 21st, and 28th days to obtain

[0246] Furthermore, if the neural network is trained using only 5 measured data points on "Day 0, 7, 14, 21, and 28", even if the fitting degree is high at the measured data points, the shape of the entire fitting curve is not unique. Without actual experimental data for reference, it is impossible to judge which fitting curve is better, and thus impossible to guide the training of the neural network. Therefore, by using traditional mathematical fitting methods, data augmentation is achieved as a reference value for the neural network.

[0247] S4212. Yield strength degradation from Day 1 to 28

[0248] S4213. Using the sequence of integrated square values of corrosion current density obtained under the condition of -1 fluoride ion concentration of 8 mmol·L as the input, and using the obtained in step S4212 as the output reference value, a radial basis neural network (RBF) with 28 radial basis neurons is selected for approximation.

[0249] S422. The data from Day 28 to 600 is the second segment.

[0250] S4221. Perform Gaussian function fitting with an adjustment factor on the yield strength data of Day 28 and 42 and extrapolate to Day 600 to obtain the selected Gaussian function with an adjustment factor as shown in the following formula (5):

[0251]

[0252] where K = 0.015.

[0253] Furthermore, the Gaussian function has an inflection point (where the second derivative function is 0), and it gradually slows down over time. The training set should cover all the behaviors of the function to be approximated (similar to the interpretation in probability statistics, the data distribution of the training set needs to be close to the true probability distribution of the pattern). Therefore, the training set needs to include the inflection point and the slowing-down trend. The inflection point is To include the trend of the function value slowing down over time in the training set.

[0254] It is recommended to select the interval as In the present invention, it is selected as [28, 600].

[0255] S4222. Yield strength degradation from Day 28 to 600

[0256] S4223. Using the sequence of integrated square values of corrosion current density obtained under the condition of fluoride ion concentration of 8 mmol·L -1 as the input, and as the input, and use the one obtained in step S4222 as the output reference value for training a five-layer (3 hidden layers) feedforward neural network.

[0257] Furthermore, the above steps S3 and S4 are supplemented and explained as follows:

[0258] I cor and σ cor do not necessarily have a one-to-one mapping relationship. Even if an approximate relationship is fitted from a pure data perspective, it may not conform to the true physical causal relationship. The corrosion current characterizes the transfer of electric charge in an electrochemical reaction and thus reflects the increase or decrease in the amount of substances of reactants and products. The transformation of substances will inevitably cause changes in their intrinsic properties, such as changes in mechanical properties. Therefore, the present invention considers fitting the process quantity rather than the state quantity, that is, using the square integral of the corrosion current and the degradation amount of the yield strength -Δσ within the same time period i = σ i-1 -σ i for fitting. For analytical operations such as integration, in order to be implemented by a computer, numerical integration processing is required. The present invention selects the commonly used trapezoidal numerical integration formula to replace

[0259] For example, 1. The degradation amount first concentration first stage model is the first RBF (Radial Basis Function) neural network model.

[0260] As Figure 7 shown, fit the data of "square integral of corrosion current density - yield strength degradation amount" from 0 to 21 days under the condition of [F - = 4 mmol·L -1 :

[0261] Radial basis neural network.

[0262] Network creation method: The network is created by adding one neuron at a time, that is, the network structure is determined iteratively. Each time during iteration, the input vector that is most effective in reducing the network output error is selected to generate a radial basis neuron. Subsequently, the error of the newly generated network is checked. If it does not exceed the expected error, the network creation process ends; otherwise, new neurons are continuously added until the mean square error of the network reaches the set error target or the network reaches the maximum number of neurons.

[0263] Network structure: Input - Radial basis neuron layer (hidden layer) - Linear output layer 1 - 21 - 1.

[0264] Both the radial basis neuron layer and the linear output layer have bias values.

[0265] Transfer function of the radial basis neuron layer: Gaussian function (a type of radial basis function)

[0266] Transfer function of the output layer: linear function y = x.

[0267] 2. The first concentration second stage model of the degradation amount is the first degradation amount feedforward neural network model.

[0268] As Figure 8 shown, fitting [F - = 4 mmol·L -1 under the condition, the data of "square integral of corrosion current density - degradation amount of yield strength" from 21 to 600 days:[[]]

[0269] Feedforward neural network, and the neural network established here is denoted as net_Jcor2_Sigma_F4:

[0270] Network structure: 1 - 10 - 8 - 5 - 1.

[0271] Training algorithm: Levenberg - Marquardt algorithm.

[0272] The training parameters are as follows:

[0273] net_Jcor2_Sigma_F4.trainParam.epochs = 8000

[0274] net_Jcor2_Sigma_F4.trainParam.max_fail = 20

[0275] net_Jcor2_Sigma_F4.trainParam.min_grad = 10 -20

[0276] net_Jcor2_Sigma_F4.trainParam.goal = 0

[0277] net_Jcor2_Sigma_F4.trainParam.lr = 0.01

[0278] net_Jcor2_Sigma_F4.trainParam.time = inf

[0279] 3. The first concentration first stage model of the degradation amount is the second radial basis neural network model.

[0280] As Figure 9 shown, fitting [F - = 8 mmol·L -1Data of "Integral of Square of Corrosion Current Density - Degradation of Yield Strength" from 0 to 28 days under certain conditions:

[0281] Radial basis neural network.

[0282] Network creation method: The network is created by adding one neuron each time, that is, the network structure is determined iteratively. Each time during iteration, the input vector that is most effective in reducing the network output error is selected to generate a radial basis neuron. Subsequently, the error of the newly generated network is checked. If it does not exceed the expected error, the network creation process ends; otherwise, new neurons are continuously added until the mean square error of the network reaches the set error target or the network reaches the maximum number of neurons.

[0283] Network structure: Input - Radial Basis Neuron Layer (Hidden Layer) - Linear Output Layer 1 - 28 - 1.

[0284] Both the radial basis neuron layer and the linear output layer have bias values.

[0285] Transfer function of the radial basis neuron layer: Gaussian function (a type of radial basis function)

[0286] Transfer function of the output layer: Linear function y = x.

[0287] 4. The second - concentration second - stage model of the degradation amount is a feed - forward neural network model for the second degradation amount.

[0288] Such as Figure 10 shown, fitting [F - = 8 mmol·L -1 Data of "Integral of Square of Corrosion Current Density - Degradation of Yield Strength" from 28 to 600 days under certain conditions:

[0289] Feed - forward neural network, and the neural network established here is denoted as net_Jcor2_Sigma_F8:

[0290] Network structure: 1 - 10 - 8 - 5 - 1.

[0291] Training algorithm: Levenberg - Marquardt algorithm.

[0292] Training parameters are as follows:

[0293] net_Jcor2_Sigma_F8.trainParam.epochs = 8000

[0294] net_Jcor2_Sigma_F8.trainParam.max_fail = 20

[0295] net_Jcor2_Sigma_F8.trainParam.min_grad = 10 -20

[0296] net_Jcor2_Sigma_F8.trainParam.goal = 0

[0297] net_Jcor2_Sigma_F8.trainParam.lr = 0.01

[0298] net_Jcor2_Sigma_F8.trainParam.time = inf

[0299] S5. Obtain the remaining life of the titanium alloy with the to-be-predicted time length according to the sequence of the integral values of the square of the corrosion current density and the sequence of the yield strength degradation amounts.

[0300] Optionally, obtaining the remaining life of the titanium alloy with the to-be-predicted time length according to the sequence of the integral values of the square of the corrosion current density and the sequence of the yield strength degradation amounts in S5 includes:

[0301] S51. Obtain the degradation curve of the yield strength of the titanium alloy according to the sequence of the integral values of the square of the corrosion current density and the sequence of the yield strength degradation amounts.

[0302] Optionally, the degradation curve of the yield strength of the titanium alloy in S51 is as shown in the following formula (6):

[0303]

[0304] wherein, σ(t m ) is the Gaussian function of the yield strength of the titanium alloy on the m-th day; σ0 is the yield strength of the uncorroded titanium alloy; -Δσ m is the sequence of the yield strength degradation amounts of the titanium alloy.

[0305] In a feasible implementation manner, by integrating the discrete time series in step S3 and the sequence of the yield strength degradation amounts in step S4 , the degradation amount -Δσ of the yield strength of the titanium alloy from the start of corrosion (t0 = 0) to any day t m during this period can be obtained. Using the calculation formula "σ m = σ0 - (-Δσ m )", the remaining yield strength of the titanium alloy on any day after corrosion can be obtained, and thus the degradation curve of the yield strength of the titanium alloy with respect to time can be plotted

[0306] Figures 11 - 14 For example, the degradation curve is as Figures 11 - 14 shown.

[0307] S52. According to the degradation curve, approximately calculate the formula for the distribution of the yield strength of titanium alloy with respect to the fluoride ion concentration in segments.

[0308] Optionally, the approximate calculation formula for the distribution of the yield strength of titanium alloy with respect to the fluoride ion concentration in S52 is as shown in the following formula (7):

[0309]

[0310] where is the fluoride ion concentration; is the yield strength of the titanium alloy at time t after the start of corrosion under the condition that the fluoride ion concentration is ; σ(ρ m ; t ref ; t m ) is the yield strength of the titanium alloy at time t after the start of corrosion under the condition of the reference fluoride ion concentration; t m is the m-th day; ρ m is the reference fluoride ion concentration, which is 4 mmol·L ref or 8 mmol·L -1 or 8 mmol·L -1 ; σ m (ρ ref ) is a shorthand notation for σ(ρ ref ; t m ); is the average power-law factor, and under different reference fluoride ion concentration conditions and different time periods, the values are as shown in Table 1 below:

[0311] Table 1

[0312] Time period Reference fluoride ion concentration Average power law factor 0-7 <![CDATA[4 mmol·L -1 > 0.02 7-14 <![CDATA[4 mmol·L -1 > 0.001 14-21 <![CDATA[4 mmol·L -1 > 0.1 21-28 <![CDATA[8 mmol·L -1 > 0.45732 After 28 days <![CDATA[4 mmol·L -1 > 0.015

[0313] At the same moment, the yield strength shows a power-law distribution with respect to the fluoride ion concentration; construct an extrapolation format in segments, and the power exponent K(t c ) takes the average value

[0314] In a feasible implementation manner, the neural network of the present invention fits the experimental data under the conditions of 4 mmol·L -1 and 8 mmol·L -1 fluoride ion concentrations. To obtain the yield strength degradation curve of the titanium alloy under other fluoride ion concentration conditions, this approximate calculation formula needs to be used.

[0315] That is, the usage steps are as follows: First, use the neural network model in the present invention to obtain the experimental data under 4 mmol·L -1 and 8 mmol·L -1The yield strength degradation curve of the titanium alloy under certain conditions is obtained, and then this formula is used to calculate the yield strength degradation curves of the titanium alloy under the remaining fluoride ion concentration conditions.

[0316] S53. According to the approximate calculation formula, the degradation of the yield strength of the titanium alloy over time is obtained.

[0317] In a feasible implementation, from steps S1 to S4, the yield strength degradation curves of the titanium alloy at any time length under the conditions of fluoride ion concentration of 4 mmol / L -1 or 8 mmol / L -1 are obtained; on this basis, using the formula in step S52, the degradation of the yield strength of the titanium alloy over time under any fluoride ion concentration conditions can be further estimated.

[0318] S54. According to the degradation curve and the pass line required for the yield strength by the service safety of the material, the remaining life of the titanium alloy for the to-be-predicted time length is obtained.

[0319] Optionally, obtaining the remaining life of the titanium alloy for the to-be-predicted time length according to the degradation curve and the pass line required for the yield strength by the service safety of the material in S54 includes:

[0320] Determine the pass line for the service safety of the titanium alloy based on the yield strength.

[0321] According to the time corresponding to the intersection point of the pass line and the degradation curve, the maximum allowable service period T of the titanium alloy is obtained s .

[0322] According to the maximum allowable service period T of the titanium alloy s and the to-be-predicted time length T cor , the remaining life T of the titanium alloy is obtained r .

[0323] In a feasible implementation, according to the actual engineering needs, the pass line required for the yield strength by the service safety of the material is determined; combined with the yield strength degradation curve of the titanium alloy obtained in step S51, the time corresponding to the intersection point of the two is the maximum allowable service period T for continuous immersion in a solution with a certain fluoride ion concentration s , subtracting the already served time length T cor , the remaining life T is obtained r .

[0324] For example, such as Figure 15 , 16As shown, the present invention is mainly used for monitoring the service status of titanium alloy materials for submersibles in the deep - sea environment. Taking a submarine as an example for prediction: Looking back at the history of submarine development, the steel used in submarines before 1940 was low - carbon steel with a yield strength of 220 MPa. Therefore, in this example, the present invention selects 220 MPa as the qualified line for material service safety. Under different fluoride ion concentration conditions, the maximum allowable service period of the titanium alloy predicted by the present invention is shown in Table 2 below:

[0325] Table 2

[0326]

[0327]

[0328] In a feasible implementation manner, the present invention comprehensively considers the requirement for the minimum time scale (1 day) in the problem to be solved, the actual experimental sampling period (7 days), and the trends of current and yield strength data over time inspired or constrained by prior knowledge of materials science (as the corrosion time extends infinitely, the measured data should gradually slow down and tend to a very small fixed value). First, the present invention uses traditional mathematical fitting methods to achieve data enhancement and supplement necessary training data to cover all the behaviors of the function to be fitted, and then guides and constrains the next - step neural network training.

[0329] When performing piece - wise fitting, different neural network models are used for different segments of data to ensure a sufficiently accurate fitting degree. The present invention combines the use of multi - layer feed - forward neural networks, radial basis neural networks, and generalized regression neural networks.

[0330] Finally, based on the common - sense constraint that "as the fluoride ion concentration increases, the yield strength of the titanium alloy after the same corrosion duration should gradually decrease to a very small non - negative value", the present invention uses the negative power - exponential function to estimate the degradation curve of the yield strength of the titanium alloy under other fluoride ion concentration conditions that are not measured in the experiment, with the degradation curve of the yield strength fitted and predicted by the above - mentioned neural network as the benchmark.

[0331] In the embodiment of the present invention, based on dimensional analysis, the square value of the corrosion current density is selected as an intermediate variable. First, the relationship between "time - square value of the corrosion current density" is fitted, then the composite trapezoidal quadrature formula is used to numerically integrate the square value of the corrosion current density over time, and then the relationship between "integrated value of the square of the corrosion current density - degradation amount of the yield strength" is fitted. Finally, the degradation curve of the yield strength of the titanium alloy over time is obtained for engineering service reference.

[0332] The prediction model designed by the present invention has good performance and excellent fitting effect. It can be used as a monitoring method for the service safety of titanium alloy materials for submersibles under fluoride ion corrosion conditions in the deep - sea environment, providing a reference value for the maximum allowable service period and supplementing the existing prediction methods for the remaining life of titanium alloy materials for submersibles.

[0333] Under the condition of fluoride ion corrosion, according to the experimental data fitting, the invention obtains the variation laws of the corrosion current density and yield strength of titanium alloy with time (the minimum time scale δT is set to 1 day), so as to realize the prediction of the remaining life of titanium alloy; in addition, the invention further estimates the influence of different fluoride ion concentrations on the degradation degree of the yield strength of titanium alloy.

[0334] As Figure 17 shown, an embodiment of the invention provides a neural network prediction device 1700 for the remaining life of titanium alloy in a fluoride ion environment. The device 1700 is applied to implement the neural network prediction method for the remaining life of titanium alloy in a fluoride ion environment. The device 1700 includes:

[0335] An acquisition module 1710, configured to acquire the time length to be predicted Discretize the time length into

[0336] A squared value sequence module 1720, configured to input the discretized time length into the constructed time-corrosion current density squared value model to obtain a corrosion current density squared value sequence of the time length to be predicted.

[0337] A squared integral value sequence module 1730, configured to perform a composite trapezoidal numerical integration on the corrosion current density squared value sequence to obtain a corrosion current density squared integral value sequence.

[0338] A degradation amount sequence module 1740, configured to input the corrosion current density squared integral value sequence into the constructed corrosion current density squared integral-yield strength degradation amount model to obtain a yield strength degradation amount sequence.

[0339] An output module 1750, configured to obtain the remaining life of the titanium alloy of the time length to be predicted according to the corrosion current density squared integral value sequence and the yield strength degradation amount sequence.

[0340] Figure 18 is a schematic structural diagram of an electronic device 1800 provided by an embodiment of the invention. The electronic device 1800 may vary greatly due to different configurations or performances, and may include one or more processors (central processing units, CPUs) 1801 and one or more memories 1802. Among them, at least one instruction is stored in the memory 1802, and at least one instruction is loaded and executed by the processor 1801 to implement the above-mentioned neural network prediction method for the remaining life of titanium alloy in a fluoride ion environment.

[0341] In an exemplary embodiment, a computer-readable storage medium is also provided, such as a memory including instructions that can be executed by a processor in a terminal to complete the above neural network prediction method for the remaining life of a titanium alloy in a fluoride ion environment. For example, the computer-readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, magnetic tape, a floppy disk, and an optical data storage device, etc.

[0342] Those of ordinary skill in the art can understand that all or part of the steps to implement the above embodiments can be completed by hardware or by a program instructing relevant hardware. The program can be stored in a computer-readable storage medium. The storage medium mentioned above can be a read-only memory, a magnetic disk, or an optical disc, etc.

[0343] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included in the protection scope of the present invention.

Claims

1. A neural network prediction method for the remaining life of titanium alloy in a fluoride ion environment, characterized in that The method includes: S1. Obtain the time length to be predicted Discretize the time length into S2. Input the discrete time length into the constructed time - corrosion current density squared value model to obtain the corrosion current density squared value sequence for the time length to be predicted; S3. Perform complex trapezoidal numerical integration on the corrosion current density squared value sequence to obtain the corrosion current density squared integral value sequence; S4. Input the corrosion current density squared integral value sequence into the constructed corrosion current density squared integral - yield strength degradation amount model to obtain the yield strength degradation amount sequence; S5. Obtain the remaining life of the titanium alloy for the time length to be predicted based on the corrosion current density squared integral value sequence and the yield strength degradation amount sequence; The obtaining of the remaining life of the titanium alloy for the time length to be predicted according to the corrosion current density squared integral value sequence and the yield strength degradation amount sequence in S5 includes: S51. Obtain the degradation curve of the yield strength of the titanium alloy based on the corrosion current density squared integral value sequence and the yield strength degradation amount sequence; S52. Piecewise construct an approximate calculation formula for the distribution of the yield strength of the titanium alloy with respect to the fluoride ion concentration according to the degradation curve; S53. Obtain the degradation situation of the yield strength of the titanium alloy with respect to time according to the approximate calculation formula; S54. Obtain the remaining life of the titanium alloy for the time length to be predicted based on the degradation curve and the qualified line required for the yield strength by the service safety of the material.

2. The method according to claim 1, characterized in that, The construction process of the time - corrosion current density squared value model in S2 includes: S21. Obtain the experimental data of titanium alloy corrosion; S22. Preprocess the experimental data; S23. Construct the time - corrosion current density squared value model according to the preprocessed experimental data.

3. The method according to claim 2, wherein The time - corrosion current density squared value model in S2 includes: the squared value first concentration first stage model, the squared value first concentration second stage model, the squared value second concentration first stage model, and the squared value second concentration second stage model; The squared value first concentration first stage model is the first squared value backpropagation BP neural network model; The squared value first concentration second stage model is the squared value feedforward neural network model FNN; The squared value second concentration first stage model is the second squared value backpropagation BP neural network model; The squared value second concentration second stage model is the generalized regression neural network model GRNN.

4. The method according to claim 2, wherein The preprocessing of the experimental data in S22 includes: Performing averaging, calculating the squared value, removing the order of magnitude, and discarding abnormal data processing on the experimental data.

5. The method according to claim 1, characterized in that, The corrosion current density squared integral - yield strength degradation amount model in S4 includes: the degradation amount first concentration first stage model, the degradation amount first concentration second stage model, the degradation amount second concentration first stage model, and the degradation amount second concentration second stage model; The degradation amount first concentration first stage model is the first degradation amount radial basis function neural network model RBF; The degradation amount first concentration second stage model is the first degradation amount feedforward neural network model FNN; The degradation amount second concentration first stage model is the second degradation amount radial basis function neural network model RBF; The second concentration and second stage model of the degradation amount is the second degradation amount feedforward neural network model FNN.

6. The method according to claim 1, wherein The degradation curve of the yield strength of the titanium alloy in S51 is shown by the following formula (1): Among them, σ(t m ) is the Gaussian function of the yield strength of titanium alloy on the m-th day; σ0 is the yield strength of the uncorroded titanium alloy; -Δσ m is the sequence of the degradation amount of the yield strength of titanium alloy.

7. The method according to claim 1, wherein The approximate calculation formula of the yield strength of the titanium alloy varying with the fluoride ion concentration in S52 is shown by the following formula (2): in, is the fluoride ion concentration; The fluoride ion concentration is After corrosion begins under the conditions of m Yield strength of titanium alloy at time; σ(ρ ref ;t m ) is the time after corrosion starts under the standard fluoride ion concentration condition. m Yield strength of titanium alloy at time t m is the mth day; ρ ref is the baseline fluoride ion concentration, which is 4mmol·L -1 or 8mmol·L -1 ; σ m (ρ ref ) is σ(ρ ref ;t m ) is the average power law factor. Under different baseline fluoride ion concentration conditions and in different time periods, the values are as follows: 0 - 7 days: ρ ref = 4 mmol·L -1 ; 7 - 14 days: ρ ref = 4 mmol·L -1 ; 14 - 21 days: ρ ref = 4 mmol·L -1 ; 21 - 28 days: ρ ref = 8 mmol·L -1 ; After 28 days: ρ ref = 4 mmol·L -1 .

8. The method according to claim 1, characterized in that The remaining life of the titanium alloy for the predicted time length obtained according to the qualified line required for the service safety of the material based on the yield strength from the degradation curve includes: Determine the qualified line for the service safety of the titanium alloy based on the yield strength; Based on the time corresponding to the intersection point of the qualified line and the degradation curve, the maximum allowable service period T of the titanium alloy is obtained s ; According to the maximum allowable service life T of the titanium alloy s and the length of the time period T to be predicted cor , the remaining life T of the titanium alloy is obtained r .

9. A neural network prediction device for the remaining life of a titanium alloy in a fluoride ion environment, characterized in that, The device includes: An acquisition module for acquiring the time length to be predicted Discretize the time length into A squared value sequence module, configured to input the discretized time length into the constructed time-corrosion current density squared value model to obtain a corrosion current density squared value sequence for the predicted time length; A squared integral value sequence module, configured to perform a composite trapezoidal numerical integration on the corrosion current density squared value sequence to obtain a corrosion current density squared integral value sequence; A degradation amount sequence module, configured to input the corrosion current density squared integral value sequence into the constructed corrosion current density squared integral-yield strength degradation amount model to obtain a yield strength degradation amount sequence; An output module, configured to obtain the remaining life of the titanium alloy for the predicted time length according to the corrosion current density squared integral value sequence and the yield strength degradation amount sequence; The obtaining of the remaining life of the titanium alloy for the predicted time length according to the corrosion current density squared integral value sequence and the yield strength degradation amount sequence includes: S51. Obtain the degradation curve of the yield strength of the titanium alloy according to the corrosion current density squared integral value sequence and the yield strength degradation amount sequence; S52. Piecewise construct an approximate calculation formula for the yield strength of the titanium alloy varying with the fluoride ion concentration according to the degradation curve; S53. Obtain the degradation of the yield strength of the titanium alloy over time according to the approximate calculation formula; S54. Obtain the remaining life of the titanium alloy for the predicted time length according to the degradation curve and the qualified line required for the service safety of the material for the yield strength.

Images