A fan vibration online diagnosis method and system based on generalized extreme value distribution

Abstract

The application discloses a fan vibration online diagnosis method and system based on generalized extreme value distribution, and the method filters old historical vibration time series data, extracts characteristic values and calculates fitting distribution conditions, obtains a generalized extreme value distribution curve of the fan, calculates a cumulative probability density curve of the curve, that is, a fitting model, calculates key point extreme value data according to the curve, enables the fan master control to directly judge a theoretical occurrence probability of a vibration condition at the moment according to the fitting model, flexibly adjusts a control strategy based on the probability, and realizes real-time vibration online diagnosis; the application is simple to implement and low in cost, does not need to additionally add other equipment on the basis of the original, does not need internet and other facilities, only needs to modify a fan master control program code, and meanwhile, the bottom logic is simple and clear, so that the master control corresponding operation logic is convenient to debug and modify.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of online detection of a fan, and particularly to a fan vibration online diagnosis method and system based on generalized extreme value distribution. BACKGROUND

[0002] At present, a fan is generally equipped with a vibration sensor and a collector capable of collecting vibration, which can collect and record the vibration of each component of the fan in real time. The vibration collection system serves as an auxiliary control system and is independent of the main control system of the fan.

[0003] The vibration detection system collects vibration time series data at intervals and calculates characteristic values such as effective values, and uploads them to the main control system of the fan. The main control of the fan simply and roughly sets alarm limits and shutdown limits according to the limits, and the fan will shut down to ensure safety when the shutdown limit is exceeded.

[0004] However, the following problems exist in the simple limit division: first, the limit is artificially divided and is not rigorous and scientific, and cannot adapt to different types of fans; second, the environment of the wind field is complex and there is signal interference, which causes the collected vibration value to be larger than the actual value, thereby triggering false shutdown; third, the simple limit method can only qualitatively analyze the vibration situation and cannot quantitatively analyze it, and cannot flexibly adjust the control strategy according to the vibration value.

[0005] In order to avoid the above problems, the existing technology extracts the collected vibration signal through a machine learning algorithm, uses a supervised learning algorithm, pre-trains a neural network model, and then arranges it on a fan dedicated server. The output result of the neural network model is used to determine whether shutdown is required; however, this method requires high computing power, requires an additional server, and increases costs; and machine learning is a black box, and engineers cannot determine the internal rationality of each output result, making it difficult to flexibly adjust the main control program code according to the output result.

[0006] In addition, data is also returned to a cloud server through a network, and the cloud server uniformly calculates the vibration judgment result and returns the result to the main control system of the fan; however, when there are a large number of fans in the wind field, there are disadvantages such as insufficient real-time performance, large bandwidth demand, and no network signal for offshore fans. SUMMARY

[0007] The present application aims to overcome the shortcomings of the prior art and provides a fan vibration online diagnosis method and system based on generalized extreme value distribution. The method filters and extracts characteristic values and fits the distribution of historical vibration data to obtain a generalized extreme value distribution curve of the fan. The cumulative probability density curve of the curve can be used by the main control of the fan to directly determine the theoretical probability of the vibration condition at this time, and the condition can be quantitatively analyzed. The control strategy can be flexibly adjusted according to the model.

[0008] The object of the present application is achieved by the following technical solution: a fan vibration online diagnosis method based on generalized extreme value distribution, comprising the following steps:

[0009] S1, classifying according to a preset condition, obtaining historical vibration time series data of each measuring point of several fans, and eliminating abnormal data therein;

[0010] S2, filtering the obtained historical vibration time series data to exclude interference;

[0011] S3, performing characteristic value extraction calculation on the filtered time series data;

[0012] S4, uniformly distributing and extracting the characteristic value data of the several fans calculated in step S3, and the number of extractions is greater than the original sample number, to form a matrix, so as to ensure that the data weights of the fans are equal;

[0013] S5, fitting a cumulative probability density curve of the generalized extreme value distribution according to the distribution, obtaining parameters describing the cumulative probability density curve, and obtaining a fitting model;

[0014] S6, calculating key point extreme value data according to the cumulative probability density curve of the generalized extreme value distribution;

[0015] S7, writing related parameters of the generalized extreme value distribution into a fan main control program, and the fan main control program calculates the corresponding cumulative probability density value according to the actually measured vibration time series characteristic value;

[0016] S8, the fan main control program diagnoses the vibration condition of the fan according to the calculated corresponding cumulative probability density value, and performs an operation corresponding to the vibration condition on the fan.

[0017] Further, the step S1 comprises the following steps:

[0018] According to the vicinity of the rated wind speed interval, the historical vibration time series data of each measuring point of multiple fans of the same type are obtained, and the abnormal historical vibration data are deleted; wherein the historical vibration time series data include acceleration, speed and displacement, and the measuring points include the main bearing, gear box, generator and cabin tower of the fan.

[0019] Further, the step S2 comprises the following steps:

[0020] The obtained historical vibration time series data are filtered in a preset frequency band to remove data of other frequencies; wherein the frequency band includes 0.1-10Hz, 10-1000Hz and 10-2000Hz.

[0021] Further, the step S3 comprises the following steps:

[0022] The characteristic value data of the same type but different numbered fans are uniformly distributed extracted, the number of extraction is greater than the original sample number, and a matrix is formed, each column of the matrix is the uniformly distributed extracted characteristic value of the corresponding fan; at the same time, the weight of each fan needs to be consistent, that is, the input sample number of each fan needs to be consistent.

[0023] Further, the step S4 comprises the following steps:

[0024] The characteristic value data of the same type but different numbered fans are uniformly distributed extracted, the number of extraction is greater than the original sample number, and a matrix is formed, each column of the matrix is the uniformly distributed extracted characteristic value of the corresponding fan; at the same time, the weight of each fan needs to be consistent, that is, the input sample number of each fan needs to be consistent.

[0025] Further, the step S5 comprises the following steps:

[0026] After extracting the characteristics of the vibration time series data of the same type of fan in history, the cumulative probability density curve of the generalized extreme value distribution is fitted according to the distribution, and three parameters including the position parameter μ, the scale parameter σ and the shape parameter k describing the cumulative probability density curve are obtained, and the cumulative probability density function of the generalized extreme value distribution is defined as:

[0027]

[0028] Wherein, x is the sample value, exp() is the exponential function with natural constant e as the base, μ is the position parameter of the cumulative probability density curve, σ is the scale parameter of the cumulative probability density curve and k is the shape parameter of the cumulative probability density curve.

[0029] Further, the step S6 comprises the following steps:

[0030] According to the cumulative probability density curve of the generalized extreme value distribution, the key point extreme value data is calculated; the key point extreme value data includes 75% quantile, 95% quantile, 99.7% quantile and quartile.

[0031] Further, the step S7 comprises the following steps:

[0032] The related parameters of the generalized extreme value distribution are written into the fan main control program, according to the actually measured vibration time series characteristic value x of the fan, the generalized extreme value distribution formula is substituted, and the cumulative probability density value corresponding to the vibration characteristic value is calculated by the main control program Wherein, exp() is the exponential function with natural constant e as the base, μ is the position parameter of the cumulative probability density curve, σ is the scale parameter of the cumulative probability density curve and k is the shape parameter of the cumulative probability density curve.

[0033] Further, the step S8 comprises the following steps:

[0034] According to the cumulative probability density value y, the main control program controls the fan to perform corresponding operation; if the cumulative probability density of the corresponding vibration characteristic value is less than the 75% quantile of the generalized extreme value distribution fitting model, i.e., y<=p75, it is considered that the vibration belongs to normal condition, and the fan is in normal operation; if y>p75 and y<=p95, it is indicated that the vibration value is relatively large, and the fan is not stopped but needs to be recorded and reminded; if y>p99, it is indicated that the vibration value is large, and the latest vibration data needs to be re-collected immediately, a new y value is calculated to avoid false triggering, and if the cumulative probability density value corresponding to the new characteristic value is still y>p99, the fan is operated at a reduced power until the y value meets the requirement; if y>p75+1.5xp_iqr, it is indicated that the vibration is very violent, and the latest vibration data is re-collected, and if it is still in the interval, a shutdown strategy is adopted to ensure the safe operation of the fan.

[0035] A fan vibration online diagnosis system based on generalized extreme value distribution is used to realize the above-mentioned fan vibration online diagnosis method based on generalized extreme value distribution, and the system comprises:

[0036] A data acquisition module is used to acquire historical vibration time series data of each measuring point of a plurality of fans, and to eliminate abnormal data therein;

[0037] A data filtering module is used to filter the acquired historical vibration time series data to exclude interference;

[0038] A characteristic value extraction module is used to perform characteristic value extraction calculation on the filtered time series data;

[0039] A uniform distribution extraction module is used to perform uniform distribution extraction on the characteristic value data of the plurality of fans to form a matrix;

[0040] A fitting model construction module is used to fit a cumulative probability density curve of the generalized extreme value distribution according to the distribution, to obtain parameters describing the cumulative probability density curve, and to obtain a fitting model;

[0041] A key point extreme value data calculation module is used to calculate key point extreme value data according to the cumulative probability density curve of the generalized extreme value distribution;

[0042] A cumulative probability density value calculation module is used to calculate the corresponding cumulative probability density value according to the actually measured vibration time series characteristic value;

[0043] A diagnosis module is used to diagnose the vibration condition of the fan according to the calculated corresponding cumulative probability density value, and to perform corresponding vibration condition operation on the fan.

[0044] Compared with the prior art, the present application has the following advantages and beneficial effects:

[0045] 1、The application extracts characteristic values by filtering old historical vibration data, fits the distribution, and obtains the generalized extreme value distribution curve of the fan, wherein the generalized extreme value distribution has three undetermined parameters, which has higher adaptability than other 2-parameter models, and the fitting model has better modeling and analysis effect on extreme variability, so the modeling and fitting effect of extreme vibration value is better, and through the cumulative probability density curve of the curve, the fan master control can flexibly adjust the control strategy according to the model.

[0046] 2、The application is simple to implement and low in cost, does not need to additionally add other equipment on the original basis, does not need internet and other facilities, and only needs to modify the fan master control program code.

[0047] 3、The bottom logic of the application is simple and clear, and the corresponding operation logic of the master control is convenient to debug and modify. BRIEF DESCRIPTION OF DRAWINGS

[0048] Fig. 1 It is a flowchart of the fan vibration online diagnosis method.

[0049] Fig. 2 It is a flowchart of the fan vibration online diagnosis method.

[0050] Fig. 3 It is a probability density curve diagram of the fitting model.

[0051] Fig. 4 It is a cumulative probability density curve diagram of the generalized extreme value distribution. DETAILED DESCRIPTION

[0052] The application will be further described below in combination with specific embodiments.

[0053] Embodiment 1

[0054] Referring to Figs. 1 to 4 The fan vibration online diagnosis method based on the generalized extreme value distribution provided by the embodiment includes the following steps:

[0055] S1, classified according to the preset condition, obtain historical vibration time series data of each measuring point of a plurality of fans, and eliminate abnormal data, including the following steps:

[0056] According to the rated wind speed interval, the generator front bearing measuring point of the A type fan is obtained, and the abnormal data historical vibration data is deleted, and the vibration time series data_1, data_2, data_3……data_n are obtained; in addition, the vibration time series includes physical quantities such as acceleration, speed or displacement, and the measuring points include main bearing, gear box, generator, and cabin tower.

[0057] S2, filtering the obtained historical vibration time series data to exclude interference, including the following steps:

[0058] The generator historical vibration data is band-pass filtered at 10-5000 Hz to obtain filtered time series data fdata_1, fdata_2, fdata_3,..., fdata_n. The filtered frequency range of each measuring point includes 0.1-10 Hz, 10-1000 Hz, and 10-2000 Hz.

[0059] S3, performing feature value extraction calculation on the filtered time series data, including the following steps:

[0060] Performing feature value calculation on the filtered data to obtain a feature value array of each unit. In this example, the feature value is the root mean square value. Using the above method, the root mean square value arrays of several fans of the same type are obtained, such as arr_1 = [rms1, rms2, rms3,..., rms_n] for fan 1, arr_2 for fan 2, and arr_n for fan n. The feature values include root mean square value, peak-to-peak value, and kurtosis.

[0061] S4, uniformly distributing and extracting the feature value data of several fans calculated in step S3, and the number of extracted samples is greater than the original sample number, to form a matrix to ensure equal data weight of the fans, including the following steps:

[0062] Uniformly distributing and extracting the feature value array of each fan, and the number of extracted samples is n_samp, which is greater than the maximum length of the root mean square array, to obtain an equal-weight data matrix matrix. Each column of the matrix is the uniformly distributed and extracted feature value of the corresponding fan. For example, if the maximum length of the arr array is 100, the sampling number n_smap is 500. Uniformly distribute and extract 500 root mean square data of arr1 and place them in the first column of the matrix matrix. Similarly, uniformly distribute and extract 500 root mean square data of arr_n and place them in the n-th column of the matrix matrix.

[0063] The code is as follows:

[0064] idx = randi([1, length(arr_n)], n_samp, 1);

[0065] matrix(:, n) = arr_n(idx);

[0066] S5, fitting a cumulative probability density curve of the generalized extreme value distribution according to the distribution, obtaining parameters describing the cumulative probability density curve, and obtaining a fitting model, including the following steps:

[0067] Fitting all data in the matrix matrix as a whole sample to obtain the generalized extreme value distribution object pd and three parameters of the curve: location parameter μ, scale parameter σ, and shape parameter k.

[0068] The code is as follows: pd = fitdist(matrix(:), 'Generalized Extreme Value').

[0069] S6, according to the cumulative probability density curve of the generalized extreme value distribution, the key point extreme value data is calculated, including the following steps:

[0070] The key point limit value data is calculated: the 75th percentile, the 95th percentile and the 99.7th percentile are calculated; the functions p_75 = prctile(pd, 75), p_95 = prctile(pd, 95) and p_99 = prctile(pd, 99.7) are used; the interquartile range p_iqr = iqr(pd) is calculated.

[0071] S7, write the related parameters of the generalized extreme value distribution into the fan main control program, and the fan main control program calculates the corresponding cumulative probability density value according to the actually measured vibration time sequence characteristic value, including the following steps:

[0072] The related parameters are written into the main control program, and the cumulative probability density value corresponding to the vibration characteristic value is calculated by substituting the actually measured vibration time sequence characteristic value x into the generalized extreme value distribution formula.

[0073] S8, the fan main control program diagnoses the vibration condition of the fan according to the calculated corresponding cumulative probability density value, and performs corresponding vibration condition operation on the fan, including the following steps:

[0074] According to the cumulative probability density value y, the main control program controls the fan to perform corresponding operation; if the cumulative probability density of the corresponding vibration characteristic value is less than the 75th percentile of the generalized extreme value distribution fitting model, that is, y≤p75, it is considered that the vibration belongs to normal condition and normal operation; if y>p75 and y≤p95, it is indicated that the vibration value is relatively large, and the machine is not stopped but needs to be recorded and reminded; if y>p99, it is indicated that the vibration value is large, and the latest vibration data needs to be re-collected immediately to calculate a new y value to avoid false triggering, if the cumulative probability density value corresponding to the new characteristic value is still y>p99, the power is reduced to run until the y value meets the requirements; if y>p75+1.5×p_iqr, it is indicated that the vibration is very intense, and the latest vibration data is re-collected, if it is still in the interval, the machine is stopped to ensure the safe operation of the fan.

[0075] Example 2

[0076] The embodiment discloses a fan vibration online diagnosis system based on generalized extreme value distribution, which is used to realize the fan vibration online diagnosis method based on generalized extreme value distribution in the embodiment 1, and the system comprises:

[0077] The data acquisition module is used to obtain the historical vibration time series data of each measuring point of several wind turbines and eliminate abnormal data;

[0078] Data filtering module, used to filter the acquired historical vibration time series data to eliminate interference;

[0079] The eigenvalue extraction module is used to perform eigenvalue extraction calculation on the filtered time series data;

[0080] The uniform distribution extraction module is used to perform uniform distribution extraction on the calculated characteristic value data of several wind turbines to form a matrix;

[0081] The fitting model construction module fits the cumulative probability density curve of the generalized extreme value distribution according to the distribution situation, obtains the parameters describing the cumulative probability density curve, and obtains the fitting model;

[0082] The key point extreme value data calculation module calculates the key point extreme value data based on the cumulative probability density curve of the generalized extreme value distribution;

[0083] The cumulative probability density value calculation module calculates the corresponding cumulative probability density value based on the actual measured vibration time series characteristic value;

[0084] The diagnosis module diagnoses the vibration condition of the fan according to the corresponding cumulative probability density value obtained by calculation, and performs an operation corresponding to the vibration condition on the fan.

[0085] Example 3

[0086] This embodiment discloses a non-transitory computer-readable medium storing instructions. When the instructions are executed by a processor, the steps of the online diagnosis method for fan vibration based on generalized extreme value distribution according to embodiment 1 are performed.

[0087] The non-transitory computer-readable medium in this embodiment can be a disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), a USB flash drive, a mobile hard disk, or other media.

[0088] Example 4

[0089] This embodiment discloses a computing device, including a processor and a memory for storing a program executable by the processor. When the processor executes the program stored in the memory, the online diagnosis method for fan vibration based on generalized extreme value distribution described in Example 1 is implemented.

[0090] The computing device described in this embodiment can be a desktop computer, a notebook computer, a smart phone, a PDA handheld terminal, a tablet computer, a programmable logic controller (PLC), or other terminal device with processor function.

[0091] The above-described embodiments are only preferred embodiments of the present application, and are not intended to limit the scope of the present application. Any changes made in the shape or principle of the present application should be covered within the scope of the present application.

Claims

1. A fan vibration online diagnosis method based on generalized extreme value distribution, characterized in that: The following steps are involved: S1. Classify according to preset conditions, obtain historical vibration time series data of each measuring point of several wind turbines, and eliminate abnormal data; S2. Filter the acquired historical vibration time series data to eliminate interference; S3, performing feature value extraction calculation on the filtered time series data; S4. Evenly distribute and extract the characteristic value data of several wind turbines calculated in step S3, and the number of extractions is greater than the number of original samples, to form a matrix to ensure that the data weights of the wind turbines are equal; S5. Fitting a cumulative probability density curve of the generalized extreme value distribution according to the distribution, obtaining parameters describing the cumulative probability density curve, and obtaining a fitting model; S6. Calculate the extreme value data of key points based on the cumulative probability density curve of the generalized extreme value distribution; S7. Writing the relevant parameters of the generalized extreme value distribution into the fan main control program, and the fan main control program calculating the corresponding cumulative probability density value based on the actual measured vibration time series characteristic value; S8. The fan main control program diagnoses the vibration condition of the fan according to the corresponding cumulative probability density value obtained by calculation, and performs operations corresponding to the vibration condition on the fan.

2. The online diagnosis method for fan vibration based on generalized extreme value distribution according to claim 1, characterized in that: The step S1 comprises the following steps: Based on the rated wind speed range, historical vibration time series data of each measuring point of multiple wind turbines of the same model are obtained, and abnormal historical vibration data are deleted; wherein the historical vibration time series data includes acceleration, velocity and displacement, and the measuring points include the main bearing, gearbox, generator, and nacelle tower of the wind turbine.

3. The online diagnosis method for fan vibration based on generalized extreme value distribution according to claim 1, characterized in that: The step S2 comprises the following steps: The acquired historical vibration time series data is filtered by a preset frequency band to remove data of other frequencies; wherein the frequency band includes 0.1-10 Hz, 10-1000 Hz and 10-2000 Hz.

4. The online diagnosis method for fan vibration based on generalized extreme value distribution according to claim 1, characterized in that: The step S3 comprises the following steps: The filtered time series data is subjected to characteristic value extraction and calculation to obtain characteristic value data of each wind turbine; wherein the characteristic values ​​include root mean square value, peak-to-peak value and kurtosis.

5. The online diagnosis method for fan vibration based on generalized extreme value distribution according to claim 1, characterized in that: The step S4 comprises the following steps: The eigenvalue data of fans of the same model but different numbers are uniformly extracted. The number of extractions is greater than the original number of samples to form a matrix. Each column of the matrix is ​​the eigenvalue of the corresponding fan after uniform distribution extraction. At the same time, the weight of each fan must be consistent, that is, the number of input samples for each fan must be consistent.

6. The online diagnosis method for fan vibration based on generalized extreme value distribution according to claim 1, characterized in that: The step S5 comprises the following steps: After extracting all the features from the historical vibration time series data of the same model of fan, the cumulative probability density curve of the generalized extreme value distribution is fitted according to its distribution. The three parameters describing the cumulative probability density curve are obtained, including the location parameter μ, the scale parameter σ, and the shape parameter k. The cumulative probability density function of the generalized extreme value distribution is defined as: Where x is the sample value, exp() is the exponential function with the natural constant e as the base, μ is the location parameter of the cumulative probability density curve, σ is the scale parameter of the cumulative probability density curve, and k is the shape parameter of the cumulative probability density curve.

7. The online diagnosis method for fan vibration based on generalized extreme value distribution according to claim 1, characterized in that: The step S6 comprises the following steps: According to the cumulative probability density curve of the generalized extreme value distribution, the key point extreme value data are calculated; the key point extreme value data include the 75% quantile, the 95% quantile, the 99.7% quantile and the interquartile range.

8. The online diagnosis method for fan vibration based on generalized extreme value distribution according to claim 1, characterized in that: The step S7 comprises the following steps: The relevant parameters of the generalized extreme value distribution are written into the fan main control program. According to the actual measured vibration time series characteristic value x of the fan, it is substituted into the generalized extreme value distribution formula. The main control program calculates the cumulative probability density value corresponding to the vibration characteristic value Where exp() is an exponential function with the natural constant e as the base, μ is the location parameter of the cumulative probability density curve, σ is the scale parameter of the cumulative probability density curve, and k is the shape parameter of the cumulative probability density curve.

9. The online diagnosis method for fan vibration based on generalized extreme value distribution according to claim 1, characterized in that: The step S8 comprises the following steps: Based on the cumulative probability density value y, the main control program controls the fan to perform corresponding operations; if the cumulative probability density of the corresponding vibration eigenvalue is less than the 75% quantile of the generalized extreme value distribution fitting model, that is, y≤p75, the vibration is considered normal and normal operation is required; if y>p75 and y≤p95, it means that the vibration value is relatively large and the machine will not be shut down but a record reminder is required; if y>p99, it means that the vibration value is too large and it is necessary to immediately re-collect the latest vibration data and calculate a new y value to avoid false triggering. If the cumulative probability density value corresponding to the new eigenvalue is still y>p99, the power is reduced until the y value meets the requirements; if y>p75+1.5×p_iqr, p_iqr is the four-quartile interval, it means that the vibration is very severe and the latest vibration data is collected again. If it is still within this range, a shutdown strategy is adopted to ensure the safe operation of the fan.

10. A fan vibration online diagnosis system based on generalized extreme value distribution, characterized in that: The system for implementing the online diagnosis method for fan vibration based on generalized extreme value distribution according to any one of claims 1 to 9 comprises: The data acquisition module is used to obtain the historical vibration time series data of each measuring point of several wind turbines and eliminate abnormal data; Data filtering module, used to filter the acquired historical vibration time series data to eliminate interference; The eigenvalue extraction module is used to perform eigenvalue extraction calculation on the filtered time series data; The uniform distribution extraction module is used to perform uniform distribution extraction on the calculated characteristic value data of several wind turbines to form a matrix; The fitting model construction module fits the cumulative probability density curve of the generalized extreme value distribution according to the distribution situation, obtains the parameters describing the cumulative probability density curve, and obtains the fitting model; The key point extreme value data calculation module calculates the key point extreme value data based on the cumulative probability density curve of the generalized extreme value distribution; The cumulative probability density value calculation module calculates the corresponding cumulative probability density value based on the actual measured vibration time series characteristic value; The diagnosis module diagnoses the vibration condition of the fan according to the corresponding cumulative probability density value obtained by calculation, and performs an operation corresponding to the vibration condition on the fan.

Images