Magnetic suspension bearing life prediction method and system based on machine learning

Abstract

The invention discloses a magnetic suspension bearing life prediction method and system based on machine learning, and belongs to the technical field of machine learning, and the method comprises the steps: collecting the operation response data and working environment data of a magnetic suspension bearing in each operation period; predicting the degradation degree of each component in the magnetic suspension bearing through the operation response data, and determining the current life cycle stage of the magnetic suspension bearing; and according to the operation response data and the working environment data, predicting a life accelerated degradation point of the magnetic suspension bearing, constructing a stable period characteristic and a degradation period characteristic through the life accelerated degradation point, and predicting life degradation conditions of the magnetic suspension bearing in a stable period and a degradation period respectively to obtain a life degradation result. Therefore, by implementing the method and the device, the problem that the predicted bearing life degradation result is not accurate due to the fact that self-adaptive prediction is not carried out on the stage degradation of the magnetic bearing life in the prior art can be solved.

Description

Technical Field

[0001] This application belongs to the field of medium- and long-term load forecasting of power grids, and specifically relates to a method and system for predicting the life of magnetic levitation bearings based on machine learning. Background Technology

[0002] Magnetic levitation bearings are bearing devices that dynamically adjust the rotor position using a closed-loop control system composed of electromagnets, sensors, and controllers. They offer advantages such as frictionless operation, low energy consumption, and long lifespan, and are widely used in high-speed motors, compressors, and flywheel energy storage. Therefore, predicting the remaining service life of magnetic levitation bearings is crucial for ensuring safe equipment operation and reducing maintenance costs.

[0003] Currently, the remaining service life of magnetic levitation bearings is often predicted by modeling their life cycle. This typically involves using operational data to predict the degradation trend over a given period. However, this method fails to account for the phased degradation of bearing lifespan: during the stable phase, bearing performance shows no obvious signs of degradation, and the theoretical lifespan is relatively long; during the degradation phase, the bearing exhibits significant performance decline, and its theoretical lifespan decreases rapidly. Therefore, existing technology does not consider the different data indicating lifespan decline at different stages, resulting in a significant discrepancy between the predicted and actual lifespan changes. This makes it difficult for maintenance personnel to perform timely maintenance. Summary of the Invention

[0004] This application proposes a machine learning-based method and system for predicting the lifespan of magnetic levitation bearings, which can solve the problem that existing technologies do not adaptively predict the stage-by-stage degradation of the lifespan of magnetic levitation bearings, resulting in inaccurate predictions of bearing life degradation.

[0005] The first aspect of this application provides a machine learning-based method for predicting the lifespan of magnetic levitation bearings, the method comprising: The system collects operational response data and working environment data of the magnetic levitation bearing during various operating phases; wherein, the operational response data includes operating current, operating temperature, and vibration amplitude; and the working environment data includes humidity and start / stop frequency. Based on preset degradation indicators, the degradation degree of each component in the magnetic levitation bearing is predicted by the operational response data, and the current life cycle stage of the magnetic levitation bearing is determined. Based on the operational response data and working environment data, the accelerated degradation point of the magnetic levitation bearing after the current life cycle stage is predicted, and stable period characteristics and degradation period characteristics are constructed through the accelerated degradation point. Using stable period features and degradation period features, the life degradation of magnetic levitation bearings during the stable period and degradation period are predicted respectively, and the life degradation results are obtained.

[0006] The above scheme not only collects the response data of the magnetic levitation bearing during operation but also collects surrounding environmental data, providing data support for subsequent consideration of the impact of the external environment on the bearing. First, based on pre-set degradation indicators, it preliminarily determines which point in the magnetic levitation bearing's life cycle it is currently at, directly determining the corresponding technical means for subsequent feature extraction and life prediction, achieving adaptive and dynamic life prediction, and ensuring that the most relevant and sensitive feature data can be extracted. Then, the operational response data and working environment data are used to predict the overall life degradation process of the magnetic levitation bearing, identifying the accelerated degradation point. Using the accelerated degradation point as a boundary, the remaining life cycle of the bearing is divided into a stable period and a degradation period, and corresponding stable period features and degradation period features are constructed respectively, providing more correlated and sensitive data for life degradation prediction at different stages. Finally, by combining the life degradation predictions in the two stages, a more accurate life degradation result is obtained, providing a foundation for subsequent equipment maintenance.

[0007] In one possible implementation of the first aspect, based on a preset degradation index, the degradation degree of each component in the magnetic levitation bearing is predicted using the operational response data to determine the current life cycle stage of the magnetic levitation bearing, specifically: Based on the degradation index corresponding to each device, the operational sequence characteristics of each device are extracted from the operational response data; wherein, the operational sequence characteristics include operational current characteristics, temperature characteristics, and amplitude characteristics; Based on the normal operating standards of each component, a corresponding sliding window is generated for the runtime sequence characteristics; The sliding window is used to identify anomalies in the runtime sequence features, and the frequency of anomalies for each runtime sequence feature is counted. The degradation level of each device is predicted based on the frequency of the anomalies, and the current life cycle stage is determined by combining the degradation level.

[0008] The above scheme first sets up a sliding window for detecting anomalies in timing data based on the normal operating standards of various components, and then detects the frequency of anomalies in each data, thereby quantifying the more abstract degree of degradation into the specific frequency of anomalies.

[0009] In one possible implementation of the first aspect, the degradation degree of each component is predicted based on the frequency of the anomalies, and the current lifecycle stage is determined in conjunction with the degradation degree, specifically as follows: Based on the contribution of each runtime timing feature to the degradation of the component, a weighting factor for each runtime timing feature is set; wherein, the degradation contribution is the degree of impact on the component performance when the runtime timing feature is abnormal; The degradation degree of each component is predicted using the weighting factor and the frequency of outliers. Based on the degree of degradation, the impact of each component on the performance degradation of the magnetic levitation bearing control system is quantified, and the current life cycle stage is determined.

[0010] The above scheme further refines the calculation of degradation degree by introducing a weighting factor based on "degradation contribution" to characterize the impact of degradation of different components on the control system. Then, the degradation of the components is correlated with the performance degradation of the entire control system to roughly assess the life cycle stage of the entire magnetic levitation bearing, providing a basis for more precise life prediction in the future.

[0011] In one possible implementation of the first aspect, based on the operational response data and operating environment data, the accelerated degradation point of the magnetic levitation bearing after the current life cycle stage is predicted, and stable period characteristics and degradation period characteristics are constructed through the accelerated degradation point, specifically as follows: Based on the current lifecycle stage, feature fusion is performed on the runtime response data and working environment data to obtain fused features; The overall life degradation cycle of the magnetic levitation bearing is predicted using fusion features, and the accelerated degradation point is determined. The period between the current life cycle stage and the accelerated life degradation point is defined as the stable period, and the period after the accelerated life degradation point is defined as the degradation period. Based on the strong correlation indicators between the stable period and the degradation period, stable period features and degradation period features are constructed by fusing features respectively.

[0012] The above scheme divides the future lifespan of magnetic levitation bearings into stages by identifying accelerated degradation points, providing a foundation for more precise capture of the degradation patterns of bearings at different life stages and avoiding the low accuracy caused by single predictions. Furthermore, it selects more correlated features for different stages and ignores irrelevant features, reducing the impact of redundant data on predictions and making the prediction results more realistic.

[0013] In one possible implementation of the first aspect, based on the current lifecycle stage, feature fusion is performed on the runtime response data and the working environment data to obtain fused features, specifically as follows: Internal degradation features are extracted from the operational response data, and external influence features are extracted from the working environment data; The influence of the external influence features on the internal degradation features is quantified by the chi-square value to obtain the corresponding feature correlation degree. If the correlation degree of the features is greater than the first threshold, then data point matching is performed on the external influence features and the internal degradation features to obtain a pair of related features; The fluctuation of the associated features within each preset time interval is described, and the state label for each preset time interval is generated; The corresponding state label is inserted into the internal degradation feature, and the external influence feature corresponding to the state label is concatenated to the internal degradation feature to obtain the fused feature.

[0014] The above scheme uses the chi-square value to quantify the correlation between internal degradation features and external environmental features, thereby assessing whether feature data from two different sources are completely independent, making subsequent feature fusion based on a solid foundation. Finally, based on the feature correlation, the two features are fused together to generate a more informative and discriminative fused feature.

[0015] In one possible implementation of the first aspect, the overall lifespan degradation cycle of the magnetic levitation bearing is predicted using fused features to determine the accelerated degradation point, specifically: Using fused features, the remaining lifespan of the magnetic levitation bearing at each prediction time point is predicted through a pre-set long short-term memory network. Based on the adjacent remaining lifetime values, the lifetime degradation rate at each predicted time point is calculated in chronological order. The point at which the lifetime degradation rate first exceeds the second threshold is defined as the point of accelerated lifetime degradation.

[0016] The above scheme learns long-term dependencies in historical data through a long short-term memory network, thereby making a more accurate prediction of the remaining lifespan of the magnetic levitation bearing. It uses an accelerated rate of lifespan degradation as an indicator of entering the degradation phase, thus achieving stage-based segmentation.

[0017] In one possible implementation of the first aspect, based on the strong correlation indicators between the stable period and the degradation period, stable period features and degradation period features are constructed separately by fusing features, specifically as follows: Based on the strong correlation indicators during the stable period, the first key feature was selected from the fusion features; based on the health indicators during the deterioration period, the second key feature was selected from the fusion features. Increase the weight factor of the first key feature in the fusion features to obtain the stable period features; By increasing the weight factor of the second key feature in the fusion features, the degradation period features are obtained.

[0018] The above scheme increases the weight of features that are more strongly correlated with the stable / degradation period, making subsequent lifespan degradation predictions pay more attention to these features, enhancing predictive ability, guiding the prediction model to learn the most important features, and improving prediction accuracy.

[0019] In one possible implementation of the first aspect, the strongly correlated metric is specifically: Based on the analysis of historical operating data, the performance degradation characteristics of the magnetic levitation bearing were obtained, and the response data characteristics of the magnetic levitation bearing during the stable period and the degradation period were obtained respectively. Based on the type of response data characteristics during the stable period, the corresponding strongly correlated indicators are extracted from the operational response data; Based on the type of response data characteristics during the degradation period, corresponding intrinsic indicators are extracted from the operational response data, and the working environment data that affects the intrinsic indicators are selected as extrinsic indicators. By combining the intrinsic indicators and the extrinsic indicators, a strongly correlated indicator for the degradation period is obtained.

[0020] The above scheme uses historical operating data as prior knowledge to identify key degradation characteristics of bearings at different stages. Furthermore, it elevates the importance of environmental factors during the degradation period, providing more comprehensive data support for understanding the aggravating effects of the external environment on the degradation process.

[0021] In one possible implementation of the first aspect, the life degradation of the magnetic levitation bearing during the stable period and the degradation period is predicted using stable period features and degradation period features, respectively, to obtain the life degradation result, specifically: Based on the characteristics of the steady-state period, the life degradation rate of the magnetic levitation bearing is predicted, and the first remaining life degradation curve is obtained. Based on the characteristics of the degradation period, the life degradation rate of the magnetic levitation bearing is predicted, and the second remaining life degradation curve is obtained. Select the curve segment in the first remaining lifetime degradation curve that is related to the time period of the stable period as the stable period curve, and select the curve segment in the second remaining lifetime degradation curve that is related to the time period of the degradation period as the degradation period curve. The stable period curve and the degradation period curve are spliced ​​together, and the splicing point is adjusted using a preset historical degradation curve to obtain the lifetime degradation result.

[0022] The above scheme avoids the dilemma that a single model cannot fit two different degradation modes at the same time. By predicting separately and extracting the relevant prediction results, higher prediction accuracy can be obtained at each stage, thereby improving the overall prediction accuracy.

[0023] The second aspect of this application provides a magnetic levitation bearing life prediction system based on machine learning, the system comprising: a data acquisition module, a degradation degree prediction module, an accelerated degradation point identification module, and a life prediction module; The data acquisition module is used to collect the operating response data and working environment data of the magnetic levitation bearing during each operating phase; the operating response data includes operating current, operating temperature, and vibration amplitude; the working environment data includes humidity and start / stop frequency. The degradation degree prediction module is used to predict the degradation degree of each component in the magnetic levitation bearing based on preset degradation indicators and the operation response data, so as to determine the current life cycle stage of the magnetic levitation bearing. The accelerated degradation point identification module is used to predict the accelerated degradation point of the magnetic levitation bearing after the current life cycle stage based on the operation response data and working environment data, and to construct stable period characteristics and degradation period characteristics through the accelerated degradation point. The life prediction module is used to predict the life degradation of magnetic levitation bearings during the stable period and the degradation period using stable period features and degradation period features, respectively, and obtain the life degradation results. Attached Figure Description

[0024] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0025] Figure 1 This is a schematic diagram of a specific process for predicting the life of a magnetic levitation bearing based on machine learning, provided in an embodiment of this application. Figure 2 This is a structural diagram of a machine learning-based magnetic levitation bearing life prediction system provided in an embodiment of this application. Detailed Implementation

[0026] 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 this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0027] It should be understood that the step numbers used in the text are for ease of description only and are not intended to limit the order in which the steps are performed.

[0028] First Embodiment Existing technologies often use a single model to fit the entire lifecycle degradation process of magnetic levitation bearings, ignoring the significant differences in the expression of performance degradation characteristics at different degradation stages. Therefore, predictions using a single model are prone to insufficient accuracy. This application's embodiments employ different feature data to segmentally model and predict the lifecycle degradation of magnetic levitation bearings, improving the accuracy of bearing lifecycle degradation prediction.

[0029] like Figure 1 As shown, to address the problem in existing technologies where the lack of adaptive prediction for the phased degradation of magnetic levitation bearing life leads to inaccurate predicted bearing life degradation results, the first embodiment of this application provides a detailed flowchart of a machine learning-based magnetic levitation bearing life prediction method. This embodiment's machine learning-based magnetic levitation bearing life prediction method includes steps S1 to S4, detailed below: Step S1: Collect the operating response data and working environment data of the magnetic levitation bearing during each operating phase.

[0030] Unlike conventional mechanical bearings, which only exhibit performance degradation due to fatigue failure of the bearing material, magnetic levitation bearings, because they have no moving parts in contact, exhibit performance degradation more in the form of control system failure or low sensitivity. Therefore, it is necessary to collect the operating response data of magnetic levitation bearings in order to accurately predict the performance degradation of each component in the system.

[0031] In addition, the external environment can also affect the operation of magnetic levitation bearings. For example, strong external vibrations can cause the rotor to rub against the backup bearing or stator, affecting the adjustment capability of the control system; excessive dust or humidity in the operating environment can contaminate the coils of the magnetic levitation bearing, leading to a decrease in the coil insulation performance. Therefore, it is also necessary to collect operating environment data to assist in prediction.

[0032] Therefore, operational response data and working environment data of the magnetic levitation bearing are collected during each operating phase. The operational response data is primarily collected from the controller, the electromagnet coils of the stator and rotor, and sensors. Therefore, the operational response data includes operating current (used to measure whether there are any abnormalities in the current commands issued by the controller), operating temperature (used to measure the aging degree of the insulation material), vibration amplitude (used to measure the wear degree of the components), and sensor response speed (used to measure the sensitivity of the sensors and their ability to provide timely and accurate rotor position feedback to the control system). The collection of working environment data takes into account the interference of external vibration on the magnetic levitation bearing, the burden on the control system caused by environmental cleanliness and excessively high temperatures, and the risk of failure due to improper operation. Therefore, the working environment data includes temperature and humidity, start-stop frequency, pollution levels, and the frequency of overload operation.

[0033] Therefore, the life degradation prediction of magnetic levitation bearings cannot only look at the degree of structural damage. Considering that it corresponds to a complex closed-loop control system composed of electromagnets, sensors and controllers, it is also necessary to evaluate each component of the system individually in order to obtain a comprehensive degradation situation.

[0034] Step S2: Based on preset degradation indicators, the degradation degree of each component in the magnetic levitation bearing is predicted using the operational response data to determine the current life cycle stage of the magnetic levitation bearing.

[0035] First, based on the components and their corresponding degradation modes in the existing magnetic levitation bearing system, the degradation index of each component is determined.

[0036] For example, for an electromagnet coil, the typical degradation mode is insulation material aging, inter-turn short circuit and core performance degradation. Therefore, its degradation index is related to changes in inductance or bias current error. For a rolling device, the typical degradation mode is material wear or lubrication failure. Therefore, its degradation index is related to vibration frequency.

[0037] The aforementioned degradation indicators are benchmark values ​​obtained from a large number of samples, which can cover most device degradation scenarios and achieve accurate prediction of the degree of device degradation.

[0038] The degradation index is used to extract the operational sequence characteristics of each component from the operational response data. For example, the operational sequence characteristics of a displacement sensor are the displacement signals of each degree of freedom, the operational sequence characteristics of a rolling bearing are the bearing amplitude signal and the temperature of the bearing base, and the operational sequence characteristics of an electromagnet coil are the feedback value of the coil current.

[0039] Then, based on the normal operating standards of each component, a corresponding sliding window is generated for each runtime timing feature. When generating the sliding window, the window size is set according to the frequency of abnormal values ​​defined by the normal operating standards; and the sliding step size of the window is set according to the feedback frequency of the runtime timing feature.

[0040] Using the sliding window, anomalies are identified in the corresponding runtime sequence features by setting a sliding step size. Anomalies are determined by detecting whether the data within each window exceeds a set normal threshold, and the number of anomalies during each sliding step is counted to obtain the anomaly frequency for each runtime sequence feature.

[0041] As an improvement to the above scheme, when setting the normal threshold, the statistics from the previous step are used to dynamically adjust the normal threshold for the next time step, providing a larger correction space for anomaly detection.

[0042] Because the components in a magnetic levitation bearing system have varying degrees of importance, severe performance degradation in more critical components leads to more significant bearing life degradation, while performance degradation in less critical components sometimes has negligible impact on the bearing. Considering these factors, this application's embodiments predict the degradation level of each component to accurately pinpoint the current performance status of the magnetic levitation bearing and determine its current lifecycle stage. This provides precise time points for subsequent lifecycle degradation predictions targeting stages with different degradation characteristics.

[0043] First, considering the impact of abnormal operating sequence characteristics on the operating state of components, the contribution of each operating sequence characteristic to component degradation is set. For example, when the coil overheats severely, the insulation material of the electromagnet coil is more prone to failure, leading to inter-turn short circuits, thus the corresponding impact is relatively large; when the voltage signal output by the controller is converted into current, there is a deviation from the preset normal current, but the impact on the rotor's magnetic force is not significant, thus the corresponding impact is relatively small.

[0044] Then, the weighting factor for the runtime sequence features is set based on the degradation contribution. The greater the degradation contribution, the greater the corresponding weighting factor.

[0045] The degradation degree of each component is predicted using the weighting factors and the outlier frequencies. Because the data collection frequencies differ, all outlier frequencies are standardized before calculating the degradation degree, bringing them to the same dimension to ensure accurate prediction. Furthermore, considering that the dimensionality of the runtime sequence features for each component may differ, the weighted data is averaged according to the feature dimensions.

[0046] Finally, based on the degree of degradation of each component and considering the importance of each component in the magnetic levitation bearing system, the impact of each component on the performance degradation of the magnetic levitation bearing control system is quantified, and the current life cycle stage of the magnetic levitation bearing is determined.

[0047] Therefore, the embodiments of this application link the degradation of components with the performance degradation of the entire control system, thereby roughly assessing the current performance of the entire magnetic levitation bearing and providing a basis for predicting subsequent performance degradation based on the current performance.

[0048] Step S3: Based on the operational response data and working environment data, predict the accelerated degradation point of the magnetic levitation bearing after the current life cycle stage, and construct stable period characteristics and degradation period characteristics through the accelerated degradation point.

[0049] In this embodiment of the application, the operating response data reflects the internal degradation of the magnetic levitation bearing, and the working environment data reflects the external influence on the magnetic levitation bearing. Therefore, the internal degradation features are extracted from the operating response data, and the external influence features are extracted from the working environment data.

[0050] To detect whether external influences are related to internal degradation, the chi-square value between the two features is calculated by statistically analyzing whether the corresponding internal degradation feature exhibits an extreme value at an adjacent time point when the external influence feature exhibits an extreme value within a preset time interval. This quantifies whether there is a correlation or influence between the external influence feature and the internal degradation feature, thereby obtaining the corresponding feature correlation degree.

[0051] If the correlation degree of the features is greater than the first threshold, it indicates that there is a connection between the corresponding external influence feature and the internal degradation feature. Then, the two features are matched in time sequence to obtain a pair of related features. The fluctuation of the related feature pair within each preset time interval is then described, and a state label is generated for each time interval.

[0052] According to a preset time interval, the corresponding state label is inserted into the internal degradation feature, and then the external influence feature corresponding to the state label (i.e. the external influence feature of the associated feature pair) is spliced ​​into the internal degradation feature to obtain the fused feature of each associated feature pair.

[0053] The fused features are input into a preset long short-term memory network to predict the remaining lifetime of the magnetic levitation bearing at each predicted time point. Using adjacent remaining lifetime values, the lifetime degradation rate of the magnetic levitation bearing at each predicted time point is calculated in chronological order to obtain an overall lifetime degradation trend. This trend does not consider the different degradation characteristics of the bearing's stable period and degradation period, therefore the prediction accuracy differs from the actual situation. Therefore, it is not directly used in this embodiment. Instead, the time point when the lifetime degradation rate first exceeds a set rate threshold is identified and designated as the lifetime acceleration degradation point.

[0054] After the accelerated degradation point, magnetic levitation bearings begin to age comprehensively, the failure rate increases significantly, and the system may become unstable, requiring more frequent maintenance and component replacement. Therefore, after the accelerated degradation point, it is necessary to incorporate more degradation characteristics for lifespan prediction to improve prediction accuracy.

[0055] In this embodiment of the application, the period between the current life cycle stage and the accelerated life degradation point is defined as the stable period, and the period after the accelerated life degradation point is defined as the degradation period.

[0056] During the stable period, the bearing system has a low failure rate and reliable performance. The components are basically operating stably, with only a few abnormal situations. During the degradation period, the failure rate of the bearing system increases, the components show obvious degradation, and the environment has a greater impact on the components.

[0057] To address the above phenomena, we identified the characteristics that best reflect the performance changes of magnetic levitation bearings in the fusion profile by using strong correlation indicators between the stable and degradation periods.

[0058] Specifically, based on historical operating data analysis, the performance degradation characteristics of the magnetic levitation bearing were obtained, yielding response data characteristics during the stable and degradation periods. For example, during the stable period, the amplitude signal characteristics show slight shifts in the frequency domain and peak-valley fluctuations in the time domain; harmonics appear in the current of the electromagnetic coil. During the degradation period, the rolling bearing exhibits significant wear, producing noticeable friction noise during operation; the electromagnetic coil may experience electrical breakdown due to insulation degradation, resulting in a significant increase in current.

[0059] Then, based on the type of response data characteristics during the stable period, corresponding strongly correlated indicators are extracted from the operational response data; based on the type of response data characteristics during the degradation period, corresponding intrinsic indicators are extracted from the operational response data, and the working environment data that affects the intrinsic indicators are selected as extrinsic indicators. Finally, by combining the intrinsic indicators and the extrinsic indicators, the strongly correlated indicators during the degradation period are obtained.

[0060] The above scheme mainly takes into account that during the degradation period, changes in the external environment have a greater impact on magnetic levitation bearings, and some momentary disturbances may accelerate bearing aging. Therefore, when predicting the life changes during the degradation period, more attention will be paid to the environmental impact.

[0061] Based on the strong correlation indicators during the stable period, the first key feature was selected from the fusion features; based on the health indicators during the deterioration period, the second key feature was selected from the fusion features.

[0062] Since key features are learned more effectively in subsequent lifetime prediction, the weight coefficients of the first and second key features are increased to make lifetime predictions for the stable and deterioration periods focus more on these key features, resulting in predictions with better fit. Based on this, the stable and deterioration period features are obtained.

[0063] Step S4: Using the stable period features and degradation period features, the life degradation of the magnetic levitation bearing during the stable period and degradation period are predicted respectively, and the life degradation results are obtained.

[0064] Using the steady-state characteristics, the life degradation rate of the magnetic levitation bearing in its subsequent life cycle is predicted to obtain the first remaining life degradation curve. Using the degradation period characteristics, the life degradation rate of the magnetic levitation bearing in its subsequent life cycle is predicted to obtain the second remaining life degradation curve.

[0065] Because the two curves are obtained using weighting coefficients with different emphases, the curve that better fits the stable period is selected from the first remaining lifetime degradation curve, and the curve that better fits the degradation period is selected from the second remaining lifetime degradation curve.

[0066] Specifically, the curve segment in the first remaining lifetime degradation curve that is related to the time period of the stable period is selected as the stable period curve, and the curve segment in the second remaining lifetime degradation curve that is related to the time period of the degradation period is selected as the degradation period curve.

[0067] Considering that the steady-state curve and the degradation curve may not match at the accelerated degradation point, a historical degradation curve is also introduced to adjust the splicing point at the accelerated degradation point, so that the remaining bearing life at the accelerated degradation point is more in line with historical conditions.

[0068] Finally, the stable period curve and the degradation period curve are spliced ​​together at the point of accelerated degradation to obtain a characterizing result of the magnetic levitation bearing's future lifespan degradation. Based on this lifespan degradation result, maintenance personnel can more accurately schedule maintenance work on the magnetic levitation bearing, improving system operational stability.

[0069] Implementing the embodiments of this application has the following beneficial effects: This application embodiment not only collects the response data of the magnetic levitation bearing during operation but also collects surrounding environmental data, providing data support for subsequent consideration of the impact of the external environment on the bearing. First, based on pre-set degradation indicators, it preliminarily determines which point in the magnetic levitation bearing's life cycle it is currently at, directly determining the corresponding technical means for subsequent feature extraction and life prediction, achieving adaptive and dynamic life prediction, and ensuring that the most relevant and sensitive feature data can be extracted subsequently. Then, the overall life degradation process of the magnetic levitation bearing is predicted based on the operational response data and working environment data, identifying the accelerated degradation point. Using the accelerated degradation point as a boundary, the remaining life cycle of the bearing is divided into a stable period and a degradation period, and corresponding stable period features and degradation period features are constructed respectively, providing more correlated and sensitive data for life degradation prediction at different stages. Finally, by combining the life degradation predictions in the two stages, a more accurate life degradation result is obtained, providing a foundation for subsequent equipment maintenance.

[0070] Second Embodiment Furthermore, in order to implement the machine learning-based magnetic levitation bearing life prediction system corresponding to the above method embodiments, and to achieve the corresponding functions and technical effects, Figure 2 A structural diagram of a machine learning-based magnetic levitation bearing life prediction system is provided. For ease of explanation, only the parts relevant to this embodiment are shown. The machine learning-based magnetic levitation bearing life prediction system provided in this application embodiment includes: The data acquisition module 201 is used to collect the operating response data and working environment data of the magnetic levitation bearing during each operating phase; wherein, the operating response data includes operating current, operating temperature and vibration amplitude; and the working environment data includes humidity and start-stop frequency.

[0071] Unlike conventional mechanical bearings, which only exhibit performance degradation due to fatigue failure of the bearing material, magnetic levitation bearings, because they have no moving parts in contact, exhibit performance degradation more in the form of control system failure or low sensitivity. Therefore, it is necessary to collect the operating response data of magnetic levitation bearings in order to accurately predict the performance degradation of each component in the system.

[0072] In addition, the external environment can also affect the operation of magnetic levitation bearings. For example, strong external vibrations can cause the rotor to rub against the backup bearing or stator, affecting the adjustment capability of the control system; excessive dust or humidity in the operating environment can contaminate the coils of the magnetic levitation bearing, leading to a decrease in the coil insulation performance. Therefore, it is also necessary to collect operating environment data to assist in prediction.

[0073] Therefore, operational response data and working environment data of the magnetic levitation bearing are collected during each operating phase. The operational response data is primarily collected from the controller, the electromagnet coils of the stator and rotor, and sensors. Therefore, the operational response data includes operating current (used to measure whether there are any abnormalities in the current commands issued by the controller), operating temperature (used to measure the aging degree of the insulation material), vibration amplitude (used to measure the wear degree of the components), and sensor response speed (used to measure the sensitivity of the sensors and their ability to provide timely and accurate rotor position feedback to the control system). The collection of working environment data takes into account the interference of external vibration on the magnetic levitation bearing, the burden on the control system caused by environmental cleanliness and excessively high temperatures, and the risk of failure due to improper operation. Therefore, the working environment data includes temperature and humidity, start-stop frequency, pollution levels, and the frequency of overload operation.

[0074] Therefore, the life degradation prediction of magnetic levitation bearings cannot only look at the degree of structural damage. Considering that it corresponds to a complex closed-loop control system composed of electromagnets, sensors and controllers, it is also necessary to evaluate each component of the system individually in order to obtain a comprehensive degradation situation.

[0075] The degradation degree prediction module 202 is used to predict the degradation degree of each component in the magnetic levitation bearing based on the preset degradation index and the operation response data, and to determine the current life cycle stage of the magnetic levitation bearing.

[0076] In this embodiment of the application, the operational sequence characteristics of each component are extracted from the operational response data based on the degradation index corresponding to each component; wherein, the operational sequence characteristics include operational current characteristics, temperature characteristics, and amplitude characteristics; Based on the normal operating standards of each component, a corresponding sliding window is generated for the runtime sequence characteristics; The sliding window is used to identify anomalies in the runtime sequence features, and the frequency of anomalies for each runtime sequence feature is counted. The degradation level of each device is predicted based on the frequency of the anomalies, and the current life cycle stage is determined by combining the degradation level.

[0077] The accelerated degradation point identification module 203 is used to predict the accelerated degradation point of the magnetic levitation bearing after the current life cycle stage based on the operation response data and working environment data, and to construct stable period characteristics and degradation period characteristics through the accelerated degradation point.

[0078] In this embodiment of the application, based on the current lifecycle stage, feature fusion is performed on the runtime response data and the working environment data to obtain fused features; The overall life degradation cycle of the magnetic levitation bearing is predicted using fusion features, and the accelerated degradation point is determined. The period between the current life cycle stage and the accelerated life degradation point is defined as the stable period, and the period after the accelerated life degradation point is defined as the degradation period. Based on the strong correlation indicators between the stable period and the degradation period, stable period features and degradation period features are constructed by fusing features respectively.

[0079] The life prediction module 204 is used to predict the life degradation of the magnetic levitation bearing during the stable period and the degradation period using stable period characteristics and degradation period characteristics, respectively, and obtain the life degradation result.

[0080] In this embodiment, the stationary period feature is used to predict the life degradation rate of the magnetic levitation bearing in its subsequent life cycle, resulting in a first remaining life degradation curve. The degradation period feature is then used to predict the life degradation rate of the magnetic levitation bearing in its subsequent life cycle, yielding a second remaining life degradation curve.

[0081] Because the two curves are obtained using weighting coefficients with different emphases, the curve that better fits the stable period is selected from the first remaining lifetime degradation curve, and the curve that better fits the degradation period is selected from the second remaining lifetime degradation curve.

[0082] Specifically, the curve segment in the first remaining lifetime degradation curve that is related to the time period of the stable period is selected as the stable period curve, and the curve segment in the second remaining lifetime degradation curve that is related to the time period of the degradation period is selected as the degradation period curve.

[0083] Considering that the steady-state curve and the degradation curve may not match at the accelerated degradation point, a historical degradation curve is also introduced to adjust the splicing point at the accelerated degradation point, so that the remaining bearing life at the accelerated degradation point is more in line with historical conditions.

[0084] Finally, the stable period curve and the degradation period curve are spliced ​​together at the point of accelerated degradation to obtain a characterizing result of the magnetic levitation bearing's future lifespan degradation. Based on this lifespan degradation result, maintenance personnel can more accurately schedule maintenance work on the magnetic levitation bearing, improving system operational stability.

[0085] In some embodiments, the degradation degree prediction module 202 specifically comprises: First, based on the components and their corresponding degradation modes in the existing magnetic levitation bearing system, the degradation index of each component is determined.

[0086] For example, for an electromagnet coil, the typical degradation mode is insulation material aging, inter-turn short circuit and core performance degradation. Therefore, its degradation index is related to changes in inductance or bias current error. For a rolling device, the typical degradation mode is material wear or lubrication failure. Therefore, its degradation index is related to vibration frequency.

[0087] The aforementioned degradation indicators are benchmark values ​​obtained from a large number of samples, which can cover most device degradation scenarios and achieve accurate prediction of the degree of device degradation.

[0088] The degradation index is used to extract the operational sequence characteristics of each component from the operational response data. For example, the operational sequence characteristics of a displacement sensor are the displacement signals of each degree of freedom, the operational sequence characteristics of a rolling bearing are the bearing amplitude signal and the temperature of the bearing base, and the operational sequence characteristics of an electromagnet coil are the feedback value of the coil current.

[0089] Then, based on the normal operating standards of each component, a corresponding sliding window is generated for each runtime timing feature. When generating the sliding window, the window size is set according to the frequency of abnormal values ​​defined by the normal operating standards; and the sliding step size of the window is set according to the feedback frequency of the runtime timing feature.

[0090] Using the sliding window, anomalies are identified in the corresponding runtime sequence features by setting a sliding step size. Anomalies are determined by detecting whether the data within each window exceeds a set normal threshold, and the number of anomalies during each sliding step is counted to obtain the anomaly frequency for each runtime sequence feature.

[0091] As an improvement to the above scheme, when setting the normal threshold, the statistics from the previous step are used to dynamically adjust the normal threshold for the next time step, providing a larger correction space for anomaly detection.

[0092] Because the components in a magnetic levitation bearing system have varying degrees of importance, severe performance degradation in more critical components leads to more significant bearing life degradation, while performance degradation in less critical components sometimes has negligible impact on the bearing. Considering these factors, this application's embodiments predict the degradation level of each component to accurately pinpoint the current performance status of the magnetic levitation bearing and determine its current lifecycle stage. This provides precise time points for subsequent lifecycle degradation predictions targeting stages with different degradation characteristics.

[0093] First, considering the impact of abnormal operating sequence characteristics on the operating state of components, the contribution of each operating sequence characteristic to component degradation is set. For example, when the coil overheats severely, the insulation material of the electromagnet coil is more prone to failure, leading to inter-turn short circuits, thus the corresponding impact is relatively large; when the voltage signal output by the controller is converted into current, there is a deviation from the preset normal current, but the impact on the rotor's magnetic force is not significant, thus the corresponding impact is relatively small.

[0094] Then, the weighting factor for the runtime sequence features is set based on the degradation contribution. The greater the degradation contribution, the greater the corresponding weighting factor.

[0095] The degradation degree of each component is predicted using the weighting factors and the outlier frequencies. Because the data collection frequencies differ, all outlier frequencies are standardized before calculating the degradation degree, bringing them to the same dimension to ensure accurate prediction. Furthermore, considering that the dimensionality of the runtime sequence features for each component may differ, the weighted data is averaged according to the feature dimensions.

[0096] Finally, based on the degree of degradation of each component and considering the importance of each component in the magnetic levitation bearing system, the impact of each component on the performance degradation of the magnetic levitation bearing control system is quantified, and the current life cycle stage of the magnetic levitation bearing is determined.

[0097] Therefore, the embodiments of this application link the degradation of components with the performance degradation of the entire control system, thereby roughly assessing the current performance of the entire magnetic levitation bearing and providing a basis for predicting subsequent performance degradation based on the current performance.

[0098] In some embodiments, the accelerated degradation point identification module 203 specifically comprises: The operational response data reflects the internal degradation of the magnetic levitation bearing, while the working environment data reflects the external influences on the magnetic levitation bearing. Therefore, the internal degradation characteristics are extracted from the operational response data, and the external influence characteristics are extracted from the working environment data.

[0099] To detect whether external influences are related to internal degradation, the chi-square value between the two features is calculated by statistically analyzing whether the corresponding internal degradation feature exhibits an extreme value at an adjacent time point when the external influence feature exhibits an extreme value within a preset time interval. This quantifies whether there is a correlation or influence between the external influence feature and the internal degradation feature, thereby obtaining the corresponding feature correlation degree.

[0100] If the correlation degree of the features is greater than the first threshold, it indicates that there is a connection between the corresponding external influence feature and the internal degradation feature. Then, the two features are matched in time sequence to obtain a pair of related features. The fluctuation of the related feature pair within each preset time interval is then described, and a state label is generated for each time interval.

[0101] According to a preset time interval, the corresponding state label is inserted into the internal degradation feature, and then the external influence feature corresponding to the state label (i.e. the external influence feature of the associated feature pair) is spliced ​​into the internal degradation feature to obtain the fused feature of each associated feature pair.

[0102] The fused features are input into a preset long short-term memory network to predict the remaining lifetime of the magnetic levitation bearing at each predicted time point. Using adjacent remaining lifetime values, the lifetime degradation rate of the magnetic levitation bearing at each predicted time point is calculated in chronological order to obtain an overall lifetime degradation trend. This trend does not consider the different degradation characteristics of the bearing's stable period and degradation period, therefore the prediction accuracy differs from the actual situation. Therefore, it is not directly used in this embodiment. Instead, the time point when the lifetime degradation rate first exceeds a set rate threshold is identified and designated as the lifetime acceleration degradation point.

[0103] After the accelerated degradation point, magnetic levitation bearings begin to age comprehensively, the failure rate increases significantly, and the system may become unstable, requiring more frequent maintenance and component replacement. Therefore, after the accelerated degradation point, it is necessary to incorporate more degradation characteristics for lifespan prediction to improve prediction accuracy.

[0104] In this embodiment of the application, the period between the current life cycle stage and the accelerated life degradation point is defined as the stable period, and the period after the accelerated life degradation point is defined as the degradation period.

[0105] During the stable period, the bearing system has a low failure rate and reliable performance. The components are basically operating stably, with only a few abnormal situations. During the degradation period, the failure rate of the bearing system increases, the components show obvious degradation, and the environment has a greater impact on the components.

[0106] To address the above phenomena, we identified the characteristics that best reflect the performance changes of magnetic levitation bearings in the fusion profile by using strong correlation indicators between the stable and degradation periods.

[0107] Specifically, based on historical operating data analysis, the performance degradation characteristics of the magnetic levitation bearing were obtained, yielding response data characteristics during the stable and degradation periods. For example, during the stable period, the amplitude signal characteristics show slight shifts in the frequency domain and peak-valley fluctuations in the time domain; harmonics appear in the current of the electromagnetic coil. During the degradation period, the rolling bearing exhibits significant wear, producing noticeable friction noise during operation; the electromagnetic coil may experience electrical breakdown due to insulation degradation, resulting in a significant increase in current.

[0108] Then, based on the type of response data characteristics during the stable period, corresponding strongly correlated indicators are extracted from the operational response data; based on the type of response data characteristics during the degradation period, corresponding intrinsic indicators are extracted from the operational response data, and the working environment data that affects the intrinsic indicators are selected as extrinsic indicators. Finally, by combining the intrinsic indicators and the extrinsic indicators, the strongly correlated indicators during the degradation period are obtained.

[0109] The above scheme mainly takes into account that during the degradation period, changes in the external environment have a greater impact on magnetic levitation bearings, and some momentary disturbances may accelerate bearing aging. Therefore, when predicting the life changes during the degradation period, more attention will be paid to the environmental impact.

[0110] Based on the strong correlation indicators during the stable period, the first key feature was selected from the fusion features; based on the health indicators during the deterioration period, the second key feature was selected from the fusion features.

[0111] Since key features are learned more effectively in subsequent lifetime prediction, the weight coefficients of the first and second key features are increased to make lifetime predictions for the stable and deterioration periods focus more on these key features, resulting in predictions with better fit. Based on this, the stable and deterioration period features are obtained.

[0112] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of this application. It should be understood that the above descriptions are merely specific embodiments of this application and are not intended to limit the scope of protection of this application. In particular, it should be noted that any modifications, equivalent substitutions, or improvements made by those skilled in the art within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for predicting the lifespan of magnetic levitation bearings based on machine learning, characterized in that, include: The operation response data and working environment data of the magnetic levitation bearing are collected during each operating phase; wherein, the operation response data includes operating current, operating temperature and vibration amplitude; The operating environment data includes humidity and start / stop frequency; Based on preset degradation indicators, the degradation degree of each component in the magnetic levitation bearing is predicted by the operational response data, and the current life cycle stage of the magnetic levitation bearing is determined. Based on the operational response data and working environment data, the accelerated degradation point of the magnetic levitation bearing after the current life cycle stage is predicted, and stable period characteristics and degradation period characteristics are constructed through the accelerated degradation point. Using stable period features and degradation period features, the life degradation of magnetic levitation bearings during the stable period and degradation period are predicted respectively, and the life degradation results are obtained.

2. The method for predicting the lifespan of magnetic levitation bearings based on machine learning according to claim 1, characterized in that, The method, based on preset degradation indicators, uses operational response data to predict the degree of degradation of each component in the magnetic levitation bearing, thereby determining the current life cycle stage of the magnetic levitation bearing. Specifically: Based on the degradation index corresponding to each device, the operational sequence characteristics of each device are extracted from the operational response data; wherein, the operational sequence characteristics include operational current characteristics, temperature characteristics, and amplitude characteristics; Based on the normal operating standards of each component, a corresponding sliding window is generated for the runtime sequence characteristics; The sliding window is used to identify anomalies in the runtime sequence features, and the frequency of anomalies for each runtime sequence feature is counted. The degradation level of each device is predicted based on the frequency of the anomalies, and the current life cycle stage is determined by combining the degradation level.

3. The method for predicting the lifespan of magnetic levitation bearings based on machine learning according to claim 2, characterized in that, The step of predicting the degradation degree of each component based on the frequency of anomalies, and determining the current life cycle stage based on the degradation degree, specifically involves: Based on the contribution of each runtime timing feature to the degradation of the component, a weighting factor for each runtime timing feature is set; wherein, the degradation contribution is the degree of impact on the component performance when the runtime timing feature is abnormal; The degradation degree of each component is predicted using the weighting factor and the frequency of outliers. Based on the degree of degradation, the impact of each component on the performance degradation of the magnetic levitation bearing control system is quantified, and the current life cycle stage is determined.

4. The method for predicting the lifespan of magnetic levitation bearings based on machine learning according to claim 1, characterized in that, Based on the operational response data and working environment data, the accelerated degradation point of the magnetic levitation bearing after the current life cycle stage is predicted. Stable period characteristics and degradation period characteristics are then constructed using these accelerated degradation points. Specifically: Based on the current lifecycle stage, feature fusion is performed on the runtime response data and working environment data to obtain fused features; The overall life degradation cycle of the magnetic levitation bearing is predicted using fusion features, and the accelerated degradation point is determined. The period between the current life cycle stage and the accelerated life degradation point is defined as the stable period, and the period after the accelerated life degradation point is defined as the degradation period. Based on the strong correlation indicators between the stable period and the degradation period, stable period features and degradation period features are constructed by fusing features respectively.

5. The method for predicting the lifespan of magnetic levitation bearings based on machine learning according to claim 4, characterized in that, The step of fusing features between the runtime response data and the working environment data based on the current lifecycle stage to obtain fused features is as follows: Internal degradation features are extracted from the operational response data, and external influence features are extracted from the working environment data; The influence of the external influence features on the internal degradation features is quantified by the chi-square value to obtain the corresponding feature correlation degree. If the correlation degree of the features is greater than the first threshold, then data point matching is performed on the external influence features and the internal degradation features to obtain a pair of related features; The fluctuation of the associated features within each preset time interval is described, and the state label for each preset time interval is generated; The corresponding state label is inserted into the internal degradation feature, and the external influence feature corresponding to the state label is concatenated to the internal degradation feature to obtain the fused feature.

6. The method for predicting the lifespan of magnetic levitation bearings based on machine learning according to claim 4, characterized in that, The method of using fusion features to predict the overall lifespan degradation cycle of the magnetic levitation bearing and determining the accelerated degradation point specifically involves: Using fused features, the remaining lifespan of the magnetic levitation bearing at each prediction time point is predicted through a pre-set long short-term memory network. Based on the adjacent remaining lifetime values, the lifetime degradation rate at each predicted time point is calculated in chronological order. The point at which the lifetime degradation rate first exceeds the second threshold is defined as the point of accelerated lifetime degradation.

7. The method for predicting the lifespan of magnetic levitation bearings based on machine learning according to claim 4, characterized in that, The strongly correlated indicators based on the stable period and the degradation period are used to construct stable period features and degradation period features respectively by fusing features, specifically as follows: Based on the strong correlation indicators during the stable period, the first key feature was selected from the fusion features; based on the health indicators during the deterioration period, the second key feature was selected from the fusion features. Increase the weight coefficient of the first key feature in the fusion features to obtain the stable period features; By increasing the weight coefficient of the second key feature in the fusion features, the degradation period features are obtained.

8. The method for predicting the lifespan of magnetic levitation bearings based on machine learning according to claim 7, characterized in that, The strongly correlated indicators are specifically: Based on the analysis of historical operating data, the performance degradation characteristics of the magnetic levitation bearing were obtained, and the response data characteristics of the magnetic levitation bearing during the stable period and the degradation period were obtained respectively. Based on the type of response data characteristics during the stable period, the corresponding strongly correlated indicators are extracted from the operational response data; Based on the type of response data characteristics during the degradation period, corresponding intrinsic indicators are extracted from the operational response data, and the working environment data that affects the intrinsic indicators are selected as extrinsic indicators. By combining the intrinsic indicators and the extrinsic indicators, a strongly correlated indicator for the degradation period is obtained.

9. The method for predicting the lifespan of magnetic levitation bearings based on machine learning according to claim 1, characterized in that, The method uses stable period characteristics and degradation period characteristics to predict the life degradation of magnetic levitation bearings during the stable period and degradation period, respectively, and obtains the life degradation results, specifically: Based on the characteristics of the steady-state period, the life degradation rate of the magnetic levitation bearing is predicted, and the first remaining life degradation curve is obtained. Based on the characteristics of the degradation period, the degradation rate of the magnetic levitation bearing is predicted, and the second remaining life degradation curve is obtained. Select the curve segment in the first remaining lifetime degradation curve that is related to the time period of the stable period as the stable period curve, and select the curve segment in the second remaining lifetime degradation curve that is related to the time period of the degradation period as the degradation period curve. The stable period curve and the degradation period curve are spliced ​​together, and the splicing point is adjusted using a preset historical degradation curve to obtain the lifetime degradation result.

10. A machine learning-based system for predicting the lifespan of magnetic levitation bearings, characterized in that, include: Data acquisition module, degradation degree prediction module, accelerated degradation point identification module, and lifetime prediction module; The data acquisition module is used to collect the operating response data and working environment data of the magnetic levitation bearing during each operating phase; the operating response data includes operating current, operating temperature, and vibration amplitude; the working environment data includes humidity and start / stop frequency. The degradation degree prediction module is used to predict the degradation degree of each component in the magnetic levitation bearing based on preset degradation indicators and the operation response data, so as to determine the current life cycle stage of the magnetic levitation bearing. The accelerated degradation point identification module is used to predict the accelerated degradation point of the magnetic levitation bearing after the current life cycle stage based on the operation response data and working environment data, and to construct stable period characteristics and degradation period characteristics through the accelerated degradation point. The life prediction module is used to predict the life degradation of magnetic levitation bearings during the stable period and the degradation period using stable period features and degradation period features, respectively, and obtain the life degradation results.

Images