A mechanical component life prediction method, device, equipment and medium

Abstract

This application discloses a method, apparatus, equipment, and medium for predicting the lifespan of mechanical components, relating to the field of mechanical component testing technology. The method includes: acquiring condition monitoring data of the mechanical component to be predicted during operation; extracting a target feature set from the condition monitoring data that satisfies preset constraints and preset evaluation conditions; inputting the target feature set into a preset lifespan prediction model to output the remaining service life of the mechanical component to be predicted; wherein the preset lifespan prediction model is a model established by integrating the degradation trajectories of several historical mechanical components to map the correlation between the target feature set and the health status of the component. By extracting the target feature set that satisfies the preset constraints and preset evaluation conditions, it is ensured that the features selected from the target feature set must be related to the physical mechanism of component failure. Selecting features through data indicators ensures that the selected target feature set reflects the complex degradation process in actual operation, making lifespan prediction more accurate.

Description

Technical Field

[0001] This invention relates to the field of mechanical component testing technology, and in particular to a method, apparatus, equipment and medium for predicting the lifespan of mechanical components. Background Technology

[0002] In the field of mechanical equipment condition monitoring and health management, performance degradation assessment and remaining service life prediction of key mechanical components (such as bearings and gears) are core technologies for achieving predictive maintenance and ensuring safe operation. Existing prediction methods can be mainly divided into three categories: 1. Physical mechanism model-based approach: This approach relies on a deep understanding of the physical process of component failure. It describes the performance degradation by establishing differential equations based on fault dynamics and analytical models of degradation trajectories. The advantage is that the model has a clear physical meaning and strong interpretability. However, it is often difficult to cover the complex and variable failure modes and individual differences in actual operation, resulting in insufficient characterization ability and poor adaptability of the constructed feature set for the actual degradation process.

[0003] 2. Data-driven approach: This approach does not rely on prior physical models but directly learns the degradation patterns and lifespan characteristics of components from historical monitoring data (e.g., vibration and temperature signals) through statistical analysis, machine learning, or deep learning algorithms. Its advantage lies in its strong adaptability, capable of handling complex and nonlinear degradation processes. However, it lacks a correlation with the physical processes of component failure, resulting in poor interpretability. This makes the prediction results difficult for domain experts to understand and trust, and it is prone to overfitting or unstable performance with small sample sizes.

[0004] 3. Hybrid-driven approach: This approach attempts to integrate the two approaches mentioned above. For example, it may use data-driven methods to calibrate the parameters of the mechanistic model or perform weighted fusion of the prediction results of the two models in order to balance interpretability and prediction accuracy. However, most hybrid approaches only combine the models at the model level (such as result fusion) without achieving the essential integration and mutual constraint of mechanistic knowledge and statistical data laws at the fundamental stage of feature engineering.

[0005] In summary, how to ensure that the extracted feature set can be closely related to the physical mechanism of component failure to guarantee interpretability and fully conform to the statistical laws of data, so as to make the prediction of the remaining life of mechanical components accurate, is a technical problem to be solved in this field. Summary of the Invention

[0006] In view of this, the purpose of this invention is to provide a method, apparatus, device, and medium for predicting the remaining life of mechanical components, which can ensure interpretability by extracting a feature set that is closely related to the physical mechanism of component failure and fully conforms to statistical data patterns, thereby enabling accurate prediction of the remaining life of mechanical components. The specific solution is as follows: In a first aspect, this application discloses a method for predicting the lifespan of mechanical components, including: Acquire status monitoring data of the mechanical component to be predicted during operation; Extract target feature sets that satisfy preset constraints and preset evaluation conditions from the condition monitoring data; wherein, the preset constraints are set based on the laws of the physical process of component failure, and the preset evaluation conditions are conditions set using data statistical indicators determined based on historical monitoring data; The target feature set is input into a preset life prediction model to output the remaining life of the mechanical component to be predicted; wherein, the preset life prediction model is a model established by integrating the degradation trajectories of several historical mechanical components to map the correlation between the target feature set and the health status of the component.

[0007] Optionally, before inputting the target feature set into the preset lifetime prediction model, the method further includes: The initial preset life prediction model is trained using the historical feature set of the historical mechanical component operation process to obtain the preset life prediction model.

[0008] Optionally, before training the initial preset life prediction model using the historical feature set from the historical operation of mechanical components, the method further includes: Acquire full lifecycle monitoring data of several historical mechanical components as raw training data; And / or, acquire monitoring data of several historical mechanical components from the occurrence of fault characteristics to the period of failure and disassembly as raw training data; Based on the health index corresponding to the starting point and failure point of the original training data, the original training data is labeled with corresponding health status labels to form a training sample set carrying health status labels. Extract a set of historical features from the training sample set that are related to the physical processes of historical mechanical component failures and to the characteristics driven by operational data.

[0009] Optionally, the step of using historical feature sets from the historical operation of mechanical components to train the initial preset lifespan prediction model to obtain the preset lifespan prediction model includes: The learner is trained using the historical feature set carrying health status labels to learn the mapping relationship from historical features to health status, and the degradation trajectory of the corresponding individual historical mechanical parts is obtained to acquire the degradation trajectory of each historical mechanical part. The several degradation trajectories are fused using an ensemble learning strategy to construct and obtain a preset lifetime prediction model.

[0010] Optionally, the health status label is associated with operating time or mileage through an S-shaped function curve that conforms to the degradation law of mechanical parts.

[0011] Optionally, acquiring the status monitoring data of the mechanical component to be predicted during operation includes: The vibration, impact, temperature, and operating condition information of the mechanical component to be predicted during operation are acquired to obtain condition monitoring data.

[0012] Optionally, after acquiring the vibration signal, impact signal, temperature signal, and operating condition information of the mechanical component to be predicted during operation to obtain condition monitoring data, the method further includes: The status monitoring data is checked for outliers or missing values ​​to obtain the corresponding detection results. If the detection result shows outliers, the outliers in the status monitoring data are removed by using the number of sample entries or the operating mileage to obtain preprocessed status monitoring data. If the detection result indicates the presence of missing values, historical status monitoring data is selected for filling to obtain preprocessed status monitoring data. The preprocessed state monitoring data is normalized to obtain the target state monitoring data.

[0013] Optionally, extracting the target feature set that satisfies preset constraints and preset evaluation conditions from the state monitoring data includes: Based on the preset constraints, a candidate feature set including time-domain statistical features, trend change features, and same-position comparison features is extracted from the state monitoring data; The target feature set is selected from the candidate feature set using the preset evaluation conditions.

[0014] Optionally, the step of extracting a candidate feature set from the state monitoring data based on the preset constraints, including time-domain statistical features, trend change features, and same-position comparison features, includes: Extract the corresponding state data values ​​representing different state monitoring indicators and the total number of state data values ​​per day from the state monitoring data, and calculate and determine the corresponding daily statistical characteristics of each state monitoring indicator to obtain the time domain statistical characteristics. Based on the daily statistical characteristics of each of the aforementioned status monitoring indicators, the daily average trend and monthly daily average trend of each of the aforementioned status monitoring indicators are determined to obtain the trend change characteristics. The same-location comparison feature is determined based on the difference between the first statistic of the current location status monitoring index and the second statistic of the other status monitoring indexes at the same location; wherein, the current location is the mechanical component to be predicted and the corresponding evaluation time, and the other same locations are other mechanical components of the same model and located in the same equipment that operate symmetrically or equivalently to the mechanical component to be predicted and the same evaluation time.

[0015] Optionally, the step of using the preset evaluation criteria to filter the target feature set from the candidate feature set includes: Calculate the monotonicity index, predictability index, stability index, and separability index for each candidate feature in the candidate feature set; The target feature set is obtained by filtering target features whose monotonicity index value, predictability index value, and separability index value are all greater than or equal to a first preset threshold and whose stability index value is greater than or equal to a second preset threshold.

[0016] Optionally, calculating the monotonicity index, predictability index, stability index, and separability index for each candidate feature in the candidate feature set includes: The monotonicity index of the current candidate feature is calculated based on the current candidate feature and the respective feature values ​​of the adjacent candidate features preceding the current candidate feature. The predictability index value of the candidate predictive feature is calculated based on the first mean of the candidate predictive feature representing the performance degradation feature at the failure time, the second mean of the candidate predictive feature at the initial time, and the standard deviation of the candidate predictive feature at the failure time. The stability index value of the current candidate feature is calculated based on the degradation trend parameter sequence of the current candidate feature and the candidate predictive feature; Calculate the Fisher ratio for each candidate feature between healthy and faulty states to obtain the separability index.

[0017] Optionally, inputting the target feature set into a preset life prediction model to output the remaining life of the mechanical component to be predicted includes: The target feature set is input into a preset lifespan prediction model, and multiple initial health indices are calculated and obtained using all the degradation trajectories in the preset lifespan prediction model. The multiple initial health indices are smoothed to obtain the predicted health index; The remaining service life of the mechanical component to be predicted is calculated based on the predicted health index, the model parameters corresponding to the degradation trajectory, and the average total lifespan of historically failed components.

[0018] Secondly, this application discloses a mechanical component life prediction device, comprising: The data monitoring module is used to acquire status monitoring data of the mechanical component to be predicted during operation; The feature extraction module is used to extract a set of target features that satisfy preset constraints and preset evaluation conditions from the condition monitoring data; wherein, the preset constraints are set based on the laws of the physical process of component failure, and the preset evaluation conditions are conditions set using data statistical indicators determined based on historical monitoring data; The life prediction module is used to input the target feature set into a preset life prediction model to output the remaining life of the mechanical component to be predicted; wherein, the preset life prediction model is a model established by integrating the degradation trajectories of several historical mechanical components to map the correlation between the target feature set and the health status of the component.

[0019] Thirdly, this application discloses an electronic device, including: Memory, used to store computer programs; A processor is configured to execute the computer program to implement the steps of the aforementioned disclosed method for predicting the lifespan of mechanical components.

[0020] Fourthly, this application discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the steps of the aforementioned disclosed method for predicting the lifespan of mechanical components.

[0021] As can be seen, this application discloses the following: acquiring condition monitoring data of a mechanical component to be predicted during operation; extracting a target feature set that satisfies preset constraints and preset evaluation conditions from the condition monitoring data; wherein the preset constraints are set based on the physical process law of component failure, and the preset evaluation conditions are conditions set using data statistical indicators determined based on historical monitoring data; inputting the target feature set into a preset life prediction model to output the remaining service life of the mechanical component to be predicted; wherein the preset life prediction model is a model established by integrating the degradation trajectories of several historical mechanical components to map the correlation between the target feature set and the health status of the component. Therefore, by extracting the target feature set that meets the preset constraints and evaluation conditions, it is ensured that the features selected from the target feature set must be related to the physical mechanism of component failure. This fundamentally guarantees that the target feature set used for prediction has clear physical meaning. At the same time, through data index screening, it is ensured that the features can reflect the complex degradation process in actual operation, and features that, although they conform to the ideal physical model, fluctuate greatly and have unclear trends in actual data are discarded. Furthermore, the preset life prediction model established by integrating the degradation trajectories of several historical mechanical components is based on the target feature set. It finds the most similar or performs weighted synthesis from the multiple integrated historical trajectories to make accurate life predictions. This improves the model's generalization ability when facing individual differences in components, changes in operating conditions, and small sample failure data, and directly solves the problem of over-reliance on a single historical sample. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0023] Figure 1 This is a flowchart of a method for predicting the lifespan of mechanical components disclosed in this application; Figure 2 This application discloses a flowchart of a preset life prediction model training process and a mechanical component life prediction method. Figure 3 This is a schematic diagram of the mechanical component life prediction device disclosed in this application; Figure 4 This is a structural diagram of an electronic device disclosed in this application. Detailed Implementation

[0024] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0025] In the field of mechanical equipment condition monitoring and health management, performance degradation assessment and remaining service life prediction of key mechanical components (such as bearings and gears) are core technologies for achieving predictive maintenance and ensuring safe operation. Existing prediction methods can be mainly divided into three categories: 1. Physical mechanism model-based approach: This approach relies on a deep understanding of the physical process of component failure. It describes the performance degradation by establishing differential equations based on fault dynamics and analytical models of degradation trajectories. The advantage is that the model has a clear physical meaning and strong interpretability. However, it is often difficult to cover the complex and variable failure modes and individual differences in actual operation, resulting in insufficient characterization ability and poor adaptability of the constructed feature set for the actual degradation process.

[0026] 2. Data-driven approach: This approach does not rely on prior physical models but directly learns the degradation patterns and lifespan characteristics of components from historical monitoring data (e.g., vibration and temperature signals) through statistical analysis, machine learning, or deep learning algorithms. Its advantage lies in its strong adaptability, capable of handling complex and nonlinear degradation processes. However, it lacks a correlation with the physical processes of component failure, resulting in poor interpretability. This makes the prediction results difficult for domain experts to understand and trust, and it is prone to overfitting or unstable performance with small sample sizes.

[0027] 3. Hybrid-driven approach: This approach attempts to integrate the two approaches mentioned above. For example, it may use data-driven methods to calibrate the parameters of the mechanistic model or perform weighted fusion of the prediction results of the two models in order to balance interpretability and prediction accuracy. However, most hybrid approaches only combine the models at the model level (such as result fusion) without achieving the essential integration and mutual constraint of mechanistic knowledge and statistical data laws at the fundamental stage of feature engineering.

[0028] To this end, the present invention provides a mechanical component life prediction scheme that can ensure that the extracted feature set is closely related to the physical mechanism of component failure to ensure interpretability and fully conforms to the statistical law of data, so as to make the prediction of the remaining life of mechanical components accurate.

[0029] like Figure 1 As shown, the present invention provides a method for predicting the lifespan of mechanical components, comprising: Step S11: Obtain the status monitoring data of the mechanical component to be predicted during operation.

[0030] In this embodiment, vibration signals, impact signals, temperature signals, and operating condition information of the mechanical component to be predicted during operation are acquired to obtain condition monitoring data. It can be understood that the vibration signals, impact signals, and temperature signals of the mechanical component to be predicted during operation are monitored in real time using sensors, while simultaneously acquiring the current operating condition information to obtain condition monitoring data.

[0031] In this embodiment, outliers or missing values ​​are detected in the status monitoring data to obtain corresponding detection results. If the detection result indicates the presence of outliers, the status monitoring data is processed by removing outliers using the number of sample entries or operating mileage to obtain preprocessed status monitoring data. If the detection result indicates the presence of missing values, historical status monitoring data is selected for filling to obtain preprocessed status monitoring data. The preprocessed status monitoring data is then normalized to obtain the target status monitoring data. It can be understood that, taking the train bearing as an example of the mechanical component to be predicted, if outliers or missing values ​​are found in the recorded bearing status monitoring data, personalized preprocessing is performed to repair the data. Simultaneously, to ensure that all types of data are on the same order of magnitude for ease of subsequent model building and use, the data is normalized. Specifically, if outliers are detected in the status monitoring data due to insufficient data caused by inactivity or low operating speed, resulting in low calculated feature values, outlier removal is performed based on the number of collected sample entries or operating mileage. Analysis of existing case data revealed that some cases had fewer than 20 samples or an operating mileage of less than 100km on a given day, indicating that the train was not in operation that day. Therefore, this strategy was used to remove the feature values ​​for that day to obtain preprocessed status monitoring data. In health management, when analyzing current data, if missing data is detected, the most recent historical data is considered reliable. Therefore, in practice, missing data is filled with the most recent historical data to obtain preprocessed status monitoring data. Furthermore, all preprocessed status monitoring data is normalized and dimensionless to prevent the model from biasing towards features with larger learning values ​​as important features. Therefore, all status data needs to be normalized to the same data magnitude. Data repair is performed using the maximum-minimum value normalization criterion, expressed as: ; in, This represents target status monitoring data. This represents the preprocessed status monitoring data. This represents the minimum value of the data. This indicates the maximum value of the data.

[0032] Step S12: Extract the target feature set that satisfies the preset constraints and preset evaluation conditions from the condition monitoring data; wherein, the preset constraints are set based on the laws of the physical process of component failure, and the preset evaluation conditions are conditions set using data statistical indicators determined based on historical monitoring data.

[0033] In this embodiment, a candidate feature set, including time-domain statistical features, trend change features, and same-position comparison features, is extracted from the status monitoring data based on the preset constraints. Specifically, the screening process of the candidate feature set is as follows: The corresponding status data values ​​representing different status monitoring indicators and the daily total number of the status data values ​​are extracted from the status monitoring data, and the corresponding daily statistical features of each status monitoring indicator are calculated to obtain time-domain statistical features; the daily average trend and monthly average trend of each status monitoring indicator are determined based on the daily statistical features of each status monitoring indicator to obtain trend change features; and the same-position comparison features are determined based on the difference between the first statistic of the status monitoring indicator at the current location and the second statistic of other status monitoring indicators at the same location; wherein, the current location is the mechanical component to be predicted and the corresponding evaluation time, and the other same locations are other mechanical components of the same model and located in the same equipment that operate symmetrically or equivalently to the mechanical component to be predicted and the same evaluation time. Understandably, further exploration of the failure mechanism reveals its characteristics. Considering value magnitude, trend changes, and in-situ comparison changes, and combining these with lifespan prediction applications, feature transformations are performed on daily statistics (daily mean, median, mode), trend changes (weekly and monthly daily mean trends), and in-situ comparisons (differences between daily means). This yields time-domain statistical features, trend change features, and in-situ comparison features, ultimately forming 142-dimensional features. The calculation formula for the daily statistics feature within the time-domain statistical features is as follows: ; in, This represents the status data value corresponding to the target status monitoring data for each day, where N is the total number of status data values ​​for the target status monitoring data for each day.

[0034] The formula for calculating the median is as follows: ; in, Indicates the first digit after ascending order. Each status data value.

[0035] Based on the above-mentioned daily statistical characteristics and median, the time-domain statistical characteristics are determined.

[0036] Then, the trend change value includes the daily average trend and the monthly daily average trend. The formula for calculating the trend change characteristics is as follows: ; in, For the first The daily average value of the day For the first Daily smoothed value or monthly smoothed value.

[0037] The same-position comparison value refers to the difference between the statistic at the current position and other statistics at the same position. The formula for calculating the same-position comparison feature is as follows: ; in, For the first The daily average value of the day same position j In each position The mean.

[0038] Accordingly, an example table of 142-dimensional bearing features is shown in Table 1: Table 1 142-dimensional bearing characteristics

[0039] In this embodiment, the target feature set is selected from the candidate feature set using the preset evaluation conditions. Specifically, the monotonicity index value, predictability index value, stability index value, and separability index value of each candidate feature in the candidate feature set are calculated. Target features whose monotonicity index value, predictability index value, and separability index value are all greater than or equal to a first preset threshold, and whose stability index value is greater than or equal to a second preset threshold, are selected to obtain the target feature set. It can be understood that the feature selection method combines the advantages of model-based feature selection methods (which can prioritize features that conform to the model trend and combined features with interactive effects) and feature selection methods based on feature evaluation indicators (selecting features based on monotonicity, predictability, robustness, and separability, avoiding model bias, and providing strong interpretability). Feature selection includes global feature importance analysis, feature interaction effect analysis, and feature indicator analysis. Specifically, global feature importance analysis uses the average SHAP value to select features that improve model accuracy; feature interaction effect analysis selects combined features that improve model accuracy; and feature indicator analysis selects features based on monotonicity, predictability, robustness, and separability.

[0040] Specifically, the process for determining the monotonicity index is as follows: Based on the current candidate feature and the respective feature values ​​of the adjacent candidate features preceding the current candidate feature, the monotonicity index of the current candidate feature is calculated. It can be understood that the monotonicity index characterizes the consistency between the feature and the bearing performance degradation state. Its value is within the range of [0,1]. The closer the monotonicity value is to 1, the better the monotonicity trend of the feature, and the better it can be used to construct the bearing performance degradation feature set. The calculation formula is as follows: ; in, Represents the unit step function. Indicates the first The feature values ​​of each feature (the current candidate feature). Indicates the first The feature values ​​of each feature (adjacent candidate features).

[0041] Specifically, the predictability index determination process is as follows: Based on the first mean of the candidate predictive features characterizing performance degradation at the failure time, the second mean of the candidate predictive features at the initial time, and the standard deviation of the candidate predictive features at the failure time, the predictability index value of the candidate predictive features is calculated; the stability index value of the current candidate feature is calculated based on the degradation trend parameter sequence of the current candidate feature and the candidate predictive features; the Fisher ratio between the healthy state and the fault state of each candidate feature is calculated to obtain the separability index. It can be understood that the predictability index is defined based on real bearing data, and its value ranges from [0,1]. The larger the degradation feature amplitude and the smaller the standard deviation at the failure time, the closer its value is to 1, indicating better predictability of the feature and a better representation of the separation between the normal state and the fault state. The calculation formula is as follows: ; in, It is the first mean of the degradation characteristic y at the time of failure; It is the second mean of the performance degradation characteristic y at the initial time; It is the standard deviation of the performance degradation characteristic r at the time of failure.

[0042] Furthermore, the stability index is determined as follows: The stability index describes the resistance of performance degradation characteristics to external environmental interference, and its value range remains within [0,1]. The smoother the fluctuation of the characteristic curve over time, the closer its robustness index value is to 1. The higher the accuracy of using this feature for bearing performance degradation and its participation in condition assessment / life prediction, the more accurate it is. The calculation method is as follows: ; in, A sequence of trend parameters representing performance degradation characteristics.

[0043] The separability index is determined as follows: A higher Fisher ratio indicates stronger feature identifiability. Using the start time of the fault data provided by the analysis group as the dividing point, the features of each bearing are divided into two segments: healthy and faulty. The separability of the extracted features with the bearing degradation label is then analyzed using the following formula: ; in, j Indicates the first j dimensional features, Indicates the first j Fisher ratio value for dimensional features.

[0044] In this way, the features extracted based on the fusion feature selection method are combined, and the score index value of the model convergence for each feature combination is calculated iteratively. The feature with the best model score is selected as the optimal feature for the bearing. Finally, the optimal feature combination (11 features) is selected: impact sample sharpness - outer loop, impact sample - frequency kurtosis, dB average - outer loop, SV average, vibration sample multi-order - outer loop, vibration sample - frequency kurtosis, impact sample - root mean square value, impact sample sharpness - inner loop, vibration sample - average frequency, SV median, and the proportion of outer loop dB at 900 r / min, to obtain the target feature set.

[0045] Step S13: Input the target feature set into the preset life prediction model to output the remaining life of the mechanical component to be predicted; wherein, the preset life prediction model is a model established by integrating the degradation trajectories of several historical mechanical components to map the correlation between the target feature set and the health status of the component.

[0046] In this embodiment, before inputting the target feature set into the preset lifespan prediction model, the method further includes: acquiring full lifespan monitoring data of several historical mechanical components as raw training data; and / or acquiring monitoring data of several historical mechanical components from the occurrence of fault characteristics to the failure disassembly interval as raw training data; labeling the raw training data with corresponding health status tags based on the health index corresponding to the starting point and failure point of the raw training data to form a training sample set carrying health status tags; the health status tags are associated with operating time or mileage through an S-shaped function curve that conforms to the degradation law of mechanical components. A set of historical features related to the physical process of failure of historical mechanical components and related to the driving characteristics of operating data is extracted from the training sample set. The extraction process of the set of historical features related to the physical process of failure of historical mechanical components and related to the driving characteristics of operating data is the same as the extraction process of the target feature set, and will not be described again here. Then, the initial preset lifespan prediction model is trained using the historical feature set from the operation process of the historical mechanical components to obtain the preset lifespan prediction model. Specifically, a learner is trained using the historical feature set carrying health status labels to learn the mapping relationship from historical features to health status, thereby obtaining the degradation trajectory of the corresponding individual historical mechanical component and acquiring the degradation trajectory of each historical mechanical component; the several degradation trajectories are fused using an ensemble learning strategy to construct and obtain a preset lifespan prediction model.

[0047] Understandably, based on the data from n failed bearing records, the following phenomena were observed: Failed bearing data includes monitoring data for the entire lifecycle (from commissioning to failure removal) with a defined start and end point, and monitoring data for failed bearings with an uncertain start point but a defined end point (from level 3 / 4 repair to failure removal). The data for failed bearings with an uncertain start point but a defined end point (from level 3 / 4 repair to failure removal) spans the level 3 repair. Since level 3 repair involves inspecting and replacing failed bearings, while healthy bearings are either reinstalled as old bearings or replaced (the reinstalled bearing is not the original one), the data collected before and after level 3 repair is not for the same bearing. This leads to an uncertainty about the current health status of the installed bearing, hindering the determination of its initial health index during degradation model training. To address this issue, for faulty bearing data with incomplete full lifecycle data, the data for these cases is truncated from the start time of the fault characteristic or the alarm time of each bearing. Specifically, data from the time of fault characteristic occurrence to the time of dismantling is used as the original training data. Then, based on the health index corresponding to the start and failure points of the original training data, a health status label is determined, further generating a training sample set with these labels. Specifically, full lifecycle monitoring data of several historical mechanical components from commissioning to failure and dismantling is acquired, or at least interval monitoring data from the time of fault characteristic occurrence to the time of failure and dismantling is included. For each historical mechanical component, the health index (HI) corresponding to the start and end points of its degradation trajectory is determined as the health status label. For samples with complete full lifecycle data, the health index at the initial commissioning time is set as the first preset value, and the health index at the time of failure and dismantling is set as the second preset value. For samples with only fault interval data (from after level 3 maintenance to failure and dismantling), the health index at the time the fault characteristic first appears is set as the third preset value, and the health index at the time of failure and dismantling is set as the second preset value.

[0048] Specifically, a health index degradation model is established. The above process only mentions determining the health status label of the original training data based on the health index at the starting point and failure point. However, analysis of the bearing mechanism's characteristic trends reveals that the early characteristics are stable, followed by rapid development, and then a significant shift in the rate of change towards a more stable characteristic in the later stages. This aligns with the Sigmoid function graph, a derivative model of the index model. Therefore, it is necessary to utilize the sigmoid function to describe the degradation process of the mechanical component's health index over operating time (or mileage). The mathematical expression of the health index degradation model is as follows: ; Where t represents the running time or mileage, Let X represent the health index, and let a, b, and c represent the model parameters, which are determined by the feature X.

[0049] Therefore, for each historical mechanical component, once the health index at the starting point and the health index at the failure point are determined, substituting them into the mathematical expression of the health index degradation model above, we can obtain the expressions of model parameters a and b with respect to model parameter c. Since the value of c is not determined, different c values ​​correspond to different degradation trajectory curves. Therefore, within the range of preset c values, we iterate through all possible c values ​​according to a preset step size. For each candidate c value, we combine the expressions of model parameters a and b with respect to model parameter c, the mathematical expression of the health index degradation model, and the HI values ​​corresponding to each time point of the historical mechanical component from the starting point to the failure point to form a set of candidate health status label sequences. Then, we obtain several sets of candidate health status label sequences formed by the HI values ​​corresponding to each time point of the historical mechanical component from the starting point to the failure point under different candidate c values. We train the learner and select the optimal set of candidate health status label sequences. For each candidate c value, we use the generated corresponding candidate health status label sequence as the training target and the historical feature set of the corresponding time point of the component as the input to train the XGBoost regression model to learn the mapping relationship from the feature space to the health index. After training, the model's fitting error (Root Mean Square Error, RMSE) on the component data is calculated. All candidate c values ​​are iterated, and the c value that minimizes the fitting error is selected as the optimal model parameter for the component. The degradation model corresponding to this optimal model parameter is the optimal degradation trajectory of the current historical mechanical component. The health index value at each time point on this optimal degradation trajectory is the final health status label for the component at each time point. The XGBoost model trained with this optimal c value is the base learner for the component, used to describe its feature-health status mapping relationship. Multiple base learners are integrated to construct a preset lifespan prediction model. The above steps are repeated for each historical mechanical component to obtain several base learners and their corresponding optimal degradation trajectories. An ensemble learning strategy (such as weighted averaging, Stacking, etc.) is used to fuse these base learners to construct the final preset lifespan prediction model. This ensemble model, when predicting new components, can comprehensively utilize multiple historical degradation experiences to output a robust health index estimate, and then infer the remaining lifespan based on the degradation model.

[0050] The core of the above training process lies in not pre-fixing all model parameters, but rather using start and end point constraints to first determine the analytical relationship between a, b, and c, and then selecting the optimal value of c through a data-driven approach that minimizes the fitting error. This allows the model to both satisfy the physical degradation law (S-shaped function form) and adaptively adapt to the individual differences of different components. The coupling between label generation and model training: Health status labels are not directly observed, but are generated in reverse through model assumptions, parameter search, and goodness-of-fit evaluation. This solves the problem that health indices cannot be directly measured in practical engineering. By integrating the degradation experience of multiple components, the model's generalization ability and predictive stability on new components are improved.

[0051] like Figure 2 As shown, the health index of n historical feature data starting points and dismantling points is determined (for cases where the dismantling point is before level 3 repair, training is performed based on the full life cycle data, with the initial health index set to 0.999; for cases where the dismantling point is after level 3 repair, training is performed based on the bearing failure feature starting data, with the initial health index set to 0.15; the health index of the dismantling point is determined by the degree of failure), which is also the health status label. After determining the parameter values ​​of the bearing data according to the above-mentioned health label method for each time point, the XGBoost regression model is used for training to map the relationship between the health index HI and the bearing degradation characteristics, thereby obtaining the optimal degradation trajectory of the health index of the training faulty bearing. After training to obtain the optimal degradation trajectory of the health index of n faulty bearings, the optimal degradation process of the remaining life is further obtained, forming an integrated application model of health assessment and life prediction, that is, the preset life prediction model. Furthermore, the preset life prediction model derives the following formula for calculating the remaining life: ; Where HI is the predicted health index, a, b, and c are the parameters solved by the model, T is the average total life of all bearing failure cases, and RUL is the predicted remaining life of the bearing.

[0052] It should be noted that the remaining life of the bearing corresponds one-to-one with the bearing condition (health index). Therefore, the establishment of the condition degradation process model also indirectly establishes the preset life prediction model. Based on the definition of the health index, the smaller the health index, the worse the bearing condition, and the larger the health index, the better the bearing condition. Its value is usually between [0,1]

[49] .

[0053] In this embodiment, the target feature set is input into a preset lifespan prediction model. Multiple initial health indices are calculated using all degradation trajectories within the preset lifespan prediction model. These initial health indices are then smoothed to obtain a predicted health index. Based on the predicted health index, the model parameters corresponding to the degradation trajectories, and the average total lifespan of historically faulty components, the remaining service life of the mechanical component to be predicted is calculated. It can be understood that when new bearing data is used for prediction, the bearing's health index is calculated based on the bearing's target feature set, using all degradation trajectories in the preset lifespan prediction model, and then smoothed to determine the final health index. The parameter c is determined using all optimal degradation trajectories, and the remaining lifespan value is calculated using the following formula: , , ; in, m The health index of the component when it is first put into use. n This is a health index during bearing disassembly. Since the S-shaped function value approaches 1 infinitely, therefore... m The value should be less than 1 and cannot be equal to 1.

[0054] As can be seen, this application discloses the following: acquiring condition monitoring data of a mechanical component to be predicted during operation; extracting a target feature set that satisfies preset constraints and preset evaluation conditions from the condition monitoring data; wherein the preset constraints are set based on the physical process law of component failure, and the preset evaluation conditions are conditions set using data statistical indicators determined based on historical monitoring data; inputting the target feature set into a preset life prediction model to output the remaining service life of the mechanical component to be predicted; wherein the preset life prediction model is a model established by integrating the degradation trajectories of several historical mechanical components to map the correlation between the target feature set and the health status of the component. Therefore, by extracting the target feature set that meets the preset constraints and evaluation conditions, it is ensured that the features selected from the target feature set must be related to the physical mechanism of component failure. This fundamentally guarantees that the target feature set used for prediction has clear physical meaning. At the same time, through data index screening, it is ensured that the features can reflect the complex degradation process in actual operation, and features that, although they conform to the ideal physical model, fluctuate greatly and have unclear trends in actual data are discarded. Furthermore, the preset life prediction model established by integrating the degradation trajectories of several historical mechanical components is based on the target feature set. It finds the most similar or performs weighted synthesis from the multiple integrated historical trajectories to make accurate life predictions. This improves the model's generalization ability when facing individual differences in components, changes in operating conditions, and small sample failure data, and directly solves the problem of over-reliance on a single historical sample.

[0055] like Figure 3 As shown, the present invention provides a mechanical component life prediction device, comprising: Data monitoring module 11 is used to acquire status monitoring data of the mechanical component to be predicted during operation; Feature extraction module 12 is used to extract a set of target features that satisfy preset constraints and preset evaluation conditions from the status monitoring data; wherein, the preset constraints are set based on the laws of the physical process of component failure, and the preset evaluation conditions are conditions set using data statistical indicators determined based on historical monitoring data; The life prediction module 13 is used to input the target feature set into a preset life prediction model to output the remaining service life of the mechanical component to be predicted; wherein, the preset life prediction model is a model established by integrating the degradation trajectories of several historical mechanical components to map the correlation between the target feature set and the health status of the component.

[0056] Therefore, by establishing feature extraction that meets preset constraints and evaluation conditions, further mining fault mechanism features, and constructing a life prediction model for the performance degradation process driven by both mechanism and data, this model mines the mapping relationship between high-dimensional feature data and bearing degradation state, fully utilizes historical information of small sample faulty bearings, achieves integrated intelligent decision-making, supports dynamic iteration, and realizes the integrated application of the life prediction model that has transformed from the limitation of single samples to the optimal integrated generalization.

[0057] Furthermore, embodiments of this application also disclose an electronic device, Figure 4 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content of the diagram should not be construed as limiting the scope of this application.

[0058] Figure 4 This is a schematic diagram of the structure of an electronic device 20 provided in an embodiment of this application. Specifically, the electronic device 20 may include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the mechanical component life prediction method disclosed in any of the foregoing embodiments. Alternatively, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0059] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0060] The processor 21 may include one or more processing cores, such as a quad-core processor or an octa-core processor. The processor 21 may be implemented using at least one hardware form selected from DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). The processor 21 may also include a main processor and a coprocessor. The main processor, also known as a CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, the processor 21 may integrate a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the screen. In some embodiments, the processor 21 may also include an AI (Artificial Intelligence) processor, which is used to handle computational operations related to machine learning.

[0061] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or optical disk, etc. The resources stored thereon can include operating system 221, computer program 222, etc., and the storage method can be temporary storage or permanent storage.

[0062] The operating system 221 manages and controls the various hardware devices and computer programs 222 on the electronic device 20 to enable the processor 21 to perform calculations and processing on the massive amounts of data 223 in the memory 22. The operating system 221 can be Windows Server, Netware, Unix, Linux, etc. The computer program 222, in addition to including a computer program capable of performing the mechanical component life prediction method executed by the electronic device 20 as disclosed in any of the foregoing embodiments, may further include computer programs capable of performing other specific tasks. The data 223 may include data received by the electronic device from external devices, as well as data collected by its own input / output interface 25.

[0063] Furthermore, this application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned method for predicting the lifespan of mechanical components. Specific steps of this method can be found in the corresponding content disclosed in the foregoing embodiments, and will not be repeated here.

[0064] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.

[0065] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application. The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly in hardware, software modules executed by a processor, or a combination of both. The software module may be located in random access memory (RAM), memory, read-only memory (ROM), electrically programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, removable disks, CD-ROMs (Compact Disc-Read Only Memory), or any other form of storage medium known in the art.

[0066] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0067] The solution provided by the present invention has been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only for helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for predicting the lifespan of mechanical components, characterized in that, include: Acquire status monitoring data of the mechanical component to be predicted during operation; Extract target feature sets that satisfy preset constraints and preset evaluation conditions from the condition monitoring data; wherein, the preset constraints are set based on the laws of the physical process of component failure, and the preset evaluation conditions are conditions set using data statistical indicators determined based on historical monitoring data; The target feature set is input into a preset life prediction model to output the remaining life of the mechanical component to be predicted; wherein, the preset life prediction model is a model established by integrating the degradation trajectories of several historical mechanical components to map the correlation between the target feature set and the health status of the component.

2. The method for predicting the lifespan of mechanical components according to claim 1, characterized in that, Before inputting the target feature set into the preset lifetime prediction model, the method further includes: The initial preset life prediction model is trained using the historical feature set of the historical mechanical component operation process to obtain the preset life prediction model.

3. The method for predicting the lifespan of mechanical components according to claim 2, characterized in that, Before training the initial preset life prediction model using historical feature sets from the operation of historical mechanical components, the method further includes: Acquire full lifecycle monitoring data of several historical mechanical components as raw training data; And / or, acquire monitoring data of several historical mechanical components from the occurrence of fault characteristics to the period of failure and disassembly as raw training data; Based on the health index corresponding to the starting point and failure point of the original training data, the original training data is labeled with corresponding health status labels to form a training sample set carrying health status labels. Extract a set of historical features from the training sample set that are related to the physical processes of historical mechanical component failures and to the characteristics driven by operational data.

4. The method for predicting the lifespan of mechanical components according to claim 3, characterized in that, The process of training an initial preset lifespan prediction model using historical feature sets from the operation of historical mechanical components to obtain the preset lifespan prediction model includes: The learner is trained using the historical feature set carrying health status labels to learn the mapping relationship from historical features to health status, and the degradation trajectory of the corresponding individual historical mechanical parts is obtained to acquire the degradation trajectory of each historical mechanical part. The several degradation trajectories are fused using an ensemble learning strategy to construct and obtain a preset lifetime prediction model.

5. The method for predicting the lifespan of mechanical components according to claim 4, characterized in that, The health status label is associated with operating time or mileage through an S-shaped function curve that conforms to the degradation law of mechanical parts.

6. The method for predicting the lifespan of mechanical components according to claim 1, characterized in that, The acquisition of the status monitoring data of the mechanical component to be predicted during operation includes: The vibration, impact, temperature, and operating condition information of the mechanical component to be predicted during operation are acquired to obtain condition monitoring data.

7. The method for predicting the lifespan of mechanical components according to claim 6, characterized in that, After acquiring vibration signals, impact signals, temperature signals, and operating condition information of the mechanical component to be predicted during operation to obtain condition monitoring data, the process further includes: The status monitoring data is checked for outliers or missing values ​​to obtain the corresponding detection results. If the detection result shows outliers, the outliers in the status monitoring data are removed by using the number of sample entries or the operating mileage to obtain preprocessed status monitoring data. If the detection result indicates the presence of missing values, historical status monitoring data is selected for filling to obtain preprocessed status monitoring data. The preprocessed state monitoring data is normalized to obtain the target state monitoring data.

8. The method for predicting the lifespan of mechanical components according to claim 1, characterized in that, The step of extracting the target feature set that satisfies preset constraints and preset evaluation conditions from the status monitoring data includes: Based on the preset constraints, a candidate feature set including time-domain statistical features, trend change features, and same-position comparison features is extracted from the state monitoring data; The target feature set is selected from the candidate feature set using the preset evaluation conditions.

9. The method for predicting the lifespan of mechanical components according to claim 8, characterized in that, The extraction of a candidate feature set from the state monitoring data based on the preset constraints, including time-domain statistical features, trend change features, and same-position comparison features, includes: Extract the corresponding state data values ​​representing different state monitoring indicators and the total number of state data values ​​per day from the state monitoring data, and calculate and determine the corresponding daily statistical characteristics of each state monitoring indicator to obtain the time domain statistical characteristics. Based on the daily statistical characteristics of each of the aforementioned status monitoring indicators, the daily average trend and monthly daily average trend of each of the aforementioned status monitoring indicators are determined to obtain the trend change characteristics. The same-location comparison feature is determined based on the difference between the first statistic of the current location status monitoring index and the second statistic of the other status monitoring indexes at the same location; wherein, the current location is the mechanical component to be predicted and the corresponding evaluation time, and the other same locations are other mechanical components of the same model and located in the same equipment that operate symmetrically or equivalently to the mechanical component to be predicted and the same evaluation time.

10. The method for predicting the lifespan of mechanical components according to claim 9, characterized in that, The step of selecting the target feature set from the candidate feature set using the preset evaluation criteria includes: Calculate the monotonicity index, predictability index, stability index, and separability index for each candidate feature in the candidate feature set; The target feature set is obtained by filtering target features whose monotonicity index value, predictability index value, and separability index value are all greater than or equal to a first preset threshold and whose stability index value is greater than or equal to a second preset threshold.

11. The method for predicting the lifespan of mechanical components according to claim 10, characterized in that, The calculation of the monotonicity index, predictability index, stability index, and separability index for each candidate feature in the candidate feature set includes: The monotonicity index of the current candidate feature is calculated based on the current candidate feature and the respective feature values ​​of the adjacent candidate features preceding the current candidate feature. The predictability index value of the candidate predictive feature is calculated based on the first mean of the candidate predictive feature representing the performance degradation feature at the failure time, the second mean of the candidate predictive feature at the initial time, and the standard deviation of the candidate predictive feature at the failure time. The stability index value of the current candidate feature is calculated based on the degradation trend parameter sequence of the current candidate feature and the candidate predictive feature; Calculate the Fisher ratio for each candidate feature between healthy and faulty states to obtain the separability index.

12. The method for predicting the lifespan of mechanical components according to any one of claims 1 to 11, characterized in that, The step of inputting the target feature set into a preset life prediction model to output the remaining life of the mechanical component to be predicted includes: The target feature set is input into a preset lifespan prediction model, and multiple initial health indices are calculated and obtained using all the degradation trajectories in the preset lifespan prediction model. The multiple initial health indices are smoothed to obtain the predicted health index; The remaining service life of the mechanical component to be predicted is calculated based on the predicted health index, the model parameters corresponding to the degradation trajectory, and the average total lifespan of historically failed components.

13. A device for predicting the lifespan of mechanical components, characterized in that, include: The data monitoring module is used to acquire status monitoring data of the mechanical component to be predicted during operation; The feature extraction module is used to extract a set of target features that satisfy preset constraints and preset evaluation conditions from the condition monitoring data; wherein, the preset constraints are set based on the laws of the physical process of component failure, and the preset evaluation conditions are conditions set using data statistical indicators determined based on historical monitoring data; The life prediction module is used to input the target feature set into a preset life prediction model to output the remaining life of the mechanical component to be predicted; wherein, the preset life prediction model is a model established by integrating the degradation trajectories of several historical mechanical components to map the correlation between the target feature set and the health status of the component.

14. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the steps of the mechanical component life prediction method as described in any one of claims 1 to 12.

15. A computer-readable storage medium, characterized in that, Used to store a computer program; wherein, when the computer program is executed by a processor, it implements the steps of the mechanical component life prediction method as described in any one of claims 1 to 12.

Images