Fault diagnosis method for transmission chain of direct-driven wind generating set

Abstract

The invention provides a fault diagnosis method for a transmission chain of a direct-driven wind generating set. The fault diagnosis method comprises the following steps: acquiring sensing data of a plurality of measurement positions on the transmission chain of the direct-driven wind generating set; performing feature analysis on the acquired sensing data to obtain time domain feature data and frequency domain feature data; and inputting the time domain characteristic data and the frequency domain characteristic data into a pre-constructed fault diagnosis model to obtain a fault diagnosis result representing whether the transmission chain of the direct-driven wind generating set has a fault or not. The problem that accurate fault diagnosis is difficult to carry out on the transmission chain of the unit is solved, and fault diagnosis can be carried out based on more comprehensive data and characteristics by carrying out time domain and frequency domain comprehensive analysis on the data of the multiple measurement positions on the transmission chain, so that the transmission chain abnormity can be accurately identified, maintenance is carried out in time, and the fault diagnosis efficiency is improved. And the safety of the unit is ensured.

Description

Technical Field

[0001] The present disclosure relates to the field of wind power generation, and more specifically, to a fault diagnosis method for a transmission chain of a direct-drive wind power generator set. Background Art

[0002] With the rapid development of wind power generation technology and the widespread application of wind turbines, the requirements for the safety and reliability of wind turbines are becoming increasingly higher, which requires more accurate and reliable fault diagnosis solutions for components such as transmission chains.

[0003] In related technologies, the time series analysis of the effective vibration value can be used to identify whether the vibration state of the unit is abnormal, thereby performing fault diagnosis and monitoring. However, in such a scheme, due to the single analysis method, there may be major drawbacks in the identification of complex features. In particular, the simple effective vibration value can effectively analyze the long-term vibration trend of the unit, but there are certain blind spots for failure models with local abnormalities and complex features, and the model has poor adaptability and is difficult to identify failure models with complex features. Summary of the invention

[0004] In view of the problem in the related art that it is difficult to accurately diagnose faults in the transmission chain of a unit, the present disclosure provides a method for diagnosing faults in the transmission chain of a direct-drive wind turbine generator set.

[0005] A first aspect of the present disclosure provides a fault diagnosis method for a transmission chain of a direct-drive wind turbine generator set, wherein the transmission chain includes a main bearing and a generator, wherein the fault diagnosis method includes: acquiring sensing data of multiple measurement positions on the transmission chain of the direct-drive wind turbine generator set, wherein the sensing data includes vibration data, wherein the multiple measurement positions include a main bearing measurement position and a generator measurement position; performing feature analysis on the acquired sensing data to obtain time domain feature data and frequency domain feature data; inputting the time domain feature data and the frequency domain feature data into a pre-constructed fault diagnosis model to obtain a fault diagnosis result characterizing whether there is a fault in the transmission chain of the direct-drive wind turbine generator set.

[0006] Optionally, the sensing data also includes generator speed data, the main bearing measurement position includes a front bearing horizontal vibration measurement position, a front bearing vertical vibration measurement position and a front bearing axial vibration measurement position on the front bearing of the main bearing, and the generator measurement position includes a stator measurement position on the stator of the generator and a speed measurement position of the generator, wherein the sensing data of the main bearing measurement position is sensed by a first type of sensor, and the sensing data of the generator measurement position is sensed by a second type of sensor, and the sensing frequency band of the first type of sensor is within the sensing frequency band of the second type of sensor.

[0007] Optionally, the step of performing feature analysis on the acquired sensing data to obtain time domain feature data and frequency domain feature data includes: performing envelope analysis on the sensing data in the frequency domain to obtain frequency domain envelope features; performing empirical mode decomposition on the sensing data to obtain modal decomposition features, wherein the modal decomposition features include time domain modal features and frequency domain modal features; obtaining the time domain feature data based on the time domain modal features; and obtaining the frequency domain feature data based on the frequency domain envelope features and the frequency domain modal features.

[0008] Optionally, the modal decomposition characteristics are obtained in the following manner: extracting time series data within a specific time period from the sensing data; determining a data upper limit envelope and a data lower limit envelope of the time series data; performing vibration decomposition on the time series data based on the data upper limit envelope and the data lower limit envelope to obtain a plurality of vibration sub-components, wherein each vibration sub-component corresponds to a different vibration mode of the main bearing; performing time domain analysis and frequency domain analysis on each vibration sub-component to respectively obtain a time domain modal characteristic and a frequency domain modal characteristic of each vibration sub-component.

[0009] Optionally, the frequency domain envelope characteristics are obtained in the following manner: determining the frequency spectrum data of the vibration data; filtering the vibration data based on the frequency spectrum data and a preset fault vibration frequency range to obtain filtered data in the vibration data that is within the fault vibration frequency range; performing envelope analysis on the filtered data to determine the frequency domain envelope characteristics.

[0010] Optionally, the frequency domain feature data includes frequency domain analysis data, wherein the frequency domain analysis data is obtained by: determining the frequency spectrum data of the vibration data; performing high amplitude analysis and uniform interval vibration analysis on the frequency spectrum data to obtain the frequency domain analysis data, wherein the high amplitude analysis refers to determining whether there is data with an amplitude higher than a preset amplitude threshold in the frequency spectrum data, and the uniform interval vibration analysis refers to determining whether there is data with frequency peaks with equal intervals in the frequency spectrum data.

[0011] Optionally, the pre-built fault diagnosis model includes multiple models, wherein the fault diagnosis result is obtained in the following manner: based on each model of the multiple models, the sensed data is fault diagnosed to obtain a single model diagnosis result corresponding to each model; based on preset model weights, the single model diagnosis results of the corresponding models are weighted to obtain a weighted diagnosis result corresponding to each model; based on each weighted diagnosis result, the fault diagnosis result is obtained.

[0012] Optionally, the multiple models are constructed based on historical sensing data of multiple sample units at the multiple measurement locations, the multiple sample units include faulty units in multiple dimensions, the multiple dimensions include different unit dimensions, different wind farm dimensions, different unit type dimensions and different unit region dimensions, and the multiple models include statistical models, mechanism models and machine learning models.

[0013] A second aspect of the present disclosure provides an electronic device, comprising: at least one processor; and at least one memory storing processor-executable instructions, wherein the processor-executable instructions, when executed by the at least one processor, prompt the at least one processor to execute the fault diagnosis method for the transmission chain of a direct-drive wind turbine generator set according to an embodiment of the present disclosure.

[0014] A third aspect of the present disclosure provides a computer program product, which includes computer instructions. When the computer instructions are executed by a processor, the fault diagnosis method of the transmission chain of the direct-drive wind turbine generator set according to the embodiment of the present disclosure is implemented.

[0015] According to the fault diagnosis method of the transmission chain of the direct-drive wind turbine generator set disclosed in the present invention, the sensing data of multiple measurement positions on the transmission chain of the direct-drive wind turbine generator set, including the main bearing measurement position and the generator measurement position, can be obtained, and the sensing data can be feature analyzed to obtain time domain feature data and frequency domain feature data. Finally, a pre-built model can be used to diagnose whether there is a fault in the transmission chain to obtain a fault diagnosis result. In this way, by performing a comprehensive analysis of the data at multiple measurement positions on the transmission chain in the time domain and frequency domain, fault diagnosis can be performed based on more comprehensive data and features, so that the transmission chain abnormality can be accurately identified, maintenance can be performed in a timely manner, and the safety of the unit can be ensured. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a schematic flow chart showing a fault diagnosis method of a transmission chain of a direct-drive wind turbine generator system according to an exemplary embodiment of the present disclosure.

[0017] Figure 2 is a schematic diagram showing measurement positions in a fault diagnosis method for a transmission chain of a direct-drive wind turbine generator system according to an exemplary embodiment of the present disclosure.

[0018] Figure 3 Detailed description is a schematic block diagram showing an overall analysis flow of a fault diagnosis method for a transmission chain of a direct-drive wind turbine generator system according to an exemplary embodiment of the present disclosure.

[0019] Figure 4A is a schematic diagram illustrating a filtering process in a fault diagnosis method for a transmission chain of a direct-drive wind turbine generator system according to an exemplary embodiment of the present disclosure.

[0020] Figure 4B is a schematic diagram showing sliding statistics of a filtered signal in a fault diagnosis method for a transmission chain of a direct-drive wind turbine generator system according to an exemplary embodiment of the present disclosure.

[0021] Figure 5 It is a schematic flow chart showing the steps of obtaining time domain characteristic data and frequency domain characteristic data in a fault diagnosis method for a transmission chain of a direct-drive wind turbine generator set according to an exemplary embodiment of the present disclosure.

[0022] Figure 6 Schematic diagram showing a fault waveform obtained by envelope demodulation analysis of a faulty unit in a fault diagnosis method for a transmission chain of a direct-drive wind turbine generator set according to an exemplary embodiment of the present disclosure.

[0023] Figure 7 is a schematic diagram showing vibration subcomponents obtained through empirical mode decomposition in a fault diagnosis method for a transmission chain of a direct-drive wind turbine generator system according to an exemplary embodiment of the present disclosure.

[0024] Figure 8 Schematic diagram showing a time domain impact waveform of a faulty unit in a fault diagnosis method for a transmission chain of a direct-drive wind turbine generator set according to an exemplary embodiment of the present disclosure.

[0025] Fig. 9 1 is a schematic diagram showing a comparison of vibration indices of a faulty unit and a normal unit in a fault diagnosis method for a transmission chain of a direct-drive wind turbine generator set according to an exemplary embodiment of the present disclosure.

[0026] Fig. 10A and Fig. 10B Schematic diagram showing a fault waveform obtained by analyzing SCADA data in a fault diagnosis method for a transmission chain of a direct-drive wind turbine generator system according to an exemplary embodiment of the present disclosure.

[0027] Fig.11 is a schematic diagram showing a fault waveform obtained by CMS time domain and frequency domain analysis in a fault diagnosis method for a transmission chain of a direct-drive wind turbine generator set according to an exemplary embodiment of the present disclosure.

[0028] Fig.12 and Fig.13 1 is a schematic diagram showing a fault waveform obtained by performing time domain and frequency domain analysis on a faulty unit in a fault diagnosis method for a transmission chain of a direct-drive wind turbine generator set according to an exemplary embodiment of the present disclosure.

[0029] Fig.14 is a schematic block diagram illustrating model federated learning in a fault diagnosis method for a transmission chain of a direct-drive wind turbine generator system according to an exemplary embodiment of the present disclosure.

[0030] Fig.15 is a schematic flowchart illustrating model federated learning in a fault diagnosis method for a transmission chain of a direct-drive wind turbine generator system according to an exemplary embodiment of the present disclosure.

[0031] Fig.16 is a schematic diagram showing an example of feature importance in a fault diagnosis method for a transmission chain of a direct-drive wind turbine generator system according to an exemplary embodiment of the present disclosure.

[0032] Fig.17 is a schematic diagram showing the quantification of feature importance in a fault diagnosis method for a transmission chain of a direct-drive wind turbine generator system according to an exemplary embodiment of the present disclosure.

[0033] Fig.18 is a schematic diagram showing an example of characteristic depth in a fault diagnosis method for a transmission chain of a direct-drive wind turbine generator system according to an exemplary embodiment of the present disclosure.

[0034] Fig.19 is a schematic diagram showing an example of a confusion matrix in a fault diagnosis method for a transmission chain of a direct-drive wind turbine generator system according to an exemplary embodiment of the present disclosure.

[0035] Fig. 20 is a schematic diagram showing an example of visualization of model prediction results in a fault diagnosis method for a transmission chain of a direct-drive wind turbine generator system according to an exemplary embodiment of the present disclosure. DETAILED DESCRIPTION

[0036] The following specific embodiments are provided to help the reader obtain a comprehensive understanding of the methods, devices and / or systems described herein. However, after understanding the disclosure of the present application, various changes, modifications and equivalents of the methods, devices and / or systems described herein will be clear. For example, the order of operations described herein is only an example and is not limited to those orders set forth herein, but can be changed as will be clear after understanding the disclosure of the present application, except for operations that must occur in a specific order. In addition, for greater clarity and simplicity, the description of features known in the art may be omitted.

[0037] The features described herein can be implemented in different forms and should not be construed as being limited to the examples described herein. Rather, the examples described herein have been provided to illustrate only some of the many possible ways to implement the methods, devices, and / or systems described herein, which will be clear after understanding the disclosure of the present application.

[0038] As used herein, the term "and / or" includes any one of the associated listed items and any combination of any two or more.

[0039] Although terms such as "first", "second", and "third" may be used herein to describe various members, components, regions, layers, or portions, these members, components, regions, layers, or portions should not be limited by these terms. Instead, these terms are only used to distinguish one member, component, region, layer, or portion from another member, component, region, layer, or portion. Therefore, without departing from the teachings of the examples described herein, the first member, first component, first region, first layer, or first portion referred to in the examples may also be referred to as the second member, second component, second region, second layer, or second portion.

[0040] In the specification, when an element (such as a layer, a region, or a substrate) is described as being “on”, “connected to”, or “coupled to” another element, the element may be directly “on”, “connected to”, or “coupled to” another element, or one or more other elements may be present therebetween. Conversely, when an element is described as being “directly on”, “directly connected to”, or “directly coupled to” another element, there may be no other elements present therebetween.

[0041] The terms used herein are only used to describe various examples and are not intended to limit the disclosure. Unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. The terms "comprise", "include" and "have" indicate the presence of the described features, quantities, operations, components, elements and / or combinations thereof, but do not exclude the presence or addition of one or more other features, quantities, operations, components, elements and / or combinations thereof.

[0042] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as those generally understood by a person of ordinary skill in the art to which the present disclosure belongs after understanding the present disclosure. Unless explicitly defined as such herein, terms (such as those defined in a general dictionary) should be interpreted as having a meaning consistent with their meaning in the context of the relevant art and the present disclosure, and should not be interpreted in an idealized or overly formal manner.

[0043] Furthermore, in the description of examples, when it is considered that a detailed description of a well-known related structure or function would cause vague interpretation of the present disclosure, such a detailed description will be omitted.

[0044] In view of the above-mentioned problems, the present disclosure provides a fault diagnosis method for a transmission chain of a direct-drive wind turbine generator set, an electronic device and a computer program product to solve or at least alleviate the above-mentioned problems.

[0045] According to a first aspect of an embodiment of the present disclosure, a fault diagnosis method for a transmission chain of a direct-drive wind turbine generator set is provided. The method can be performed by an electronic device with computing capabilities. The electronic device can be, for example, a terminal device or a server, wherein the terminal device can be, for example, a tablet computer, a laptop computer, a digital assistant, etc.; the server can be an independent server, a server cluster, a cloud computing platform, or a virtualization center.

[0046] In an example application scenario, an electronic device can obtain sensing data from multiple measurement positions on a transmission chain of a direct-drive wind turbine generator set, wherein the sensing data includes vibration data, wherein the multiple measurement positions include a main bearing measurement position and a generator measurement position; perform feature analysis on the acquired sensing data to obtain time domain feature data and frequency domain feature data; input the time domain feature data and the frequency domain feature data into a pre-built fault diagnosis model to obtain a fault diagnosis result that characterizes whether there is a fault in the transmission chain of the direct-drive wind turbine generator set.

[0047] The electronic device can be communicatively connected to a data center or a measuring device that can provide a wind turbine generator set, so that the sensing data required to execute the above method can be obtained. According to the fault diagnosis method of the transmission chain of a direct-drive wind turbine generator set of the embodiment of the present disclosure, the data of multiple measurement positions on the transmission chain can be comprehensively analyzed in the time domain and frequency domain, and the fault diagnosis can be performed based on more comprehensive data and features, so that the abnormality of the transmission chain can be accurately identified, and maintenance can be performed in time to ensure the safety of the unit.

[0048] The following will refer to Figures 1 to 20 A fault diagnosis method for a transmission chain of a direct-drive wind turbine generator system according to an embodiment of the present disclosure is described. The method may include the following steps:

[0049] like Figure 1 As shown, in step S110, sensing data of multiple measurement positions on the transmission chain of the direct-drive wind turbine generator system may be acquired.

[0050] Here, the sensed data may include vibration data, and the plurality of measurement positions may include a main bearing measurement position and a generator measurement position.

[0051] Specifically, a direct-drive wind turbine generator set may include blades, hubs, pitch control systems, main bearings, generator stators, wind measurement systems, hoists, bases, generator rotors, yaws, nacelles, cooling systems, converters, impeller systems, control systems and lightning protection devices. The transmission chain of the unit may include main bearings and generators. The main bearing is the main component that supports the weight and centrifugal force of the generator rotor, so accurate and timely fault monitoring of the main bearing is particularly important. In particular, for large-megawatt direct-drive wind turbine generator sets, the diameter of the main bearing may exceed 3 meters, and its operating status directly affects the stability and safety of the entire unit.

[0052] In the embodiments of the present disclosure, a transmission chain monitoring system (CMS) may be provided. Specifically, the transmission chain monitoring system may collect vibration signals of the transmission chain and utilize intelligent models to extract and analyze key information in the vibration signals of each component, thereby realizing status monitoring and fault warning of key components. For example, the common failure modes and failure types of key components such as generators and main bearings in transmission chains of different models and configurations may be monitored and warned, thereby effectively avoiding the occurrence of serious accidents and reducing unnecessary economic losses.

[0053] Here, measurement positions can be set on the main bearing and the generator respectively to monitor the operating status of the main bearing. Figure 2 As shown, the main bearing measurement position may include a front bearing horizontal vibration measurement position 1H, a front bearing vertical vibration measurement position 1V and a front bearing axial vibration measurement position 1A on the front bearing of the main bearing; the generator measurement position includes a stator measurement position 3H on the stator of the generator and a generator speed measurement position 1N; in a dual-bearing unit, the main bearing measurement position may also include a rear bearing vertical vibration measurement position 2V.

[0054] The sensing data can be obtained by sensing the sensors arranged at the above-mentioned measuring positions of the transmission chain. For example, the sensors can be arranged corresponding to the above-mentioned measuring positions, so that the sensing data can cover multiple positions such as the horizontal radial, vertical radial, axial, front bearing, and rear bearing of the direct-drive unit transmission chain, so that the vibration characteristics of different dimensions of the transmission chain can be monitored and early warning can be performed.

[0055] As an example, when the generator measurement position includes the generator speed measurement position 1N, the sensing data may also include generator speed data. The generator speed data may be acquired in time sequence with the vibration data, for example. In this way, the operating condition of the unit during the measurement process may be determined based on the generator speed data, thereby performing a more accurate analysis of the vibration data.

[0056] In addition, in the embodiments of the present disclosure, different sensors may be provided for different measurement positions and different measurement objects to sense data more accurately. For example, the sensing data of the main bearing measurement position may be obtained by sensing the first type of sensor, and the sensing data of the generator measurement position may be obtained by sensing the second type of sensor. Here, the sensing frequency band of the first type of sensor may be within the sensing frequency band of the second type of sensor.

[0057] For example, for vibration sensors used to sense vibration data, two types of sensors can be used: a wideband accelerometer and a low-frequency accelerometer. The wideband accelerometer can have a sensitivity of 100mv / g and a measurement range of ±80g; the low-frequency accelerometer can have a sensitivity of 500mv / g and a measurement range of ±10g. In addition, the sensor can be selected based on parameters such as the frequency range and operating temperature of the sensor.

[0058] As an example, as shown in Table 1 below, a direct-drive unit can use 4 low-frequency vibration sensors and 1 through-frequency vibration sensor. Since the direct-drive unit does not have gearbox-related structural components (such as input shaft, primary gear ring, secondary gear ring, tertiary gear ring, medium-speed shaft, high-speed shaft, etc.) compared to the semi-direct-drive unit and the doubly-fed unit, the direct-drive unit can reduce 3 low-frequency sensors and 7 through-frequency sensors respectively compared to the semi-direct-drive unit and the doubly-fed unit. Here, in the direct-drive unit, the main bearing measurement position can use a low-frequency sensor to sense vibration data, and the sensor can be installed on the front bearing to measure the vibration characteristics of the front bearing horizontal vibration, the front bearing vertical vibration, and the front bearing axial vibration; the generator stator measurement position can use a through-frequency sensor to sense the stator vibration data, one is located near the stator flange to measure the vertical radial vibration of the main bearing, and one sensor is located near the rear bearing of the generator.

[0059] Table 1

[0060]

[0061] According to the embodiments of the present disclosure, by measuring the vibration data on the main bearing and the stator data and speed data on the generator, the unit operating condition in which the corresponding vibration data is located can be determined based on the stator data and speed data, thereby facilitating a more accurate analysis of the features in the vibration data and determining whether there is a transmission chain fault. The process of analyzing the transmission chain vibration data in combination with the stator data and speed data will be described in detail below.

[0062] In step S120, feature analysis may be performed on the acquired sensing data to obtain time domain feature data and frequency domain feature data.

[0063] In this step, multiple feature analyses may be performed on the sensing data to obtain time domain features and frequency domain features to comprehensively analyze fault indicators of the sensing data.

[0064] Specifically, the vibration data can be synthesized from a series of vibration waveforms, each of which represents or corresponds to the vibration frequency of a component or a vibration form in the transmission chain, and the algebraic addition of these individual vibration waveforms obtains the waveform of the total vibration, such as the sensed vibration data.

[0065] Figure 3 FIG. 1 is a schematic diagram showing an example of the overall process of the fault diagnosis method according to an embodiment of the present disclosure. Figure 3 As shown, the data can be accurately processed through data analysis and processing, thereby improving the accuracy and timeliness of the analysis results. Through data analysis, the parameter characteristics of the time domain and frequency domain of the vibration data of each channel (such as the above-mentioned measurement positions) can be calculated, and the abnormal characteristics (or patterns) can be analyzed and predicted through appropriate algorithms and integration of expert experience. In this way, a fault diagnosis conclusion can be drawn based on the analysis results. For example, a comprehensive judgment can be made based on the vibration diagnosis results combined with the diagnosis results obtained in various ways. In the above process, data analysis can help to extract the vibration waveforms corresponding to the various components on the transmission chain or the various vibration forms of the transmission chain, and analyze the fault characteristics from the vibration data for subsequent modeling and diagnosis.

[0066] The following will be combined Figure 3 The overall process of the data analysis process is described in detail. Figure 3 The overall process example of the fault diagnosis method is shown, but the embodiments of the present disclosure are not limited to Figure 3 , for example, you can also delete, replace, or add individual steps or operations in the process.

[0067] Before describing the steps of data analysis, the process of data parsing and processing is illustrated.

[0068] As an example, during the data parsing and processing process, the raw sensed data can be parsed, processed and analyzed through tools such as PYTHON, R, and Dbeaver.

[0069] Data analysis and processing may include, for example, data analysis, data preprocessing and extraction, data aggregation, data filtering, etc.

[0070] Data analysis may include, for example, online automatic analysis and offline manual analysis. Online automatic analysis may directly transmit data from the wind farm back to the big data platform for automatic analysis, storage, calling and analysis. In the embodiments of the present disclosure, offline data analysis and processing are mainly described.

[0071] During offline data analysis and processing, the original data may be data in the form of CMS sensors or other types of database formats. For example, the data in the form of databases may be converted into data in common formats, such as excel, csv, txt, etc., through tools such as PYTHON, R, and Dbeaver for subsequent processing.

[0072] During data preprocessing and extraction, vibration data that is obviously beyond the predetermined normal vibration range, or is obviously lower than the predetermined normal amplitude, or is continuously unchanged can be eliminated. Data that is obviously not in line with the normal vibration amplitude range of the corresponding model and the corresponding sensor measurement position can also be eliminated. For example, the acceleration data measured by the radial sensor of the outer ring of the front bearing corresponding to a certain model is continuously unchanged and lower than 0.03m / s 2 This vibration feature obviously does not conform to the normal data feature of the unit during operation, so this part of the data can be eliminated.

[0073] During the data aggregation process, the original signal waveform data can be aggregated into different indicators according to different time dimensions. For example, the data set can be aggregated according to the aggregation time of every 1 minute (min) to obtain the corresponding effective value, average value, peak-to-peak value, etc.

[0074] In the analysis of the sensing data, for example, a bandpass filter can be used to perform bandpass filtering on the sensing data (such as vibration data). During the bandpass filtering process, for example, signals within a certain frequency range can be allowed to pass through, retaining vibration signals suitable for analyzing specific frequency components, and suppressing frequencies that are not of concern to a value close to 0. In this process, the amount of data will not decrease.

[0075] Here, in the process of configuring the bandpass filter parameters, the algorithms and parameters of multiple filters can be tested and verified to select the optimal filter design. For example, by analyzing and verifying the failure data of specific bearing components of the direct-drive unit, it can be found that using a 4th-order bandpass Butterworth filter to perform bandpass filtering on preset frequencies such as 10-1000HZ can more clearly reflect the frequency domain fault characteristics of the component. Figure 4A A schematic diagram of an example filtering result for comprehensive verification and early warning through a filter is shown, wherein the horizontal axis is the vibration frequency (in Hertz Hz) and the vertical axis is the normalized amplitude (dimensionless).

[0076] Butterworth filter is an electronic filter, also known as a maximum flat filter. The characteristic of Butterworth filter is that the frequency response curve in the passband is maximally flat, without ripples, and gradually decreases to zero in the stopband. Butterworth low-pass filter can be expressed as the square of the amplitude |H(ω)| 2 To express:

[0077]

[0078] Where n = filter order, ω c = Cut-off frequency = frequency at which the amplitude drops to -3 dB, ω p = passband edge frequency.

[0079] In addition, as an example, the parameter configuration of the filtering process can also take into account the operating conditions corresponding to the sensing data. For example, based on the sensing data of the generator measurement position, the vibration data under operating conditions such as the grid-connected state at a preset speed and / or preset power can be determined.

[0080] In the bandpass filtering process, the bandpass range may include a frequency band range of 10-1000HZ, 600-1200HZ and / or 50-800HZ. Specifically, the fault characteristic frequency of the faulty unit is within the above frequency band range, so the timing characteristics and spectrum characteristics can be evaluated based on the vibration data within the range to see whether they are obvious and prominent, thereby performing fault diagnosis. In addition, it can also be determined that within a predetermined generator speed band (e.g., 8-9RPM), the fault characteristic frequency is more prominent. If the effect is obvious, the next step of analysis can be performed directly; secondly, if it is not obvious, the filtering range can be further limited and tried. As an example, the filter order can be selected as 4th order, or 2-6th order or other filters can be selected for bandpass filtering according to the predetermined machine model or abnormal vibration mode.

[0081] Examples of various analysis methods for performing data analysis will be described in detail below.

[0082] As an example, in step S120, the time domain feature data and the frequency domain feature data may be obtained in the following manner:

[0083] like Figure 5 As shown, in step S510, envelope analysis may be performed on the sensing data in the frequency domain to obtain frequency domain envelope features.

[0084] In step S520, empirical mode decomposition may be performed on the sensing data to obtain modal decomposition features. Here, the modal decomposition features may include time domain modal features and frequency domain modal features.

[0085] In step S530, time domain feature data may be obtained based on the time domain modal features.

[0086] In step S540, frequency domain feature data may be obtained based on the frequency domain envelope features and the frequency domain modal features.

[0087] pass Figure 5 The example process shown can combine frequency domain envelope analysis and empirical mode decomposition to obtain time domain feature data and frequency domain feature data, wherein the time domain feature data can be derived from empirical mode decomposition, and the frequency domain feature data can be derived from both envelope analysis and empirical mode decomposition. In this way, the time domain and frequency domain characteristics of the vibration data can be reflected, so as to facilitate the evaluation of whether vibration abnormalities occur from multiple dimensions, thereby determining whether there is a fault in the transmission chain.

[0088] The example processes of envelope analysis and empirical mode decomposition are described in detail below.

[0089] For envelope analysis, after a rolling bearing fails, modulation phenomenon will generally be manifested. The manifestation of modulation phenomenon in the spectrum can include frequency components with equal intervals on both sides of the center frequency, which can also be called side frequencies or sidebands. The sidebands and their amplitudes can be used as indicators for fault judgment. The carrier signal (or frequency signal) is complex and difficult to demodulate and analyze. There are obvious modulation phenomena in multiple frequency bands, and multiple sidebands appear. Due to the existence of multiple modulation frequencies (such as sum frequency and difference frequency), mixed modulation, etc., the distribution of sidebands is extremely irregular, and it is difficult to demodulate the modulation frequency directly from the sidebands, so envelope analysis diagnosis is required.

[0090] As an example, the frequency domain envelope characteristics can be obtained in the following ways: the spectrum data of the vibration data can be determined; the vibration data can be filtered based on the spectrum data and a preset fault vibration frequency range to obtain filtered data in the vibration data that is within the fault vibration frequency range; and envelope analysis can be performed on the filtered data to determine the frequency domain envelope characteristics.

[0091] Specifically, the original signal of the vibration data can be subjected to fast Fourier transform (FFT) analysis to obtain spectrum data. As an example, the spectrum data can be subjected to frequency band screening, for example, the frequency segment whose peak is located at the preset fault frequency can be screened out, or a specific frequency band can be modulated and screened out, etc. Based on the spectrum data and the possible vibration frequency band range or the frequency band range that may be involved in the bearing failure (for example, it can be the above-mentioned frequency segment or specific frequency band), the vibration data is filtered, and the data within these vibration frequency bands is screened out to obtain the above-mentioned filtered data. As an example, for the main bearing of the direct-drive unit, the screened frequency band can be 100-400Hz. For the filtered data, the filtered data can be transformed and analyzed by algorithms such as the Hilbert transform, the envelope can be extracted, and the envelope FFT can be calculated to obtain the above-mentioned frequency domain envelope features for fault identification.

[0092] In mathematics and signal processing, the Hilbert transform is a linear operator that produces a function with the same domain as a function. The Hilbert transform of u can be considered as the convolution of u(t) with the function h(t) = 1 / (πt). Since h(t) is not integrable, the integral defining the convolution does not converge. Therefore, the Hilbert transform can be defined using the Cauchy principal value (denoted here as pv). To be precise, the Hilbert transform of a function (or signal) μ(t) can be expressed as follows:

[0093]

[0094] Here, t represents time, and T represents the period of the signal. The essence of the Hilbert transform is to use the filter in the imaginary circuit to realize the analysis of the signal, that is, to convert a real-valued signal into a complex-valued signal, which contains all the information of the original signal, including its phase, amplitude and frequency. Through the Hilbert transform, the envelope and phase information in the signal can be separated. Therefore, in the embodiment of the present disclosure, the use of the Hilbert transform can help to more accurately separate the vibration information.

[0095] As an example, the frequency domain envelope feature can be further calculated and plotted by an envelope analysis method such as Mod to obtain an envelope curve, and the frequency domain envelope feature or the envelope curve can reflect the frequency domain characteristics of the vibration data, so as to facilitate further frequency domain analysis. Figure 6 An example of a fault waveform obtained by envelope demodulation analysis of a faulty unit is shown, where Figure 6 As shown in the figure above, regular (or equally spaced) frequency peaks appear in the envelope, and these peaks can be considered as the characteristic frequencies of the main bearing fault.

[0096] For empirical mode decomposition, the empirical mode decomposition (EMD) method is an adaptive, data-driven decomposition method and signal time-frequency processing method, which is particularly suitable for the analysis and processing of nonlinear and non-stationary signals. Empirical mode decomposition is essentially a method of decomposing signals, decomposing signals into superpositions of independent components, and decomposing signals based on the time scale characteristics of the data itself. For each separated component, spectrum analysis, statistical model, and machine learning analysis and prediction are performed to extract fault characteristics, fault levels, etc., and the results are fused and superimposed. Examples of vibration sub-components extracted by empirical mode decomposition of the vibration data of a faulty unit are as follows: Figure 7 shown.

[0097] This method extracts intrinsic mode functions (IMFs) from the original signal in an iterative manner. These IMFs reflect the oscillation modes of the signal at different time scales. Since most of the data to be analyzed are not intrinsic mode functions, at any point in time, the data may contain multiple fluctuation modes. Therefore, a simple Hilbert transform cannot fully characterize the frequency characteristics of general data. In this regard, in an embodiment of the present disclosure, the eigenmode function can be obtained by performing EMD decomposition on the original data. After obtaining different components of the vibration signal, the corresponding statistical model and machine learning model analysis are performed, and further improved and optimized to extract more obvious abnormal features.

[0098] As an example, the modal decomposition characteristics can be obtained in the following manner: extracting time series data within a specific time period from the sensing data; determining the data upper limit envelope and the data lower limit envelope of the time series data; based on the data upper limit envelope and the data lower limit envelope, performing vibration decomposition on the time series data to obtain multiple vibration sub-components; performing time domain analysis and frequency domain analysis on each vibration sub-component to obtain the time domain modal characteristics and frequency domain modal characteristics of each vibration sub-component, respectively.

[0099] Here, each vibrator component may correspond to a different vibration mode of the main bearing, and the vibration mode may be, for example, the vibration of different components in the transmission chain.

[0100] For example, the amplitude maximum and minimum values ​​within a specific time period can be extracted from the original signal of the vibration data. As an example, there can be multiple specific time periods, and the corresponding maximum and minimum values ​​can be obtained for each time period. Based on these maximum and minimum values, the data upper envelope and the data lower envelope can be constructed. Based on the data upper envelope and the data lower envelope, the vibration can be decomposed in the time domain. For example, the mean function between the data upper envelope and the data lower envelope can be calculated, and the mean function can be iteratively decomposed to obtain the intrinsic mode function (IMF), and the vibration decomposition signal can be obtained. For example, each component of the above-mentioned multiple vibration sub-components can correspond to a corresponding IMF. Further sub-component analysis can be performed on each vibration sub-component (such as each IMF) to obtain the time domain modal characteristics and frequency domain modal characteristics of each vibration sub-component. Here, the sub-component analysis can include, for example, signal reconstruction, denoising, IMF frequency analysis, feature extraction, signal prediction, vibration mode recognition, vibration anomaly detection, multi-dimensional comparative analysis, frequency component and its time dimension change analysis, etc.

[0101] By obtaining the modal decomposition features through the above-mentioned empirical mode decomposition, the total vibration data can be decomposed to obtain the vibration sub-components under different vibration modes of the main bearing, thereby facilitating the evaluation of the specific fault type of the main bearing. For example, different fault types can correspond to different vibration modes.

[0102] In addition, as an example, the frequency domain modal features may include modal matching features. Specifically, for each vibration sub-component, the modal matching features can be obtained in the following manner: the frequency of the vibration sub-component is matched with the preset fault frequencies of multiple components in the transmission chain (such as the main bearing) to obtain a frequency matching result; in response to the frequency matching result indicating that the frequency of the vibration sub-component matches the preset fault frequency of a target component among the multiple components, a modal matching feature is generated, wherein the modal matching feature characterizes that the vibration sub-component is related to the fault of the target component.

[0103] Here, the modal matching feature may include marking the frequency of the vibration subcomponent, wherein the marking of the frequency indicates that the vibration subcomponent matches the fault frequency of the target component. In addition, the above-mentioned multiple components may include, for example, an inner ring, an outer ring, a cage, a rolling element, and the like.

[0104] In the above process, frequency matching may refer to that the difference between the frequency of the vibration subcomponent and the preset fault frequency of the component is less than a preset threshold value, and the preset threshold value may be 1 Hz, for example.

[0105] In addition, as an example, the modal matching feature may also include a mark of the amplitude of the vibration subcomponent, and the mark of the amplitude indicates that the vibration subcomponent matches the target fault level. Here, multiple fault levels may be preset for each component, each fault level corresponds to a vibration amplitude range, and multiple fault levels correspond to different vibration amplitude ranges. Different fault levels correspond to different fault intensities.

[0106] Specifically, the modal matching feature can be obtained in the following manner: in response to the frequency matching result indicating that the frequency of the vibration sub-component matches the preset fault frequency of the target component, the vibration sub-component can be amplitude matched with the vibration amplitude range corresponding to the preset fault level of the target component to obtain an amplitude matching result; based on the frequency matching result and the amplitude matching result, the modal matching feature is generated.

[0107] Here, the amplitude matching result can represent the target fault level corresponding to the vibration subcomponent.

[0108] As an example, the modal matching feature can be input into a subsequent model as part of the frequency domain feature data together with the time domain feature data and other frequency domain feature data (such as frequency domain envelope features and frequency domain analysis data for the original vibration data) for fault diagnosis. In this way, the modal matching feature can help identify the fault characteristics of specific components in the transmission chain and quantify the fault intensity for fault level classification.

[0109] The above describes an example of using envelope analysis and empirical mode decomposition in the frequency domain to perform data feature analysis. In the embodiments of the present disclosure, time domain analysis and frequency domain analysis can also be used. The time domain analysis and frequency domain analysis will be described exemplarily below.

[0110] Time domain analysis is a method to directly observe the changes in the time history of vibration signals. By performing time domain analysis on vibration signals, we can understand the amplitude, frequency, phase and other information of the unit vibration, and analyze the abnormal increase in time domain periodicity, intermittent impact, and vibration energy (RMS) trend.

[0111] As an example, predetermined time domain analysis indicators of vibration data may be calculated, and the time domain analysis indicators may include, for example, root mean square value, peak value, effective value, absolute amplitude, kurtosis value, etc. These indicators may reflect the difference between a faulty unit and a normal unit, and therefore, fault diagnosis may be performed based on these indicators.

[0112] For time domain analysis indicators, the larger the amplitude of the indicator, the more abnormal the vibration. For example, Figure 8 The timing diagram shows the raw vibration data waveform examples of the faulty unit and the normal unit, Fig. 9An example of comparison of the effective value, absolute amplitude, and kurtosis value of the faulty unit and the normal unit is shown. Figure 8 In the data, the sudden change shows vibration shock, which is much higher than the index value of other normal units, so it can be judged as an abnormal feature. Fig. 9 The comparison of effective value, absolute value and kurtosis value of vibration between faulty unit and normal unit is shown. Fig. 9 It can be seen that the amplitudes of various indicators of the faulty unit are much higher than those of the normal unit. Taking the effective value amplitude as an example, Fig. 9 The faulty unit exceeds 0.075m / s 2 , while the normal unit does not exceed 0.050m / s 2 The above comparison shows the typical characteristics of the faulty unit, which can be used to determine whether the unit is faulty and the severity of the fault. The above time domain analysis indicators are only examples, and other time domain values ​​can also be used as indicators.

[0113] In addition, as an example, an improved filter can be used in the embodiments of the present disclosure. Specifically, when using a filter such as a 4th-order bandpass Butterworth filter, after bandpass filtering a preset frequency such as 10-1000 Hz, the filtered signal can be subjected to sliding statistics according to the window length and step size to obtain time domain analysis indicators for multiple sliding windows.

[0114] Specifically, the window length can refer to the number of elements contained in each window. For example, a window of length 3 means that each window contains 3 consecutive elements. In sliding statistics, the step size can refer to the number of elements that the window moves forward each time the window is moved. The size of the step size determines the speed and frequency of the window movement.

[0115] In this example, if Figure 4B As shown, the sliding statistics of the filtered signal can be performed in the following manner: determine the window length and step size; create an initial window (e.g. Figure 4B The sliding window of the first moment in ); by looping through the time series data to be counted, each time the window is moved by the number of elements of the step size (for example, moving to Figure 4B The time domain analysis index is obtained by performing statistics on each window to obtain the time domain analysis index for the multiple sliding windows. Here, the statistics may include, for example, determining the root mean square value, peak value, effective value, absolute amplitude, kurtosis value, average value, maximum value, minimum value, etc. in each sliding window, or performing data preprocessing, feature extraction, etc. for each sliding window.

[0116] Here, the initial window can be represented by a list, an array or other data structures, for example, the window length can be 14, and the step length can be 5. In the above manner, when counting various indicators, the data can be made smoother, effectively reducing the interference of noise on the signal and the impact on the analysis effect of the model.

[0117] In frequency domain analysis, for example, a time domain signal can be converted into a frequency domain signal through Fourier transform (FFT), and various frequency components are arranged from low to high in frequency with vibration amplitude as the vertical coordinate and frequency as the horizontal coordinate to obtain a spectrum diagram for frequency domain analysis.

[0118] Specifically, the frequency domain feature data may also include frequency domain analysis data, wherein the frequency domain analysis data may be obtained by: determining the frequency spectrum data of the vibration data; performing high amplitude analysis and uniform interval vibration analysis on the frequency spectrum data to obtain the frequency domain analysis data.

[0119] Here, high amplitude analysis may refer to determining whether there is data with an amplitude higher than a preset amplitude threshold in the spectrum data, and evenly spaced vibration analysis may refer to determining whether there is data with frequency peaks at equal intervals in the spectrum data.

[0120] For high amplitude analysis, when the vibration amplitude of the unit or bearing is large, the vibration amplitude may be relatively high. Here, high amplitude can refer to relatively high and absolutely high. Relative high is determined by comparing with the preset reference benchmark amplitude, and absolute high is determined by comparing with the preset fixed threshold. For example, under specific working conditions of a specific model, the effective value calculated by the vibration signal exceeds 0.5r / m after filtering, calculation and aggregation. 2 , which may be higher than most units.

[0121] For uniform interval vibration analysis, uniform interval (or “equal interval”) may, for example, mean that the intervals of frequency peaks or sub-peaks exceeding a predetermined number are equal.

[0122] In frequency domain analysis, data related to preset modes or preset features can be searched, such as searching for high-amplitude peaks and evenly spaced series of peaks in the spectrum data, and matching the data with the fault mode of interest. Fault modes may include, for example, locating ring movement, raceway peeling, micro-motion wear, ring rib fracture, grease leakage, high temperature, generator bearing jamming, etc. Failure features corresponding to these fault modes can be preset, so as to analyze the frequency spectrum of the vibration data, determine whether these failure features exist, and thus determine the specific fault mode or fault type.

[0123] In the above example, the preset mode may include the vibration mode of the main bearing under different models, different manufacturers, different structures and other dimensions, for example, including: under specific working conditions (for example, near full power generation of grid connection, near speed of 8-9RPM, etc.), the passing frequency of the bearing rolling element through the inner ring or outer ring of the bearing, the frequency of the cage, the frequency of the inner ring, the frequency of the outer ring, etc. The data characteristics of the bearing abnormality under specific working conditions at different serious stages are different, that is, the trend changes in time series. As an example, the preset mode may include a single mode, or a combination of multiple superimposed modes.

[0124] In frequency domain analysis, the occurrence and development of faults generally cause frequency changes in vibration signals, such as the generation of new frequencies and the increase of original frequencies as the fault worsens. Therefore, the type and severity of the fault can be determined by confirming the characteristics and changes of the data and analyzing the relationship between the main vibration frequency and the operating frequency.

[0125] In addition, as an example, the severity of the fault can be determined based on the increase in the frequency amplitude in the vibration data. Specifically, relative to the vibration frequency in a normal state, when a fault occurs, the vibration frequency will increase, and as the severity of the fault increases, the frequency will further increase. In this regard, for example, based on a preset reference vibration frequency, the reference vibration frequency can be compared with the frequency in the current vibration data, and the current fault degree can be determined based on the difference between the two. For example, multiple difference intervals can be preset, each difference interval corresponds to a different degree of fault. In general, the larger the difference, the more serious the frequency offset and the higher the degree of fault.

[0126] like Fig. 10A and Fig. 10B As shown in the figure, the direct-drive model 47# unit reported a high bearing temperature fault on March 26. The SCADA data and CMS vibration data of the unit were analyzed, and there was an obvious fault characteristic frequency of the bearing-related components (inner ring, outer ring, retainer, etc.), specifically 3.75HZ. As the fault time approaches, the bearing temperature shows abnormal temperature characteristics such as intermittent overheating, outliers, and a larger temperature rise rate. For example, the temperature of the outer ring of the bearing exceeds 60 degrees under certain working conditions. The severity of the unit abnormality can be determined through the analysis results of temperature, vibration, etc. For example, when the fault frequency occurs, it can be determined that the main bearing is abnormal. When the frequency amplitude increases, the severity of the main bearing can be determined. Fig. 10A and Fig. 10B The example is a visualization of the vibration, temperature and other characteristics of fault cases of different models, different loss parts and different severity levels. Fig. 10A and Fig. 10B It can be seen that the temperature data has intermittent mutations and abnormal increases over time, so it is judged that the main bearing is faulty.

[0127] Fig.11 It is a schematic diagram showing the fault waveforms obtained by CMS time domain and frequency domain analysis in the fault diagnosis method of the transmission chain of a direct-drive wind turbine generator set according to an exemplary embodiment of the present disclosure, which includes the time domain and frequency domain waveforms of the original signal, the time domain and frequency domain waveforms of the filtered signal, and the time domain and frequency domain waveforms of the envelope signal.

[0128] Fig.12 The waveform diagram of the abnormality of the main bearing of the generator in the 11# unit of a certain model of a certain project is shown. It can be seen that there is a certain overlap in the abnormal frequencies of the various components of the main bearing, and the fault characteristic frequency is abnormally obvious. Fig.13 The main bearing abnormality in the 5# unit of a certain project is shown. It can be seen that there are frequency characteristics of the main shaft front bearing fault in the spectrum, and the time domain impact characteristics of the main bearing measurement point are prominent under low speed conditions.

[0129] In step S130, the time domain characteristic data and the frequency domain characteristic data may be input into a pre-built fault diagnosis model to obtain a fault diagnosis result characterizing whether there is a fault in the transmission chain of the direct-drive wind turbine generator set.

[0130] In the above step S120, the sensing data (such as vibration data) is analyzed by a variety of analysis methods to obtain corresponding characteristic indicators (such as time domain characteristic data and frequency domain characteristic data), which can reflect the current state of the transmission chain of the unit, such as normal or abnormal. In this step S130, the time domain characteristic data and the frequency domain characteristic data can be input into a pre-built fault diagnosis model to obtain a fault diagnosis result that characterizes whether there is a fault in the transmission chain of the direct-drive wind turbine generator set.

[0131] As one example, only the time domain feature data and the frequency domain feature data may be input into the model, and the model may be used to output the fault diagnosis result; as another example, the time domain feature data and the frequency domain feature data may be input into the model together with the original sensing data to obtain the fault diagnosis result.

[0132] As an example, historical sensing data of multiple sample units at multiple measurement locations can be obtained, and the faulty units and normal units of these historical sensing data can be labeled, so that a fault diagnosis model can be constructed using the historical sensing data and the corresponding annotations. Here, the fault diagnosis model can include, for example, a statistical model, a mechanism model, and a machine learning model.

[0133] like Figure 3As shown, for example, historical sensing data can be used to perform statistical analysis based on statistical methods to obtain statistical indicators to build a statistical model; for another example, historical sensing data can be used to train a machine learning model using supervised learning, unsupervised learning, or reinforcement learning; for another example, historical sensing data can be used to train a deep learning model, such as a convolutional neural network, a recurrent neural network, etc.

[0134] Here, in the process of building the model, Figure 3 As shown, data analysis can be performed on the historical sensing data (such as the above-mentioned envelope analysis, empirical mode decomposition, frequency domain analysis, and time domain analysis, etc.) to obtain various analysis indicators (such as the above-mentioned frequency domain feature data and time domain feature data). A model can be constructed based on these analysis indicators or based on both the analysis indicators and the original sensing data. For example, these data can be input into a machine learning model for model training, so that the model can learn the fault characteristics in these historical data for use in fault diagnosis after training.

[0135] In addition, as an example, a pre-built fault diagnosis model may include multiple models. For multiple models, fault diagnosis results can be obtained in the following ways: based on each model in the multiple models, fault diagnosis is performed on the sensing data to obtain a single model diagnosis result corresponding to each model; based on preset model weights, the single model diagnosis results of the corresponding models are weighted to obtain a weighted diagnosis result corresponding to each model; based on each weighted diagnosis result, a fault diagnosis result is obtained.

[0136] As an example, the preset model weight may be an average weight. As another example, the preset model weight may also be obtained by testing each model using test data, for example, the accuracy of each model in fault diagnosis of different units may be tested, and the weight of each model may be determined according to the accuracy, specifically, a model with better prediction performance and higher accuracy in the test data may be given a higher weight.

[0137] As an example, the weighted diagnosis results may be combined or intersected to determine the final fault diagnosis result.

[0138] In this way, multiple models can be used for fault diagnosis, thus avoiding the inaccurate and insufficient diagnostic results of a single model, balancing the capabilities of different models in the diagnostic process, and obtaining more comprehensive and comprehensive fault diagnosis results, which is conducive to improving the accuracy of the final diagnostic results.

[0139] In addition, the multiple models described here can be constructed based on the historical sensing data of multiple sample groups at the multiple measurement locations described above. The multiple models can include, for example, statistical models, mechanism models, and machine learning models.

[0140] Here, in order to improve the model's ability to identify faults, multiple sample units may include faulty units in multiple dimensions. The multiple dimensions mentioned here may include dimensions of different units, dimensions of different wind farms, dimensions of different unit types, and dimensions of regions to which different units belong. Different unit dimensions may refer to different models of units of the same type, so as to reflect fault characteristics from the individual dimensions of the units; different wind farm dimensions may refer to different geographical locations of wind farms and / or different scales of wind farms. Compared with the unit dimension, the wind farm dimension can reflect fault characteristics from the level of the unit group of the wind farm; different unit region dimensions may refer to different geographical locations of the units and / or different environmental characteristics of the units (such as temperature, wind speed, humidity, etc.), thereby introducing the potential connection between geographical location and environmental factors and fault characteristics.

[0141] By selecting samples of different dimensions as mentioned above, sample data of multiple dimensions can be taken into account when building the model, so that the model can learn or introduce richer sample data, which is conducive to improving the diagnostic effect of the model.

[0142] As an example, in an embodiment of the present disclosure, a federated machine learning approach may be used to build a model and perform fault prediction.

[0143] Federated machine learning (Federated Machine Learning) can also be called federated learning, joint learning or federated learning. Federated machine learning is a machine learning framework that can effectively help multiple institutions use data and conduct machine learning modeling while meeting the requirements of user privacy protection, data security and relevant regulations.

[0144] In the embodiments of the present disclosure, from the sample data side, a federated learning model can be constructed by at least one of the following methods:

[0145] Based on different units (such as 1#, 2#, 3# and other units of different models), a random forest federated learning model framework is constructed between units;

[0146] Based on different wind farms, a random forest federated learning model framework is constructed between wind farms;

[0147] Based on different models (such as 2MW model, 16MW model, etc.), a random forest federated learning model framework is built between models;

[0148] Based on different users (such as user A, user B, etc.), a random forest federated learning model framework is constructed between users;

[0149] Based on different geographical regions (such as North China, Northwest China, etc.), a random forest federated learning model framework is constructed between geographical regions.

[0150] From the model side, a random forest federated learning model framework can be built between models based on different models (such as SVM, RANDOM FOREST, etc.).

[0151] Here, the Random Forest algorithm is a classifier that contains many decision trees. It can be used to deal with classification and regression problems, as well as dimensionality reduction problems. The algorithm is also very tolerant to outliers and noise, and has better prediction and classification performance than decision trees. The core concepts of random forests are randomness and integration, which include the randomness of samples and features. Based on the results of multiple decision trees, a comprehensive scoring vote is performed to obtain a comprehensive model with good stability and generalization.

[0152] Fig.14 The application logic of federated learning in the above fault diagnosis process is shown. Fig.15 FIG. 4 shows an example process of applying federated learning to the above fault diagnosis process. Fig.14 and Fig.15 As shown, through federated learning, the original sensing data, time domain feature parameters (including EMD component results), frequency domain feature parameters (including envelope demodulation, EMD component results) can be trained and tuned separately on different units, different wind farms, different models, different users, different regions, different models and other different targets, and then the average weight method is used to perform fusion testing and verification on the model. The fused model is tested, verified, and deployed, which can accurately predict and warn the working conditions and scenarios of multiple targets.

[0153] By verifying the data of faulty direct-drive units and normal direct-drive units, the accuracy of the model can reach more than 90%. The importance of some features of the model is sorted, quantified, and illustrated, which can effectively support business decision-making.

[0154] Figures 16 to 20 The performance of the model constructed using the fault diagnosis method according to the embodiment of the present disclosure is shown. Specifically, Fig.16 The figure shows the ranking of the importance of features (or indicators). The vertical axis is the feature name, and the larger the horizontal axis value, the greater the impact of the feature on the accuracy of the model in identifying positive and negative cases. Fig.17 The importance of the features is shown, and the importance of each feature is quantified, where the larger the number, the more important the feature. Fig.18The minimum depth distribution of each feature is shown. From this figure, we can see the influence and importance of each feature on the prediction results, so that we can quickly lock in and optimize important features and improve model performance. The horizontal axis represents the number of trees and the vertical axis represents the minimum depth. Here, the smaller the vertical axis value is, the smaller the number of splits and depth in the model is, and the more important it is to the model. Fig.19 The confusion matrix shows the cases where the model predicts accurately or incorrectly. The rows represent the model predicting that the unit is a normal unit (0) or an abnormal unit (1), and the columns represent the actual situation of the unit, which is a normal unit (0) or an abnormal unit (1). Through this confusion matrix, we can observe whether the model predicts that the unit is abnormal (1) when the unit is actually abnormal (1). Fig. 20 The prediction results of the model are shown from a three-dimensional perspective. The three-dimensional view can be used to compare the characteristics of the vibration data of normal units and abnormal units in different indicators in multiple dimensions. The data analysis and results shown in the embodiments of the present disclosure are all data of direct-drive units with actual faults.

[0155] Wind turbines generally operate in relatively harsh environments and may need to face complex wind conditions, high temperature, humidity, extreme wind conditions, complex wind shear, variable upwind and downwind loads, and the superposition of complex operating conditions under different surrounding terrain parameters.

[0156] According to the fault diagnosis method of the embodiment of the present disclosure, based on the improved vibration analysis technology (including data extraction, transformation, processing, statistical model, machine learning model, etc.), for different models (such as 2MW, 3MW, 4MW, etc.) and configurations of direct-drive units, the vibration data of multiple measurement positions under multiple working conditions (such as grid connection, yaw, pitch change, specific speed, etc.) are analyzed for multiple (for example, more than 10) indicators in the time domain and frequency domain, such as comparison with absolute thresholds, longitudinal comparison with historical data of target units, comparison with the same working conditions across units, etc., and simple diagnostic methods such as identification based on complex features of the model are used. The faults can be comprehensively determined by identification and quantification, so as to realize accurate diagnosis of the transmission chain and related components of the unit. In the above fault diagnosis, for example, the loads of the front bearing, rear bearing, rolling element and retaining frame of the generator under different working conditions (front bearing load, rear bearing load, etc.) can be obtained, and the corresponding components may have typical failure characteristics such as locating ring movement, raceway peeling, micro-motion wear, ring rib fracture, grease leakage, high temperature, generator bearing jamming, etc., and avoid further damage to other components such as generators, blades, towers and power generation losses caused by long-term shutdown due to main bearing abnormalities, and develop an early warning model with an accuracy rate of more than 90%.

[0157] According to the fault diagnosis method of the embodiment of the present disclosure, by performing multiple analyses on the sensing data of the transmission chain including vibration data, different technologies can be integrated, enriching and supplementing the means for more accurate and timely diagnosis and prediction of various components and abnormal modes of the transmission chain.

[0158] In addition, according to the fault diagnosis method of the embodiment of the present disclosure, the vibration signal data in the time domain and frequency domain at each measurement position under various working conditions for direct-drive units of different types and configurations can be used to obtain the optimal model parameters and hyperparameters based on improved data processing, extraction, transformation and effective fusion of improved multiple models, and establish a frequency anomaly feature library for different models and failure modes, for example, including various components of the transmission chain and abnormal modes, failure types, corresponding frequency characteristics, abnormal diagnosis model parameters, hyperparameters, iterative mechanisms, etc., to achieve accurate fault diagnosis.

[0159] In addition, according to the fault diagnosis method of the embodiment of the present invention, machine learning, federated learning and bearing vibration signal analysis technology can be combined to obtain the transmission chain vibration signal data of the direct-drive wind turbine under specific working conditions (such as a specific speed), and filter the frequency band within a specific range through a suitable filter (such as a 2-4 order bandpass Butterworth filter), screen the frequency band related to the transmission chain (such as the main bearing), perform envelope analysis by extracting the envelope line, and calculate the spectrum characteristics and time-frequency domain analysis indicators.

[0160] Here, for the filtered data, a sliding statistics method can be used to perform sliding statistics on the data according to a predetermined window length and step size to obtain time domain analysis indicators for multiple sliding windows, thereby making the statistical data smoother and effectively reducing the interference of noise on the signal and the impact on the analysis effect of subsequent models.

[0161] In addition, empirical mode decomposition can be used to obtain the IMF components of the bearing signal, sort the IMF components according to factors such as machine model and manufacturer, extract specific IMF components for feature analysis and reconstruction, extract components that can effectively identify bearing fault characteristics and then input them into a group of machine learning models (such as random forests, support vector machines, K nearest neighbor algorithms, etc.) for fault classification and diagnosis.

[0162] Here, for each vibration sub-component obtained by empirical mode decomposition, the frequency of each vibration sub-component can be matched with the preset fault frequency of each component in the transmission chain, and the amplitude of each vibration sub-component can be matched with the amplitude range of the preset fault level, so that the fault source or fault severity corresponding to different components can be accurately decomposed. Such matching results are input into the subsequent model, which not only helps the model to accurately identify the occurrence of faults, but also enables the model to locate the type or level of the fault and obtain more specific fault diagnosis results.

[0163] In the above steps, the order of execution of each step is required to be high, and the model is fused to output the fault diagnosis conclusion. In addition, through the above method, the encrypted training of the model can be completed without leaking the real data, and a more universal sum model can be obtained.

[0164] According to a second aspect of an embodiment of the present disclosure, an electronic device is provided, comprising: at least one processor; and at least one memory storing processor-executable instructions, wherein the processor-executable instructions, when executed by the at least one processor, prompt the at least one processor to execute the fault diagnosis method for the transmission chain of a direct-drive wind turbine generator set according to the embodiment of the present disclosure.

[0165] As an example, the electronic device may be a PC, a tablet device, a personal digital assistant, a smart phone, or other device capable of executing the above instruction set. Here, the electronic device is not necessarily a single electronic device, but may also be any device or circuit capable of executing the above instruction (or instruction set) individually or in combination. The electronic device may also be part of an integrated control system or system manager, or may be configured as a portable electronic device interconnected with a local or remote (e.g., via wireless transmission) interface.

[0166] In electronic devices, the processor may include a central processing unit (CPU), a graphics processing unit (GPU), a programmable logic device, a dedicated processor system, a microcontroller or a microprocessor. As an example and not limitation, the processor may also include an analog processor, a digital processor, a microprocessor, a multi-core processor, a processor array, a network processor, etc.

[0167] The processor can execute instructions or codes stored in the memory, wherein the memory can also store data. Instructions and data can also be sent and received through the network via the network interface device, wherein the network interface device can adopt any known transmission protocol.

[0168] The memory may be integrated with the processor, for example, RAM or flash memory is arranged within an integrated circuit microprocessor or the like. In addition, the memory may include a separate device, such as an external disk drive, a storage array, or any other storage device that can be used by a database system. The memory and the processor may be operatively coupled, or may communicate with each other, such as through an I / O port, a network connection, etc., so that the processor can read files stored in the memory.

[0169] In addition, the electronic device may also include a video display (such as a liquid crystal display) and a user interaction interface (such as a keyboard, a mouse, a touch input device, etc.) All components of the electronic device may be connected to each other via a bus and / or a network.

[0170] According to a third aspect of an embodiment of the present disclosure, a computer program product is provided, which includes computer instructions, which, when executed by a processor, implement the fault diagnosis method for the transmission chain of a direct-drive wind turbine generator set according to the embodiment of the present disclosure.

[0171] In addition, the fault diagnosis method of the transmission chain of the direct-drive wind turbine generator set according to the embodiment of the present disclosure can be written as a computer program and stored on a computer-readable storage medium. When the instructions in the computer-readable storage medium are executed by at least one processor, the at least one processor is prompted to execute the fault diagnosis method of the transmission chain of the direct-drive wind turbine generator set according to the exemplary embodiment of the present disclosure. Examples of computer-readable storage media include: read-only memory (ROM), random access programmable read-only memory (PROM), electrically erasable programmable read-only memory (EEPROM), random access memory (RAM), dynamic random access memory (DRAM), static random access memory (SRAM), flash memory, non-volatile memory, CD-ROM, CD-R, CD+R, CD-RW, CD+RW, DVD-ROM, DVD-R, DVD+R, DVD-RW, DVD+RW, DVD-RAM, BD-ROM, BD-R, BD-R LTH, BD-RE, Blu-ray or optical disk storage, hard disk drive (HDD), solid state drive (SSD), card storage (such as, multimedia card, secure digital (SD) card or extreme digital (XD) card), magnetic tape, floppy disk, magneto-optical data storage device, optical data storage device, hard disk, solid state disk and any other device, any other device is configured to store computer program and any associated data, data files and data structures in a non-transitory manner and provide the computer program and any associated data, data files and data structures to a processor or computer so that the processor or computer can execute the computer program. In one example, the computer program and any associated data, data files and data structures are distributed on a networked computer system so that the computer program and any associated data, data files and data structures are stored, accessed and executed in a distributed manner by one or more processors or computers.

[0172] According to an embodiment of the present disclosure, a computer program product is also provided. The computer program product includes computer instructions. When the computer instructions are executed by a processor, a fault diagnosis method for a transmission chain of a direct-drive wind turbine generator set according to an embodiment of the present disclosure can be implemented.

[0173] The specific implementation methods of the present disclosure have been described in detail above. Although some embodiments have been shown and described, those skilled in the art should understand that these embodiments may be modified and varied without departing from the principles and spirit of the present disclosure, the scope of which is defined by the claims and their equivalents. These modifications and variations should also be within the scope of protection of the claims of the present disclosure.

Claims

1. A fault diagnosis method for a transmission chain of a direct-drive wind turbine generator set, characterized in that: The transmission chain includes a main bearing and a generator, wherein the fault diagnosis method includes: Acquire sensing data of a plurality of measurement positions on a transmission chain of a direct-drive wind turbine generator set, wherein the sensing data includes vibration data, wherein the plurality of measurement positions include a main bearing measurement position and a generator measurement position; Performing feature analysis on the acquired sensing data to obtain time domain feature data and frequency domain feature data; The time domain characteristic data and the frequency domain characteristic data are input into a pre-built fault diagnosis model to obtain a fault diagnosis result indicating whether there is a fault in the transmission chain of the direct-drive wind turbine generator set.

2. The fault diagnosis method according to claim 1, characterized in that: The sensing data also includes generator speed data, The main bearing measurement positions include a front bearing horizontal vibration measurement position, a front bearing vertical vibration measurement position and a front bearing axial vibration measurement position on the front bearing of the main bearing. The generator measurement position includes a stator measurement position on the stator of the generator and a rotation speed measurement position of the generator, The sensing data of the main bearing measurement position is obtained by sensing a first type of sensor, and the sensing data of the generator measurement position is obtained by sensing a second type of sensor. The sensing frequency band of the first type of sensor is within the sensing frequency band of the second type of sensor.

3. The fault diagnosis method according to claim 1 or 2, characterized in that: The step of performing feature analysis on the acquired sensing data to obtain time domain feature data and frequency domain feature data comprises: Performing envelope analysis on the sensing data in the frequency domain to obtain frequency domain envelope features; Performing empirical mode decomposition on the sensing data to obtain mode decomposition features, wherein the mode decomposition features include time domain mode features and frequency domain mode features; Based on the time domain modal features, obtaining the time domain feature data; The frequency domain feature data is obtained based on the frequency domain envelope feature and the frequency domain modal feature.

4. The fault diagnosis method according to claim 3, characterized in that: The modal decomposition characteristics are obtained in the following way: extracting time series data within a specific time period from the sensing data; Determine a data upper limit envelope and a data lower limit envelope of the time series data; Based on the data upper limit envelope and the data lower limit envelope, the time series data is subjected to vibration decomposition to obtain a plurality of vibration subcomponents, wherein each vibration subcomponent corresponds to a different vibration mode of the main bearing; The time domain analysis and frequency domain analysis are performed on each vibration sub-component to obtain the time domain modal characteristics and frequency domain modal characteristics of each vibration sub-component respectively.

5. The fault diagnosis method according to claim 3, characterized in that: The frequency domain envelope characteristics are obtained by: determining frequency spectrum data of the vibration data; Based on the frequency spectrum data and a preset fault vibration frequency range, the vibration data is filtered to obtain filtered data in the vibration data that is within the fault vibration frequency range; Perform envelope analysis on the filtered data to determine the frequency domain envelope characteristics.

6. The fault diagnosis method according to claim 1, characterized in that: The frequency domain feature data includes frequency domain analysis data, wherein the frequency domain analysis data is obtained by: determining frequency spectrum data of the vibration data; Performing high amplitude analysis and uniform interval vibration analysis on the frequency spectrum data to obtain the frequency domain analysis data, The high amplitude analysis refers to determining whether there is data with an amplitude higher than a preset amplitude threshold in the spectrum data, and the evenly spaced vibration analysis refers to determining whether there is data with frequency peaks at equal intervals in the spectrum data.

7. The fault diagnosis method according to claim 1, characterized in that: The pre-built fault diagnosis model includes multiple models, wherein the fault diagnosis result is obtained by: Based on each model of the multiple models, performing fault diagnosis on the sensing data to obtain a single model diagnosis result corresponding to each model; Based on the preset model weights, the single model diagnosis results of the corresponding models are weighted to obtain the weighted diagnosis results corresponding to each model; Based on each weighted diagnosis result, the fault diagnosis result is obtained.

8. The fault diagnosis method according to claim 7, characterized in that: The multiple models are constructed based on historical sensing data of multiple sample units at the multiple measurement locations, the multiple sample units include faulty units in multiple dimensions, the multiple dimensions include dimensions of different units, dimensions of different wind farms, dimensions of different unit types, and dimensions of regions to which different units belong, the multiple models include statistical models, mechanism models, and machine learning models, Among them, the different unit dimensions refer to different unit models; the different wind farm dimensions refer to different geographical locations of wind farms and / or different scales of wind farms; the different unit region dimensions refer to different geographical locations of units and / or different environmental characteristics of the units.

9. An electronic device, characterized in that: include: at least one processor; at least one memory storing processor-executable instructions, Wherein, when the processor executable instructions are executed by the at least one processor, the at least one processor is prompted to execute the fault diagnosis method for the transmission chain of the direct-drive wind turbine generator set as described in any one of claims 1-8.

10. A computer program product comprising computer instructions, characterized in that: When the computer instructions are executed by the processor, the fault diagnosis method for the transmission chain of the direct-drive wind turbine generator set according to any one of claims 1-8 is implemented.

Images