Intelligent Monitoring Method for Large-Scale Marine Gearboxes Based on Multi-Source Information Fusion

By using multi-source information fusion and adaptive noise separation technology, the problem of fault identification in offshore wind turbine gearboxes under complex environments has been solved, achieving effective separation of environmental interference and accurate reconstruction of fault signals, thus improving the reliability and accuracy of fault diagnosis.

CN123109966APending Publication Date: 2026-11-17DELIJIA TRANSMISSION TECH (JIANGSU CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611323579.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-28
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Offshore megawatt wind turbine gearboxes are difficult to accurately identify faults in complex environments. Existing technologies cannot distinguish between environmental disturbances and mechanical vibrations, leading to the loss of subtle fault characteristics, false alarms, or missed alarms. Furthermore, the fixed transmission model has insufficient adaptability, and interference at local measurement points affects the overall diagnosis.

Method used

By fusing multi-source information, data such as multi-point vibration of the gearbox, reference vibration of the tower and nacelle, wind speed, wave height, and lubricating oil temperature are collected. Time-series alignment and adaptive noise separation are performed to establish normalized path transmission relationships. Weighted constraint reconstruction is carried out in combination with lubricating oil temperature and planetary carrier angle position to generate fault characteristic indicators.

Benefits of technology

It achieves effective separation of environmental interference, accurate reconstruction of fault signals, stable identification of fault location and type, reduces the risk of false alarms and missed alarms, and improves the reliability of fault diagnosis.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123109966A_ABST
    Figure CN123109966A_ABST
Patent Text Reader

Abstract

This invention discloses an intelligent monitoring method for offshore megawatt gearboxes based on multi-source information fusion, belonging to the field of wind power equipment condition monitoring technology. The method includes the following steps: collecting vibration data from multiple measurement points of the gearbox, as well as reference vibration data from the tower and nacelle; combining this data with operating information such as speed, torque, lubricating oil temperature, and planetary carrier angle position to separate environmental disturbances from mechanical vibrations; establishing vibration path transmission relationships based on different operating conditions; evaluating the reliability of multi-measurement point data and performing differential fusion to reconstruct the internal fault excitation signal of the gearbox; further extracting fault features and performing continuous window joint judgment to identify the fault location, fault type, and fault severity of the gearbox. This invention can improve the accuracy and stability of gearbox monitoring under complex offshore environments and varying operating conditions, and reduce the probability of false alarms and missed alarms.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of wind power equipment condition monitoring technology, specifically to an intelligent monitoring method for offshore megawatt gearboxes based on multi-source information fusion. Background Technology

[0002] Offshore megawatt-scale wind turbine gearboxes operate under high loads, variable speeds, and frequent load fluctuations for extended periods. They are also affected by external factors such as strong winds, waves, tower vibrations, and nacelle structural vibrations. Fault vibrations generated by internal gears and bearings are prone to attenuation, modulation, and aliasing during propagation. Due to the complex gearbox structure, long transmission path, and continuously changing operating conditions, vibration signals collected from different measuring points often exhibit significant differences. The superposition of environmental disturbances and mechanical fault components results in weak early fault characteristics and poor stability.

[0003] Existing technologies typically place vibration sensors near the gearbox housing or bearing housing. They extract features such as vibration amplitude, meshing frequency, fault frequency, and its side frequencies through time-domain analysis, spectrum analysis, envelope demodulation, order analysis, or time-frequency analysis. These features are then combined with operating parameters such as temperature, speed, and torque for threshold judgment. Some solutions also utilize multiple vibration measurement points for feature comparison, or introduce reference vibration signals from locations such as the tower or nacelle. By employing filtering, adaptive noise reduction, correlation analysis, and data fusion, the impact of environmental vibration on diagnostic results is reduced, thereby identifying abnormal conditions in gears, bearings, and planetary transmission components.

[0004] However, existing technologies still have the following problems: First, changes in wind speed and waves at sea can cause vibrations in the tower and nacelle structures, which are superimposed on the mechanical vibrations inside the gearbox. Conventional filtering methods cannot accurately distinguish between environmental disturbances and gear meshing components, easily leading to the loss of subtle fault characteristics. Second, the unit is in a state of variable speed, variable load, and temperature changes for a long time. The vibration transmission characteristics between different measuring points and internal excitation sources will change with operating conditions and planetary carrier positions, making it difficult for fixed transmission models to be consistently applicable. Third, different measuring points are affected by environmental noise to varying degrees. Existing multi-measuring-point fusion methods usually lack differentiated processing for data quality and reliability. When local measuring points are disturbed or abnormal, the overall diagnostic results are easily affected. Fourth, short-term characteristic changes caused by environmental disturbances have certain similarities to real mechanical faults. Using a single feature or a single time window for threshold judgment can easily lead to false alarms or missed alarms, making it difficult to reliably identify fault location, fault type, and fault severity.

[0005] Therefore, this invention provides an intelligent monitoring method for large-megawatt marine gearboxes based on multi-source information fusion. Summary of the Invention

[0006] The purpose of this invention is to provide an intelligent monitoring method for large-megawatt marine gearboxes based on multi-source information fusion, so as to solve the existing problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a smart monitoring method for large-megawatt marine gearboxes based on multi-source information fusion, comprising the following steps: S1. Collect multi-point vibration of gearbox, reference vibration of tower, external reference vibration of nacelle, wind speed, significant wave height, lubricating oil temperature, input torque, speed and planetary carrier angle position, and perform time alignment and angle domain resampling. S2. A mechanical component protection range is formed based on the rotational speed, number of gear teeth, and transmission relationship; environmental noise is estimated based on the wind speed, significant wave height, tower reference vibration, and nacelle external reference vibration; adaptive noise separation is performed outside the mechanical component protection range to obtain noise-reduced vibration and noise reliability. S3. Based on the amplitude ratio and phase difference of the meshing components in the noise reduction vibration, and combined with the lubricating oil temperature, the input torque and the planetary carrier angle position, establish a normalized path transmission relationship; S4. Determine the reconstruction weights based on the noise confidence level, and perform weighted constraint reconstruction of the noise-reduced vibration based on the normalized path transit relationship and the reconstruction weights to obtain the reconstructed signal; S5. Based on the order characteristics, modulation side frequency characteristics, and vibration and shock characteristics of the reconstructed signal, generate fault characteristic indicators and determine the abnormal state; if the abnormal state is not determined, update the normalized path transmission relationship; if the abnormal state is determined, stop updating, and determine the fault type, fault location, and fault degree according to the correspondence between the subsequent reconstructed signal and the theoretical fault order of the component to be detected.

[0008] A further improvement of the present invention is that the specific step S1 includes: S11. The multi-point vibration of the gearbox is collected by vibration sensors set at different vibration transmission positions of the gearbox, and the reference vibration of the tower and the reference vibration of the outside of the nacelle are collected by vibration sensors set at positions outside the tower and the nacelle, respectively. S12. Collect the wind speed, the significant wave height, the lubricating oil temperature, the input torque, the rotational speed, and the planetary carrier angle position through the corresponding monitoring device, and obtain the number of gear teeth and the transmission relationship from the gearbox design parameters; S13. Configure a unified time stamp for the collected data, use data containing at least one complete rotation cycle as the processing time window, and perform time correspondence on data from different sources according to the unified time stamp; S14. Based on the rotational speed and the planetary carrier angular position, resample the multi-point vibration of the gearbox, the reference vibration of the tower, and the external reference vibration of the nacelle to the same angular position.

[0009] A further improvement of this invention is that the protection range of the mechanical components, environmental noise estimation, adaptive noise separation, and noise confidence are determined in the following manner: The gear meshing composition is determined based on the rotational speed, the number of gear teeth, and the transmission relationship; the theoretical fault order of the component under test is determined based on the structural parameters of the component under test. The gear meshing components, the theoretical fault order of the component to be tested, and the corresponding adjacent intervals are determined as the mechanical component protection range; The environmental conditions are determined based on the wind speed and the significant wave height, and the environmentally relevant components in the multi-point vibration of the gearbox are determined based on the environmental conditions, the tower reference vibration outside the mechanical component protection range, and the external reference vibration of the nacelle. The adaptive noise separation is performed only on environmentally relevant components outside the protection range of the mechanical components. Based on the environmental noise estimation results, the correlation between the environmental components and the multi-point vibration of the gearbox, and the remaining components after adaptive noise separation, the noise confidence level corresponding to each measurement point and each analysis order band is determined. The higher the noise confidence level, the greater the degree of influence from environmental noise.

[0010] A further improvement of this invention is that the method for establishing and updating the normalized path transmission relationship includes: Extract the homogeneous meshing components belonging to the same gear meshing source from the noise reduction vibration at different measuring points, and select one of the measuring points as the comparison measuring point; Determine the amplitude ratio and phase difference of the co-electrode components at other measuring points relative to the co-electrode components at the comparison measuring point. In the normalized path propagation relationship, the first... The relative transfer coefficient of a measurement point to a comparison measurement point is determined according to the following formula: in, For lubricating oil temperature Input torque and planetary carriage angle position The corresponding number The relative transfer coefficient of each measuring point with respect to the comparison measuring point; and The first The amplitude of the homogeneous meshing component in each measuring point and the comparison measuring point; and These are the corresponding phases; The unit is the imaginary number; the normalized path transmission relationship is formed based on the relative transmission coefficients corresponding to each measurement point; Using the relative transfer coefficients corresponding to the comparison measurement points as the normalization benchmark, the normalized path transfer relationship is established based on the relative transfer coefficients in the normal state data. Establish a correspondence between the normalized path transmission relationship and the corresponding lubricating oil temperature, input torque, and planetary carrier angle position; The normalized path propagation relationship is updated only using the processing time window that is not judged as abnormal and the data is valid.

[0011] A further improvement of this invention is that the reconstruction weights and the weighted constraint reconstruction process specifically include: The noise confidence level corresponding to each measurement point and each analysis order band is associated with the corresponding noise reduction vibration. The reconstruction weight is determined according to the relationship that the greater the noise confidence level, the smaller the reconstruction weight. A weight matrix is ​​constructed based on each reconstruction weight. A transfer matrix is ​​constructed based on the normalized path transfer relationship, and the noise-reduced vibrations at each measuring point are used as observation vectors. The weighted constraint reconstruction is then performed according to the following formula: in, The reconstructed signal, Signal to be rebuilt The observation vector is composed of the noise-reduced vibrations at each measuring point. The transfer matrix is ​​composed of the normalized path transfer relations. The weight matrix is ​​composed of the reconstructed weights. This is a constraint matrix used to limit abrupt changes in the signal to be reconstructed. These are constraint coefficients used to adjust the degree of effect of the constraint matrix.

[0012] A further improvement of this invention is that the fault characteristic index generation step includes: S51. Determine the theoretical fault order of the component under test based on the number of gear teeth, the transmission relationship, and the structural parameters of the component under test. S52. Extract the order amplitude corresponding to the theoretical fault order of the component to be detected and its change relative to the normal state data from the reconstructed signal to form the order feature; S53. Extract the side frequency interval, side frequency amplitude and side frequency energy relationship around the gear meshing component to form the modulation side frequency characteristics; S54. Extract the amplitude, repetition relationship, and duration of the vibration and impact to form the vibration and impact characteristics; S55. Generate the fault characteristic index based on the degree of change of the order characteristics, the modulation side frequency characteristics, and the vibration and shock characteristics relative to the corresponding normal state data.

[0013] A further improvement of this invention is that the abnormal state judgment process includes: Select data from the normal state data that corresponds to the current rotational speed, input torque, and lubricating oil temperature, and determine the abnormality judgment threshold based on the selected data; An abnormal state is determined when at least two of the following features—order feature, modulation side frequency feature, and vibration and shock feature—reach the corresponding anomaly determination threshold within a consecutive preset number of processing time windows. When the abnormal state is not determined, the normalized path transmission relationship used in subsequent processing time windows is updated using the homogeneous meshing components in the current processing time window. When an abnormal state is determined, the normalized path transit relationship is stopped from being updated, and the noise reduction vibration in the subsequent processing time window is processed using the last valid normalized path transit relationship before the abnormal state was determined. The fault type and fault location are determined based on the correspondence between the subsequent reconstructed signal and the theoretical fault order of the component under test, and the fault degree is determined based on the degree of deviation of the fault characteristic index from the normal state data.

[0014] A further improvement of this invention is that the protection range of the mechanical component and the environmentally relevant component are determined in the following manner: Based on the order resolution of the processing time window and the order drift caused by the speed fluctuation, the adjacent intervals corresponding to the gear meshing components and the theoretical fault order of the component to be detected are determined. Vibration components within the mechanical component protection range of the tower reference vibration and the nacelle external reference vibration are prohibited from participating in the update of the adaptive noise separation; Outside the protection range of the mechanical components, determine the degree of coherence between the tower reference vibration and the external reference vibration of the nacelle and the multi-point vibration of the gearbox at the corresponding measuring points; Only vibrational components whose coherence reaches a preset coherence threshold are identified as environmentally relevant components.

[0015] A further improvement of this invention is that the specific process for establishing the correspondence between the normalized path transmission relationship and the operating conditions includes: The lubricating oil temperature is divided into multiple lubricating oil temperature ranges, the input torque is divided into multiple input torque ranges, and the planetary carrier angle position is divided into multiple planetary carrier angle position ranges. The relative transfer coefficients obtained from different processing time windows are assigned to the corresponding lubricating oil temperature range, input torque range, and planetary carrier angle position range. When the current processing time window is not determined to be abnormal and the data is valid, the normalized path transmission relationship of the corresponding interval is updated using the relative transmission coefficient of the current processing time window. When the current operating condition is between adjacent intervals, interpolation is performed according to the normalized path transmission relationship corresponding to the adjacent intervals. When the current operating condition is outside the existing range, the normalized path transmission relationship corresponding to the boundary range closest to the current operating condition is selected.

[0016] A further improvement of this invention is that when the noise confidence level of any measurement point or any analysis order band reaches a preset noise confidence level threshold, the corresponding measurement point or the corresponding analysis order band is marked as invalid, and the data marked as invalid is prohibited from participating in the weighted constraint reconstruction. A transfer matrix is ​​constructed based on the normalized path transfer relationship corresponding to the valid measurement points; The reconstructed signal for the current processing time window is output only when the number of valid measurement points is not less than the number of candidate excitation sources to be reconstructed, and the transfer matrix is ​​able to distinguish each candidate excitation source. When the output conditions of the reconstructed signal are not met, the reconstructed signal of the current processing time window is not output, and the normalized path transmission relationship is not updated using the current processing time window.

[0017] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention first solves the problem of the superposition of vibrations from sea waves and engine room structure with gearbox fault vibrations by integrating multi-point vibration of the gearbox, reference vibration of the tower, and external reference vibration of the engine room, and by protecting the mechanical characteristic components. It also solves the problem that conventional noise reduction easily weakens the fault characteristics. This invention achieves effective separation of environmental interference and reliable preservation of weak fault characteristics.

[0018] 2. By combining lubricating oil temperature, input torque and planetary carrier angle position to establish normalized path transmission relationships for different operating conditions, the problem of changes in transmission characteristics from internal excitation source to each vibration measurement point under different operating conditions and insufficient adaptability of fixed transmission model is solved, thus realizing accurate reconstruction of internal fault excitation signals and stable identification of fault locations.

[0019] 3. By assigning differentiated weights to multi-point data based on noise reliability, and combining constraint reconstruction and continuous judgment of multiple fault characteristics, the problem of strong interference or abnormality at local measurement points easily affecting the overall diagnostic results, as well as the problem of false alarms and missed alarms easily generated by single feature judgment, is solved, thus realizing reliable judgment of fault type, fault location and fault degree. Attached Figure Description

[0020] Figure 1 This is a flowchart of the intelligent monitoring method for large-megawatt marine gearboxes based on multi-source information fusion, as described in this invention. Figure 2 This is a comparison chart showing the effect of gearbox fault feature extraction under different signal processing methods of the present invention. Detailed Implementation

[0021] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations thereof. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0022] The term "and / or" is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can represent three cases: A exists alone, A and B exist simultaneously, and B exists alone.

[0023] This application belongs to the field of condition monitoring and fault diagnosis technology for large-megawatt gearboxes in offshore wind power, and is applicable to offshore wind turbine generator sets with planetary gear transmission mechanisms, especially to online monitoring of the sun gear, planet gears, gear rings, planet carrier support components and related bearings in the inner planetary gearbox.

[0024] In this embodiment, multiple vibration sensors are installed at different vibration transmission locations within the gearbox housing to collect multi-point vibrations of the gearbox; reference vibration sensors are installed at locations on the tower and the external structure of the nacelle to collect reference vibrations of the tower and the external structure of the nacelle, respectively. The vibration sensors can be acceleration sensors.

[0025] Wind speed is obtained from the nacelle wind measurement device or wind field wind measurement device; significant wave height is obtained from the marine monitoring device or wind field environmental monitoring server; lubricating oil temperature is obtained from the temperature sensor of the gearbox lubrication system; input torque and speed are obtained from the transmission chain monitoring device or the unit control system; and planetary carrier angle position is obtained from the encoder or speed measuring device with reference pulse.

[0026] The number of gear teeth, the number of planetary gears, the transmission ratio of each stage, and the structural parameters of the component to be tested are read from the gearbox design documents, equipment parameter database, or unit control system configuration file.

[0027] The above data is sent to the edge computing device on the nacelle side for processing, or it can be sent to the wind farm server for processing. A complete monitoring process includes: multi-source data acquisition and time sequence alignment, angle domain resampling, mechanical component protection and environmental noise separation, establishment of normalized path transfer relationship, weighted constraint reconstruction, abnormal state judgment, and output of fault type, fault location and fault degree.

[0028] Example 1 This embodiment uses the continuous monitoring of the planetary level within the gearbox of an offshore wind turbine during grid-connected operation as an example. The specific steps are as follows: Figure 1 As shown.

[0029] In step S1, the edge computing device acquires multi-point vibration of the gearbox, reference vibration of the tower, external reference vibration of the nacelle, wind speed, significant wave height, lubricating oil temperature, input torque, rotational speed and planetary carrier angular position, and reads the number of gear teeth and transmission relationship.

[0030] All data are configured with a unified time stamp, and vibration data are corresponding to their sampling time. For low sampling frequency data such as wind speed, significant wave height, lubricating oil temperature, and input torque, the average or median value within the processing time window can be used as the corresponding value for the current window. Interpolation can also be performed when the interval between adjacent valid sample values ​​does not exceed the allowable interval.

[0031] The edge computing device generates a continuous angular sequence based on the rotational speed and planetary carrier angular position, and establishes target angular positions according to preset angular intervals. For each target angular position, the corresponding time is determined in the original time sampling sequence, and the gearbox multi-point vibration, tower reference vibration, and nacelle external reference vibration are interpolated to the same angular position.

[0032] After completing the angular domain resampling, the vibration components related to rotational motion are represented in order form to reduce frequency drift caused by rotational speed fluctuations and to provide a unified angular reference for subsequent order selection and phase comparison.

[0033] If the processing time window cannot cover a complete rotation cycle, or if the rotation speed or planetary carrier angle position is invalid, the current window will be marked as invalid and will not proceed to the next step.

[0034] In step S2, the edge computing device determines the meshing order of each gear according to the rotational speed, the number of gear teeth and the transmission relationship, and determines the theoretical fault order of the component to be tested according to the structural parameters and relative motion relationship of the component to be tested.

[0035] Based on the order resolution of the processing time window, the order drift caused by the speed fluctuation within the window, and the allowable error of the transmission parameters, adjacent intervals are set on both sides of the gear meshing order and the theoretical fault order of the component to be tested, and the union of each adjacent interval is used as the mechanical component protection range.

[0036] The tower reference vibration and the external reference vibration of the nacelle within the mechanical component protection range do not participate in the parameter update of adaptive noise separation, so as to avoid the simultaneous weakening of gear meshing components or fault-related components.

[0037] The edge computing device determines the current environmental state based on wind speed and significant wave height, and calculates the coherence between the tower reference vibration, the nacelle external reference vibration, and the vibration measurement points of each gearbox, outside the mechanical component protection range. Only vibration components with a coherence level reaching a preset coherence threshold are identified as environmentally relevant components.

[0038] Adaptive noise estimation is performed on the environmentally relevant components, and the environmental noise estimation results are separated from the corresponding gearbox vibration to obtain the noise-reduced vibration at each measuring point. The adaptive noise estimation can be implemented using normalized least mean square adaptive filtering or recursive least squares adaptive filtering.

[0039] The edge computing device forms the noise confidence level corresponding to each measurement point and each analysis order band based on the proportion of environmental noise estimation results in the original vibration, the degree of coherence between the reference vibration and the gearbox vibration, and the magnitude of the remaining environmental components after separation.

[0040] Noise confidence is a dimensionless quantity. The higher the noise confidence, the greater the degree to which the corresponding measurement point or analysis order band is affected by environmental noise. Step S2 outputs denoised vibration and noise confidence, where denoised vibration is used to establish path transmission relationships and perform signal reconstruction, and noise confidence is used to determine reconstruction weights.

[0041] In step S3, the edge computing device extracts homogeneous meshing components belonging to the same gear meshing source from the noise reduction vibrations at different measuring points. The homogeneous meshing components can be obtained through order filtering, synchronous demodulation, or complex order spectrum, and the amplitude and phase of the corresponding components at each measuring point are obtained respectively.

[0042] One comparison measurement point is selected from multiple gearbox vibration measurement points. The comparison measurement point preferably meets the conditions of stable amplitude of co-source meshing, good phase repeatability, and low noise reliability. The same planetary carrier angle position and the starting angle of the processing time window are used as the reference for the phase of each measurement point.

[0043] No. The relative transfer coefficient of a measurement point to a comparison measurement point is determined according to the following formula: in, For the first The relative transfer coefficient of a measuring point relative to a comparison measuring point is a dimensionless complex number. This refers to the temperature of the lubricating oil, expressed in degrees Celsius. Input torque, measured in Newton-meters; This refers to the position of the planetary carrier angle; and The first The amplitude of the homogeneous meshing components at each measuring point and the comparison measuring point; and These are the corresponding phases; It is the imaginary unit.

[0044] The relative transfer coefficients of the comparison measurement points are set to unit amplitude and zero phase. The relative transfer coefficients of each measurement point are combined according to a fixed measurement point order to form a normalized path transfer relationship corresponding to the meshing source.

[0045] By establishing the transmission relationship using the amplitude ratio and phase difference of the components from the same source meshing, the influence of the absolute amplitude of the excitation and the difference in the installation position of the measuring point on the path relationship can be reduced.

[0046] When the amplitude of the homogeneous meshing component in the comparison measurement point is lower than the effective amplitude lower limit, the relative transfer coefficient of the current processing time window is not calculated, and the result is marked as invalid to avoid the amplitude ratio being abnormal due to the denominator being close to zero.

[0047] The edge computing device divides the lubricating oil temperature, input torque, and planetary carrier angle position into multiple intervals, and assigns the relative transmission coefficient to the corresponding operating condition interval based on the lubricating oil temperature, input torque, and planetary carrier angle position corresponding to each processing time window.

[0048] Multiple valid relative transfer coefficients within the same operating condition interval can form a normalized path transfer relationship for that interval through sliding statistics or recursive statistics. Updating the corresponding interval is only permitted if the current processing time window is not deemed abnormal and the data is valid.

[0049] When the current operating condition lies between adjacent intervals, the normalized path propagation relationship corresponding to the adjacent intervals is interpolated based on the distance between the current operating condition and the adjacent intervals. When planetary carrier angle positions are involved, the adjacency relationship between 0 degrees and 360 degrees is handled according to the cyclic angle.

[0050] When the current operating condition exceeds the existing range, the normalized path propagation relationship corresponding to the boundary range closest to the current operating condition is invoked, and extrapolation beyond the existing data range is not performed.

[0051] In step S4, the edge computing device maps the noise confidence level corresponding to each measurement point and each analysis order band to a reconstruction weight. The mapping relationship is monotonically decreasing, that is, the higher the noise confidence level, the smaller the corresponding reconstruction weight.

[0052] The edge computing device forms an observation vector by arranging the noise-reduced vibration of the effective measurement points in a predetermined order, forms a transfer matrix by arranging the normalized path transfer relationship corresponding to the current operating condition, and forms a weight matrix by arranging the reconstruction weights corresponding to each effective data.

[0053] Subsequently, the weighted residual between the predicted vibration and the observed vector formed after the transfer matrix is ​​applied to the signal to be reconstructed is used as the observation constraint, and a change constraint is set to limit the abnormal abrupt changes in the signal to be reconstructed. The reconstructed signal is obtained by solving the signal to be reconstructed that simultaneously satisfies the observation constraint and the change constraint.

[0054] When the noise confidence level reaches or exceeds the invalidation threshold, the corresponding measurement point or analysis order band is marked as invalid and removed from the observation vector, transfer matrix and weight matrix. Zero values ​​are not used to replace invalid data.

[0055] Before reconstruction, the edge computing device checks the number of valid measurement points and the distinguishability of the transfer matrix. If the number of valid measurement points is not less than the number of candidate excitation sources to be reconstructed, and the transfer matrix can distinguish between different candidate excitation sources, the device outputs the reconstruction signal for the current processing time window.

[0056] When the number of effective measurement points is insufficient, the transfer matrix is ​​unidentifiable, or the solution fails, the reconstruction signal of the current processing time window will not be output, and the normalized path transfer relationship will not be updated using the current window.

[0057] In step S5, the edge computing device determines the theoretical fault order of the component under test based on the number of gear teeth, transmission relationship and structural parameters of the component under test, and obtains the order characteristics, modulation side frequency characteristics and vibration and shock characteristics from the reconstructed signal.

[0058] The order characteristics include the theoretical fault order of the component under test and the order amplitude of its harmonics, as well as the amount of change of the order amplitude relative to the normal state data.

[0059] Modulation sideband characteristics include the sideband spacing, sideband amplitude, and sideband energy relationship around the gear meshing components. Vibration and shock characteristics include shock amplitude, repetition relationship of adjacent shocks, and shock duration.

[0060] The edge computing device calculates the degree of deviation of the three types of features from the corresponding normal state data, and generates fault feature indicators based on the pre-calibrated fusion relationship.

[0061] Normal operating conditions are categorized by speed, input torque, and lubricating oil temperature. Abnormal condition thresholds are derived from the normal operating data corresponding to the current speed, input torque, and lubricating oil temperature, and can be determined using quantiles, mean, dispersion, or the upper envelope of the normal data.

[0062] An abnormal state is determined when at least two of the following features—order characteristics, modulation side frequency characteristics, and vibration and shock characteristics—meet their respective anomaly criteria within a consecutive preset number of effective processing time windows. If only one feature reaches a threshold, or if multiple features only briefly reach a threshold within a single window, the current state is designated as an observation state.

[0063] When the abnormal state is not determined, the normalized path transmission relationship of the corresponding working condition interval can be updated using the valid homogeneous meshing components in the current processing time window.

[0064] Once an abnormal state is identified, the updating of the normalized path transitivity relationship using the current and subsequent processing time windows is stopped. The last valid normalized path transitivity relationship before the abnormal state was identified is used to process subsequent noise reduction and vibration analysis. Stopping the updating of the normalized path transitivity relationship does not stop data acquisition, environmental noise processing, signal reconstruction, and fault diagnosis.

[0065] The edge computing device determines the fault location based on the correspondence between the abnormal candidate excitation sources in the subsequent reconstructed signal and the theoretical fault order of the component to be detected. It determines the fault type by combining the modulation side frequency interval, impulse repetition relationship and corresponding component structure relationship, and determines the fault degree based on the degree of deviation and duration of the fault characteristic index relative to the normal state data.

[0066] The output results include abnormal status, fault type, fault location, fault severity, reconstructed signal, abnormal order, noise confidence level, and status of effective measurement points.

[0067] Example 2 This embodiment illustrates the specific implementation method for strong wind and wave environments, partial measurement point failure, and weighted constraint reconstruction. Unless otherwise specified, the processing procedures are the same as in Embodiment 1.

[0068] When both wind speed and significant wave height increase simultaneously, the tower reference vibration and the nacelle external reference vibration may intensify across multiple analysis order bands. The edge computing device still first establishes a mechanical component protection zone, within which the reference vibration does not participate in the parameter updates for adaptive noise separation.

[0069] When one reference vibration channel fails, its use is discontinued; if another reference channel is valid and meets the coherence requirements, the remaining reference channels are used for environmental noise estimation. When all reference channels are invalid, the corresponding analysis order band is marked as insufficient environmental reference, and the corresponding noise confidence level is increased or it is marked as invalid.

[0070] When some gearbox vibration measurement points or some analysis order bands become invalid due to excessively high noise confidence, the corresponding data are simultaneously removed from the observation vector, transfer matrix, and weight matrix.

[0071] For the remaining valid data, weighted constraint reconstruction is performed according to the following formula: in, For signal reconstruction; The signals to be reconstructed corresponding to each candidate excitation source; The observation vector consists of the noise-reduced vibrations of the effective measurement points and the effective analysis order band; It is a transfer matrix composed of normalized path transfer relationships corresponding to the current operating conditions; The weight matrix consists of the reconstruction weights corresponding to each valid data point; This is a constraint matrix used to limit anomalous changes in the signal to be reconstructed; These are non-negative constraint coefficients; This represents the L2 norm.

[0072] The first term is the weighted observation residual, used to ensure that the predicted vibration formed after the reconstructed signal is transmitted through the path corresponds to the denoised vibration of the effective measurement points. Data with greater environmental noise influence have smaller reconstruction weights, thereby reducing their impact on the reconstruction results.

[0073] The second term is the variation constraint, used to limit unsupported abrupt changes that occur when some measurement points are missing, there is residual noise, or the transfer matrix conditions are poor. The constraint matrix can be the first-order difference matrix of adjacent angle sampling points.

[0074] The constraint coefficient is determined based on the reconstruction residual under normal conditions, the sensor noise level, and the shortest fault impact duration angle that needs to be retained. If the constraint coefficient is too small, the residual noise may be amplified by the transfer matrix; if the constraint coefficient is too large, it may limit the local fault impact in the reconstructed signal.

[0075] Before reconstruction, check the number of remaining valid measurement points, the rank of the transfer matrix, and the singular values ​​or condition number. If the number of valid measurement points is not less than the number of candidate excitation sources, and the transfer matrix can distinguish between different candidate excitation sources, output the reconstruction signal; otherwise, do not output the reconstruction signal for the current window.

[0076] When the current lubricating oil temperature, input torque, or planetary carrier angle position is between adjacent operating ranges, the normalized path transmission relationship between adjacent ranges is interpolated. When phase interpolation is involved, the phase is first continuously expanded before interpolation is performed to avoid phase jumps near positive and negative π.

[0077] After the failed measurement point is recovered, its time synchronization status, amplitude range, noise reliability, and phase continuity of co-current meshing are checked. If the valid conditions are met for three consecutive processing time windows, it is allowed to participate in the reconstruction.

[0078] Once an abnormal state has been identified, even if environmental conditions recover or failed measuring points become valid again, the normalized path propagation relationship update will not be used to restore the data from the subsequent abnormal state. Instead, the last valid normal path relationship before the abnormal state was identified will be called to complete the subsequent monitoring.

[0079] To verify the effectiveness of the multi-source reference vibration fusion and mechanical fault feature protection method proposed in this invention for extracting weak fault features of offshore wind turbine gearboxes, this embodiment uses gearbox vibration signals as the signals to be processed and sets up three sets of comparison results: the original signal, conventional signal processing methods, and the method of this invention. The gearbox fault feature extraction effects under different signal processing methods are as follows: Figure 2 As shown.

[0080] During the simulation experiment, target vibration sensors were placed near the gear meshing area and bearing housing of the gearbox housing to collect target vibration signals including gear meshing vibration, bearing vibration, and fault impact components. Simultaneously, at least one reference vibration sensor was placed on the nacelle base, gearbox support structure, or adjacent non-fault vibration locations to collect environmental reference vibration signals generated by wind and wave excitation, nacelle swaying, tower-transmitted vibration, and the operation of other auxiliary equipment. The target vibration signals and reference vibration signals were collected synchronously to ensure the temporal correspondence between signals from different measuring points.

[0081] After obtaining the vibration signals from each measuring point, the target vibration signal and the reference vibration signal are first subjected to mean removal, outlier correction, and amplitude normalization. Then, based on the correlation, coherence, and frequency band energy distribution between each reference vibration signal and the target vibration signal, the fusion weights corresponding to different reference vibration signals are determined to form a comprehensive environmental interference reference signal. The comprehensive environmental interference reference signal is then used to estimate the non-fault-related vibration components in the target vibration signal, and the estimated environmental interference components are suppressed.

[0082] During environmental interference suppression, mechanical fault characteristic protection conditions are set for the frequency regions containing gear meshing frequency, gear meshing frequency harmonics, rotational frequency, and their sidebands. When the signal component to be suppressed has a high correlation with the reference vibration signal and does not exhibit obvious periodic impact, meshing modulation, or fault sideband characteristics, the signal component is identified as an environmental interference component and weakened. When the signal component is located within the preset mechanical fault characteristic frequency band, or has periodic impact characteristics corresponding to gear or bearing faults, the suppression degree of the signal component is reduced, thereby avoiding the loss of effective fault information while reducing background vibration.

[0083] For ease of comparison, the same raw vibration data were processed separately. The first group consisted of the raw vibration signals without interference suppression; the second group was processed using conventional filtering or noise reduction methods; and the third group was processed using the multi-source reference vibration fusion and mechanical fault feature protection method of this invention. After processing, fault feature responses were extracted from each group of signals, and the amplitudes of the fault features, background interference, and the degree of distinction between the fault features and background interference were compared.

[0084] Depend on Figure 2 It can be seen that the original vibration signal contains strong broadband background vibration and aperiodic interference. The periodic impacts and corresponding characteristic frequencies generated by the gearbox fault are easily submerged by background interference, thus the fault characteristics are less prominent. After using conventional signal processing methods, some background interference is weakened, but the impact components and characteristic frequency components related to the fault also show a certain degree of attenuation, indicating that conventional methods are difficult to effectively distinguish between environmental interference components and mechanical fault components.

[0085] After processing using the method of this invention, non-fault-related background vibrations are significantly suppressed, while periodic impact components related to gear meshing, local bearing damage, or tooth surface defects are preserved, and the fault feature response is more prominent relative to background interference. Compared to the original signal and conventional processing methods, the fault feature extraction results corresponding to the method of this invention have higher feature significance, and the difference between the fault feature peak and the adjacent background components is more obvious.

[0086] Specifically, this invention does not simply reduce the overall amplitude of the vibration signal, but rather uses multiple reference measurement points to jointly estimate environmental interference and selectively processes the components to be suppressed in conjunction with mechanical fault characteristic protection conditions. Therefore, this invention can avoid misjudging weak fault impacts as noise and eliminating them, achieving a balance between suppressing environmental interference and preserving mechanical fault characteristics.

[0087] The above results demonstrate that the method proposed in this invention can effectively improve the signal-to-noise ratio and identifiability of gearbox fault characteristics under complex conditions involving wind and wave excitation, nacelle structural vibration, and vibrations from other equipment operation. This provides a reliable vibration data foundation for subsequent gearbox fault type identification, fault location determination, and fault development trend analysis.

[0088] Example 3 This embodiment specifically illustrates the parameter representation methods in Embodiments 1 and 2, wherein: The multi-point vibration of the gearbox, the reference vibration of the tower, and the external reference vibration of the nacelle are expressed using vibration acceleration, with units of meters per second squared. Wind speed is expressed in meters per second, significant wave height in meters, lubricating oil temperature in degrees Celsius, input torque in Newton-meters, and rotational speed in revolutions per minute. The planetary carrier angle position is expressed using a cyclic angle from 0 degrees to 360 degrees, and the phase is uniformly expressed in radians.

[0089] The vibration signal sampling frequency is determined based on the gearbox sweep frequency test results, the highest order to be analyzed, and anti-aliasing requirements. Specifically, the effective vibration response frequency band of the gearbox is obtained based on the sweep frequency test, and the Nyquist frequency corresponding to the sampling frequency is made higher than the upper limit of the effective vibration response frequency band. In this embodiment, the vibration signal sampling frequency is preferably 51.2 kHz; if the sweep frequency test confirms that the highest effective analysis frequency does not exceed 10 kHz, a sampling frequency of 25.6 kHz can also be used.

[0090] The vibration channels of each gearbox preferably adopt synchronous sampling, and the time synchronization error between vibration channels is no greater than 1μs. Low sampling frequency data such as wind speed, significant wave height, lubricating oil temperature, and input torque are mapped to the current processing time window according to a unified time mark, and the allowable time correspondence error is determined according to the sampling period of the corresponding data.

[0091] The processing time window covers at least one complete rotation cycle of the selected rotating reference component. In this embodiment, the processing time window preferably covers four complete rotation cycles of the planetary carrier; when four complete rotation cycles cannot meet the resolution requirements of the theoretical fault order and modulation side frequency, the processing time window is extended to eight complete rotation cycles.

[0092] The angle domain resampling interval is determined based on the highest order to be analyzed. In this embodiment, the target angle sampling interval is preferably 0.1°, meaning that 3600 angle sampling points are formed for each complete rotation cycle. The original time-domain vibration data is interpolated to the corresponding target angle position based on the rotational speed and planetary carrier angle position.

[0093] The protection range for mechanical components is determined based on the order resolution of the processing time window, the order drift caused by speed fluctuations, and the allowable error of transmission parameters. The protection interval corresponding to each gear meshing order or the theoretical fault order of the component under test has a half-width not less than the interval between three order spectral lines within the current processing time window, and not less than the maximum order drift within the window. During the initial calibration phase, a ±0.5 order interval centered on the corresponding theoretical order can be used as the initial protection interval, which is then corrected based on frequency sweep tests and normal operation data.

[0094] The coherence level is represented by the squared amplitude coherence level, ranging from 0 to 1. The preset coherence level threshold is determined based on the coherence distribution of the tower reference vibration, the external reference vibration of the nacelle, and the gearbox measuring point vibration outside the mechanical component protection range under normal operating conditions. In this embodiment, the coherence level threshold is preferably 0.7. Vibration components with a coherence level of 0.7 or higher can proceed to further judgment of environmentally relevant components; vibration components with a coherence level below 0.7 do not participate in the adaptive noise separation update.

[0095] The noise confidence level is normalized to between 0 and 1. A higher noise confidence level indicates a greater degree of influence from environmental noise on the corresponding measurement point or analysis order band. In this embodiment, the noise confidence level invalidation threshold is preferably 0.8. When the noise confidence level is less than 0.8, the corresponding reconstruction weight is determined according to the relationship that increasing noise confidence level leads to decreasing reconstruction weight; when the noise confidence level reaches or exceeds 0.8, the corresponding measurement point or analysis order band is marked as invalid.

[0096] The effective amplitude lower limit of the homogeneous meshing component in the comparison measurement points is determined based on the local noise floor of the corresponding analysis order band. In this embodiment, the amplitude of the homogeneous meshing component in the comparison measurement points should be at least 6 dB higher than the local noise floor. If it is lower than this amplitude lower limit, the relative transfer coefficient corresponding to the current processing time window is not calculated.

[0097] The lubricating oil temperature range is determined based on gearbox thermal balance test data and normal operating temperature distribution. In this embodiment, the lubricating oil temperature range width is preferably 10°C. The input torque range is determined based on different load test conditions and rated input torque. In this embodiment, the input torque range width is preferably 10% of the rated input torque. The planetary carrier angle position range width is preferably 15°, and is divided cyclically from 0° to 360°.

[0098] Each lubricating oil temperature range, input torque range, and planetary carrier angle position range combines to form an operating condition range. For each operating condition range, preferably, at least 30 processing time windows that are not judged as abnormal and have valid data are accumulated to form the initial normalized path propagation relationship for that range. If the number of valid windows is less than 30, the normalized path propagation relationship of adjacent ranges can be used for interpolation, but ranges with insufficient data are not used as independent stable ranges.

[0099] Before performing weighted constraint reconstruction, the columns of the transfer matrix are scaled and normalized. The upper limit of the condition number of the normalized transfer matrix is ​​preferably... When the condition number of the transfer matrix is ​​greater than If the rank of the transfer matrix is ​​less than the number of candidate excitation sources, it is determined that the current transfer matrix cannot distinguish between the candidate excitation sources, and the reconstructed signal of the current processing time window is not output.

[0100] The constraint coefficients are determined based on the observation residuals of the test bench's normal operating data, sensor noise levels, and the shortest duration angle of the fault impact to be retained. After normalization of the observation vector and transfer matrix, the preferred range of values ​​for the constraint coefficients is as follows: to The initial value is preferably Using data from frequency sweep tests, tests under different loads, and normal conditions in the project, the reconstruction residuals and the degree of change in the reconstruction signal under the candidate constraint coefficients were calculated respectively. Constraint coefficients that can limit unsupported abrupt changes without weakening the calibration impact components were selected.

[0101] The anomaly detection threshold is determined based on normal state data corresponding to the current speed, input torque, and lubricating oil temperature. In this embodiment, the 99th percentile of each order feature, modulation side frequency feature, and vibration and shock feature in the corresponding normal state data can be used as their respective anomaly detection thresholds. When the number of normal state samples is insufficient, no independent threshold is formed; instead, the statistical results of normal states from adjacent operating condition intervals are used.

[0102] The number of consecutive windows for determining abnormal states is preferably 3. An abnormal state is determined when at least two of the following features—order features, modulation side frequency features, and vibration and shock features—reach their respective abnormal determination thresholds for three consecutive effective processing time windows.

[0103] After an invalid measuring point is restored, if the conditions of normal time synchronization, unsaturated vibration amplitude, noise confidence level below the invalidity threshold, and continuous phase of the meshing of the same source are met for three consecutive processing time windows, the measuring point is restored to participate in the weighted constraint reconstruction.

[0104] The fault severity classification boundary should be formed by combining the degree of deviation of fault characteristic indicators from normal state data, the duration of continuous fault, and the results of project testing and subsequent maintenance confirmation. When there are no maintenance-confirmed fault samples, a fixed damage size correspondence is not set, and the fault severity is output according to the degree of deviation from the normal state.

[0105] The threshold, weight, coefficient and other settings involved in the above embodiments can be set by default according to the present invention, or can be set by those skilled in the art according to actual engineering needs.

[0106] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0107] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0108] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0109] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0110] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A method for intelligent monitoring of a marine megawatt gearbox based on multi-source information fusion, characterized in that: Includes the following steps: S1. Collect multi-point vibration of gearbox, reference vibration of tower, external reference vibration of nacelle, wind speed, significant wave height, lubricating oil temperature, input torque, speed and planetary carrier angle position, and perform time alignment and angle domain resampling. S2. A mechanical component protection range is formed based on the rotational speed, number of gear teeth, and transmission relationship; Environmental noise is estimated based on the wind speed, the significant wave height, the tower reference vibration, and the nacelle external reference vibration. Adaptive noise separation is performed outside the mechanical component protection range to obtain noise-reduced vibration and noise confidence. S3. Based on the amplitude ratio and phase difference of the meshing components in the noise reduction vibration, and combined with the lubricating oil temperature, the input torque and the planetary carrier angle position, establish a normalized path transmission relationship; S4. Determine the reconstruction weights based on the noise confidence level, and perform weighted constraint reconstruction of the noise-reduced vibration based on the normalized path transit relationship and the reconstruction weights to obtain the reconstructed signal; S5. Based on the order characteristics, modulation side frequency characteristics, and vibration and shock characteristics of the reconstructed signal, generate fault characteristic indicators and determine the abnormal state; if the abnormal state is not determined, update the normalized path transmission relationship; if the abnormal state is determined, stop updating, and determine the fault type, fault location, and fault degree according to the correspondence between the subsequent reconstructed signal and the theoretical fault order of the component to be detected.

2. The intelligent monitoring method for large-megawatt marine gearboxes based on multi-source information fusion according to claim 1, characterized in that: The specific steps in S1 include: S11. The multi-point vibration of the gearbox is collected by vibration sensors set at different vibration transmission positions of the gearbox, and the reference vibration of the tower and the reference vibration of the outside of the nacelle are collected by vibration sensors set at positions outside the tower and the nacelle, respectively. S12. Collect the wind speed, the significant wave height, the lubricating oil temperature, the input torque, the rotational speed, and the planetary carrier angle position through the corresponding monitoring device, and obtain the number of gear teeth and the transmission relationship from the gearbox design parameters; S13. Configure a unified time stamp for the collected data, use data containing at least one complete rotation cycle as the processing time window, and perform time correspondence on data from different sources according to the unified time stamp; S14. Based on the rotational speed and the planetary carrier angular position, resample the multi-point vibration of the gearbox, the reference vibration of the tower, and the external reference vibration of the nacelle to the same angular position.

3. The intelligent monitoring method for large-megawatt marine gearboxes based on multi-source information fusion according to claim 1, characterized in that: The protection range of the mechanical components, environmental noise estimation, adaptive noise separation, and noise confidence level are determined as follows: The gear meshing composition is determined based on the rotational speed, the number of gear teeth, and the transmission relationship; the theoretical fault order of the component under test is determined based on the structural parameters of the component under test. The gear meshing components, the theoretical fault order of the component to be tested, and the corresponding adjacent intervals are determined as the mechanical component protection range; The environmental conditions are determined based on the wind speed and the significant wave height, and the environmentally relevant components in the multi-point vibration of the gearbox are determined based on the environmental conditions, the tower reference vibration outside the mechanical component protection range, and the external reference vibration of the nacelle. The adaptive noise separation is performed only on environmentally relevant components outside the protection range of the mechanical components. Based on the environmental noise estimation results, the correlation between the environmental components and the multi-point vibration of the gearbox, and the remaining components after adaptive noise separation, the noise confidence level corresponding to each measurement point and each analysis order band is determined. The higher the noise confidence level, the greater the degree of influence from environmental noise.

4. The intelligent monitoring method for large-megawatt marine gearboxes based on multi-source information fusion according to claim 1, characterized in that: The methods for establishing and updating the normalized path transit relationship include: Extract the homogeneous meshing components belonging to the same gear meshing source from the noise reduction vibration at different measuring points, and select one of the measuring points as the comparison measuring point; Determine the amplitude ratio and phase difference of the co-electrode components at other measuring points relative to the co-electrode components at the comparison measuring point. In the normalized path propagation relationship, the first... The relative transfer coefficient of a measurement point to a comparison measurement point is determined according to the following formula: in, For lubricating oil temperature Input torque and planetary carriage angle position The corresponding number The relative transfer coefficient of each measuring point with respect to the comparison measuring point; and The first The amplitude of the homogeneous meshing component in each measuring point and the comparison measuring point; and These are the corresponding phases; The unit is the imaginary number; the normalized path transmission relationship is formed based on the relative transmission coefficients corresponding to each measurement point; Using the relative transfer coefficients corresponding to the comparison measurement points as the normalization benchmark, the normalized path transfer relationship is established based on the relative transfer coefficients in the normal state data. Establish a correspondence between the normalized path transmission relationship and the corresponding lubricating oil temperature, input torque, and planetary carrier angle position; The normalized path propagation relationship is updated only using the processing time window that is not judged as abnormal and the data is valid.

5. The intelligent monitoring method for large-megawatt marine gearboxes based on multi-source information fusion according to claim 1, characterized in that: The specific process of the reconstruction weights and the weighted constraint reconstruction includes: The noise confidence level corresponding to each measurement point and each analysis order band is associated with the corresponding noise reduction vibration. The reconstruction weight is determined according to the relationship that the greater the noise confidence level, the smaller the reconstruction weight. A weight matrix is ​​constructed based on each reconstruction weight. A transfer matrix is ​​constructed based on the normalized path transfer relationship, and the noise-reduced vibrations at each measuring point are used as observation vectors. The weighted constraint reconstruction is then performed according to the following formula: in, The reconstructed signal, Signal to be rebuilt The observation vector is composed of the noise-reduced vibrations at each measuring point. The transfer matrix is ​​composed of the normalized path transfer relations. The weight matrix is ​​composed of the reconstructed weights. This is a constraint matrix used to limit abrupt changes in the signal to be reconstructed. These are constraint coefficients used to adjust the degree of effect of the constraint matrix.

6. The intelligent monitoring method for large-megawatt marine gearboxes based on multi-source information fusion according to claim 1, characterized in that: The fault characteristic index generation step includes: S51. Determine the theoretical fault order of the component under test based on the number of gear teeth, the transmission relationship, and the structural parameters of the component under test. S52. Extract the order amplitude corresponding to the theoretical fault order of the component to be detected and its change relative to the normal state data from the reconstructed signal to form the order feature; S53. Extract the side frequency interval, side frequency amplitude and side frequency energy relationship around the gear meshing component to form the modulation side frequency characteristics; S54. Extract the amplitude, repetition relationship, and duration of the vibration and impact to form the vibration and impact characteristics; S55. Generate the fault characteristic index based on the degree of change of the order characteristics, the modulation side frequency characteristics, and the vibration and shock characteristics relative to the corresponding normal state data.

7. The intelligent monitoring method for large-megawatt marine gearboxes based on multi-source information fusion according to claim 1, characterized in that: The abnormal state determination process includes: Select data from the normal state data that corresponds to the current rotational speed, input torque, and lubricating oil temperature, and determine the abnormality judgment threshold based on the selected data; An abnormal state is determined when at least two of the following features—order feature, modulation side frequency feature, and vibration and shock feature—reach the corresponding anomaly determination threshold within a consecutive preset number of processing time windows. When the abnormal state is not determined, the normalized path transmission relationship used in subsequent processing time windows is updated using the homogeneous meshing components in the current processing time window. When an abnormal state is determined, the normalized path transit relationship is stopped from being updated, and the noise reduction vibration in the subsequent processing time window is processed using the last valid normalized path transit relationship before the abnormal state was determined. The fault type and fault location are determined based on the correspondence between the subsequent reconstructed signal and the theoretical fault order of the component under test, and the fault degree is determined based on the degree of deviation of the fault characteristic index from the normal state data.

8. The intelligent monitoring method for large-megawatt marine gearboxes based on multi-source information fusion according to claim 3, characterized in that: The protection range of the mechanical component and the environmentally relevant component are determined as follows: Based on the order resolution of the processing time window and the order drift caused by the speed fluctuation, the adjacent intervals corresponding to the gear meshing components and the theoretical fault order of the component to be detected are determined. Vibration components within the mechanical component protection range of the tower reference vibration and the nacelle external reference vibration are prohibited from participating in the update of the adaptive noise separation; Outside the protection range of the mechanical components, determine the degree of coherence between the tower reference vibration and the external reference vibration of the nacelle and the multi-point vibration of the gearbox at the corresponding measuring points; Only vibrational components whose coherence reaches a preset coherence threshold are identified as environmentally relevant components.

9. The intelligent monitoring method for large-megawatt marine gearboxes based on multi-source information fusion according to claim 4, characterized in that: The specific process for establishing the correspondence between the normalized path propagation relationship and the operating conditions includes: The lubricating oil temperature is divided into multiple lubricating oil temperature ranges, the input torque is divided into multiple input torque ranges, and the planetary carrier angle position is divided into multiple planetary carrier angle position ranges. The relative transfer coefficients obtained from different processing time windows are assigned to the corresponding lubricating oil temperature range, input torque range, and planetary carrier angle position range. When the current processing time window is not determined to be abnormal and the data is valid, the normalized path transmission relationship of the corresponding interval is updated using the relative transmission coefficient of the current processing time window. When the current operating condition is between adjacent intervals, interpolation is performed according to the normalized path transmission relationship corresponding to the adjacent intervals. When the current operating condition is outside the existing range, the normalized path transmission relationship corresponding to the boundary range closest to the current operating condition is selected.

10. The intelligent monitoring method for large-megawatt marine gearboxes based on multi-source information fusion according to claim 5, characterized in that: When the noise confidence level of any measurement point or any analysis order band reaches the preset noise confidence level threshold, the corresponding measurement point or the corresponding analysis order band is marked as invalid, and the data marked as invalid is prohibited from participating in the weighted constraint reconstruction. A transfer matrix is ​​constructed based on the normalized path transfer relationship corresponding to the valid measurement points; The reconstructed signal for the current processing time window is output only when the number of valid measurement points is not less than the number of candidate excitation sources to be reconstructed, and the transfer matrix is ​​able to distinguish each candidate excitation source. When the output conditions of the reconstructed signal are not met, the reconstructed signal of the current processing time window is not output, and the normalized path transmission relationship is not updated using the current processing time window.