A method for evaluating the deterioration state of wind turbine gearboxes
By performing time alignment and evidence quality analysis on multi-point measurement data of wind turbine gearboxes, and combining the projected Kalman filter model, the stability and reliability issues of gearbox degradation status assessment were resolved, and a more accurate degradation status assessment was achieved.
Patent Information
- Application Number
- CN202611269420.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-20
- Publication Date
- 2026-11-17
AI Technical Summary
In the existing technology, the stability and reliability of the evaluation results of the deterioration status of wind turbine gearboxes are insufficient, and they are easily affected by operating condition fluctuations and noise, resulting in evaluation results that do not conform to the actual operating conditions.
By performing time alignment, validity screening, and operating condition matching on multi-point status monitoring data, a health reference phase is established, directional offset and evidence quality are calculated, a common evidence cluster is constructed, repeated deterioration evidence is suppressed, and a projection Kalman filter model is used for status prediction and observation updates to obtain a continuous latent state deterioration index.
This improves the stability and reliability of gearbox degradation status evaluation, reduces operating condition differences and noise interference, and the evaluation results are more consistent with the gradual degradation law of gearboxes.
Smart Images

Figure CN123116134A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of condition assessment technology, and in particular to a method for assessing the deterioration condition of wind turbine gearboxes. Background Technology
[0002] With the continuous growth of wind power installed capacity, wind turbines are gradually developing towards larger size, higher power, and more centralized designs. As a key component in the wind turbine's transmission chain, the gearbox is subjected to alternating loads, variable speed operation, and complex environmental conditions for extended periods. Its operating status directly affects the overall power generation efficiency and operational reliability of the turbine. To reduce economic losses caused by unplanned downtime, wind farms typically deploy multiple vibration monitoring points on the gearbox. These monitoring points, combined with data on speed, load, operating status, fault alarm records, and maintenance event records, continuously monitor the gearbox's operation, generating condition monitoring data covering multiple measuring points, operating conditions, and evaluation times. By processing this data, the degree of gearbox deterioration over time can be obtained, providing a basis for equipment maintenance planning, maintenance timing selection, and operational risk assessment. Therefore, how to utilize operating data obtained from multiple monitoring points to continuously, stably, and in accordance with actual operating patterns to evaluate the overall deterioration of the gearbox has become an important research topic in the field of intelligent wind power operation and maintenance.
[0003] Existing technologies typically use data from multiple monitoring points to directly fuse evidence to obtain the overall gearbox operating status evaluation results. Since there is a large amount of data from similar sources and with consistent trends among multiple monitoring points, duplication and accumulation can easily occur during the fusion process, causing some evidence to be repeatedly amplified. This leads to the overall degradation evaluation results deviating from the actual operating state. At the same time, instantaneous monitoring data is easily affected by operating condition fluctuations, sampling noise, and occasional shocks, causing continuous evaluation results to fluctuate drastically without conforming to the actual degradation patterns of the equipment. This reduces the stability and reliability of the gearbox degradation status evaluation results. Summary of the Invention
[0004] The purpose of this invention is to address the shortcomings of existing technologies that reduce the stability and reliability of gearbox degradation status evaluation results, and to propose a method for evaluating the degradation status of wind turbine gearboxes.
[0005] To address the problems existing in the prior art, the present invention adopts the following technical solution: A method for evaluating the deterioration state of a wind turbine gearbox includes: S1. Based on the original status monitoring data of each monitoring point of the wind turbine gearbox, invalid samples are removed from the multi-point aligned samples to obtain the multi-point valid samples; the multi-point valid samples are divided to determine the health reference stage. S2. Calculate the directional shift based on the health reference stage; calculate the quality of evidence for deterioration propositions, health propositions, and uncertain propositions based on the directional shift. S3. Based on the measurement point number, characteristic physical category and frequency band attribute corresponding to each state characterization quantity, obtain the evidence credibility of each state characterization quantity; redistribute the evidence quality of each homologous evidence cluster to obtain the healthy proposition evidence quality and uncertain proposition evidence quality of each homologous evidence cluster. S4. Based on the quality of health proposition evidence and the quality of uncertain proposition evidence in each homologous evidence cluster, the evidence of instantaneous degradation of the gearbox as a whole is obtained; S5. Based on the instantaneous degradation evidence sequence, the projected Kalman filter model is used for state prediction and observation update to obtain the unconstrained hidden state degradation index at each evaluation time, and this index is determined as the evaluation result of the wind turbine gearbox degradation state.
[0006] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention establishes a data foundation consistent with the health reference stage by performing time alignment, validity screening, and operating condition matching on multi-point status monitoring data, and forms a directional offset based on health benchmarks, health scales, and deterioration response directions, thereby reducing the interference of operating condition differences, missing data, and abnormal sampling on the evaluation results from the source.
[0007] 2. This invention constructs a cluster of evidence from the same source according to the location of the measurement point, the data category, and the frequency band attribute. It combines the correlation within the cluster and the number of state characterization quantities to suppress repeated degradation evidence and transfers the suppressed part to uncertain propositions. This avoids the repeated accumulation of data from similar sources during the fusion process and improves the objectivity and credibility of the overall degradation evidence.
[0008] 3. This invention inputs the gearbox overall instantaneous degradation evidence obtained by fusing multiple layers of evidence into the projected Kalman filter model. Through state prediction, observation update, boundary constraints and adjacent time change constraints, a continuous hidden state degradation index is obtained, reducing the impact of short-term fluctuations and occasional shocks, and making the evaluation results more consistent with the gradual degradation law of the gearbox. Attached Figure Description
[0009] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings: Figure 1 This is a flowchart illustrating a method for evaluating the deterioration state of a wind turbine gearbox, as provided in an embodiment of the present invention. Detailed Implementation
[0010] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.
[0011] This embodiment provides a method for evaluating the deterioration state of wind turbine gearboxes. (See also...) Figure 1 Specifically, including: S1. Based on the original status monitoring data of each monitoring point of the wind turbine gearbox, invalid samples are removed from the multi-point aligned samples to obtain the multi-point valid samples; the multi-point valid samples are divided to determine the health reference stage. In embodiments of the present invention, obtaining a multi-point effective sample includes: Acquire raw status monitoring data, fault alarm records, and maintenance event records for each monitoring point of the wind turbine gearbox; among which, raw status monitoring data includes sampling time, measuring point number, sampling frequency, rotational speed, load, operating status, and vibration signals or status characteristics; The evaluation time is generated according to the preset sampling period, and the original state monitoring data of each monitoring point is assigned to the corresponding evaluation time according to the sampling time. The original state monitoring data belonging to the same evaluation time are matched according to the measurement point number to form a multi-measurement point aligned sample; Based on the sensor status marker, data missing ratio, rotational speed, and vibration signal amplitude, invalid samples are removed from the multi-measurement point aligned samples to obtain valid multi-measurement point samples. Invalid samples include samples with sensor status marker indicating failure, samples with data missing ratio greater than a preset missing ratio threshold, samples with rotational speed of zero, and samples with vibration signal amplitude reaching the upper limit of the sensor range.
[0012] Specifically, for each monitoring point of the wind turbine gearbox, the vibration acquisition device obtains vibration signals, sensor status markers, sampling time, measurement point number, sampling frequency, and sensor range upper limit; the wind turbine operation control device obtains speed, load, and operating status; the fault alarm database obtains alarm occurrence time, alarm cancellation time, alarm type, and alarm-associated measurement point number; and the maintenance management database obtains maintenance start time, maintenance end time, maintenance type, and maintenance component number. The sampling time is uniformly converted to the same time base, the speed is uniformly converted to revolutions per minute, and the load is uniformly converted to power or torque values. The vibration signals are stored according to sampling time, measurement point number, and sampling frequency. The sensor status markers and sensor range upper limit are associated with vibration signals of the same measurement point number. The speed, load, and operating status are associated with the vibration signals according to time. The fault alarm records and maintenance event records are associated with the vibration signals according to occurrence time, thus obtaining the original status monitoring data, fault alarm records, and maintenance event records for each monitoring point.
[0013] Specifically, for the raw state monitoring data of each monitoring point, the data of the same monitoring point number are arranged from earliest to latest according to the sampling time. The time difference between two adjacent sampling times is calculated. The median of all time differences for the same monitoring point number is taken as the data collection interval for that monitoring point. The maximum value among the data collection intervals of each monitoring point is taken as the evaluation interval. The latest time among the earliest sampling times of each monitoring point is taken as the evaluation start time. The earliest time among the latest sampling times of each monitoring point is taken as the evaluation end time. Starting from the evaluation start time, the evaluation time is increased sequentially according to the evaluation interval until the evaluation end time. A time inclusion interval is formed by the half evaluation interval before and after each evaluation time. Raw state monitoring data whose sampling time is within the same time inclusion interval are included in the evaluation time. When there are multiple raw state monitoring data of the same monitoring point within the same time inclusion interval, the raw state monitoring data with the smallest absolute value of the difference between the sampling time and the evaluation time is selected. When there are two raw state monitoring data with the same absolute value of the time difference with the evaluation time, the raw state monitoring data with the later sampling time is selected to obtain the inclusion data of each monitoring point at each evaluation time.
[0014] Specifically, for the data collected from each monitoring point at each evaluation time, the monitoring points are arranged in ascending order of their measurement point numbers. The mapping relationship between the measurement point number and the vibration acquisition device channel number in each piece of collected data is verified. Data with inconsistent measurement point numbers and vibration acquisition device channel numbers are corrected according to the measurement point number corresponding to the vibration acquisition device channel number. Data collected at the same evaluation time but with different measurement point numbers are grouped into the same data group. The sampling time, measurement point number, sampling frequency, rotational speed, load, operating status, vibration signal, sensor status mark, and sensor range upper limit in each data group are recorded in the order of the monitoring points. The evaluation time in the data group is used as the time identifier for that data group, resulting in a multi-measurement point aligned sample.
[0015] Specifically, for each multi-point aligned sample, the sensor status markers at each monitoring point are checked. Data from monitoring points where the sensor status markers do not match the status codes indicating normal acquisition in the vibration acquisition device's communication protocol are identified as sensor failure data. The number of missing fields for each monitoring point in the multi-point aligned sample is counted: sampling time, measurement point number, sampling frequency, rotational speed, load, operating status, vibration signal, sensor status marker, and sensor range upper limit. The number of missing fields is divided by the total number of fields to obtain the data missing ratio. The total number of multi-point aligned samples for the same monitoring point is then calculated. The data at each monitoring point is sorted by missing percentage from smallest to largest, and the upper quartile is taken as the threshold for the missing percentage. Data at monitoring points with a missing percentage greater than the threshold are considered missing data, and data at monitoring points with a rotational speed of zero are considered shutdown data. The maximum absolute value of each sampled value in the vibration signal is calculated, and data at monitoring points with a maximum value greater than or equal to the upper limit of the corresponding sensor range are considered amplitude saturation data. Multi-point aligned samples containing sensor failure data, missing data, shutdown data, or amplitude saturation data are deleted to obtain valid samples from multiple monitoring points.
[0016] In embodiments of the present invention, determining the health reference stage includes: The effective samples of multiple measurement points are divided according to the preset speed range and preset load range to obtain multiple operating condition ranges; When the original state monitoring data includes vibration signals, state features are extracted from effective samples of multiple measuring points in each operating condition range. If the original condition monitoring data includes condition characteristics, read the condition characteristics within each operating condition interval; The extracted or read state features are defined as state characterization quantities; wherein, the state characterization quantities include at least one of the following: root mean square value, peak value, peak-to-peak value, standard deviation, kurtosis, peak factor, impulse factor, frequency center, target frequency band energy, frequency shift order amplitude, meshing order amplitude, envelope frequency band energy, and entropy complexity index. Configure sampling time, measurement point number, feature name, feature physical category, spectral attribute, corresponding component, degradation response direction and data validity label for each state characterization quantity; The spectral attributes include at least one of frequency band range, frequency shift order, and meshing order; the degradation response direction includes a positive response direction in which the state characterization quantity increases with the degree of degradation, and a negative response direction in which the state characterization quantity decreases with the degree of degradation. Based on fault alarm records, maintenance event records, speed change rate, and load change rate, select continuous samples from the effective samples of multiple measurement points that do not contain fault alarms or maintenance events, have a speed change rate not greater than a preset speed change rate threshold, and have a load change rate not greater than a preset load change rate threshold. The time intervals corresponding to continuous samples are defined as the health reference phase.
[0017] S2. Calculate the directional shift based on the health reference stage; calculate the quality of evidence for deterioration propositions, health propositions, and uncertain propositions based on the directional shift. In embodiments of the present invention, calculating the directional offset includes: Based on the health reference stage, health reference state characteristic data are extracted from the multi-point state characteristic data of the wind turbine gearbox; For each set of health reference state characteristic data, calculate the median of all state characteristic values in the health reference state characteristic data, and determine the median as the health benchmark value of the corresponding state characterization quantity. Calculate the absolute value of the difference between each state feature value and the corresponding health baseline value in the health reference state feature data, calculate the median of all absolute values, and determine the health scale parameter of the corresponding state characterization quantity by summing the median of all absolute values and the scale correction parameter. From the multi-point state characteristic data of the wind turbine gearbox, obtain the current state characteristic value, current operating condition range and deterioration response direction of each state characterization quantity at the current evaluation time, and match the current operating condition range with the operating condition range corresponding to the health reference state characteristic data. When the current operating condition interval is the same as the operating condition interval corresponding to the health reference state characteristic data, the directional offset is calculated based on the current state characteristic value, health benchmark value, health scale parameter and deterioration response direction.
[0018] Specifically, for the matching health reference data corresponding to the current state representation value, all state representation values in the matching health reference data are arranged in ascending order. When the number of data points is odd, the value in the middle of the arrangement is determined as the health baseline value. When the number of data points is even, the arithmetic mean of the two middle values is determined as the health baseline value. The absolute value of the difference between each state representation value and the health baseline value in the matching health reference data is calculated. All absolute values of the difference are arranged in ascending order. When the number of absolute values of the difference is odd, the absolute value of the difference in the middle of the arrangement is determined as the health scale parameter. When the number of absolute values of the difference is even, the arithmetic mean of the two middle absolute values of the difference is determined as the health scale parameter. When the health scale parameter is greater than zero and deterioration occurs... When the response direction is positive, the difference between the current state characterization value and the health baseline value is divided by the health scale parameter to obtain the directional offset. When the health scale parameter is greater than zero and the degradation response direction is negative, the difference between the health baseline value and the current state characterization value is divided by the health scale parameter to obtain the directional offset. When the health scale parameter is equal to zero and the current state characterization value is equal to the health baseline value, the directional offset is determined to be zero. When the health scale parameter is equal to zero and the current state characterization value is not equal to the health baseline value, the absolute value of the difference between the current state characterization value and the health baseline value is determined as the substitute health scale parameter, and the directional offset is calculated according to the difference calculation method corresponding to the positive or negative response direction. When the degradation response direction is the zero response direction, the directional offset is determined to be zero.
[0019] In embodiments of the present invention, calculating the quality of evidence for deterioration propositions, health propositions, and uncertain propositions includes: Based on each set of health reference state feature data, the corresponding health baseline value, the corresponding health scale parameter, and the corresponding deterioration response direction, calculate the health reference directional offset of each state feature value in the health reference state feature data. All health reference directional offsets are arranged in ascending order of value, and the 95th percentile of the arrangement is determined as the health fluctuation threshold of the corresponding state characterization quantity. Determine the evaluation time window with the current evaluation time as the cutoff time, and obtain the data validity coefficient of the corresponding state characterization quantity based on the ratio of the number of state characteristic values with validity markers within the evaluation time window to the total number of state characteristic values within the evaluation time window; Determine the minimum number of healthy samples, and determine the health sample quantity coefficient based on the number of state feature values contained in the corresponding health reference state feature data and the minimum number of healthy samples. Calculate the standard deviation of the corresponding health reference state characteristic data; The ratio of the standard deviation to the corresponding health scale parameter is calculated, and the result obtained by dividing one by the sum of one and the ratio is determined as the characteristic stability coefficient of the corresponding state characterization quantity. The product of the data validity coefficient, the healthy sample quantity coefficient, and the feature stability coefficient is determined as the credibility coefficient of the corresponding state representation quantity. If the current operating condition range is different from the operating condition range corresponding to the health reference status characteristic data, the confidence coefficient will be set to zero. Determine the evidence change range that is greater than zero; when the directional offset is not greater than the health fluctuation threshold, the degradation support is set to zero; when the directional offset is greater than the health fluctuation threshold, calculate the difference between the directional offset and the health fluctuation threshold, and determine the ratio of this difference to the evidence change range as the initial degradation support. If the initial degradation support is greater than one, the degradation support is set to one; if the initial degradation support is not greater than one, the degradation support is set to the initial degradation support. The difference between 1 and the deterioration support is determined as the health support; the product of the credibility coefficient and the deterioration support is determined as the evidence quality of the deterioration proposition; the product of the credibility coefficient and the health support is determined as the evidence quality of the health proposition; and the difference between 1 and the credibility coefficient is determined as the evidence quality of the uncertain proposition.
[0020] S3. Based on the measurement point number, characteristic physical category and frequency band attribute corresponding to each state characterization quantity, obtain the evidence credibility of each state characterization quantity; redistribute the evidence quality of each homologous evidence cluster to obtain the healthy proposition evidence quality and uncertain proposition evidence quality of each homologous evidence cluster. In embodiments of the present invention, the credibility of evidence for each state representation quantity is obtained, including: Based on the measurement point number, characteristic physical category and frequency band attribute corresponding to each state characterization quantity, the state characterization quantities are grouped to obtain multiple homologous evidence clusters. Based on the quality of deteriorated propositional evidence corresponding to each state representation quantity at multiple consecutive evaluation times, the correlation between each state representation quantity in each homogeneous evidence cluster is calculated to obtain the intra-cluster correlation of each homogeneous evidence cluster. Based on the quality of uncertain propositional evidence corresponding to each state representation quantity at the current evaluation time, the credibility of evidence for each state representation quantity is calculated, and the credibility of evidence for each state representation quantity is obtained.
[0021] Specifically, for each homologous evidence cluster, the quality of degraded propositional evidence for each state representation quantity within the cluster is obtained at the current evaluation time and at previous evaluation times. The quality of degraded propositional evidence for each state representation quantity is arranged from earliest to latest evaluation time, forming a sequence of degraded propositional evidence quality for each state representation quantity. Evaluation times where all state representation quantities within the cluster possess degraded propositional evidence quality are selected. These selected evaluation times are arranged chronologically, and the longest common time segment consecutively arranged in the evaluation time sequence is selected from the current evaluation time backwards. This longest common time segment is determined as the correlation calculation period. The sequence of degraded propositional evidence quality for each state representation quantity is aligned according to the evaluation times within the correlation calculation period. For any two different state representation quantities within the same homologous evidence cluster, two degraded propositional evidence quality sequences are calculated. The arithmetic mean of the quality of evidence for each deteriorated proposition is calculated. The difference between the quality of evidence for each deteriorated proposition and the corresponding arithmetic mean is calculated. The product of the two differences at the same evaluation time is calculated and the sum of all products over the correlation calculation period is obtained to obtain the co-variance. The sum of squares of the differences in the quality of evidence for each of the two deteriorated propositions is calculated. The square root of the product of the two sums of squares is obtained to obtain the scale of change. When the scale of change is greater than zero, the co-variance is divided by the scale of change to obtain the correlation coefficient between the two state representation quantities. When the scale of change is equal to zero, the correlation coefficient between the two state representation quantities is determined to be zero. The absolute values of all correlation coefficients within the same source evidence cluster are calculated. The arithmetic mean of all absolute values is determined to be the intra-cluster correlation of the corresponding source evidence cluster. When the source evidence cluster contains only one state representation quantity, the intra-cluster correlation of the corresponding source evidence cluster is determined to be zero.
[0022] Specifically, for each state representation quantity, the quality of uncertain proposition evidence corresponding to that state representation quantity at the current evaluation time is obtained. The sum of this uncertain proposition evidence quality and the quality of health and deterioration evidence corresponding to the same state representation quantity at the current evaluation time is checked against one. If the sum of the three quality of evidence equals one, the uncertain proposition evidence quality is retained. If the sum of the three quality of evidence is not equal to one but is greater than zero, the uncertain proposition evidence quality is divided by the sum of the three quality of evidence to obtain the normalized uncertain proposition evidence quality. The difference between one and the normalized uncertain proposition evidence quality is determined as the evidence credibility of that state representation quantity at the current evaluation time. If the sum of the three quality of evidence equals zero, the evidence credibility is determined to be zero. Based on the current evaluation time, measurement point number, homologous evidence cluster number, and state representation quantity name, the evidence credibility of each state representation quantity is correlated with the corresponding quality of health, deterioration, and uncertain proposition evidence, thus obtaining the evidence credibility of each state representation quantity at the current evaluation time.
[0023] In embodiments of the present invention, obtaining the quality of evidence for health propositions and the quality of evidence for uncertain propositions includes: Based on the evidentiary credibility of each state representation quantity, the quality of evidence for deteriorated propositions, the quality of evidence for healthy propositions, and the quality of evidence for uncertain propositions corresponding to each state representation quantity in each homologous evidence cluster at the current evaluation time are weighted and aggregated to obtain the initial quality of evidence for deteriorated propositions, the initial quality of evidence for healthy propositions, and the initial quality of evidence for uncertain propositions for each homologous evidence cluster. Based on the intra-cluster correlation and the number of state representation quantities of each homologous evidence cluster, redundancy suppression is performed on the initial deteriorated propositional evidence quality of each homologous evidence cluster to obtain the deteriorated propositional evidence quality of each homologous evidence cluster. Based on the initial healthy proposition evidence quality, the initial uncertain proposition evidence quality, and the suppressed deteriorated proposition evidence quality of each homologous evidence cluster, the evidence quality of each homologous evidence cluster is redistributed to obtain the healthy proposition evidence quality and uncertain proposition evidence quality of each homologous evidence cluster.
[0024] Specifically, for each homologous evidence cluster, the intra-cluster correlation, the number of state representations, and the initial deterioration propositional evidence quality of that homologous evidence cluster are obtained. The difference between the number of state representations and one is calculated. The product of this difference and the intra-cluster correlation is determined as the redundancy level. The sum of one and the redundancy level is calculated. The result of dividing one by this sum is determined as the redundancy suppression coefficient. The product of the initial deterioration propositional evidence quality and the redundancy suppression coefficient is determined as the deterioration propositional evidence quality of that homologous evidence cluster. The difference between the initial deterioration propositional evidence quality and the deterioration propositional evidence quality is determined as the suppressed deterioration propositional evidence quality. When the homologous evidence cluster contains only one state representation, the difference between the number of state representations and one is zero, the redundancy suppression coefficient is one, and the deterioration propositional evidence quality is equal to the initial deterioration propositional evidence quality.
[0025] Specifically, for each homologous evidence cluster, the initial healthy proposition evidence quality, the initial uncertain proposition evidence quality, and the suppressed deteriorated proposition evidence quality of the homologous evidence cluster are obtained. The initial healthy proposition evidence quality is kept unchanged and determined as the healthy proposition evidence quality of the homologous evidence cluster. The initial uncertain proposition evidence quality is added to the suppressed deteriorated proposition evidence quality, and the sum is determined as the uncertain proposition evidence quality of the homologous evidence cluster. The sum of the healthy proposition evidence quality, the deteriorated proposition evidence quality, and the uncertain proposition evidence quality of the homologous evidence cluster is calculated. When the sum is equal to one, the three evidence qualities are stored. When the sum deviates from one due to numerical calculation error, the three evidence qualities are divided by the sum respectively to obtain the normalized healthy proposition evidence quality, the deteriorated proposition evidence quality, and the uncertain proposition evidence quality.
[0026] S4. Based on the quality of health proposition evidence and the quality of uncertain proposition evidence in each homologous evidence cluster, the evidence of instantaneous degradation of the gearbox as a whole is obtained; In embodiments of the present invention, evidence of instantaneous degradation of the gearbox as a whole is obtained, including: Based on the quality of healthy propositional evidence and the quality of uncertain propositional evidence for each homologous evidence cluster, the quality of degraded propositional evidence for each homologous evidence cluster is calculated. Specifically, for each homologous evidence cluster, the quality of evidence for healthy propositions and uncertain propositions corresponding to the current evaluation time are obtained. The quality of evidence for healthy propositions and uncertain propositions is checked against a value between zero and one. Evidence quality less than zero is adjusted to zero, and evidence quality greater than one is adjusted to one. The sum of the adjusted quality of evidence for healthy propositions and uncertain propositions is calculated. If the sum is not greater than one, the difference between one and the sum is determined as the degraded quality of evidence for the corresponding homologous evidence cluster. If the sum is greater than one, the adjusted quality of evidence for healthy propositions and uncertain propositions are divided by the sum to obtain the normalized quality of evidence for healthy propositions and the normalized quality of evidence for uncertain propositions. The degraded quality of evidence is set to zero. The homologous evidence cluster number, the current evaluation time, the quality of evidence for healthy propositions, the quality of evidence for uncertain propositions, and the quality of evidence for degraded propositions are correlated to obtain the degraded quality of evidence for each homologous evidence cluster at the current evaluation time.
[0027] According to the test point number corresponding to each homologous evidence cluster, the quality of evidence for health propositions, quality of evidence for deterioration propositions, and quality of evidence for uncertain propositions corresponding to the same test point are fused to obtain the test point-level quality of evidence for health propositions, quality of evidence for deterioration propositions, and quality of evidence for uncertain propositions for each test point. Based on the quality of evidence for health propositions, degradation propositions, and uncertain propositions at each test point, evidence fusion is performed on each test point. The conflict quality between the health propositions and degradation propositions generated by the fusion is allocated to the uncertain propositions to obtain the overall quality of evidence for the gearbox health proposition, the overall quality of evidence for the gearbox degradation proposition, and the overall quality of evidence for the gearbox uncertain proposition. The quality of evidence for the proposition of overall gearbox deterioration is determined to be evidence of instantaneous overall gearbox deterioration.
[0028] S5. Based on the instantaneous degradation evidence sequence, the projected Kalman filter model is used for state prediction and observation update to obtain the unconstrained hidden state degradation index at each evaluation time, and this index is determined as the evaluation result of the wind turbine gearbox degradation state.
[0029] In embodiments of the present invention, the unconstrained hidden state degradation index at each evaluation time is obtained, including: According to the evaluation time, the evidence of instantaneous degradation of the gearbox as a whole is arranged to obtain the instantaneous degradation evidence sequence; Determine the initial state estimate, initial estimate error variance, process noise variance, and observation noise variance of the projected Kalman filter model; Based on the instantaneous degradation evidence sequence, initial state estimate, initial estimation error variance, process noise variance, and observation noise variance, the projected Kalman filter model is used for state prediction and observation update to obtain the unconstrained hidden state degradation index at each evaluation time.
[0030] Specifically, for the gearbox overall instantaneous degradation evidence sequence arranged from earliest to latest evaluation time, the gearbox overall instantaneous degradation evidence corresponding to the first evaluation time in the sequence is determined as the initial state estimate. The difference between each gearbox overall instantaneous degradation evidence in the sequence and the initial state estimate is calculated. The squares of each difference are summed, and the summation result is divided by the number of gearbox overall instantaneous degradation evidence to obtain the initial estimate error variance. When the number of gearbox overall instantaneous degradation evidence is not less than two, the difference between the gearbox overall instantaneous degradation evidence corresponding to every two adjacent evaluation times is calculated. The squares of each adjacent difference are summed, and the summation result is divided by the number of adjacent differences to obtain the process noise variance. When there are at least three pieces of evidence of gearbox degradation, for each evaluation time other than the first and last evaluation time, the arithmetic mean of the gearbox overall instantaneous degradation evidence corresponding to the previous and next evaluation times is calculated. The difference between the gearbox overall instantaneous degradation evidence corresponding to the current evaluation time and the arithmetic mean is calculated. The squares of each difference are summed, and the sum is divided by the number of evaluation times involved in the calculation to obtain the observation noise variance. When there are two pieces of evidence of gearbox overall instantaneous degradation, half of the process noise variance is determined as the observation noise variance. When there is one piece of evidence of gearbox overall instantaneous degradation, the initial estimation error variance, process noise variance, and observation noise variance are determined to be zero.
[0031] Specifically, the unconstrained hidden state degradation index corresponding to the first evaluation time is determined as the initial state estimate, and the updated estimation error variance corresponding to the first evaluation time is determined as the initial estimation error variance. For each evaluation time after the first evaluation time, the unconstrained hidden state degradation index corresponding to the previous evaluation time is determined as the predicted state value for the current evaluation time. The updated estimation error variance corresponding to the previous evaluation time is added to the process noise variance to obtain the prediction estimation error variance for the current evaluation time. The prediction estimation error variance is then added to the observation noise variance to obtain the denominator for gain calculation. When the gain calculation denominator is greater than zero, the prediction estimation error variance is divided by the gain. Calculate the denominator to obtain the Kalman gain at the current evaluation time; when the denominator of the gain calculation is equal to zero, the Kalman gain at the current evaluation time is determined to be zero; calculate the difference between the gearbox overall instantaneous degradation evidence and the predicted state value corresponding to the current evaluation time, multiply the difference by the Kalman gain, and add the multiplication result to the predicted state value to obtain the unconstrained hidden state degradation index at the current evaluation time; calculate the difference obtained by subtracting the Kalman gain from the value, multiply the difference by the prediction estimation error variance to obtain the updated estimation error variance at the current evaluation time, and repeat the state prediction and observation update in the order of evaluation times to obtain the unconstrained hidden state degradation index at each evaluation time.
[0032] In embodiments of the present invention, determining the evaluation result of the deterioration state of the wind turbine gearbox includes: Determine the lower limit, upper limit, and maximum allowable change of the latent state degradation index between adjacent evaluation times; Based on the lower limit of the index, the upper limit of the index, the maximum allowable change between adjacent evaluation times, and the hidden state degradation index of the previous evaluation time, the unconstrained hidden state degradation index of the current evaluation time is projected and constrained to obtain the hidden state degradation index of the current evaluation time.
[0033] The latent state degradation index at each evaluation time is determined as the evaluation result of the wind turbine gearbox degradation state.
[0034] Specifically, for the unconstrained latent state degradation index at each evaluation time, the lower limit of the normalized range of evidence quality (zero) is determined as the lower limit of the index, and the upper limit of the normalized range of evidence quality (one) is determined as the upper limit of the index. The non-negative square root of the process noise variance is determined as the maximum allowable change between adjacent evaluation times. For the first evaluation time, the initial state estimate is compared with the lower limit of the index. When the initial state estimate is less than the lower limit, the lower limit is determined as the latent state degradation index for the first evaluation time. When the initial state estimate is greater than the upper limit, the upper limit is determined as the latent state degradation index for the first evaluation time. When the initial state estimate is between the lower and upper limits, the initial state estimate is determined as the latent state degradation index for the first evaluation time. For each current evaluation time after the first evaluation time, the latent state degradation index of the previous evaluation time is determined as the time constraint for the current evaluation time. The lower limit is calculated by adding the latent state degradation index of the previous evaluation time to the maximum allowable change of the adjacent evaluation time to obtain the candidate time constraint upper limit for the current evaluation time. The smaller value between the candidate time constraint upper limit and the index upper limit is determined as the time constraint upper limit for the current evaluation time. The unconstrained latent state degradation index of the current evaluation time is compared with the time constraint lower limit. When the unconstrained latent state degradation index is less than the time constraint lower limit, the time constraint lower limit is determined as the latent state degradation index for the current evaluation time. When the unconstrained latent state degradation index is greater than the time constraint upper limit, the time constraint upper limit is determined as the latent state degradation index for the current evaluation time. When the unconstrained latent state degradation index is between the time constraint lower limit and the time constraint upper limit, the unconstrained latent state degradation index is determined as the latent state degradation index for the current evaluation time. The projection constraint is repeatedly executed from the earliest to the latest evaluation time to obtain the latent state degradation index for each evaluation time.
[0035] Specifically, for each evaluation moment after completing the projection constraints, the evaluation moment, the corresponding instantaneous degradation evidence of the gearbox as a whole, the corresponding unconstrained latent state degradation index, and the corresponding latent state degradation index are correlated. The correlation results are arranged from earliest to latest according to the evaluation moment to form a sequence of latent state degradation indices for the wind turbine gearbox. The sequence of latent state degradation indices is determined as the evaluation result of the wind turbine gearbox degradation state. The latent state degradation index corresponding to the last evaluation moment in the sequence is determined as the current evaluation value of the wind turbine gearbox degradation state. The difference between the latent state degradation indices of adjacent evaluation moments is determined as the degradation change amount in the corresponding time interval. The time interval with zero degradation change amount is recorded as the state maintenance interval, and the time interval with greater than zero degradation change amount is recorded as the state growth interval. The evaluation moment, the current evaluation value of the wind turbine gearbox degradation state, the degradation change amount, and the corresponding time interval are correlated to obtain the evaluation record of the wind turbine gearbox degradation state.
[0036] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.
Claims
1. A method for evaluating the deterioration state of a wind turbine gearbox, characterized in that, Includes the following steps: S1. Based on the original status monitoring data of each monitoring point of the wind turbine gearbox, invalid samples are removed from the multi-point aligned samples to obtain the multi-point valid samples; the multi-point valid samples are divided to determine the health reference stage. S2. Calculate the directional shift based on the health reference stage; calculate the quality of evidence for deterioration propositions, health propositions, and uncertain propositions based on the directional shift. S3. Based on the measurement point number, characteristic physical category and frequency band attribute corresponding to each state characterization quantity, obtain the evidence credibility of each state characterization quantity; redistribute the evidence quality of each homologous evidence cluster to obtain the healthy proposition evidence quality and uncertain proposition evidence quality of each homologous evidence cluster. S4. Based on the quality of health proposition evidence and the quality of uncertain proposition evidence in each homologous evidence cluster, the evidence of instantaneous degradation of the gearbox as a whole is obtained; S5. Based on the instantaneous degradation evidence sequence, the projected Kalman filter model is used for state prediction and observation update to obtain the unconstrained hidden state degradation index at each evaluation time, and this index is determined as the evaluation result of the wind turbine gearbox degradation state.
2. The method for evaluating the deterioration state of a wind turbine gearbox according to claim 1, characterized in that, Obtain valid samples from multiple measurement points, including: Acquire raw status monitoring data, fault alarm records, and maintenance event records for each monitoring point of the wind turbine gearbox; The evaluation time is generated according to the preset sampling period, and the original state monitoring data of each monitoring point is assigned to the corresponding evaluation time according to the sampling time. The original state monitoring data belonging to the same evaluation time are matched according to the measurement point number to form a multi-measurement point aligned sample; Based on the sensor status markers, data missing ratio, rotational speed, and vibration signal amplitude, invalid samples are removed from the multi-measurement point aligned samples to obtain valid multi-measurement point samples.
3. The method for evaluating the deterioration state of a wind turbine gearbox according to claim 2, characterized in that, Determine the health reference phase, including: The effective samples of multiple measurement points are divided according to the preset speed range and preset load range to obtain multiple operating condition ranges; When the original state monitoring data includes vibration signals, state features are extracted from effective samples of multiple measuring points in each operating condition range. If the original condition monitoring data includes condition characteristics, read the condition characteristics within each operating condition interval; The extracted or read state features are defined as state representation quantities; Configure sampling time, measurement point number, feature name, feature physical category, spectral attribute, corresponding component, degradation response direction and data validity label for each state characterization quantity; Based on fault alarm records, maintenance event records, speed change rate, and load change rate, select continuous samples from the effective samples of multiple measurement points that do not contain fault alarms or maintenance events, have a speed change rate not greater than a preset speed change rate threshold, and have a load change rate not greater than a preset load change rate threshold. The time intervals corresponding to continuous samples are defined as the health reference phase.
4. The method for evaluating the deterioration state of a wind turbine gearbox according to claim 3, characterized in that, Calculating the directional offset includes: Based on the health reference stage, health reference state characteristic data are extracted from the multi-point state characteristic data of the wind turbine gearbox; For each set of health reference state characteristic data, calculate the median of all state characteristic values in the health reference state characteristic data, and determine the median as the health benchmark value of the corresponding state characterization quantity. Calculate the absolute value of the difference between each state feature value and the corresponding health baseline value in the health reference state feature data, calculate the median of all absolute values, and determine the health scale parameter of the corresponding state characterization quantity by summing the median of all absolute values and the scale correction parameter. From the multi-point state characteristic data of the wind turbine gearbox, obtain the current state characteristic value, current operating condition range and deterioration response direction of each state characterization quantity at the current evaluation time, and match the current operating condition range with the operating condition range corresponding to the health reference state characteristic data. When the current operating condition interval is the same as the operating condition interval corresponding to the health reference state characteristic data, the directional offset is calculated based on the current state characteristic value, health benchmark value, health scale parameter and deterioration response direction.
5. The method for evaluating the deterioration state of a wind turbine gearbox according to claim 4, characterized in that, Calculate the quality of evidence for deterioration propositions, health propositions, and uncertain propositions, including: Based on each set of health reference state feature data, the corresponding health baseline value, the corresponding health scale parameter, and the corresponding deterioration response direction, calculate the health reference directional offset of each state feature value in the health reference state feature data. All health reference directional offsets are arranged in ascending order of value, and the 95th percentile of the arrangement is determined as the health fluctuation threshold of the corresponding state characterization quantity. Determine the evaluation time window with the current evaluation time as the cutoff time, and obtain the data validity coefficient of the corresponding state characterization quantity based on the ratio of the number of state characteristic values with validity markers within the evaluation time window to the total number of state characteristic values within the evaluation time window; Determine the minimum number of healthy samples, and determine the health sample quantity coefficient based on the number of state feature values contained in the corresponding health reference state feature data and the minimum number of healthy samples. Calculate the standard deviation of the corresponding health reference state characteristic data; The ratio of the standard deviation to the corresponding health scale parameter is calculated, and the result obtained by dividing one by the sum of one and the ratio is determined as the characteristic stability coefficient of the corresponding state characterization quantity. The product of the data validity coefficient, the healthy sample quantity coefficient, and the feature stability coefficient is determined as the credibility coefficient of the corresponding state representation quantity. If the current operating condition range is different from the operating condition range corresponding to the health reference status characteristic data, the confidence coefficient will be set to zero. Determine the evidence change range that is greater than zero; when the directional offset is not greater than the health fluctuation threshold, the degradation support is set to zero; when the directional offset is greater than the health fluctuation threshold, calculate the difference between the directional offset and the health fluctuation threshold, and determine the ratio of this difference to the evidence change range as the initial degradation support. If the initial degradation support is greater than one, the degradation support is set to one; if the initial degradation support is not greater than one, the degradation support is set to the initial degradation support. The difference between 1 and the deterioration support is determined as the health support; the product of the credibility coefficient and the deterioration support is determined as the evidence quality of the deterioration proposition; the product of the credibility coefficient and the health support is determined as the evidence quality of the health proposition; and the difference between 1 and the credibility coefficient is determined as the evidence quality of the uncertain proposition.
6. The method for evaluating the deterioration state of a wind turbine gearbox according to claim 5, characterized in that, The credibility of the evidence obtained for each state representation quantity includes: Based on the measurement point number, characteristic physical category and frequency band attribute corresponding to each state characterization quantity, the state characterization quantities are grouped to obtain multiple homologous evidence clusters. Based on the quality of deteriorated propositional evidence corresponding to each state representation quantity at multiple consecutive evaluation times, the correlation between each state representation quantity in each homogeneous evidence cluster is calculated to obtain the intra-cluster correlation of each homogeneous evidence cluster. Based on the quality of uncertain propositional evidence corresponding to each state representation quantity at the current evaluation time, the credibility of evidence for each state representation quantity is calculated, and the credibility of evidence for each state representation quantity is obtained.
7. The method for evaluating the deterioration state of a wind turbine gearbox according to claim 6, characterized in that, To determine the quality of evidence for health propositions and the quality of evidence for uncertain propositions, including: Based on the evidentiary credibility of each state representation quantity, the quality of evidence for deteriorated propositions, the quality of evidence for healthy propositions, and the quality of evidence for uncertain propositions corresponding to each state representation quantity in each homologous evidence cluster at the current evaluation time are weighted and aggregated to obtain the initial quality of evidence for deteriorated propositions, the initial quality of evidence for healthy propositions, and the initial quality of evidence for uncertain propositions for each homologous evidence cluster. Based on the intra-cluster correlation and the number of state representation quantities of each homologous evidence cluster, redundancy suppression is performed on the initial deteriorated propositional evidence quality of each homologous evidence cluster to obtain the deteriorated propositional evidence quality of each homologous evidence cluster. Based on the initial healthy proposition evidence quality, the initial uncertain proposition evidence quality, and the suppressed deteriorated proposition evidence quality of each homologous evidence cluster, the evidence quality of each homologous evidence cluster is redistributed to obtain the healthy proposition evidence quality and uncertain proposition evidence quality of each homologous evidence cluster.
8. The method for evaluating the deterioration state of a wind turbine gearbox according to claim 7, characterized in that, Evidence of instantaneous overall gearbox degradation was obtained, including: Based on the quality of healthy propositional evidence and the quality of uncertain propositional evidence for each homologous evidence cluster, the quality of degraded propositional evidence for each homologous evidence cluster is calculated. According to the test point number corresponding to each homologous evidence cluster, the quality of evidence for health propositions, quality of evidence for deterioration propositions, and quality of evidence for uncertain propositions corresponding to the same test point are fused to obtain the test point-level quality of evidence for health propositions, quality of evidence for deterioration propositions, and quality of evidence for uncertain propositions for each test point. Based on the quality of evidence for health propositions, degradation propositions, and uncertain propositions at each test point, evidence fusion is performed on each test point. The conflict quality between the health propositions and degradation propositions generated by the fusion is allocated to the uncertain propositions to obtain the overall quality of evidence for the gearbox health proposition, the overall quality of evidence for the gearbox degradation proposition, and the overall quality of evidence for the gearbox uncertain proposition. The quality of evidence for the proposition of overall gearbox deterioration is determined to be evidence of instantaneous overall gearbox deterioration.
9. The method for evaluating the deterioration state of a wind turbine gearbox according to claim 8, characterized in that, The unconstrained hidden state degradation index is obtained at each evaluation time, including: According to the evaluation time, the evidence of instantaneous degradation of the gearbox as a whole is arranged to obtain the instantaneous degradation evidence sequence; Determine the initial state estimate, initial estimate error variance, process noise variance, and observation noise variance of the projected Kalman filter model; Based on the instantaneous degradation evidence sequence, initial state estimate, initial estimation error variance, process noise variance, and observation noise variance, the projected Kalman filter model is used for state prediction and observation update to obtain the unconstrained hidden state degradation index at each evaluation time.
10. The method for evaluating the deterioration state of a wind turbine gearbox according to claim 1, characterized in that, Determine the assessment results of the deterioration status of the wind turbine gearbox, including: Determine the lower limit, upper limit, and maximum allowable change of the latent state degradation index between adjacent evaluation times; Based on the lower limit of the index, the upper limit of the index, the maximum allowable change between adjacent evaluation times, and the hidden state degradation index of the previous evaluation time, the unconstrained hidden state degradation index of the current evaluation time is projected and constrained to obtain the hidden state degradation index of the current evaluation time. The latent state degradation index at each evaluation time is determined as the evaluation result of the wind turbine gearbox degradation state.
