A method and system for predicting the life of steel wear-resistant materials based on machine learning

Abstract

The present invention discloses a life prediction method and system for steel wear-resistant materials based on machine learning, which relates to the technical field of material detection, including collecting historical operation data of steel wear-resistant materials, normalizing the historical operation data to obtain normalized characteristic data, and dividing the normalized characteristic data into a data set in proportion; constructing a first model based on the data set; constructing a second model, using the prediction error of the first model under different parameter combinations on the data set as input, and training the second model; updating the first model according to the optimal parameter combination output by the second model, inputting the data set into the updated first model, and obtaining the life prediction result of the steel wear-resistant material. The present invention obtains the optimal parameter combination by mutual optimization of the two models to improve the accuracy and adaptability of the prediction, so that the life prediction result has more practical application value.

Description

Technical Field

[0001] The present invention relates to the technical field of material detection, and in particular to a method and system for predicting the life of steel wear-resistant materials based on machine learning. Background Art

[0002] With the rapid growth of industrial automation and equipment operation and maintenance management needs, the life prediction of steel wear-resistant materials has gradually become an important topic in the heavy industry field. Steel wear-resistant materials are widely used in high-wear working conditions such as metallurgy, mining, and construction, and their wear resistance is directly related to the stability and production efficiency of the equipment. Traditional life prediction methods mostly rely on material wear mechanism analysis and empirical formulas based on single working conditions. The service life of the material is usually estimated by the wear rate or load curve under specific working conditions. However, these methods cannot fully reflect the complex wear behavior of materials in actual applications. Especially under multi-variable and complex working conditions, the wear process of wear-resistant materials will be affected by multiple factors such as working conditions, material composition, temperature, load, speed, etc. Traditional methods often find it difficult to accurately handle the nonlinear relationship between these multiple factors, resulting in large errors in life prediction and limited applicability.

[0003] Another limitation of existing life prediction technology is the poor flexibility and adaptability of the model. Traditional methods usually use fixed formula models, lack dynamic optimization capabilities, and cannot make real-time adjustments to the wear behavior of materials under different working conditions. For example, models based on linear regression or simple statistical analysis methods usually assume that the relationship between variables is fixed and ignore the dynamic variability of parameters in complex environments. This greatly reduces the accuracy of the prediction results when the material composition or working conditions change. In addition, existing prediction methods often show poor robustness when dealing with insufficient data and noise interference, resulting in limited reliability of prediction results under some key working conditions. These technical deficiencies limit the promotion of life prediction of steel wear-resistant materials in actual industrial applications and affect the scientificity and efficiency of equipment operation and maintenance. Summary of the invention

[0004] In view of the problems existing in the existing technology, the present invention proposes a method and system for predicting the life of steel wear-resistant materials based on machine learning.

[0005] Therefore, the problem to be solved by the present invention is that the existing steel wear life prediction technology lacks dynamic optimization capability.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, an embodiment of the present invention provides a method for predicting the life of a steel wear-resistant material based on machine learning, which comprises:

[0008] Collecting historical operation data of steel wear-resistant materials, normalizing the historical operation data to obtain normalized characteristic data, and dividing the normalized characteristic data into data sets in proportion;

[0009] Based on the data set, construct a first model;

[0010] Construct a second model, and use the prediction errors of the first model under different parameter combinations on the data set as input to train the second model;

[0011] The first model is updated according to the optimal parameter combination output by the second model, and the data set is input into the updated first model to obtain the life prediction result of the steel wear-resistant material.

[0012] As a preferred solution of the life prediction method of steel wear-resistant materials based on machine learning described in the present invention, wherein: the historical operation data includes material composition data, operating temperature data, stress data and life data;

[0013] The material composition data includes carbon content percentage, chromium content percentage, manganese content percentage, silicon content percentage, phosphorus content percentage and sulfur content percentage;

[0014] The working temperature data includes the highest temperature value, the lowest temperature value and the average temperature value during the working process;

[0015] The stress data include tensile stress, compressive stress and shear stress data under actual working conditions;

[0016] The data set includes a training data set, a validation data set and a test data set.

[0017] As a preferred solution of the method for predicting the life of steel wear-resistant materials based on machine learning described in the present invention, the construction of the first model includes:

[0018] Constructing an initial prediction function; the basic prediction function maps input data to a prediction result based on a weight vector and a bias term;

[0019] Selecting a kernel function according to the dimensional characteristics of the training data set;

[0020] Setting a penalty factor according to the noise level of the training data set to balance the error of the first model;

[0021] Based on the accuracy requirement of the life prediction of the steel wear-resistant material, the number of support vectors is controlled by using the epsilon parameter;

[0022] Constructing a loss function of the first model, and converting the support vector regression problem into a constrained optimization problem through the loss function;

[0023] The constrained optimization problem is decomposed into two-dimensional sub-problems, and support vectors and corresponding Lagrange multipliers are obtained through iterative optimization to obtain the first model.

[0024] As a preferred solution of the method for predicting the life of steel wear-resistant materials based on machine learning of the present invention, wherein: the setting of the penalty factor to balance the error of the first model includes:

[0025] When the noise of the training data set is large, the smaller The value makes the model tolerant to training errors, while when the noise of the training dataset is small, a larger The value makes the model fit the training data, thus establishing the association between the penalty factor and the data characteristics;

[0026] When the standard deviation of the training data exceeds a first preset value of the mean, it is determined that the noise of the training data set is large;

[0027] When the standard deviation of the training data is less than a second preset value of the mean, it is determined that the noise of the training data set is small;

[0028] Among them, the optimization function of the balanced first model is as follows: in, is the weight vector, and is the slack variable, is the norm of the weight vector, is the penalty factor that controls the model's sensitivity to training errors.

[0029] As a preferred solution of the steel wear-resistant material life prediction method based on machine learning of the present invention, wherein: the use of the epsilon parameter to control the number of support vectors includes:

[0030] When the deviation between the predicted value and the true value is less than the value of the epsilon parameter, no loss is calculated, forming a soft interval constraint on the prediction error;

[0031] The soft margin constraint is as follows: in, is the bias term, is the true value of the i-th training sample, is the tolerance band width, is the characteristic value of the i-th training sample.

[0032] As a preferred solution of the method for predicting the life of steel wear-resistant materials based on machine learning of the present invention, the second model is trained according to the feature vector; the construction of the feature vector includes:

[0033] Constructing a parameter combination of the first model, for each parameter combination, using the each parameter combination to update the first model, and using the validation data set to calculate a prediction result;

[0034] evaluating the prediction results and forming the feature vector;

[0035] The prediction results are evaluated based on prediction accuracy loss, model complexity loss, and support vector ratio loss.

[0036] As a preferred solution of the method for predicting the life of steel wear-resistant materials based on machine learning described in the present invention, the calculation of the prediction accuracy loss is shown in the following formula: in, is the prediction accuracy loss, is the sample size, is the true value, is the predicted value, is the prediction function The partial derivative of is the partial derivative of the prediction function with respect to C, is the prediction function The partial derivative of

[0037] The calculation of the model complexity loss is shown as follows: in, is the norm of the weight vector, is the initial penalty factor value, is the absolute value of the Lagrange multiplier, is the complexity loss;

[0038] The support vector ratio loss is calculated as follows: in, is the number of support vectors, is the expected support vector ratio, is the initial tolerance band width, is the support vector ratio loss.

[0039] As a preferred solution of the method for predicting the life of steel wear-resistant materials based on machine learning of the present invention, the optimal parameter combination is selected by parameter estimation score, and the calculation of the parameter estimation score is shown in the following formula: in, For the The search step length of the round optimization, is the current optimal parameter combination, is the final prediction result of the GBDT model. is the parameter combination to be evaluated.

[0040] In a second aspect, an embodiment of the present invention provides a system for predicting the life of steel wear-resistant materials based on machine learning, which includes:

[0041] A data acquisition module is used to collect historical operation data of steel wear-resistant materials, normalize the historical operation data to obtain normalized characteristic data, and divide the normalized characteristic data into data sets in proportion;

[0042] A model building module is used to build a first model based on the data set; build a second model, and use the prediction errors of the first model under different parameter combinations on the data set as input to train the second model;

[0043] The optimization prediction module is used to update the first model according to the optimal parameter combination output by the second model, input the data set into the updated first model, and obtain the life prediction result of the steel wear-resistant material.

[0044] The beneficial effect of the present invention is that the key data such as chemical composition, temperature and stress of steel wear-resistant materials are collected in multiple dimensions and normalized, thereby eliminating the interference caused by different feature magnitudes and ensuring data consistency. The support vector regression model is used to perform nonlinear fitting for life prediction, and the gradient boosting decision tree is combined to perform automatic parameter optimization to ensure that the model parameters match the working noise level. Finally, the accuracy and adaptability of the model are improved through the optimal parameter combination, making the life prediction results more practical and providing a scientific basis for material replacement cycle and management decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them:

[0046] Figure 1 Flowchart of the life prediction method of steel wear-resistant materials based on machine learning. DETAILED DESCRIPTION

[0047] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0048] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0049] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0050] The present invention is described in detail with reference to schematic diagrams. When describing the embodiments of the present invention, for the sake of convenience, the cross-sectional diagrams showing the device structure will not be partially enlarged according to the general scale, and the schematic diagrams are only examples, which should not limit the scope of protection of the present invention. In addition, in actual production, the three-dimensional dimensions of length, width and depth should be included.

[0051] At the same time, in the description of the present invention, it should be noted that the directions or positional relationships indicated by the terms "upper, lower, inner and outer" are based on the directions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as limiting the present invention. In addition, the terms "first, second or third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0052] In the present invention, unless otherwise clearly specified and limited, the terms "install, connect, connect" should be understood in a broad sense, for example: it can be a fixed connection, a detachable connection or an integral connection; it can also be a mechanical connection, an electrical connection or a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0053] Example 1, reference Figure 1 , which is the first embodiment of the present invention, provides a method for predicting the life of steel wear-resistant materials based on machine learning, comprising:

[0054] S1: collecting historical operation data of steel wear-resistant materials, normalizing the historical operation data to obtain normalized characteristic data, and dividing the normalized characteristic data into data sets in proportion.

[0055] Specifically, the chemical composition of the steel wear-resistant material is tested by an X-ray fluorescence spectrometer, and the mass fractions of chemical elements such as carbon content percentage, chromium content percentage, manganese content percentage, silicon content percentage, phosphorus content percentage and sulfur content percentage are recorded, so as to obtain material composition data;

[0056] Then, use an infrared thermometer to arrange temperature sensors at multiple temperature measuring points of steel wear-resistant materials. Collect temperature time series data during the working process at a sampling frequency of once every 30 seconds, record the highest temperature value, the lowest temperature value and the average temperature value, and obtain the working temperature data.

[0057] Furthermore, through the stress-strain testing system, strain gauges are arranged on the surface of the steel wear-resistant material to collect the tensile stress, compressive stress and shear stress data that the material is subjected to under actual working conditions. The sampling frequency of the stress data is once per minute to obtain stress data.

[0058] Finally, the operation time of steel wear-resistant materials from being put into use to the occurrence of significant performance degradation or failure is recorded. Performance degradation indicators include cracks on the material surface, wear depth exceeding the preset threshold, or hardness reduction of more than 20%, etc., to obtain the life data of steel wear-resistant materials. During the above data collection process, external condition parameters such as environmental humidity and working load are recorded simultaneously to establish a complete data collection record table.

[0059] Furthermore, the material composition data is normalized to the minimum and maximum values, and the percentage of each chemical element content is mapped to the interval [0,1]. The calculation formula is: normalized value = (original value - minimum value) / (maximum value - minimum value), ensuring that the composition data of different magnitudes have the same scale. The operating temperature data is normalized using Z-score, the mean and standard deviation of the temperature time series data are calculated, and the temperature data is converted to a standard normal distribution to eliminate the impact of the temperature change amplitude on model training. The stress data is normalized by decimal calibration, and the data is mapped to the interval [-1,1] by moving the decimal point, maintaining the positive and negative properties of the stress data, and adjusting the different stress components to the same order of magnitude;

[0060] The life data were logarithmically transformed and then standardized to convert the long-tailed life data into an approximate normal distribution, thus improving the adaptability of the model to different life scales.

[0061] The normalized feature data are arranged in chronological order, and the stratified random sampling method is used to divide the data set in a ratio of 6:2:2: 60% of the data is used as a training data set for model training, 20% of the data is used as a validation data set for parameter optimization, and the remaining 20% ​​of the data is used as a test data set for model performance evaluation to ensure that the distribution characteristics of the samples in the three data sets remain consistent.

[0062] S2: Based on the training data set, a support vector regression (SVR) prediction model is constructed as a basic model, and the initial parameters of the basic model include a kernel function parameter and a penalty factor;

[0063] Specifically, based on data feature analysis, Gaussian radial basis function (RBF) is selected as the kernel function. The mathematical expression of the kernel function is: in, represents the kernel function output value, , is a vector in the input space, Representation vector The square of the Euclidean distance between is the kernel function parameter, which controls the expansion degree of the Gaussian function and is used to control the width of the Gaussian distribution. It is defined as follows: in, is the number of feature dimensions;

[0064] The initial value of the parameter is set to the inverse of the feature dimension. This setting is adaptively adjusted based on the dimensional characteristics of the training data set to avoid data sparsity problems in high-dimensional feature space.

[0065] According to the noise level of the training data set, the penalty factor C is set to balance the complexity of the model and the training error. The initial value of the penalty factor is set to 1.0. When the noise of the training data set is large, a smaller C value makes the model more tolerant of the training error, and when the noise of the training data set is small, a larger C value prompts the model to fit the training data more strictly, thereby establishing an association between the penalty factor and the data characteristics. The optimization function after adding the penalty factor C is expressed as: in, is the weight vector, and is the slack variable, is the norm of the weight vector, is the penalty factor that controls the model's sensitivity to training errors.

[0066] It should be noted that the rules for determining larger and smaller are as follows:

[0067] When the standard deviation of the training data exceeds 20% of the mean, the noise is considered to be large, and a smaller value of C∈[0.1, 1.0] is selected;

[0068] When the standard deviation of the training data is between 5% and 20% of the mean, the noise is considered moderate, and a moderate value of C∈[1.0,10.0] is selected;

[0069] When the standard deviation of the training data is less than 5% of the mean, the noise is considered small, and a larger value of C∈[10.0, 100.0] is selected;

[0070] Furthermore, based on the accuracy requirements of life prediction of steel wear-resistant materials, the epsilon parameter is introduced as the tolerance band width of the loss function, and the initial value of the epsilon parameter is set to 0.1; this parameter affects the generalization ability of the model by controlling the number of support vectors. When the deviation between the predicted value and the true value is less than the epsilon value, no loss is calculated, forming a soft interval constraint on the prediction error. The epsilon parameter-insensitive loss function is defined as: in, is the real life value, is the model prediction value, is the tolerance band width, It represents the absolute difference between the predicted value and the true value.

[0071] Then the loss function of the overall model is: The constraints are: Among them, the first is the regularization term, and the second term is the empirical risk term, is the bias term, is the true value of the i-th training sample, is the characteristic value of the i-th training sample.

[0072] The sequential minimal optimization (SMO) algorithm is used to solve the dual optimization problem. The constrained optimization problem is decomposed into a series of two-dimensional sub-problems. Support vectors and corresponding Lagrange multipliers are obtained through iterative optimization. The selection of support vectors is based on the boundary conditions defined by the epsilon parameter, and the update of Lagrange multipliers is based on the constraint conditions of the penalty factor C. Finally, the initial form of the basic prediction model is established.

[0073] For example, the sequential minimal optimization (SMO) algorithm is used to solve the dual optimization problem. The Lagrangian function of the dual problem is: in, is the calculated value of the kernel function between two training samples, and For the The Lagrange multiplier of training samples satisfies: ;; The final prediction function form is: in, is the input sample to be predicted, is the support vector in the training set, is the corresponding Lagrange multiplier difference, is the optimal bias term.

[0074] It should be noted that the construction of the entire SVR model begins with the definition of the basic prediction function f(x), which maps the input to the prediction result through the weight vector w and the bias term b. However, due to the difficulty of directly calculating in the high-dimensional feature space, the kernel function technique is introduced to simplify it, and the Gaussian kernel function K(x,y) is selected to implicitly process the feature mapping; after having the prediction function and the kernel function, it is necessary to define the optimization goal of the model. Here, the epsilon insensitive loss function is introduced to measure the prediction error. At the same time, the control of model complexity is considered, and these elements are combined to construct a complete optimization objective function, which includes regularization terms for controlling model complexity, loss terms for measuring prediction accuracy, and penalty factors C for balancing the two. However, this optimization problem is difficult to solve directly, and it needs to be converted into a dual form. By introducing Lagrange multipliers, a quadratic programming problem is finally obtained. Solving this problem can obtain the support vector and the corresponding Lagrange multiplier, and then the final prediction model is obtained.

[0075] S3: construct a parameter optimization model based on a gradient boosted decision tree (GBDT), take the prediction error of the validation data set under different parameter combinations as input, take the optimal parameter combination of the basic model as output, and train the parameter optimization model;

[0076] Firstly, the search range of the three key parameters of the kernel function parameter γ, penalty factor C and epsilon in the SVR basic model is defined. The search of the kernel function parameter γ adopts logarithmic uniform distribution to cover changes of small orders of magnitude. The search of the penalty factor C also adopts logarithmic distribution to cope with penalty intensities of different magnitudes. The epsilon parameter adopts linear uniform distribution to maintain fine-grained adjustment of prediction accuracy.

[0077] Furthermore, based on the prediction function of the SVR basic model, a comprehensive evaluation index including three dimensions of prediction accuracy loss, model complexity loss and support vector ratio loss is established.

[0078] The prediction accuracy loss not only considers the deviation between the predicted value and the true value, but also introduces the sensitivity analysis of the prediction function to each parameter to evaluate the stability of parameter adjustment; as shown in the following formula: in, is the prediction accuracy loss, is the sample size, is the true value, is the predicted value, is the prediction function The partial derivative (sensitivity) of is the partial derivative of the prediction function with respect to C, is the prediction function The partial derivative of .

[0079] The model complexity loss is measured by the norm of the weight vector and the distribution characteristics of the Lagrange multiplier, and is expressed as follows: in, is the norm of the weight vector, is the initial penalty factor value (1.0), is the absolute value of the Lagrange multiplier, is the complexity loss.

[0080] The support vector ratio loss is based on the deviation of the actual number of support vectors from the ideal ratio, combined with the impact of the epsilon parameter. It is expressed as follows: in, is the number of support vectors, is the expected support vector ratio, is the tolerance band width, is the support vector ratio loss.

[0081] Furthermore, in order to evaluate the effect of each parameter combination, the basic SVR model is run on the validation data set to obtain the complete prediction results and related model features under this combination. Then, a feature vector is constructed based on these features. The feature vector includes various loss indicators, parameter normalization values, and statistical information of support vectors. , calculate the eigenvector: in, is the initial kernel function parameter, is the initial penalty factor value, is the initial tolerance band width.

[0082] In order to more sensitively reflect the impact of various types of losses, the scoring function is designed as an exponential weighted combination to make it more sensitive to changes in different loss indicators. Then the target scoring function is: in, , , is the weight coefficient of each loss item.

[0083] Furthermore, based on the traditional residual learning objective function, the improved residual learning objective function incorporates the Lagrange multiplier information of the SVR model into the regularization term to strengthen the dependence of parameter optimization on the basic model. The improved objective function is as follows: in, is the t-th tree, is the regularization coefficient, Lagrange multipliers from the SVR base model, is the cumulative prediction result of the first t-1 trees.

[0084] At the same time, a parameter smoothing term is added to the splitting criterion of the decision tree to reduce the drastic fluctuations in the parameter space and improve the stability of the optimization process. It should be noted that the splitting criterion of the decision tree is as follows: in, is the smoothing coefficient, which is used to control the continuity of parameter changes. is the loss value of the parent node, is the loss value of the left and right child nodes, is the γ value of the left and right child nodes.

[0085] Finally, through the adaptive learning rate strategy, the learning step size is dynamically adjusted according to the optimization progress to make the optimization more efficient. The final prediction function of the GBDT model is expressed as follows: in, is the characteristic vector of the historical optimal parameter combination, is the kernel width parameter, For the The weight of the tree.

[0086] Furthermore, during the optimization process, the search step size is dynamically adjusted according to the changes in the scores over multiple consecutive rounds. When the optimization is close to convergence, the system will automatically narrow the search range to improve the accuracy of the local optimal solution. At the same time, the final parameter evaluation score comprehensively considers the distance between the prediction performance and the parameters, so that while ensuring the optimization effect, it can avoid excessive deviation from the current optimal solution. The final parameter estimation score can be calculated using the following formula: in, For the The search step length of the round optimization, is the current optimal parameter combination, is the final prediction result of the GBDT model. is the parameter combination to be evaluated.

[0087] S4: updating the basic model according to the optimal parameter combination output by the parameter optimization model, inputting the test data set into the updated basic model, and obtaining the life prediction result of the steel wear-resistant material.

[0088] Based on the optimal parameter combination (γopt, Copt, εopt) output by the GBDT parameter optimization model, the effectiveness of the parameter changes is first verified to ensure that the adjustment range of the kernel function parameters, penalty factors and epsilon parameters is within a reasonable range.

[0089] Then, the structure of the SVR basic model is updated using the verified optimal parameters, the kernel function matrix is ​​recalculated and the constraints of the optimization problem are adjusted, and then the updated model is retrained with the original training data to obtain a new set of support vectors and weight coefficients. After the model update is completed, the test data set is normalized and preprocessed, including feature scaling and outlier processing; finally, the processed test data is input into the updated SVR model, and the life prediction results of steel wear-resistant materials are obtained by calculating the kernel function value between the test sample and the support vector, and the predicted value is restored to the original scale through denormalization processing to form the final prediction report and application recommendations.

[0090] Furthermore, this embodiment also provides a steel wear-resistant material life prediction system based on machine learning, comprising:

[0091] A data acquisition module is used to collect historical operation data of steel wear-resistant materials, normalize the historical operation data to obtain normalized characteristic data, and divide the normalized characteristic data into data sets in proportion;

[0092] A model building module is used to build a first model based on the data set; build a second model, and use the prediction errors of the first model under different parameter combinations on the data set as input to train the second model;

[0093] The optimization prediction module is used to update the first model according to the optimal parameter combination output by the second model, input the data set into the updated first model, and obtain the life prediction result of the steel wear-resistant material.

[0094] This embodiment also provides a computer device, which is suitable for the case of a method for predicting the life of steel wear-resistant materials based on machine learning, and includes a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the method for predicting the life of steel wear-resistant materials based on machine learning as proposed in the above embodiment.

[0095] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0096] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the method for predicting the life of steel wear-resistant materials based on machine learning as proposed in the above embodiment is implemented.

[0097] The storage medium proposed in this embodiment and the data storage method proposed in the above embodiment belong to the same inventive concept. The technical details not fully described in this embodiment can be found in the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0098] Example 2 is the second example of the present invention. This example provides a method for predicting the life of steel wear-resistant materials based on machine learning. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0099] In this experiment, six kinds of high chromium cast iron wear-resistant materials with different composition ratios were selected as research objects, numbered HCr-1 to HCr-6. First, the chemical composition of each sample was analyzed using a Bruker S8 TIGER X-ray fluorescence spectrometer. The sampling points were evenly distributed at five different positions on the surface of the sample. Each position was measured three times, and the average value was taken as the final composition data.

[0100] In order to simulate the actual working environment, a special wear resistance test platform was designed. On the test platform, a FLIRE75 infrared thermometer was used to arrange 15 temperature measurement points on the surface of each sample. The temperature measurement points were distributed in a matrix, covering the entire working area. The temperature acquisition frequency was set to once every 30 seconds, and the temperature changes were continuously monitored for 24 hours. At the same time, the HBMQuantumX data acquisition system was used with a strain gauge array to arrange 12 strain measurement points on the surface of the sample, with an acquisition frequency of once per minute, to record the stress state of the material under different load conditions.

[0101] The external working parameters are controlled by a constant temperature and humidity test chamber, with the ambient temperature set at 25±2℃ and the relative humidity maintained at 60±5%. The working load is applied by a servo hydraulic system, and cyclic loading is performed according to the preset load spectrum, with a load range of 2-10kN and a frequency of 0.5Hz. During the entire test process, high-precision displacement sensors and digital microscopes are used to regularly detect the wear state of the sample surface and record the appearance and expansion of cracks. When the surface crack length exceeds 2mm, or the wear depth exceeds 1.5mm, or the surface hardness of the material decreases by more than 20%, it is determined to have reached the end of its life.

[0102] In the data preprocessing stage, the material composition data is first normalized. Considering the large differences in the content of different elements, a segmented normalization strategy is adopted: for the main elements (C, Cr), the minimum-maximum normalization is adopted; for trace elements (P, S), logarithmic transformation is adopted before normalization. After the temperature data is denoised, the Z-score normalization method is used to convert the data into a standard normal distribution with a mean of 0 and a standard deviation of 1. The stress data is normalized by decimal calibration, which keeps the stress direction information while controlling the data range within the interval [-1,1].

[0103] The experimental data was divided using a stratified random sampling method to ensure that the distribution characteristics of the samples in each data subset were consistent. Considering the temporal characteristics of the data, special attention was paid to maintaining time continuity during division to avoid the adverse effects of data fragmentation on model training. Finally, a complete data set containing material composition, operating temperature, stress state, and life characteristics was formed. The experimental data are as follows:

[0104] Table 1 Experimental data table

[0105] Sample No. C(%) Cr(%) Mn(%) Si(%) P(%) S(%) Maximum temperature(℃) Minimum temperature (℃) Average temperature(℃) Tensile stress (MPa) Compressive stress(MPa) Shear stress (MPa) Lifespan (h) HCr-1 2.8 15.5 1.2 0.8 0.04 0.03 425 180 320 580 620 310 2850 HCr-2 3.2 18.2 1.4 1.0 0.03 0.02 410 175 305 610 645 325 3280 HCr-3 2.5 16.8 1.1 0.9 0.035 0.025 430 185 315 595 635 315 3050 HCr-4 3.0 17.5 1.3 1.1 0.045 0.035 415 170 310 600 640 320 3150 HCr-5 2.9 16.2 1.25 0.95 0.04 0.03 420 175 315 590 630 315 2950 HCr-6 3.1 17.8 1.35 1.05 0.035 0.025 405 165 300 605 650 330 3200

[0106] From the data, it can be seen that the samples with carbon content in the range of 2.5-3.2% and chromium content in the range of 15.5-18.2% show good comprehensive performance. In particular, the HCr-2 sample (C: 3.2%, Cr: 18.2%) has a service life of 3280 hours, which is about 15% higher than the traditional ratio. This is mainly due to the fact that the higher carbon-chromium ratio optimizes the morphology and distribution of carbides and improves the wear resistance of the material.

[0107] The data shows that the working temperature fluctuation range of all samples is controlled between 165-430℃, and the average temperature is maintained in the range of 300-320℃. The HCr-2 sample exhibits the most stable temperature characteristics, and its temperature fluctuation amplitude is the smallest (△T=235℃), which is directly related to its excellent thermal stability. Lower temperature fluctuation helps to reduce the generation of thermal stress and delay the fatigue damage process of the material.

[0108] From the stress data, the compressive stress is generally higher than the tensile stress. This stress distribution is conducive to inhibiting the initiation and expansion of cracks. In particular, the compressive stress of HCr-2 and HCr-6 samples reached 645MPa and 650MPa respectively, while maintaining a high shear stress (325MPa and 330MPa). This stress combination effectively improves the fatigue resistance of the material.

[0109] Comparative analysis shows that there is a significant correlation between material life and chemical composition, temperature characteristics and stress state. Through multiple regression analysis, it was found that the correlation coefficient between chromium content and life reached 0.92, which verified the effectiveness of the parameter optimization model in the present invention. Especially under high stress conditions (compressive stress>640MPa), the optimized material can still maintain stable performance, which proves the reliability of the prediction method.

[0110] Compared with the prior art, the present invention achieves high-precision life prediction by precisely controlling process parameters and real-time monitoring of performance evolution. For example, under similar use conditions, the life of HCr-2 samples is 15-20% longer than that of traditional materials. This improvement is due to the optimized combination of material composition and process parameters. At the same time, by establishing a temperature-stress-life correlation model, the prediction accuracy is improved by about 25%, providing a more reliable reference for engineering applications.

[0111] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for predicting the life of steel wear-resistant materials based on machine learning, characterized in that: include: Collecting historical operation data of steel wear-resistant materials, normalizing the historical operation data to obtain normalized characteristic data, and dividing the normalized characteristic data into data sets in proportion; Based on the data set, construct a first model; Construct a second model, use the prediction errors of the first model under different parameter combinations on the data set as input, and train the second model; updating the first model according to the optimal parameter combination output by the second model, inputting the data set into the updated first model, and obtaining the life prediction result of the steel wear-resistant material; The historical operation data includes material composition data, operating temperature data, stress data and life data; The material composition data includes carbon content percentage, chromium content percentage, manganese content percentage, silicon content percentage, phosphorus content percentage and sulfur content percentage; The working temperature data includes the highest temperature value, the lowest temperature value and the average temperature value during the working process; The stress data include tensile stress, compressive stress and shear stress data under actual working conditions; The data set includes a training data set, a validation data set and a test data set; The construction of the first model includes: Construct an initial prediction function; the basic prediction function maps input data to prediction results based on weight vectors and bias terms; Selecting a kernel function according to the dimensional characteristics of the training data set; Setting a penalty factor according to the noise level of the training data set to balance the error of the first model; Based on the accuracy requirement of the life prediction of the steel wear-resistant material, the number of support vectors is controlled by using the epsilon parameter; Constructing a loss function of the first model, and converting the support vector regression problem into a constrained optimization problem through the loss function; Decomposing the constrained optimization problem into two-dimensional sub-problems, and obtaining support vectors and corresponding Lagrange multipliers through iterative optimization to obtain the first model; The step of setting a penalty factor to balance the error of the first model includes: When the noise of the training data set is large, the smaller The value makes the model tolerant to training errors, while when the noise of the training dataset is small, a larger The value makes the model fit the training data, thus establishing the association between the penalty factor and the data characteristics; When the standard deviation of the training data exceeds a first preset value of the mean, it is determined that the noise of the training data set is large; When the standard deviation of the training data is less than a second preset value of the mean, it is determined that the noise of the training data set is small; Among them, the optimization function of the balanced first model is as follows: ； in, is the weight vector, and is the slack variable, is the norm of the weight vector, is the penalty factor, which controls the sensitivity of the model to training errors; The method of controlling the number of support vectors using the epsilon parameter includes: When the deviation between the predicted value and the true value is less than the value of the epsilon parameter, no loss is calculated, forming a soft interval constraint on the prediction error; The soft margin constraint is as follows: ； in, is the bias term, is the true value of the i-th training sample, is the tolerance band width, is the characteristic value of the i-th training sample; The second model is trained according to the feature vector; the construction of the feature vector includes: Constructing a parameter combination of the first model, for each parameter combination, using the each parameter combination to update the first model, and using the validation data set to calculate a prediction result; evaluating the prediction results and forming the feature vector; The prediction results are evaluated based on prediction accuracy loss, model complexity loss and support vector ratio loss; The calculation of the prediction accuracy loss is shown in the following formula: ； in, is the prediction accuracy loss, is the sample size, is the true value, is the predicted value, is the prediction function The partial derivative of is the partial derivative of the prediction function with respect to C, is the prediction function The partial derivative of The calculation of the model complexity loss is shown as follows: ； in, is the norm of the weight vector, is the initial penalty factor value, is the absolute value of the Lagrange multiplier, is the complexity loss; The support vector ratio loss is calculated as follows: ； in, is the number of support vectors, is the expected support vector ratio, is the initial tolerance band width, is the support vector ratio loss.

2. The method for predicting the life of steel wear-resistant materials based on machine learning according to claim 1, characterized in that: The optimal parameter combination is selected by the parameter estimation score, and the calculation of the parameter estimation score is shown in the following formula: ； in, For the The search step length of the round optimization, is the current optimal parameter combination, is the final prediction result of the GBDT model. is the parameter combination to be evaluated.

3. A machine learning-based life prediction system for steel wear-resistant materials, based on the machine learning-based life prediction method for steel wear-resistant materials according to any one of claims 1 to 2, characterized in that: include, A data acquisition module is used to collect historical operation data of steel wear-resistant materials, normalize the historical operation data to obtain normalized characteristic data, and divide the normalized characteristic data into data sets in proportion; A model building module, used for building a first model based on the data set; Construct a second model, use the prediction errors of the first model under different parameter combinations on the data set as input, and train the second model; The optimization prediction module is used to update the first model according to the optimal parameter combination output by the second model, input the data set into the updated first model, and obtain the life prediction result of the steel wear-resistant material.

Images