Machine equipment life prediction method based on digital-analog linkage

Through the digital-to-analog linkage of machine equipment life prediction method, combined with mechanism model and data driving method, vibration signal analysis and neural network model are used to solve the prediction and physical interpretability of wear processes throughout the life cycle, achieving more accurate Equipment life forecasting and maintenance strategy formulation.

CN118981948APending Publication Date: 2024-11-19BENGANG STEEL PLATES CO LTD +1
View PDF 0 Cites 3 Cited by

Patent Information

Application Number
CN202411041651.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-31
Publication Date
2024-11-19

AI Technical Summary

Technical Problem

The prior art is difficult to describe the wear process of mechanical equipment in a comprehensive and detailed manner throughout the life cycle, and lacks physical interpretability, and it is difficult for a single wear formula to accurately predict the multi-mechanical coupling process.

Method used

The life prediction method of machine equipment based on digital-to-analog linkage is adopted, combined with mechanism model and data-driven method, and the life prediction is performed through vibration signal analysis, neural network model and mechanism model fusion, and integrated learning algorithms are used.

Benefits of technology

It realizes rapid predictions throughout the life cycle and improves the physical interpretability of predictions, reduces maintenance costs, and improves the level of equipment failure prevention.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118981948A_ABST
    Figure CN118981948A_ABST
Patent Text Reader

Abstract

The invention provides a machine equipment life prediction method based on digital-analog linkage, and the method comprises the following steps: S1, collecting an equipment vibration signal under a target working condition through a vibration sensor, obtaining a working condition parameter through an equipment system, and obtaining a tool wear loss through a direct measurement method as a residual life label; s2, extracting key features of the equipment vibration signal; s3, selecting features with correlation coefficient absolute values greater than a rated value to form a feature set; s4, outputting a prediction value of the residual life; s5, constructing a degradation or failure mechanism model, and describing a degradation process from a mechanism level; s6, a mechanism model changing along with operation time under a certain working condition is obtained; and S7, outputting a final life prediction result. The model constructed by the invention can predict the residual life more accurately, integrates the advantages of a mechanism model and a data driving method, not only keeps interpretability, but also ensures the grasp of the overall degradation trend.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of digital twin and remaining life prediction, and in particular, to a method for predicting the life of machinery and equipment based on digital-analog linkage. Background Art

[0002] In modern industrial production processes, the stable operation of mechanical equipment is essential to ensure the continuity of the production process, improve production efficiency, and ensure workplace safety. The failure of mechanical production equipment often leads to the stagnation of the production line, which not only affects production efficiency, but also may lead to high maintenance costs and potential safety risks. Therefore, accurate prediction of the remaining useful life (RUL) of these mechanical equipment is of great economic and safety significance for optimizing maintenance plans, preventing unexpected downtime, reducing maintenance costs, and avoiding potential safety accidents.

[0003] In the prior art, the numerical simulation of the cutting process of nickel-based high-temperature alloys is a prediction method based on a mechanism model. The orthogonal experimental design is used to estimate the parameters of the mechanism model constructed based on the Archard formula, complete the model construction, and predict the tool wear. The shortcoming of this method is that the degradation of the entire life cycle is a process of coupling multiple mechanisms. For example, the tool wear process is divided into three stages. The wear mechanism of each stage may be a combined influence of one or more of chemical wear, adhesive wear and diffusion wear. The mechanism model constructed by a single wear formula can only achieve relatively accurate calculations in a certain stage, and the construction of a multi-mechanism coupling model to comprehensively and comprehensively describe the degradation process of the entire life cycle is a huge workload and is not practical. The prior art also has a backtracking search algorithm to optimize the LS_SVM parameters. The method for identifying the wear state of milling tools is still a migration of algorithms in other fields, and there are deficiencies in physical interpretability. Summary of the invention

[0004] In view of this, the purpose of the present invention is to propose a machine equipment life prediction method based on digital-analog linkage. Since the interpretability of the mechanism model and the rapid prediction of the degradation process of the whole life cycle by the data-driven method can make up for each other's shortcomings, the digital-analog linkage method that integrates the mechanism model and the data-driven method can simultaneously solve the rapid prediction reflecting the degradation process of the whole life cycle and the problem of poor interpretability. The present invention aims to simultaneously solve the problems of rapid prediction and lack of physical interpretability in the degradation process of the whole life cycle.

[0005] The technical means adopted by the present invention are as follows:

[0006] A method for predicting the life of machinery and equipment based on digital-analog linkage includes the following steps:

[0007] S1. The vibration signal of the equipment under the target working condition is collected by the vibration sensor, the working condition parameters are obtained by the equipment system, and the tool wear amount is obtained by direct measurement as the remaining life label;

[0008] S2. Perform time domain and frequency domain analysis on the equipment vibration signal, and use the wavelet packet decomposition method to perform time and frequency domain analysis on the equipment vibration signal to extract key features of the equipment vibration signal;

[0009] S3, calculating the correlation coefficient between the key feature and the remaining life label, and selecting the features whose absolute value of the correlation coefficient is greater than the rated value to form a feature set;

[0010] S4. Based on the feature set, a radial basis neural network is used to construct the first neural network model, and the predicted value M of the remaining life is output. 1 ; Use convolutional neural network to build the second neural network model and output the final prediction result M 2 ;

[0011] S5. Construct a degradation or failure mechanism model based on Archard wear theory to describe the degradation process from a mechanism level;

[0012] S6. Based on the working condition parameters, through orthogonal experimental design, multiple sets of simulation experimental data are obtained to estimate the degradation or failure mechanism model, obtain the mechanism model that changes with the running time under certain working conditions, and obtain the output M of the timing conditions. 3 ;

[0013] S7, the first neural network model, the second neural network model, and the mechanism model that changes with the running time under certain working conditions are used as sub-models, and the various sub-models are integrated by using the ensemble learning principle. 1 、M 2 and M 3 Classify and take the average to obtain the synthesis of the prediction results of different sub-models, and output the final life prediction result.

[0014] Furthermore, the key features described in S2 include:

[0015] Mean value mean: where x i is the vibration value at a certain time point within the signal length range, N is the signal length;

[0016] Standard deviation std:

[0017] Root mean square value rms:

[0018] Maximum value max: max(x i );

[0019] Minimum value min: min(x i );

[0020] Peak to peak: max(x i )-min(x i );

[0021] Kurt:

[0022] Skewness:

[0023] Crest factor):

[0024] Shape factor: in

[0025] Clearance factor:

[0026] Impulse factor:

[0027] Peak frequency: freqs[argmax(PSD(f))], where Where, f is the frequency;

[0028] Total power: ∑PSD(f);

[0029] Spectral centroid:

[0030] Spectral kurt: psd i is the power spectral density value of the i-th frequency point, and psd is the average value of the power spectral density;

[0031] Spectral skew:

[0032] The features extracted by the wavelet packet decomposition method in S2 include the energy in the corresponding frequency bands of the decomposed approximate coefficients and detail coefficients. The number of layers of wavelet packet decomposition is 6, and the db3 wavelet basis function is used.

[0033] Further, in S3, the correlation coefficient is calculated as follows:

[0034]

[0035] Among them, x n is the nth element of the eigenvalue sequence, y n is the nth element in the sequence of remaining life value or life index, is the mean of the eigenvalue sequence, is the remaining life value or the mean of the life indicator series;

[0036] When |ρ xy The closer | is to 1, the stronger the linear correlation between X and Y is. The rated value is 0.9. Select |ρ xy The features with |>0.9 are taken as the feature set.

[0037] Furthermore, in S4, the steps of constructing the first neural network model are as follows:

[0038] The input data X is transmitted to the hidden layer and nonlinearly mapped by the radial basis function:

[0039]

[0040] The output of the hidden layer is used as input again and passed to the output layer through linear calculation to map the feature to an actual predicted value to form the final network output M 1 , which is the predicted value of remaining life.

[0041] Furthermore, in S4, the steps of constructing the second neural network model are as follows:

[0042] The input data X is transmitted to the convolution layer for feature map extraction and convolution processing:

[0043] X′=Conv 1 (X)

[0044] Then X′ is normalized by layer:

[0045]

[0046] Where E[X′] represents the mean of X′, Var[X′] represents the variance of X′; ∈ is a positive number to avoid division by zero; γ represents a scaling parameter used to readjust the scale of the normalized data; β represents an offset parameter used to readjust the center position of the normalized data; ⊙ represents element-wise multiplication;

[0047] The normalized data X LN Transformed by a linear layer:

[0048] X Linear =W Linear X LN +b Linear

[0049] Where W Linear represents the weight matrix of the linear layer, b Linear represents the bias term of the linear layer;

[0050] The data X after linear transformation Linear Apply the Relu activation function:

[0051] X Relu =X Linear ·Relu(X Linear )

[0052] In the formula, Relu(X Linear ) represents the Relu activation function;

[0053] Data X after Relu activation Relu The Dropout layer reduces overfitting:

[0054] X Dropout = Dropout(X Relu )

[0055] Data X after Dropout Dropout The final prediction result M is obtained through the fully connected layer and the regression layer. 2 .

[0056] Furthermore, the degradation or failure mechanism model in S5 is as follows:

[0057] O={σ,τ,∈,α}

[0058] L = f(t, O)

[0059] Among them, O is a set of factors affecting the remaining life, σ represents stress-related parameters, τ represents temperature-related parameters, ∈ represents deformation-related parameters, t represents operating time, L represents the remaining life or direct variables related to life, and α represents other influencing parameters.

[0060] Furthermore, S6 specifically includes:

[0061] Through the orthogonal experimental design method, the multi-factor and multi-level simulation experimental data P = {p 1 ,p 2 ,…}, the specific steps are as follows:

[0062] Determine the number of factors, which are determined as three factors here according to the set O in S5: speed, feed rate and axial cutting depth;

[0063] Determine the number of levels. According to the allowable range of process parameters, each factor is set to four levels;

[0064] The number of factors and levels were orthogonalized, and 16 groups of experiments were conducted;

[0065] Input the simulation experiment data P into the S5 degradation model, estimate the parameters contained in the parameter set O of the degradation or failure mechanism model, complete the construction of the mechanism model that changes with the running time under certain working conditions, and obtain the output M of the timing condition 3 .

[0066] Furthermore, S7 specifically includes the following steps:

[0067] Use each sub-model as a sub-model learner in the idea of ​​ensemble learning bagging algorithm;

[0068] Combining the comprehensive strategy of regression task and classification task, the output results M of each sub-model are 1 , M 2 and M 3 After classification, the weighted average is used for fusion and the result is used as the final output.

[0069] The present invention also provides a storage medium, which includes a stored program, wherein when the program is run, any of the above-mentioned machine equipment life prediction methods based on digital-analog linkage is executed.

[0070] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes any one of the above-mentioned machine equipment life prediction methods based on digital-analog linkage through the computer program.

[0071] Compared with the prior art, the present invention has the following advantages:

[0072] This invention proposes a machine equipment life prediction method based on digital-analog linkage, adopts the idea of ​​ensemble learning bagging algorithm, takes the mechanism model and data-driven model as its different sub-model learners, and ensures that the fused result has both the advantage of strong interpretability of the mechanism model and the advantage of the data-driven model in predicting and controlling the degradation trend of the equipment throughout its life cycle. It solves the shortcomings of a single method, uses the mechanism model as a high-fidelity model for interpretable calculation, and uses the neural network model as a low-fidelity model to grasp the direction of the degradation trend throughout the life cycle, forming a multi-fidelity prediction model.

[0073] This method can predict the life of different equipment and formulate different maintenance strategies according to different health states. That is, the present invention can be used in the field of predictive maintenance of mechanical equipment, effectively reducing maintenance costs and improving the level of prevention of equipment failures. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0075] Figure 1 The figure is a flow chart of the method of the present invention.

[0076] Figure 2 A process for constructing the mechanism model of the present invention.

[0077] Figure 3 This is a diagram showing the predicted effect of an embodiment of the present invention. DETAILED DESCRIPTION

[0078] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0079] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0080] like Figure 1 As shown, the present invention provides a machine equipment life prediction method based on digital-analog linkage. The method adopts the idea of ​​ensemble learning bagging algorithm to integrate the mechanism model and the neural network model as sub-models, including 1 physical sub-model and 2 neural network sub-models, and includes the following steps:

[0081] S1: The vibration signal of the equipment under the target working condition is collected by the vibration sensor as the neural network model data, and the working condition parameters are obtained by the equipment system as the physical model construction data.

[0082] S2: The collected vibration signal X = [x 1 ,x 2 ,…,x N ] Perform time domain and frequency domain analysis, and use wavelet packet decomposition method to perform time and frequency domain analysis to extract key features:

[0083] Mean:

[0084] Standard deviation (std):

[0085] Root mean square value (rms):

[0086] Maximum value (max): max(x i );

[0087] Minimum value (min): min(x i );

[0088] Peak to peak: max(x i )-min(x i );

[0089] Kurt:

[0090] Skewness:

[0091] Crest factor:

[0092] Shape factor: in

[0093] Clearance factor:

[0094] Impulse factor:

[0095] Peak frequency: freqs[argmax(PSD(f)))], where Where f is the frequency.

[0096] Total power: ∑PSD(f);

[0097] Spectral centroid:

[0098] Spectral kurt: psd i is the power spectral density value of the i-th frequency point, and psd is the average value of the power spectral density;

[0099] Spectral skew:

[0100] Wavelet packet decomposition (WPD) method extracts features including the energy in the corresponding frequency bands of the decomposed approximate coefficients and detail coefficients. The number of layers of wavelet packet decomposition is 6, and the 'db3' wavelet basis function is used.

[0101] S3: Calculate the correlation coefficient between the extracted time domain, frequency domain and time-frequency domain features and the remaining life , where x n ,y n are the nth element in the eigenvalue sequence and the remaining life value or life indicator sequence, respectively. is the mean of each sequence; |ρ xy The closer | is to 1, the better the linear correlation between X and Y. xy The features with |>0.9 are taken as the feature set.

[0102] S4: Based on the extracted feature set, a deep learner and a shallow machine learner are constructed using a convolutional neural network and a radial basis neural network respectively as sub-models of ensemble learning; first, the radial basis neural network model is constructed, the input data X is transmitted to the hidden layer, and nonlinear mapping is performed through the radial basis function:

[0103]

[0104] The output of the hidden layer is then used as input and passed to the output layer through linear calculation to map the feature to an actual predicted value to form the final network output M 1 , which is the predicted value of remaining life.

[0105] Then the convolutional neural network model is constructed. Similarly, the data X is transmitted to the convolution layer for feature map extraction, and the final prediction result M is output through the normalization layer, activation layer and full connection layer. 2 The data X is processed by the following steps:

[0106] First, the data X is convolved:

[0107] X′=Conv1 (X)

[0108] Then X′ is normalized by layer:

[0109]

[0110] Where E[X′] represents the mean of X′, Var[X′] represents the variance of X′, ∈ is a small positive number used to avoid division by zero, γ represents the scaling parameter used to readjust the scale of the normalized data, β represents the offset parameter used to readjust the center position of the normalized data, and ⊙ represents element-wise multiplication;

[0111] Normalized data X LN Transformed by a linear layer:

[0112] X Linear =W Linear X LN +b Linear

[0113] Where W Linear represents the weight matrix of the linear layer, b Linear represents the bias term of the linear layer;

[0114] The data X after linear transformation Linear Apply the Relu activation function:

[0115] X Relu =X Linear ·Relu(X Linear )

[0116] In the formula, Relu(X Linear ) represents the Relu activation function;

[0117] Data X after Relu activation Relu The Dropout layer reduces overfitting:

[0118] X Dropout = Dropout(X Relu )

[0119] Data X after Dropout Dropout The final output is obtained through the fully connected layer and the regression layer.

[0120] S5: Construct a degradation or failure mechanism model to describe the degradation process from the mechanism level as one of the sub-models of ensemble learning. The mechanism model is constructed as follows:

[0121] O={σ,τ,∈,α}

[0122] L = f(t, O)

[0123] Among them, O is a set of factors affecting the remaining life, σ represents stress-related parameters, τ represents temperature-related parameters, ∈ represents deformation-related parameters, t represents operating time, L represents the remaining life or direct variables related to life, and α represents other influencing parameters.

[0124] S6: Based on the operating parameters of the equipment, the multi-factor and multi-level simulation experimental data are obtained through the orthogonal experimental design method. The factor level is determined by the factor set O in S5. The obtained data P = {p 1 ,p 2 ,…}, taking the data P as input, the parameters contained in the parameter set O of the mechanism model constructed in S5 are estimated, the construction of the mechanism model that changes with the running time under certain working conditions is completed, and the output M of the timing condition is obtained. 3 .

[0125] S7: First, the neural network model and mechanism model constructed by S2, 3, 4 and 5 are used as sub-model learners in the idea of ​​ensemble learning bagging algorithm. Secondly, the comprehensive strategy of regression task and classification task is combined to classify the sub-models and then take the average according to the weight as the final output result.

[0126] Example

[0127] The process of the present invention is as follows Figure 1 As shown in the figure, the physical parameters of the mechanism model are estimated using the experimental simulation data to complete the construction of the mechanism model under certain working conditions. The sensor data set D is divided into a training set D train and the test set D test , the radial basis neural network (RBF) and convolutional neural network (CNN) were trained using the training set. After the training was completed, the mechanism model and the neural network model were used as sub-models of ensemble learning for comprehensive strategy integration. Based on the comprehensive strategy of classification and regression tasks, the sub-models were first classified, and the model results with too large errors or not in line with the degradation trend were excluded. The remaining models were averaged as the final output result. In this process, the mechanism model was used as a high-fidelity model for interpretable calculations, and the neural network model was used as a low-fidelity model to guide the degradation trend of the entire life cycle.

[0128] This embodiment uses the public data set of the 2010 high-speed CNC machine tool tool health prediction competition of the New York Prognostic and Health Management Association (PHM) to verify the proposed method. The experimental platform consists of a high-speed CNC machine tool, a ball-end carbide milling cutter, a vibration sensor, a wear meter, a charge amplifier and a collector. The experimental working conditions are a spindle speed of 10400r / min, a feed rate of 1555mm / min, a cutting width of 0.125mm and a cutting depth of 0.2mm. The experimental machine tool is Roders Tech RFM760, and the milling material is inconel718. The experiment collected three groups of vibration signals and cutting force signals of the same tool under the same cutting conditions and the corresponding cutting wear values ​​for each time.

[0129] A machine equipment life prediction method based on digital-analog linkage, the overall steps include:

[0130] S1. Collect the cutting force signal X under the target working condition through the cutting force sensor. 1 ,x 2 ,…,x N ], as the neural network model data, and the working condition parameters obtained by the numerical control system are used as the physical model construction data.

[0131] S2. The collected cutting force signals are analyzed in time domain, frequency domain and time-frequency domain to extract multiple key features.

[0132] S3. Calculate the correlation coefficient between the extracted time domain, frequency domain and time-frequency domain features and tool wear In the formula, x n ,y n are the nth elements in the eigenvalue sequence and the wear value sequence, respectively. is the mean of each sequence; |ρ xy The closer | is to 1, the better the linear correlation between X and Y. xy The features with |>0.9 are taken as the feature set.

[0133] S4. Use convolutional neural networks and radial basis neural networks to build deep learners and shallow machine learning machines respectively as sub-models of ensemble learning.

[0134] S5. Describe the degradation process from the mechanism level and construct its mechanism model: O = {σ, τ,∈, α}; L = f(t, O).

[0135] S6. Based on the operating parameters of the equipment, multiple sets of simulation experimental data are obtained through orthogonal experimental design, and the model parameters in step S5 are estimated, such as Figure 2 shown.

[0136] S7. The mechanism model and neural network model obtained above are used as sub-models, and integrated using the principle of ensemble learning. The prediction results of different sub-models are synthesized through the classification and averaging strategy to output the final life prediction result.

[0137] Next, some steps are described in detail:

[0138] In S2, the extracted features include:

[0139] Mean:

[0140] Standard deviation (std):

[0141] Root mean square value (rms):

[0142] Maximum value (max): max(x i );

[0143] Minimum value (min): min(x i );

[0144] Peak to peak: max(x i )-min(x i );

[0145] Kurt:

[0146] Skewness:

[0147] Crest factor:

[0148] Shape factor: in

[0149] Clearance factor:

[0150] Impulse factor:

[0151] Peak frequency: freqs[argmax(PSD(f))], where

[0152] Total power: ∑PSD(f);

[0153] Spectral centroid:

[0154] Spectral kurt: psd i is the power spectral density value of the i-th frequency point, and psd is the average value of the power spectral density;

[0155] Spectral skew:

[0156] Wavelet packet decomposition (WPD), the extracted features include the energy in the corresponding frequency bands of the decomposed approximate coefficients and detail coefficients, the number of wavelet packet decomposition layers is 6, and the 'db3' wavelet basis function is used;

[0157] In S4, the radial basis neural network model is first constructed, the input data X is transmitted to the hidden layer, and nonlinear mapping is performed through the radial basis function:

[0158]

[0159] The output of the hidden layer is then used as input and passed to the output layer through linear calculation to map the feature to an actual predicted value to form the final network output M 1 , which is the predicted value of remaining life.

[0160] Then the convolutional neural network model is constructed. Similarly, the data N is transmitted to the convolution layer for feature map extraction, and the final prediction result M is output through the normalization layer, activation layer and full connection layer. 2 The data X is processed by the following steps:

[0161] First, the data X is convolved:

[0162] X′=Conv 1 (X)

[0163] Then X′ is normalized by layer:

[0164]

[0165] Where E[X′] represents the mean of X′, Var[X′] represents the variance of X′, ∈ is a small positive number used to avoid division by zero, γ represents the scaling parameter used to readjust the scale of the normalized data, β represents the offset parameter used to readjust the center position of the normalized data, and ⊙ represents element-wise multiplication;

[0166] Normalized data X LN Transformed by a linear layer:

[0167] X Linear =W Linear X LN +b Linear

[0168] Where W Linear represents the weight matrix of the linear layer, b Linear represents the bias term of the linear layer;

[0169] The data X after linear transformation Linear Apply the Relu activation function:

[0170] X Relu =X Linear ·Relu(X Linear )

[0171] In the formula, Relu(X Linear ) represents the Relu activation function;

[0172] Data X after Relu activation Relu The Dropout layer reduces overfitting:

[0173] X Dropout = Dropout(X Relu )

[0174] Data X after Dropout Dropout The final output is obtained through the fully connected layer and the regression layer.

[0175] In this step, the relevant parameters of this embodiment are set as follows: setting a convolutional neural network, a convolution kernel of 3*1, performing two convolutions, an initial learning rate of 0.01, and a learning rate reduction factor of 0.5.

[0176] In step S5, O is a set of factors affecting the remaining life, which is determined by experiments.

[0177] In step S6, the parameter estimation process of the mechanism model is as follows:

[0178] S6.1. Construct a simulation model of the physical equipment for the research object;

[0179] S6.2. Obtain simulation data: Use orthogonal experimental design to conduct three-factor four-level simulation experiments, obtain multiple groups of simulation experimental data, and estimate and fit the mechanism model parameters.

[0180] The final result of this embodiment is as follows Figure 3As shown in the figure, its prediction curve is more consistent with the actual wear results than the prediction results of a single model, highlighting its significant ability in grasping task-related information and adapting to new environments. In this task, the root mean square error (RMSE) between the predicted value and the true RUL is 22.6412, which shows the accuracy and reliability of this life prediction method in actual industrial scenarios.

[0181] Finally, 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 aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A machine equipment life prediction method based on digital-analog linkage, characterized in that: The steps include: S1. The vibration signal of the equipment under the target working condition is collected by the vibration sensor, the working condition parameters are obtained by the equipment system, and the tool wear amount is obtained by direct measurement as the remaining life label; S2. Perform time domain and frequency domain analysis on the equipment vibration signal, and use the wavelet packet decomposition method to perform time and frequency domain analysis on the equipment vibration signal to extract key features of the equipment vibration signal; S3, calculating the correlation coefficient between the key feature and the remaining life label, and selecting features with an absolute value of the correlation coefficient greater than the rated value to form a feature set; S4, based on the feature set, a radial basis neural network is used to construct a first neural network model, and a predicted value M1 of the remaining life is output; a convolutional neural network is used to construct a second neural network model, and a final prediction result M2 is output; S5. Construct a degradation or failure mechanism model based on Archard wear theory to describe the degradation process from a mechanism level; S6. Based on the operating condition parameters, multiple sets of simulation experimental data are obtained through orthogonal experimental design, the mechanism model of degradation or failure is estimated, the mechanism model that changes with the operating time under certain operating conditions is obtained, and the output M3 of the timing condition is obtained; S7. The first neural network model, the second neural network model, and the mechanism model that changes with the operating time under certain working conditions are used as sub-models. The sub-models are integrated using the principle of integrated learning. The outputs M1, M2, and M3 of the sub-models are classified and averaged to obtain a comprehensive prediction result of different sub-models, and the final life prediction result is output.

2. The machine equipment life prediction method based on digital-analog linkage according to claim 1 is characterized in that: The key features described in S2 include: Mean value mean: where x i is the vibration value at a certain time point within the signal length range, N is the signal length; Standard deviation std: Root mean square value rms: Maximum value max: max(x i ); Minimum value min: min(x i ); Peak to peak: max(x i )-min(x i ); Kurt: Skewness: Crest factor): Shape factor: in Clearance factor: Impulse factor: Peak frequency: freqs[argmax(PSD(f))], where Where, f is the frequency; Total power: ∑PSD(f); Spectral centroid: Spectral kurt: psd i is the power spectral density value of the i-th frequency point, and psd is the average value of the power spectral density; Spectral skew: The features extracted by the wavelet packet decomposition method in S2 include the energy in the corresponding frequency bands of the decomposed approximate coefficients and detail coefficients. The number of layers of wavelet packet decomposition is 6, and the db3 wavelet basis function is used.

3. The machine equipment life prediction method based on digital-analog linkage according to claim 1 is characterized in that: In S3, the correlation coefficient is calculated as follows: Among them, x n is the nth element of the eigenvalue sequence, y n is the nth element in the sequence of remaining life value or life index, is the mean of the eigenvalue sequence, is the remaining life value or the mean of the life indicator series; When |ρ xy The closer | is to 1, the stronger the linear correlation between X and Y. The rated value is 0.

9. Select |ρ xy The features with |>0.9 are taken as the feature set.

4. The machine equipment life prediction method based on digital-analog linkage according to claim 1 is characterized in that: In S4, the steps for building the first neural network model are as follows: The input data X is transmitted to the hidden layer and nonlinearly mapped by the radial basis function: The output of the hidden layer is used as input again and passed to the output layer through linear calculation, mapping the features to an actual predicted value to form the final network output M1, which is the predicted value of the remaining life.

5. The machine equipment life prediction method based on digital-analog linkage according to claim 1 is characterized in that: In S4, the steps for building the second neural network model are as follows: The input data X is transmitted to the convolution layer for feature map extraction and convolution processing: X′=Conv1(X) Then X′ is normalized by layer: Where E[X′] represents the mean of X′, Var[X′] represents the variance of X′; ∈ is a positive number to avoid division by zero; γ represents a scaling parameter used to readjust the scale of the normalized data; β represents an offset parameter used to readjust the center position of the normalized data; ⊙ represents element-wise multiplication; The normalized data X LN Transformed by a linear layer: X Linear =W Linear X LN +b Linear Where W Linear represents the weight matrix of the linear layer, b Linear represents the bias term of the linear layer; The data X after linear transformation Linear Apply the Relu activation function: X Relu =X Linear ·Relu(X Linear ) In the formula, Relu(X Linear ) represents the Relu activation function; Data X after Relu activation Relu The Dropout layer reduces overfitting: X Dropout =Dropout(X Relu ) Data X after Dropout Dropout The final prediction result M2 is obtained through the fully connected layer and the regression layer.

6. The machine equipment life prediction method based on digital-analog linkage according to claim 1 is characterized in that: The degradation or failure mechanism model in S5 is as follows: O={σ,τ,∈,α} L = f(t, O) Among them, O is a set of factors affecting the remaining life, σ represents stress-related parameters, τ represents temperature-related parameters, ∈ represents deformation-related parameters, t represents operating time, L represents the remaining life or direct variables related to life, and α represents other influencing parameters.

7. The machine equipment life prediction method based on digital-analog linkage according to claim 6 is characterized in that: S6 specifically includes: Through the orthogonal experimental design method, the multi-factor and multi-level simulation experimental data P = {p1, p2, ...} are obtained. The specific steps are as follows: Determine the number of factors, which are determined as three factors here according to the set O in S5: speed, feed rate and axial cutting depth; Determine the number of levels. According to the allowable range of process parameters, each factor is set to four levels; The number of factors and levels were orthogonalized, and 16 groups of experiments were conducted; The simulation experiment data P is input into the S5 degradation model, the parameters contained in the parameter set O of the degradation or failure mechanism model are estimated, the construction of the mechanism model that changes with the operating time under certain working conditions is completed, and the output M3 of the timing condition is obtained.

8. The machine equipment life prediction method based on digital-analog linkage according to claim 1 is characterized in that: S7 specifically includes the following steps: Use each sub-model as a sub-model learner in the idea of ​​ensemble learning bagging algorithm; Combining the comprehensive strategy of regression task and classification task, the output results M1, M2 and M3 of each sub-model are classified and then fused by taking the weighted average method as the final output result.

9. A storage medium, characterized in that: The storage medium includes a stored program, wherein when the program is run, the machine equipment life prediction method based on digital-analog linkage as described in any one of claims 1 to 8 is executed.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: The processor executes the machine equipment life prediction method based on digital-analog linkage as described in any one of claims 1 to 8 through the computer program.

Citation Information

Cited By

  • Battery health state determination method and device

    CN119780770A

  • A method and device for determining the state of health of a battery

    CN119780770B

  • A method and system for controlling the inspection of fan component materials

    CN121187119B