Machine residual life prediction method, system and equipment based on hybrid data driving model

Abstract

The invention belongs to the technical field of machine life prediction, and particularly relates to a machine residual life prediction method, system and equipment based on a mixed data driving model, and the method comprises the steps: collecting vibration, temperature and sound data of equipment, extracting time domain and frequency domain features, and carrying out the preliminary prediction of the deterioration degree of the equipment through a support vector machine (SVM) model; and the prediction result and the feature data are input into a neural network model, a fatigue life formula is combined as a physical constraint, the physical consistency of the model is enhanced, and finally the residual life of the equipment is predicted. According to the method, data driving and a physical model are fused, the prediction accuracy and interpretability are improved, and the method is suitable for health management of complex industrial equipment. The system can be deployed on an edge computing device or a cloud, supports real-time prediction and dynamic updating, effectively prolongs the service life of the device, reduces sudden faults, and optimizes a maintenance strategy.

Description

Technical Field

[0001] The present invention belongs to the technical field of machine life prediction, and specifically relates to a method, system and device for predicting the remaining life of a machine based on a hybrid data-driven model. Background Art

[0002] Remaining Useful Life (RUL) prediction of a machine refers to predicting how long the equipment can continue to work under the current use status by analyzing the monitoring data and operating status of the equipment. RUL prediction is of great significance in extending the service life of the equipment, reducing sudden failures, and optimizing maintenance strategies. With the popularization of industrial automation, the operating efficiency and stability of various mechanical equipment have become important indicators of industrial production. Many equipment are at risk of wear, aging or failure during use, which makes the remaining life prediction of the equipment particularly important. Traditional equipment maintenance and replacement often rely on experience, resulting in inaccurate maintenance timing, increased costs and downtime.

[0003] Therefore, the remaining life prediction technology (RUL prediction) came into being, which mainly evaluates the health status of the equipment by monitoring its operating data (such as temperature, vibration, pressure, etc.) and predicts the remaining service life of the equipment through data modeling. Effective RUL prediction can help manufacturers, service providers and operators make more accurate maintenance decisions, optimize the scheduling of maintenance resources, maximize equipment availability and reduce unexpected downtime.

[0004] Research in the field of machine remaining life prediction mainly involves the following aspects: Data-driven method: By analyzing the historical data and sensor data of the equipment, machine learning, deep learning and other technologies are used to build models to predict the RUL of the equipment. Common algorithms include random forest, support vector machine (SVM), neural network, convolutional neural network (CNN), long short-term memory network (LSTM), etc. The advantage of these methods is that they can automatically learn rules from data and do not rely on the physical model of the equipment.

[0005] Physical model-based approach: By analyzing the structure and working principle of the equipment, combined with physical models to predict the aging process of the equipment. This method is often used for complex equipment, especially in high-risk industries such as aviation and nuclear power, although it usually requires more expert knowledge and data.

[0006] Hybrid models: Combine data-driven models with physical models to improve prediction accuracy using the strengths of both. Hybrid methods are often better at handling complex nonlinear systems and dynamic environments.

[0007] Deep learning and time series data: In recent years, deep learning has been widely used in RUL prediction, especially in processing time series data. For example, neural networks such as LSTM (Long Short-Term Memory Network) and GRU (Gated Recurrent Unit) can effectively capture the time series characteristics of equipment health data, thereby improving the accuracy of RUL prediction.

[0008] Existing methods or studies have one or more of the following defects and deficiencies: 1. The accuracy of RUL prediction often relies on a large amount of high-quality historical data, but in many practical applications, the data may be missing, noisy, or unbalanced. In particular, in some equipment operating environments, it is difficult to collect sufficient health data, which limits the training effect of the prediction model.

[0009] 2. Machine learning and deep learning algorithms perform well on specific data sets, but the generalization ability of the model under different devices, environments or operating conditions is still a problem. In many cases, the trained model cannot be well applied to new devices or new environments, resulting in unsatisfactory prediction results.

[0010] 3. Deep learning has achieved remarkable results in image processing and natural language processing, but its application in fault diagnosis is still in its early stages and there are few related patents.

[0011] 4. In industrial production, RUL prediction usually needs to be performed in real time, and deep learning models (especially complex models such as LSTM and CNN) often require large amounts of computation and a long training process. Although many patents have proposed optimization algorithms to improve computing efficiency, in scenarios where high-frequency real-time predictions are required, the limitation of computing resources is still a bottleneck.

[0012] 5. In a high-speed, dynamically changing environment, existing technologies still have limitations in data processing speed and real-time performance, and it is difficult to meet the needs of industrial sites.

[0013] 6. Machine learning and deep learning models are often considered “black box” models that lack sufficient interpretability. In practical applications, especially for predictions of key equipment (such as aircraft engines, nuclear power facilities, etc.), it is very important to understand why the model makes a certain prediction.

[0014] 7. Many patents propose prediction methods based on physical degradation models, but in practical applications, accurately modeling the degradation process of equipment is still a major challenge. Equipment degradation is usually complex and affected by many factors, including load changes, working environment, maintenance history, etc., which makes the construction of degradation models very difficult. Summary of the invention

[0015] The purpose of the present invention is to provide a machine remaining life prediction method, system and equipment based on a hybrid data-driven model to solve the technical problems of limited prediction accuracy and lack of physical mechanism considerations in existing equipment remaining life prediction methods.

[0016] The present invention achieves the above-mentioned purpose through the following technical solutions: The present invention proposes a method for predicting the remaining life of a machine based on a hybrid data-driven model, the method comprising: Collect and fuse vibration, temperature and sound data from the equipment under test to extract target feature indicators; Input the target characteristic index into a trained first prediction model to obtain a prediction result of the degradation degree of the device under test; wherein the first prediction model is based on an SVM model and is trained by target characteristic indexes and degradation degree labels of multiple devices; The degradation degree prediction result and the target characteristic index are input into a trained second prediction model to obtain the remaining life prediction result of the equipment under test; wherein, the second prediction model is obtained after the introduction of a neural network model based on physical constraints and a fatigue life formula.

[0017] Optionally, the collecting and fusing of vibration, temperature and sound data from the device under test to extract target characteristic indicators includes: Collect data from vibration sensors, sound sensors, and temperature sensors when the equipment under test is running; Clean and standardize or normalize the collected data to extract effective features; Extract the time domain and frequency domain features of vibration data, the time domain and frequency domain features of sound data, and the time series features of temperature data; Among them, the time domain characteristics of vibration data include vibration acceleration amplitude, acceleration effective value, margin, kurtosis, skewness, peak-to-peak value, and pulse; the frequency domain characteristics of vibration data include resonance frequency amplitude; The time domain features of sound data include effective value, loudness, voiceprint roughness, frequency center of gravity, sharpness, high and low frequency band energy ratio, peak-to-peak value, pulse, margin, kurtosis, and skewness; the frequency domain features of sound data include Mel spectrum coefficient, Mel cepstral coefficient, GFCC coefficient, peak value of power spectrum and corresponding frequency.

[0018] Optionally, before inputting the target feature index into the trained first prediction model, the method further includes: Constructing a training data set including historical data and degradation degree labels of multiple devices; the historical data includes target feature indicators corresponding to the devices; The training data set is used to train the SVM model to learn the degradation law of the equipment and obtain a trained first prediction model.

[0019] Optionally, before inputting the degradation degree prediction result and the target characteristic index into the trained second prediction model, the method further includes: Determining parameters of the fatigue life formula; Design the structure of the neural network model, including input layer, hidden layer and output layer; The degradation degree prediction result is used as an input feature of the neural network model, and is input into the neural network together with the target feature index; The fatigue life formula is embedded in the neural network model as a constraint condition to form a physical constraint layer to enhance the physical consistency of the model; The neural network model is trained using the preprocessed dataset and the model parameters are adjusted to optimize the prediction performance.

[0020] Optionally, the fatigue life formula is specifically the Manson-Coffin formula, which is expressed as follows: ; in, Δε / 2 represents the total strain amplitude; σ' f represents fatigue strength coefficient; E represents the elastic modulus; b represents fatigue strength index; ε' f represents the fatigue ductility coefficient; c represents the fatigue ductility index; N f Indicates the number of failure cycles.

[0021] Optionally, fatigue life formulas are introduced into the physically constrained neural network model, including: Add the residual term of the Manson-Coffin formula to the loss function to force the predicted life to satisfy the physical relationship, and its expression is: ; Among them, λ represents the constraint weight, which is used to balance data-driven and physical consistency; MES represents the mean square error, N pred represents the value predicted by the model, N true Represents the actual observed value or labeled value; A physical calculation layer is added after the network output layer. The theoretical strain amplitude is calculated by the Manson-Coffin formula and compared with the input measured strain amplitude. The error is back-propagated to obtain the updated second prediction model.

[0022] Optionally, the method further includes: Input real-time sensor data for feature extraction and model prediction, calculate theoretical life and compare it with the remaining life prediction results to verify physical consistency; The prediction performance was evaluated using the MSE and MAE indicators.

[0023] The present invention also proposes a machine remaining life prediction system based on a hybrid data-driven model, the system comprising: The data acquisition module is used to collect and fuse the vibration, temperature and sound data from the equipment under test to extract the target characteristic indicators; A first prediction module is used to input the target characteristic index into a trained first prediction model to obtain a prediction result of the degradation degree of the device under test; wherein the first prediction model is based on an SVM model and is trained by target characteristic indicators and degradation degree labels of multiple devices; The second prediction module is used to input the degradation degree prediction result and the target characteristic index into a trained second prediction model to obtain the remaining life prediction result of the equipment under test; wherein, the second prediction model is obtained after the introduction of a neural network model based on physical constraints and a fatigue life formula.

[0024] The present invention also provides an electronic device, comprising: A processor; a memory for storing instructions executable by the processor; The processor is configured to execute the instructions to implement the machine remaining life prediction method as described in any one of the above items.

[0025] The present invention also proposes a computer-readable storage medium. When the instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device can execute the machine remaining life prediction method as described in any one of the above items.

[0026] The beneficial effects of the present invention are: 1. The present invention proposes a method for predicting the remaining life of a machine based on a hybrid data-driven model, which can describe the degradation process of the equipment by using a physical model, and predict the actual operation data of the equipment by using an SVM model and introducing it, and finally obtain a more accurate remaining life prediction. In the field of mechanical fault diagnosis, introducing the fatigue life formula as a constraint condition in neural network training can improve the physical interpretability and consistency of the model, while optimizing the model performance and prediction accuracy.

[0027] 2. The present invention can comprehensively consider the characteristics of various types of data, extract more effective fault indicators, achieve more comprehensive fault identification, integrate the diagnostic results of multiple sensors, and ensure the final diagnostic accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] Figure 1 It is a schematic diagram of the process of the present invention; Figure 2 It is a schematic diagram of vibration data in case 1 of a specific embodiment of the present invention; Figure 3 It is a schematic diagram of sound data in Example 1 of a specific implementation mode of the present invention; Figure 4 It is a schematic diagram of temperature timing data in case 1 of a specific implementation method of the present invention. DETAILED DESCRIPTION

[0029] The following description provides specific application scenarios and requirements of this specification, with the purpose of enabling those skilled in the art to make and use the contents of this specification. Various local modifications to the disclosed embodiments will be apparent to those skilled in the art, and the general principles defined herein may be applied to other embodiments and applications without departing from the spirit and scope of this specification. Therefore, this specification is not limited to the embodiments shown, but to the widest scope consistent with the claims.

[0030] The terms used herein are only used for the purpose of describing specific example embodiments and are not restrictive. For example, unless the context clearly indicates otherwise, as used herein, the singular forms "a", "an" and "the" may also include plural forms. When used in this specification, the terms "include", "comprise" and / or "contain" mean that the associated integers, steps, operations, elements and / or components exist, but do not exclude the existence of one or more other features, integers, steps, operations, elements, components and / or groups or that other features, integers, steps, operations, elements, components and / or groups may be added in the system / method.

[0031] In view of the following description, these and other features of the present specification, as well as the operation and function of the related elements of the structure, and the economy of the combination and manufacture of the parts can be significantly improved. Reference is made to the accompanying drawings, all of which form a part of this specification. However, it should be clearly understood that the drawings are for illustration and description purposes only and are not intended to limit the scope of this specification. It should also be understood that the drawings are not drawn to scale.

[0032] The flowcharts used in this specification illustrate the operations implemented by the system according to some embodiments in this specification. It should be clearly understood that the operations of the flowcharts may not be implemented in sequence. On the contrary, the operations may be implemented in reverse order or simultaneously. In addition, one or more other operations may be added to the flowchart. One or more operations may be removed from the flowchart.

[0033] Example 1

[0034] See also Figure 1The present application embodiment proposes a method for predicting the remaining life of a machine based on a hybrid data-driven model. The method integrates multi-source data, machine learning models and physical constraints to achieve accurate prediction of the remaining life of the equipment. The following are specific implementation details of the present invention: Step 1: Data collection and feature extraction The present invention first collects vibration, temperature and sound data of the equipment to be tested. These data reflect the operating status and potential fault information of the equipment. Subsequently, these data are fused and preprocessed to extract target characteristic indicators. These characteristic indicators can reflect the degree of degradation and health status of the equipment, providing key input for subsequent predictions.

[0035] In specific implementation, this step includes data collection and preprocessing: (1) Data collection: Collect real-time operation data from the vibration sensor, sound sensor and temperature sensor of the device under test. The vibration sensor is used to collect the vibration signal of the device, the sound sensor is used to collect the sound signal when the device is running, and the temperature sensor is used to collect the temperature change of the device.

[0036] The collected data include vibration acceleration, sound spectrum, temperature change, etc., to ensure that the data covers different operating conditions of the equipment (such as normal, slightly degraded, severely degraded, etc.).

[0037] (2) Data preprocessing: Clean the collected data to remove noise and outliers. For example, use filtering algorithms to remove high-frequency noise and use interpolation to fill in missing values.

[0038] Standardize or normalize the data to ensure that the data from different sensors are on the same scale. For example, use the Z-score normalization method to transform the data into a distribution with a mean of 0 and a variance of 1.

[0039] In specific implementation, this step also includes feature extraction: (1) Vibration data feature extraction: Extract the time domain features of vibration data, including vibration acceleration amplitude, acceleration effective value, margin, kurtosis, skewness, peak-to-peak value, pulse, etc.

[0040] Extract frequency domain features of vibration data, including resonance frequency amplitude, frequency domain energy distribution, etc. Use Fast Fourier Transform (FFT) to convert time domain signals into frequency domain signals and extract the amplitude of key frequency bands.

[0041] (2) Sound data feature extraction: Extract the time domain features of sound data, including effective value, loudness, voiceprint roughness, frequency center of gravity, sharpness, high and low frequency band energy ratio, peak-to-peak value, pulse, margin, kurtosis, skewness, etc.

[0042] Extract frequency domain features of sound data, including Mel Spectral Coefficients (MFSC), Mel Cepstral Coefficients (MFCC), GFCC coefficients, peak values ​​of power spectrum and corresponding frequencies, etc. Use Mel filter banks to process sound signals and extract frequency domain features.

[0043] (3) Temperature data feature extraction: Extract the time series features of temperature data, including the mean, variance, peak value, etc. of temperature changes. Use the sliding window method to calculate the statistical features of temperature data.

[0044] Step 2: Preliminary SVM degradation prediction Next, the extracted target characteristic index is input into the first prediction model based on SVM. The SVM model has been trained with a training data set containing multiple equipment historical data and degradation degree labels, and has the ability to learn the equipment degradation law. Through the calculation of the SVM model, the preliminary degradation degree prediction result of the equipment to be tested can be obtained. This step provides important reference information for the subsequent neural network prediction.

[0045] Accordingly, a training dataset containing historical data and degradation degree labels of multiple devices is constructed. The historical data includes characteristic indicators of vibration, sound and temperature, and the degradation degree label is the health status of the device (such as normal, slightly degraded, and severely degraded).

[0046] Accordingly, train the SVM model: use the training data set to train the SVM model, select the radial basis function (RBF) as the kernel function, and optimize the model parameters (such as C value and gamma value) through grid search and cross-validation. After training, the SVM model can predict the degree of equipment degradation based on the input feature indicators.

[0047] Correspondingly, preliminary prediction: input the real-time collected equipment characteristic indicators into the trained SVM model, and output the preliminary prediction results of the equipment degradation degree. For example, output the health status of the equipment (such as normal, slightly degraded, and severely degraded).

[0048] Step 3: Neural Network Remaining Life Prediction After obtaining the initial degradation prediction results of SVM, they are input into the neural network model based on physical constraints together with the target characteristic index. The neural network model combines the nonlinear mapping ability of neural network and the physical constraints of fatigue life formula to improve the accuracy and physical consistency of prediction.

[0049] The construction of the neural network includes input layer, hidden layer and output layer. The input layer receives the target characteristic index and the initial degradation degree prediction result of SVM, the hidden layer realizes the nonlinear transformation of the input data through complex network connection and activation function, and the output layer outputs the remaining life prediction result of the equipment under test.

[0050] In order to enhance the physical consistency of the model, the present invention embeds the Manson-Coffin fatigue life formula as a constraint in the neural network model. During the training process, the predicted life is forced to satisfy the physical relationship by adding the formula residual term to the loss function. At the same time, a physical calculation layer is added after the network output layer to calculate the theoretical strain amplitude through the formula, and compare it with the input measured strain amplitude, and back-propagate the error to optimize the model parameters.

[0051] Step 4: Model validation and optimization After the model is built, the present invention also provides a model verification and optimization step. By inputting real-time sensor data for feature extraction and model prediction, the theoretical life is calculated and compared with the remaining life prediction results to verify the physical consistency of the model. At the same time, the prediction performance is evaluated using indicators such as MSE (mean square error) and MAE (mean absolute error), and the model is optimized and adjusted according to the evaluation results.

[0052] In the specific implementation, taking a certain type of industrial equipment as an example, the present invention first collects the vibration, temperature and sound data of the equipment, and extracts the target characteristic index through data preprocessing. Then, the SVM model is trained using a training data set containing the historical data and degradation degree labels of the equipment. After obtaining the initial degradation degree prediction result of the SVM, it is input together with the target characteristic index into the neural network model based on physical constraints for remaining life prediction.

[0053] In the construction of the neural network model, the present invention adopts a multi-layer perceptron structure and selects ReLU as the activation function to enhance the nonlinear expression ability of the model. At the same time, when embedding the Manson-Coffin fatigue life formula, the present invention balances the relationship between prediction accuracy and physical consistency by adjusting the constraint weight λ.

[0054] More specifically, the parameters of the fatigue life formula are determined. The fatigue life formula is specifically the Manson-Coffin formula, which is expressed as follows: ; in, Δε / 2 represents the total strain amplitude; σ' f represents fatigue strength coefficient; E represents the elastic modulus; brepresents fatigue strength index; ε' f represents the fatigue ductility coefficient; c represents the fatigue ductility index; N f Indicates the number of failure cycles.

[0055] During specific implementation, the parameters in the above formula are determined through experimental data.

[0056] More specifically, a neural network structure is designed, including an input layer, a hidden layer, and an output layer; the degradation degree prediction result is used as an input feature of the neural network model and input into the neural network together with the target feature index; the fatigue life formula is embedded in the neural network model as a constraint condition to form a physical constraint layer to enhance the physical consistency of the model; the neural network model is trained using the preprocessed data set, and the model parameters are adjusted to optimize the prediction performance.

[0057] More specifically, the neural network model is trained: the neural network model is trained using the preprocessed data set, and the loss function includes a data-driven part and a physical constraint part: in a specific embodiment, the fatigue life formula is introduced into the physically constrained neural network model, including: adding the residual term of the Manson-Coffin formula to the loss function, forcing the predicted life to satisfy the physical relationship, and its expression is: ; Where λ represents the constraint weight, which is used to balance data-driven and physical consistency. MES represents mean square error, N pred represents the value predicted by the model, N true Represents the actual observed value or labeled value, and the other parameters have been described above.

[0058] A physical calculation layer is added after the network output layer. The theoretical strain amplitude is calculated by the Manson-Coffin formula and compared with the input measured strain amplitude. The error is back-propagated to obtain the updated second prediction model.

[0059] In a specific embodiment, the method further includes a final prediction: inputting real-time sensor data for feature extraction and model prediction, calculating the theoretical life and comparing it with the remaining life prediction result, verifying the physical consistency, for example, the residual difference between the predicted life and the theoretical life should be less than 5%; using the mean square error (MSE) and mean absolute error (MAE) indicators to evaluate the prediction performance. For example, the MSE should be less than 100 and the MAE should be less than 10.

[0060] Example 2

[0061] This embodiment proposes a machine remaining life prediction system based on a hybrid data-driven model, the system comprising: The data acquisition module is used to collect and fuse the vibration, temperature and sound data from the equipment under test to extract the target characteristic indicators; A first prediction module is used to input the target characteristic index into a trained first prediction model to obtain a prediction result of the degradation degree of the device under test; wherein the first prediction model is based on an SVM model and is trained by the target characteristic indexes and degradation degree labels of multiple devices; The second prediction module is used to input the degradation degree prediction result and the target characteristic index into the trained second prediction model to obtain the remaining life prediction result of the equipment to be tested; wherein the second prediction model is obtained after the introduction of the neural network model based on physical constraints and the fatigue life formula.

[0062] It can be understood that the life prediction system in this embodiment can be deployed on an edge computing device to achieve real-time prediction; it can also be deployed in the cloud to dynamically update model parameters.

[0063] Case 1 Step 1: Data Collection The sensor collects the vibration data, sound and image data of the actual blades at a wind power site and inputs them into the server. Figure 2-Figure 4 shown.

[0064] Step 2: Data feature extraction 2.1 Extract the characteristic indicators of vibration data. The extracted indicators are 12 indicators: acceleration effective value, acceleration peak-to-peak value, acceleration pulse, acceleration margin, acceleration kurtosis, acceleration effective value, velocity effective value, velocity peak-to-peak value, velocity pulse, velocity margin, velocity kurtosis, and velocity skewness. The vibration characteristic indicator array is calculated as follows: a = [12, 106, 5.47, 10.48, 3.05, -0.048, 21, 72.89, 1.52, 0.01,19.34, 1.99] Select multiple devices, including healthy and faulty device data, and measure multiple sets of data to establish vibration data set A.

[0065] A= [a1,a2,a3,...an]; 2.2 Extract the characteristic indicators of sound data. The extracted indicators are effective value, loudness, voiceprint roughness, frequency center of gravity, sharpness, high-low frequency energy ratio, peak-to-peak value, pulse, margin, kurtosis, and skewness. The calculation indicators are as follows: s = [435, 51, 976, 890, 0.264, 0.359, 3670, 8.042, 6.459, 3.69, -0.018]; Select multiple devices, including healthy and faulty device data, and measure multiple groups of data to establish a vibration data set S, S = [s1, s2, s3, ... sn].

[0066] 2.3 Extract the corresponding temperature data and establish a time series data set.

[0067] Step 3: Machine Learning Modeling and Training According to the measured data, a data set is constructed containing the historical data of the equipment and the corresponding degradation degree labels. The structure of the data set is shown in Table 1: Table 1

[0068] "Equipment characteristics" are different indicators of the equipment, such as temperature, pressure, vibration, etc. "Degradation" is the target variable, usually a categorical label, indicating the degradation of the equipment.

[0069] According to step 2, the decision boundary corresponding to the data in Table 1, that is, the hyperplane, is obtained.

[0070] The decision boundary function has the following form: ; in, , is the dual variable, controlling the contribution of samples to the regression hyperplane; b is the offset of the hyperplane relative to the origin, and this model is the RBF kernel; K ( x i , x ) is the corresponding kernel function, N Indicates the number of samples.

[0071] Step 4: Fatigue Life Formula for Physical Constraints 4.1 Data Acquisition Test bench: Apply cyclic loads to the blades and record vibration signals, strain gauge data and failure times.

[0072] Material parameters: σf′=1200MPa εf′=0.25 b=-0.12 c=-0.55 4.2 Data Acquisition Obtain the corresponding vibration, sound, and temperature data from step 2 4.3 Network Structure Input: strain amplitude, mean stress, temperature, decision boundary function parameters.

[0073] Hidden layers: 3 fully connected layers (256-128-64 neurons, ReLU activation).

[0074] Physical constraint layer: The prediction parameters are put into the loss function formula for calculation.

[0075] Output: Remaining life N pred .

[0076] Loss function: MSE + Manson-Coffin constraint (λ=0.5).

[0077] 4.4 Results comparison Combined with Table 2, it shows the physical consistency of the prediction results of the pure data-driven model and the physical constraint model proposed in this application. It can be seen that the physical constraint model significantly improves the prediction accuracy, and the residual error of the predicted life and the theoretical value of the Manson-Coffin formula is less than 5%.

[0078] Table 2

[0079] In another embodiment of the present application, an electronic device is proposed, comprising: a processor; a memory for storing processor executable instructions; wherein the processor is configured to execute instructions to implement the machine remaining life prediction method as described above.

[0080] In another embodiment of the present application, a computer-readable storage medium is provided. When instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the machine remaining life prediction method as described above.

[0081] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0082] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A method for predicting the remaining life of a machine based on a hybrid data-driven model, characterized in that: The method comprises: Collect and fuse vibration, temperature and sound data from the equipment under test to extract target feature indicators; Input the target characteristic index into a trained first prediction model to obtain a prediction result of the degradation degree of the device under test; wherein the first prediction model is based on an SVM model and is trained by target characteristic indexes and degradation degree labels of multiple devices; The degradation degree prediction result and the target characteristic index are input into a trained second prediction model to obtain the remaining life prediction result of the equipment under test; wherein, the second prediction model is obtained after the introduction of a neural network model based on physical constraints and a fatigue life formula.

2. The method for predicting the remaining life of a machine based on a hybrid data-driven model according to claim 1, characterized in that: The collection and fusion of vibration, temperature and sound data from the equipment under test to extract target characteristic indicators include: Collect data from vibration sensors, sound sensors, and temperature sensors when the equipment under test is running; Clean and standardize or normalize the collected data to extract effective features; Extract the time domain and frequency domain features of vibration data, the time domain and frequency domain features of sound data, and the time series features of temperature data; Among them, the time domain characteristics of vibration data include vibration acceleration amplitude, acceleration effective value, margin, kurtosis, skewness, peak-to-peak value, and pulse; the frequency domain characteristics of vibration data include resonance frequency amplitude; The time domain features of sound data include effective value, loudness, voiceprint roughness, frequency center of gravity, sharpness, high and low frequency band energy ratio, peak-to-peak value, pulse, margin, kurtosis, and skewness; the frequency domain features of sound data include Mel spectrum coefficient, Mel cepstral coefficient, GFCC coefficient, peak value of power spectrum and corresponding frequency.

3. The method for predicting the remaining life of a machine based on a hybrid data-driven model according to claim 1, characterized in that: Before inputting the target feature index into the trained first prediction model, the method further includes: Constructing a training data set including historical data and degradation degree labels of multiple devices; the historical data includes target feature indicators corresponding to the devices; The training data set is used to train the SVM model to learn the degradation law of the equipment and obtain a trained first prediction model.

4. The method for predicting the remaining life of a machine based on a hybrid data-driven model according to claim 1, characterized in that: Before inputting the degradation degree prediction result and the target characteristic index into the trained second prediction model, the method further includes: Determining parameters of the fatigue life formula; Design the structure of the neural network model, including input layer, hidden layer and output layer; The degradation degree prediction result is used as an input feature of the neural network model, and is input into the neural network together with the target feature index; The fatigue life formula is embedded in the neural network model as a constraint condition to form a physical constraint layer to enhance the physical consistency of the model; The neural network model is trained using the preprocessed dataset and the model parameters are adjusted to optimize the prediction performance.

5. The method for predicting the remaining life of a machine based on a hybrid data-driven model according to claim 4, characterized in that: The fatigue life formula is specifically the Manson-Coffin formula, which is expressed as follows: ； in, Δε / 2 represents the total strain amplitude; σ' f represents fatigue strength coefficient; E represents the elastic modulus; b represents fatigue strength index; ε' f represents the fatigue ductility coefficient; c represents the fatigue ductility index; N f Indicates the number of failure cycles.

6. The method for predicting the remaining life of a machine based on a hybrid data-driven model according to claim 5, characterized in that: The fatigue life formula is introduced into the physical constraint neural network model, including: Add the residual term of the Manson-Coffin formula to the loss function to force the predicted life to satisfy the physical relationship, and its expression is: ； Where λ represents the constraint weight, which is used to balance data-driven and physical consistency; MES represents mean square error, N pred Represents the value predicted by the model, N true Represents the actual observed value or labeled value; A physical calculation layer is added after the network output layer. The theoretical strain amplitude is calculated by the Manson-Coffin formula and compared with the input measured strain amplitude. The error is back-propagated to obtain the updated second prediction model.

7. The method for predicting the remaining life of a machine based on a hybrid data-driven model according to claim 1, characterized in that: The method further comprises: Input real-time sensor data for feature extraction and model prediction, calculate theoretical life and compare with remaining life prediction results to verify physical consistency; The prediction performance was evaluated using the MSE and MAE indicators.

8. A machine remaining life prediction system based on a hybrid data-driven model, characterized in that: The system comprises: The data acquisition module is used to collect and fuse the vibration, temperature and sound data from the equipment under test to extract the target characteristic indicators; A first prediction module is used to input the target characteristic index into a trained first prediction model to obtain a prediction result of the degradation degree of the device under test; wherein the first prediction model is based on an SVM model and is trained by target characteristic indicators and degradation degree labels of multiple devices; The second prediction module is used to input the degradation degree prediction result and the target characteristic index into a trained second prediction model to obtain the remaining life prediction result of the equipment under test; wherein, the second prediction model is obtained after the introduction of a neural network model based on physical constraints and a fatigue life formula.

9. An electronic device, characterized in that: include: processor; a memory for storing instructions executable by the processor; The processor is configured to execute the instructions to implement the machine remaining life prediction method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: When the instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the machine remaining life prediction method according to any one of claims 1 to 7.

Images