Turbine bearing vibration fault early warning method based on machine learning

Abstract

The invention discloses a turbine bearing vibration fault early warning method based on machine learning, and the method comprises the steps: presetting a plurality of monitoring points of a turbine bearing, building a monitoring point reference diagram according to all monitoring points, the monitoring point reference diagram comprises a plurality of monitoring points, and each monitoring point is mapped with a preset fault type; determining a vibration characteristic and a vibration risk value of each monitoring point in the historical monitoring period, and generating a vibration fault early warning model of the corresponding monitoring point according to the plurality of vibration characteristics and the vibration risk values; the real-time vibration characteristics of each monitoring point in the current monitoring period are obtained, a predicted vibration risk value of the real-time vibration characteristics is generated based on the vibration fault early warning model of the corresponding monitoring point, a corresponding early warning signal is generated according to the predicted vibration risk value, and a corresponding operation and maintenance strategy is configured, so that timely early warning of the vibration fault of the steam turbine bearing is realized. The operation and maintenance cost is reduced, and a powerful guarantee is provided for safe and stable power supply of power generation enterprises.

Description

Technical Field

[0001] This application relates to the technical field of bearing vibration, and particularly to a method for early warning of steam turbine bearing vibration faults based on machine learning. Background Art

[0002] In the maintenance logs of supercritical condensing generator sets, there have always been problems of relatively large vibrations in steam turbine bearings, which are likely to cause secondary faults such as broken bearings, corrosion, and wear and hydrogen leakage of the sealing bearings. Early warning of vibration faults in bearings is an important task for unit monitoring. However, the existing technologies cannot achieve timely early warning of steam turbine bearing vibration faults, cannot quickly carry out vibration fault diagnosis and operation and maintenance work, resulting in the inability to use or failure of other equipment, and cannot ensure the safe and stable power supply of power generation enterprises. Summary of the Invention

[0003] To solve the above technical problems, this application provides a method for early warning of steam turbine bearing vibration faults based on machine learning. By setting multiple monitoring points and extracting the vibration characteristics of each monitoring point, the vibration risk value of each vibration characteristic is obtained, and a vibration fault early warning model for each monitoring point is constructed. The predicted vibration risk value of the real-time vibration characteristics is quickly obtained, and early warning signals of different levels and reasonable operation and maintenance strategies are generated, realizing timely early warning of steam turbine bearing vibration faults, determining the fault type and formulating operation and maintenance strategies, reducing operation and maintenance costs, and providing a strong guarantee for the safe and stable power supply of power generation enterprises.

[0004] In some embodiments of this application, a method for early warning of steam turbine bearing vibration faults based on machine learning is provided, including:

[0005] Preset multiple monitoring points of the steam turbine bearing in advance, and construct a monitoring point reference diagram according to all the monitoring points. The monitoring point reference diagram includes several monitoring points, and each monitoring point is mapped with a preset fault type;

[0006] Determine the vibration characteristics and vibration risk values of each monitoring point in the historical monitoring period, and generate a vibration fault early warning model for the corresponding monitoring point according to the multiple vibration characteristics and vibration risk values;

[0007] Obtain the real-time vibration characteristics of each monitoring point in the current monitoring period, generate the predicted vibration risk value of the real-time vibration characteristics based on the vibration fault early warning model of the corresponding monitoring point, generate the corresponding early warning signal according to the predicted vibration risk value, and configure the corresponding operation and maintenance strategy.

[0008] In some embodiments of this application, it is characterized in that presetting multiple monitoring points of the steam turbine bearing includes:

[0009] Obtain the historical vibration fault logs of the steam turbine bearings, analyze the historical vibration fault logs, determine the historical fault areas corresponding to each historical vibration fault log, as well as the historical fault levels and historical operation and maintenance costs of the historical fault areas;

[0010] Generate a fault evaluation value for the corresponding historical fault area according to the historical fault level of each historical fault area;

[0011] Generate a correction coefficient for the corresponding historical fault area according to the historical operation and maintenance cost, and correct the corresponding fault evaluation value according to the correction coefficient;

[0012] Compare the corrected fault evaluation value of each historical fault area with the preset fault evaluation value threshold, and set the historical fault area greater than the preset fault evaluation value threshold as the area of concern;

[0013] Set the number of monitoring points and the positions of the corresponding monitoring points for the area of concern according to the area of the area of concern;

[0014] Judge the relative distance between adjacent areas of concern. If the relative distance is greater than the preset relative distance threshold and generate a relative distance difference, set the number of monitoring points to be supplemented and the positions of the corresponding monitoring points according to the relative distance difference.

[0015] In some embodiments of the present application, a monitoring point reference diagram is constructed according to all monitoring points, including:

[0016] Construct a monitoring point reference diagram according to the position information of all monitoring points. The monitoring point reference diagram includes several monitoring points, and each monitoring point is mapped with a corresponding preset fault type;

[0017] Each preset fault type is associated with multiple preset fault characteristics, and each preset fault characteristic is configured with a corresponding preset fault evaluation value and a preset operation and maintenance strategy.

[0018] In some embodiments of the present application, determine the vibration characteristics and vibration risk values of each monitoring point within the historical monitoring period, including:

[0019] Set multiple monitoring time nodes at preset time intervals, and generate multiple monitoring periods according to the multiple monitoring time nodes;

[0020] Obtain the historical vibration parameters of each monitoring point at multiple monitoring time nodes in each historical monitoring period, and generate a vibration parameter change curve for each historical monitoring period;

[0021] Perform a variation gradient analysis on each vibration parameter variation curve, mark the monitoring time nodes with a variation gradient greater than the preset variation gradient, and set the vibration amplitude, vibration amplitude value, and vibration frequency between adjacent and marked monitoring time nodes in the same vibration parameter variation curve as the vibration characteristics of the corresponding historical monitoring period;

[0022] Compare the vibration characteristics of each historical monitoring period with the standard vibration characteristics of the corresponding monitoring point, and generate a vibration risk value according to the comparison result.

[0023] In some embodiments of the present application, generating a vibration risk value according to the comparison result includes:

[0024] The standard vibration characteristics include a standard vibration amplitude range, a vibration amplitude threshold, and a vibration frequency threshold;

[0025] Divide the corresponding vibration characteristics into multiple vibration sub-characteristics according to the duration length between the monitoring time nodes of each vibration characteristic, and divide the standard vibration characteristics into multiple standard vibration sub-characteristics according to the time interval of the monitoring time sub-nodes;

[0026] Compare each vibration sub-characteristic with the corresponding standard vibration sub-characteristic, and the comparison result includes a number of vibration amplitude differences, vibration amplitude differences, and vibration frequency differences;

[0027] If the vibration amplitude difference is greater than the preset vibration amplitude difference, the vibration amplitude difference is greater than the preset vibration amplitude difference, and the vibration frequency difference is greater than the preset vibration frequency difference, then obtain the corresponding first duration, second duration, and third duration respectively;

[0028] The calculation formula for the vibration risk value is:

[0029]

[0030] Wherein, F is the vibration risk value, y1 is the first vibration risk conversion coefficient, a1 is the weight coefficient of the vibration amplitude, △Z1i is the i-th vibration amplitude difference, t1i is the first duration of the i-th vibration amplitude difference, n is the total number of vibration sub-characteristics and standard vibration sub-characteristics, △Z2i is the i-th vibration amplitude difference, t2i is the second duration of the i-th vibration amplitude difference, △Z3i is the i-th vibration frequency difference, t3i is the third duration of the i-th vibration frequency difference, y2 is the second vibration risk conversion coefficient, a2 is the weight coefficient of the vibration amplitude, y3 is the third vibration risk conversion coefficient, and a3 is the weight coefficient of the vibration frequency.

[0031] In some embodiments of the present application, before generating a vibration fault warning model for the corresponding monitoring point according to multiple vibration characteristics and the vibration risk value, it includes:

[0032] Preset the target vibration risk value for the preset fault type of each monitoring point;

[0033] If the vibration risk value of the vibration characteristics of the current monitoring point is greater than the target vibration risk value, determine that the vibration characteristics corresponding to the current monitoring point are fault characteristics, and filter out the preset fault characteristics of the preset fault type corresponding to the target vibration risk value;

[0034] Perform a similarity analysis on the fault characteristics of the current monitoring point and the preset fault characteristics of the preset fault type that are filtered out, set the preset fault type corresponding to the preset fault characteristic with the greatest similarity as the fault type of the fault characteristics corresponding to the current monitoring point, and set a correction coefficient according to the preset fault evaluation value of the preset fault characteristic with the greatest similarity to the fault characteristics;

[0035] Determine the vibration influence range based on the fault characteristics and fault type of the current monitoring point, and determine the influencing monitoring points of the current monitoring point in the monitoring point reference diagram based on the vibration influence range;

[0036] Generate an influence coefficient according to the influence degree of the fault characteristics and fault type of the current monitoring point on each influencing monitoring point;

[0037] Correct the vibration risk value of the corresponding fault characteristics according to the correction coefficient and the influence coefficient, and construct a vibration fault warning model for the corresponding monitoring point according to the corrected vibration risk value of the fault characteristics and the vibration risk values of other vibration characteristics.

[0038] In some embodiments of the present application, setting the correction coefficient according to the preset fault evaluation value of the preset fault characteristic with the greatest similarity to the fault characteristics includes:

[0039] Preset a first preset fault evaluation value interval, a second preset fault evaluation value interval, a third preset fault evaluation value interval, and a fourth preset fault evaluation value interval;

[0040] When the preset fault evaluation value is within the first preset fault evaluation value interval, set the first preset correction coefficient as the correction coefficient;

[0041] When the preset fault evaluation value is within the second preset fault evaluation value interval, set the second preset correction coefficient as the correction coefficient;

[0042] When the preset fault evaluation value is within the third preset fault evaluation value interval, set the third preset correction coefficient as the correction coefficient;

[0043] When the preset fault evaluation value is within the fourth preset fault evaluation value interval, set the fourth preset correction coefficient as the correction coefficient.

[0044] In some embodiments of the present application, a vibration fault warning model for a corresponding monitoring point is constructed based on the corrected vibration risk value of the fault feature and the vibration risk values of other vibration features, including:

[0045] Based on the vibration features and fault features of all historical monitoring cycles of each monitoring point as the training input data set, and taking the vibration risk value corresponding to the vibration feature and the corrected vibration risk value corresponding to the fault feature as the training output data set;

[0046] According to the training input data set and training output data of each monitoring point, neural network training is performed to obtain the vibration fault warning model of the corresponding monitoring point.

[0047] In some embodiments of the present application, a corresponding warning signal is generated according to the predicted vibration risk value, and corresponding operation and maintenance strategies are configured, including:

[0048] Obtain the real-time vibration features of each monitoring point in the current monitoring cycle, and input them into the vibration fault warning model of the corresponding monitoring point to obtain the predicted vibration risk value of the real-time vibration features;

[0049] Compare the predicted vibration risk value of the real-time vibration features of each monitoring point with the target vibration risk value of each preset fault type of the corresponding monitoring point. If the predicted vibration risk values are all less than the target vibration risk values of all corresponding monitoring points, and the differences between the predicted vibration risk values and the target vibration risk values of all target vibration risk values are all greater than the preset difference, no warning signal is sent;

[0050] If the predicted vibration risk values are all less than the target vibration risk values of all corresponding monitoring points, and there are differences in the predicted vibration risk values less than the preset difference, a first-level warning signal is sent;

[0051] If the predicted vibration risk value is greater than one or more target vibration risk values of the corresponding monitoring point, a second-level warning signal is sent, and the preset fault types corresponding to the target vibration risk values are screened out. The real-time vibration features are set as real-time fault features, and a similarity analysis is performed with the preset fault features of the screened preset fault types. The preset operation and maintenance strategy of the preset fault feature with the largest similarity is set as the operation and maintenance strategy of the current real-time fault feature.

[0052] A method for warning of steam turbine bearing vibration faults based on machine learning according to an embodiment of the present application, compared with the prior art, has the beneficial effects that:

[0053] By setting multiple monitoring points and extracting the vibration characteristics of each monitoring point, the vibration risk value of each vibration characteristic is obtained, and a vibration fault warning model for each monitoring point is constructed to quickly obtain the predicted vibration risk value of the real-time vibration characteristic, and generate warning signals at different levels and reasonable operation and maintenance strategies, realizing the timely warning of the vibration fault of the steam turbine bearing, determining the fault type and formulating operation and maintenance strategies, reducing the operation and maintenance cost, and providing a strong guarantee for the safe and stable power supply of power generation enterprises. Brief Description of the Drawings

[0054] Figure 1 It is a schematic flow chart of a method for warning vibration faults of steam turbine bearings based on machine learning in a preferred embodiment of an embodiment of the present application. Detailed Embodiments

[0055] The following will further describe in detail the specific embodiments of the present application in conjunction with the drawings and embodiments. The following embodiments are used to illustrate the present application, but are not used to limit the scope of the present application.

[0056] In the description of the present application, it should be understood that the terms "center", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present application and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus cannot be understood as a limitation to the present application.

[0057] The terms "first" and "second" are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the quantity of the indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of such features. In the description of the present application, unless otherwise specified, the meaning of "plurality" is two or more.

[0058] In the description of the present application, it should be noted that unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense. For example, it may be a fixed connection, a detachable connection, or an integral connection; it may be a mechanical connection or an electrical connection; it may be directly connected, or indirectly connected through an intermediate medium, and it may be the communication inside two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood according to specific situations.

[0059] As Figure 1 shown, a method for warning vibration faults of steam turbine bearings based on machine learning in a preferred embodiment of an embodiment of the present application includes:

[0060] Step S101: Preset multiple monitoring points of the steam turbine bearing, and construct a monitoring point reference diagram according to all the monitoring points. The monitoring point reference diagram includes several monitoring points, and each monitoring point is mapped with a preset fault type;

[0061] Step S102: Determine the vibration characteristics and vibration risk values of each monitoring point within the historical monitoring period, and generate a vibration fault warning model for the corresponding monitoring point according to the multiple vibration characteristics and vibration risk values;

[0062] Step S103: Obtain the real-time vibration characteristics of each monitoring point in the current monitoring period, generate the predicted vibration risk value of the real-time vibration characteristics based on the vibration fault warning model of the corresponding monitoring point, generate a corresponding warning signal according to the predicted vibration risk value, and configure the corresponding operation and maintenance strategy.

[0063] In some embodiments of the present application, presetting multiple monitoring points of the steam turbine bearing includes:

[0064] Obtain the historical vibration fault log of the steam turbine bearing, analyze the historical vibration fault log, determine the historical fault area corresponding to each historical vibration fault log, as well as the historical fault level and historical operation and maintenance cost of the historical fault area;

[0065] Generate a fault evaluation value for the corresponding historical fault area according to the historical fault level of each historical fault area;

[0066] Generate a correction coefficient for the corresponding historical fault area according to the historical operation and maintenance cost, and correct the corresponding fault evaluation value according to the correction coefficient;

[0067] Compare the corrected fault evaluation value of each historical fault area with the preset fault evaluation value threshold, and set the historical fault area greater than the preset fault evaluation value threshold as the concerned area;

[0068] Set the number of monitoring points and the positions of the corresponding monitoring points of the concerned area according to the area of the concerned area;

[0069] Judge the relative distance between adjacent concerned areas. If the relative distance is greater than the preset relative distance threshold, generate a relative distance difference, and set the number of monitoring points to be supplemented and the positions of the corresponding monitoring points according to the relative distance difference.

[0070] In this embodiment, by determining the areas prone to faults and the fault levels and operation and maintenance costs of each fault area from the historical fault log of the steam turbine bearing, the number and positions of the corresponding monitoring points are set, so as to achieve accurate monitoring and timely warning of each fault area.

[0071] In some embodiments of the present application, a monitoring point reference diagram is constructed based on all the monitoring points, including:

[0072] A monitoring point reference diagram is constructed according to the position information of all the monitoring points. The monitoring point reference diagram includes a number of monitoring points, and each monitoring point is mapped to a corresponding preset fault type;

[0073] Each preset fault type is associated with multiple preset fault characteristics, and each preset fault characteristic is configured with a corresponding preset fault evaluation value and a preset operation and maintenance strategy.

[0074] In this embodiment, a number of preset fault types and the preset fault characteristics corresponding to each preset fault type are extracted from the historical vibration fault logs corresponding to each monitoring point. The preset fault types include bearing damage, shaft diameter deformation, bolt loosening, etc.

[0075] In this embodiment, the preset fault characteristic refers to the vibration amplitude, vibration frequency, and vibration range when each preset fault type generates a vibration fault. The preset fault evaluation value is a comprehensive evaluation of the fault level and operation and maintenance cost of the preset fault characteristics of each preset fault type. The historical operation and maintenance strategies after the vibration fault of the monitoring point are screened, and the historical operation and maintenance strategies with a vibration reduction effect greater than the preset vibration reduction effect at the monitoring point after the historical operation and maintenance strategy are set as the preset operation and maintenance strategies for the preset fault characteristics corresponding to the current monitoring point preset fault type.

[0076] In this embodiment, by constructing a monitoring point reference diagram and setting the preset fault type, preset fault characteristic, preset fault evaluation value, and preset operation and maintenance strategy of each monitoring point in the monitoring point reference diagram, a foundation is laid for subsequent determination of the vibration risk value and operation and maintenance strategy, improving the accuracy and timeliness of the vibration risk value and operation and maintenance strategy.

[0077] In some embodiments of the present application, determining the vibration characteristics and vibration risk value of each monitoring point within the historical monitoring period includes:

[0078] Set a number of monitoring time nodes at preset time intervals and generate a number of monitoring periods based on the number of monitoring time nodes;

[0079] Obtain the historical vibration parameters of each monitoring point at the multiple monitoring time nodes of each historical monitoring period and generate a vibration parameter change curve for each historical monitoring period;

[0080] Perform a change gradient analysis on each vibration parameter change curve, mark the monitoring time nodes with a change gradient greater than the preset change gradient, and set the vibration range, vibration amplitude, and vibration frequency between adjacent and marked monitoring time nodes in the same vibration parameter change curve as the vibration characteristics of the corresponding historical monitoring period;

[0081] Compare the vibration characteristics of each historical monitoring period with the standard vibration characteristics of the corresponding monitoring points, and generate a vibration risk value according to the comparison result.

[0082] In this embodiment, by setting multiple monitoring periods, the bearing vibration monitoring is divided into multiple nodes, and the bearing vibration monitoring results of each detection period are obtained, which lays a foundation for subsequent bearing fault warning, improves the accuracy of bearing fault warning, calculates the vibration risk value of the vibration characteristics of each historical monitoring period, improves the accuracy of the vibration risk value, and lays a foundation for subsequent construction of the vibration fault warning model for each monitoring point.

[0083] In some embodiments of the present application, generating a vibration risk value according to the comparison result includes:

[0084] The standard vibration characteristics include a standard vibration amplitude range, a vibration amplitude threshold, and a vibration frequency threshold;

[0085] Divide the corresponding vibration characteristics into multiple vibration sub-characteristics according to the duration between the monitoring time nodes of each vibration characteristic, and divide the standard vibration characteristics into multiple standard vibration sub-characteristics according to the time interval of the monitoring time sub-nodes;

[0086] Compare each vibration sub-characteristic with the corresponding standard vibration sub-characteristic, and the comparison results include several vibration amplitude differences, vibration amplitude differences, and vibration frequency differences;

[0087] If the vibration amplitude difference is greater than the preset vibration amplitude difference, the vibration amplitude difference is greater than the preset vibration amplitude difference, and the vibration frequency difference is greater than the preset vibration frequency difference, respectively obtain the corresponding first duration, second duration, and third duration;

[0088] The calculation formula of the vibration risk value is:

[0089]

[0090] Where F is the vibration risk value, y1 is the first vibration risk conversion coefficient, a1 is the weight coefficient of the vibration amplitude, △Z1i is the i-th vibration amplitude difference, t1i is the first duration of the i-th vibration amplitude difference, n is the total number of vibration sub-characteristics and standard vibration sub-characteristics, △Z2i is the i-th vibration amplitude difference, t2i is the second duration of the i-th vibration amplitude difference, △Z3i is the i-th vibration frequency difference, t3i is the third duration of the i-th vibration frequency difference, y2 is the second vibration risk conversion coefficient, a2 is the weight coefficient of the vibration amplitude, y3 is the third vibration risk conversion coefficient, and a3 is the weight coefficient of the vibration frequency.

[0091] In this embodiment, the standard vibration characteristics of each monitoring point are set according to the vibration characteristics of the vibration parameters of each monitoring point during normal operation. The vibration characteristics and the standard vibration characteristics are divided into multiple vibration sub-characteristics, thereby improving the accuracy of the vibration characteristics and the standard vibration characteristics, that is, improving the accuracy of the vibration risk value.

[0092] In some embodiments of the present application, before generating a vibration fault warning model for a corresponding monitoring point according to multiple vibration characteristics and a vibration risk value, it includes:

[0093] Preset the target vibration risk value of the preset fault type for each monitoring point;

[0094] If the vibration risk value of the vibration characteristics of the current monitoring point is greater than the target vibration risk value, it is determined that the vibration characteristics corresponding to the current monitoring point are fault characteristics, and the preset fault characteristics of the preset fault type corresponding to the target vibration risk value are screened out;

[0095] Perform a similarity analysis on the fault characteristics of the current monitoring point and the preset fault characteristics of the preset fault type that are screened out. Set the preset fault type corresponding to the preset fault characteristic with the greatest similarity as the fault type of the fault characteristics corresponding to the current monitoring point, and set a correction coefficient according to the preset fault evaluation value of the preset fault characteristic with the greatest similarity to the fault characteristics;

[0096] Determine the vibration influence range according to the fault characteristics and the fault type of the current monitoring point, and determine the influencing monitoring points of the current monitoring point in the monitoring point reference diagram based on the vibration influence range;

[0097] Generate an influence coefficient according to the influence degree of the fault characteristics and the fault type of the current monitoring point on each influencing monitoring point;

[0098] Correct the vibration risk value of the corresponding fault characteristics according to the correction coefficient and the influence coefficient, and construct a vibration fault warning model for the corresponding monitoring point according to the corrected vibration risk value of the fault characteristics and the vibration risk values of other vibration characteristics.

[0099] In this embodiment, the target vibration risk value is the minimum vibration risk value of the preset fault type for each monitoring point. The influencing monitoring point refers to other monitoring points within the area affected by the vibration of the fault characteristics of the corresponding fault type. The greater the influence degree, the greater the influence coefficient. The value range of the influence coefficient is (0.8, 1.2).

[0100] In this embodiment, the similarity between the preset fault characteristics and the fault characteristics is obtained by comprehensively considering the vibration amplitude similarity, vibration range similarity, and vibration frequency similarity within the same time period.

[0101] In this embodiment, by comparing the vibration risk value of each monitoring point with multiple target vibration risk values of the corresponding monitoring point, the fault characteristics are determined. A correction coefficient is set according to the preset fault evaluation value of the fault characteristics of each monitoring point, and an influence coefficient is set according to the fault type and the degree of influence on the monitoring point, so as to correct the vibration risk value of the fault characteristics, improve the accuracy of the vibration risk value of the fault characteristics, lay a foundation for subsequent sending of warning signals and formulating of operation and maintenance strategies, improve the vibration reduction effect of the steam turbine bearing and reduce the maintenance cost, and provide a strong guarantee for the safe and stable power supply of the power generation enterprise.

[0102] In some embodiments of the present application, setting a correction coefficient according to the preset fault evaluation value of the preset fault characteristic with the highest similarity to the fault characteristic includes:

[0103] A first preset fault evaluation value interval, a second preset fault evaluation value interval, a third preset fault evaluation value interval, and a fourth preset fault evaluation value interval are preset in advance;

[0104] When the preset fault evaluation value is in the first preset fault evaluation value interval, set the first preset correction coefficient as the correction coefficient;

[0105] When the preset fault evaluation value is in the second preset fault evaluation value interval, set the second preset correction coefficient as the correction coefficient;

[0106] When the preset fault evaluation value is in the third preset fault evaluation value interval, set the third preset correction coefficient as the correction coefficient;

[0107] When the preset fault evaluation value is in the fourth preset fault evaluation value interval, set the fourth preset correction coefficient as the correction coefficient.

[0108] Among them, the first preset fault evaluation value interval < the second preset fault evaluation value interval < the third preset fault evaluation value interval < the fourth preset fault evaluation value interval, the first preset correction coefficient < the second preset correction coefficient < the third preset correction coefficient < the fourth preset correction coefficient, and the value range of the preset correction coefficient is (1, 1.35).

[0109] In this embodiment, when the interval where the preset fault evaluation value is located is larger, it indicates that the fault level of the fault characteristics of the corresponding monitoring point and the subsequent operation and maintenance cost are higher, then the corresponding vibration risk value should be appropriately increased, and the preset correction coefficient is set in advance.

[0110] In some embodiments of the present application, constructing a vibration fault warning model for the corresponding monitoring point according to the corrected vibration risk value of the fault characteristics and the vibration risk values of other vibration characteristics includes:

[0111] Using the vibration characteristics and fault characteristics of all historical monitoring cycles of each monitoring point as the training input data set, and using the vibration risk value corresponding to the vibration characteristics and the corrected vibration risk value corresponding to the fault characteristics as the training output data set;

[0112] Perform neural network training based on the training input data set and training output data of each monitoring point to obtain the vibration fault warning model of the corresponding monitoring point.

[0113] In some embodiments of the present application, corresponding warning signals are generated according to the predicted vibration risk values, and corresponding operation and maintenance strategies are configured, including:

[0114] Obtain the real-time vibration characteristics of each monitoring point in the current monitoring cycle, and input them into the vibration fault warning model of the corresponding monitoring point to obtain the predicted vibration risk value of the real-time vibration characteristics;

[0115] Compare the predicted vibration risk value of the real-time vibration characteristics of each monitoring point with the target vibration risk value of each preset fault type of the corresponding monitoring point. If the predicted vibration risk values are all less than the target vibration risk values of all corresponding monitoring points, and the differences between the predicted vibration risk values and the target vibration risk values of all corresponding monitoring points are all greater than the preset difference, no warning signal is sent;

[0116] If the predicted vibration risk values are all less than the target vibration risk values of all corresponding monitoring points, and there are differences in the predicted vibration risk values less than the preset difference, send a first-level warning signal;

[0117] If the predicted vibration risk value is greater than one or more target vibration risk values of the corresponding monitoring point, send a second-level warning signal, screen out the preset fault types corresponding to the target vibration risk values, set the real-time vibration characteristics as real-time fault characteristics, perform similarity analysis with the preset fault characteristics of the screened preset fault types, and set the preset operation and maintenance strategy of the preset fault characteristic with the greatest similarity as the operation and maintenance strategy of the current real-time fault characteristic.

[0118] The above are only the preferred embodiments of the present application. It should be noted that for those of ordinary skill in the art, without departing from the technical principle of the present application, several improvements and replacements can be made, and these improvements and replacements should also be regarded as the protection scope of the present application.

Claims

1. A turbine bearing vibration fault early warning method based on machine learning, characterized in that: include: Preset multiple monitoring points of the turbine bearing, and construct a monitoring point reference map based on all the monitoring points, wherein the monitoring point reference map includes a plurality of monitoring points, and each monitoring point is mapped with a preset fault type; Determine the vibration characteristics and vibration risk value of each monitoring point in the historical monitoring period, and generate a vibration fault early warning model for the corresponding monitoring point based on multiple vibration characteristics and vibration risk values; The real-time vibration characteristics of each monitoring point in the current monitoring period are obtained, and the predicted vibration risk value of the real-time vibration characteristics is generated based on the vibration fault warning model of the corresponding monitoring point. The corresponding warning signal is generated according to the predicted vibration risk value, and the corresponding operation and maintenance strategy is configured.

2. The method for early warning of turbine bearing vibration fault based on machine learning according to claim 1, characterized in that: Multiple monitoring points of turbine bearings are pre-set, including: Obtain historical vibration fault logs of turbine bearings, analyze the historical vibration fault logs, determine the historical fault area corresponding to each historical vibration fault log, as well as the historical fault level and historical operation and maintenance cost of the historical fault area; Generate a fault evaluation value of the corresponding historical fault area according to the historical fault level of each historical fault area; Generate a correction coefficient for the historical fault area based on the historical operation and maintenance cost, and correct the corresponding fault evaluation value based on the correction coefficient; Compare the corrected fault evaluation value of each historical fault area with a preset fault evaluation value threshold, and set the historical fault area with a value greater than the preset fault evaluation value threshold as a focus area; According to the area of ​​the concerned area, the number of monitoring points corresponding to the concerned area and the positions of the corresponding monitoring points are set; The relative distance between adjacent areas of interest is determined. If the relative distance is greater than a preset relative distance threshold, a relative distance difference is generated, and the number of monitoring points to be supplemented and the positions of the corresponding monitoring points are set according to the relative distance difference.

3. The method for early warning of turbine bearing vibration fault based on machine learning according to claim 2, characterized in that: Construct a monitoring point reference map based on all monitoring points, including: Constructing a monitoring point reference map according to the location information of all monitoring points, wherein the monitoring point reference map includes a plurality of monitoring points, and each monitoring point is mapped with a corresponding preset fault type; Each preset fault type is associated with a plurality of preset fault features, and each preset fault feature is configured with a corresponding preset fault evaluation value and a preset operation and maintenance strategy.

4. The method for early warning of turbine bearing vibration fault based on machine learning according to claim 3, characterized in that: Determine the vibration characteristics and vibration risk value of each monitoring point during the historical monitoring period, including: Setting multiple monitoring time nodes according to preset time intervals, and generating multiple monitoring cycles according to the multiple monitoring time nodes; Obtain historical vibration parameters of each monitoring point at multiple monitoring time nodes in each historical monitoring period, and generate a vibration parameter change curve for each historical monitoring period; Perform change gradient analysis on each vibration parameter change curve, mark the monitoring time nodes whose change gradient is greater than the preset change gradient, and set the vibration amplitude, vibration amplitude and vibration frequency between adjacent and marked monitoring time nodes in the same vibration parameter change curve as the vibration characteristics of the corresponding historical monitoring period; The vibration characteristics of each historical monitoring period are compared with the standard vibration characteristics of the corresponding monitoring point, and the vibration risk value is generated based on the comparison results.

5. The method for early warning of turbine bearing vibration fault based on machine learning according to claim 4, characterized in that: Generate a vibration risk value based on the comparison results, including: The standard vibration characteristics include a standard vibration amplitude range, a vibration amplitude threshold, and a vibration frequency threshold; Divide the corresponding vibration feature into a plurality of vibration sub-features according to the duration between the monitoring time nodes of each vibration feature, and divide the standard vibration feature into a plurality of standard vibration sub-features according to the time interval between the monitoring time sub-nodes; Comparing each vibration sub-feature with a corresponding standard vibration sub-feature, wherein the comparison result includes a plurality of vibration amplitude differences, vibration magnitude differences, and vibration frequency differences; If the vibration amplitude difference is greater than the preset vibration amplitude difference, the vibration amplitude difference is greater than the preset vibration amplitude difference, and the vibration frequency difference is greater than the preset vibration frequency difference, the corresponding first duration, second duration, and third duration are obtained respectively; The calculation formula of the vibration risk value is: Among them, F is the vibration risk value, y1 is the first vibration risk conversion coefficient, a1 is the weight coefficient of vibration amplitude, △Z1i is the i-th vibration amplitude difference, t1i is the first duration of the i-th vibration amplitude difference, n is the total number of vibration sub-features and standard vibration sub-features, △Z2i is the i-th vibration amplitude difference, t2i is the second duration of the i-th vibration amplitude difference, △Z3i is the i-th vibration frequency difference, t3i is the third duration of the i-th vibration frequency difference, y2 is the second vibration risk conversion coefficient, a2 is the weight coefficient of vibration amplitude, y3 is the third vibration risk conversion coefficient, and a3 is the weight coefficient of vibration frequency.

6. The method for early warning of turbine bearing vibration fault based on machine learning according to claim 5, characterized in that: Before generating a vibration fault warning model for the corresponding monitoring point based on multiple vibration characteristics and vibration risk values, it includes: Preset the target vibration risk value of the preset fault type for each monitoring point; If the vibration risk value of the vibration characteristic of the current monitoring point is greater than the target vibration risk value, the vibration characteristic corresponding to the current monitoring point is determined to be a fault characteristic, and a preset fault characteristic of a preset fault type corresponding to the target vibration risk value is screened out; Perform similarity analysis on the fault feature of the current monitoring point and the preset fault feature of the screened preset fault type, set the preset fault type corresponding to the preset fault feature with the greatest similarity as the fault type of the fault feature corresponding to the current monitoring point, and set the correction coefficient according to the preset fault evaluation value of the preset fault feature with the greatest similarity to the fault feature; Determine the vibration influence range according to the fault characteristics and fault type of the current monitoring point, and determine the affected monitoring point of the current monitoring point in the monitoring point reference map based on the vibration influence range; Generate an influence coefficient based on the fault characteristics of the current monitoring point and the degree of influence of the fault type on each monitoring point; The vibration risk value of the corresponding fault feature is corrected according to the correction coefficient and the influence coefficient, and a vibration fault early warning model of the corresponding monitoring point is constructed according to the corrected vibration risk value of the fault feature and the vibration risk values ​​of other vibration features.

7. The method for early warning of turbine bearing vibration fault based on machine learning according to claim 6, characterized in that: The correction coefficient is set according to the preset fault evaluation value of the preset fault feature with the greatest similarity to the fault feature, including: Presetting a first preset fault evaluation value interval, a second preset fault evaluation value interval, a third preset fault evaluation value interval, and a fourth preset fault evaluation value interval; When the preset fault evaluation value is within the first preset fault evaluation value interval, setting the first preset correction coefficient as the correction coefficient; When the preset fault evaluation value is within the second preset fault evaluation value interval, setting the second preset correction coefficient as the correction coefficient; When the preset fault evaluation value is within the third preset fault evaluation value interval, setting the third preset correction coefficient as the correction coefficient; When the preset fault evaluation value is within a fourth preset fault evaluation value interval, the fourth preset correction coefficient is set as the correction coefficient.

8. The method for early warning of turbine bearing vibration fault based on machine learning according to claim 7, characterized in that: According to the corrected vibration risk value of the fault feature and the vibration risk values ​​of other vibration features, a vibration fault early warning model of the corresponding monitoring point is constructed, including: Based on the vibration characteristics and fault characteristics of all historical monitoring cycles of each monitoring point as training input data sets, the vibration risk values ​​corresponding to the vibration characteristics and the corrected vibration risk values ​​corresponding to the fault characteristics are used as training output data sets; The neural network is trained according to the training input data set and training output data of each monitoring point to obtain the vibration fault early warning model of the corresponding monitoring point.

9. The method for early warning of turbine bearing vibration fault based on machine learning according to claim 8, characterized in that: Generate corresponding warning signals based on the predicted vibration risk value and configure corresponding operation and maintenance strategies, including: Obtain the real-time vibration characteristics of each monitoring point in the current monitoring period, and input them into the vibration fault early warning model of the corresponding monitoring point to obtain the predicted vibration risk value of the real-time vibration characteristics; The predicted vibration risk value of the real-time vibration characteristic of each monitoring point is compared with the target vibration risk value of each preset fault type of the corresponding monitoring point. If the predicted vibration risk values ​​are all smaller than all the target vibration risk values ​​of the corresponding monitoring point, and the difference between the predicted vibration risk value and the predicted vibration risk value of all the target vibration risk values ​​is greater than the preset difference, no warning signal is sent; If the predicted vibration risk values ​​are all less than the total target vibration risk values ​​of the corresponding monitoring points, and there is a predicted vibration risk value difference less than the preset difference, a first-level warning signal is sent; If the predicted vibration risk value is greater than one or more target vibration risk values ​​of the corresponding monitoring point, a secondary warning signal is sent, and the preset fault type corresponding to the target vibration risk value is screened out, the real-time vibration feature is set as the real-time fault feature, and a similarity analysis is performed with the preset fault feature of the screened preset fault type, and the preset operation and maintenance strategy of the preset fault feature with the greatest similarity is set as the operation and maintenance strategy of the current real-time fault feature.

Images