An early fault detection system for wind turbine generator sets

Abstract

This method belongs to the field of wind power generation and relates to an early fault detection system for wind turbine generators. The system calls a data filtering module to remove abnormal or invalid data; a feature extraction module to extract valid components and fault features from the data; a fault pre-detection module to perform preliminary fault detection; a feature enhancement module to optimize algorithm parameters, enhance data features, and detect weak signals using the feature extraction results; a fault identification module to detect the location and severity of the fault, and after detection, returns the results to the feature enhancement and fault identification modules to detect other areas; and a fault diagnosis module to summarize the conclusions from the data filtering, fault pre-detection, and fault identification modules and output the detection results. This system improves fault extraction methods, extracts weak fault features from vibration data, and achieves effective and accurate identification of fault location and severity in the early stages of a fault, facilitating early detection and timely handling of faults.

Description

Technical Field

[0001] This method belongs to the field of wind power generation and involves an early fault detection system for wind turbine generator sets. Background Technology

[0002] Wind turbine generators have a complex structure and are prone to major failures such as broken gears and stuck bearings during long-term high-load operation, which can affect the normal operation of the wind turbine.

[0003] The aforementioned faults exhibit subtle fault characteristics in their early stages of development. Therefore, early fault detection of wind turbine generators is of great significance for timely scheduling of fault inspections and repairs, preventing major faults, reducing downtime due to faults, and lowering the operation and maintenance costs of wind turbine generators.

[0004] Currently, early fault detection for wind turbine generators still faces the following challenges:

[0005] 1. Low-speed, heavy-load components such as main bearings and planetary gear sets are large in size and have complex vibration transmission paths, resulting in unclear fault characteristics, which is not conducive to early fault identification and fault severity detection.

[0006] 2. While manual fault diagnosis methods can effectively identify faults, they are not conducive to the timely detection of sudden faults;

[0007] 3. Traditional fault detection systems cannot effectively identify the location and severity of faults, nor can they achieve early fault identification;

[0008] 4. In practical diagnostic applications, methods based on fault feature extraction suffer from low data utilization and poor standardization of output results. Summary of the Invention

[0009] This method proposes an early fault detection system for wind turbine generators. The system includes six modules: data filtering, feature extraction, fault pre-detection, feature enhancement, fault identification, and conclusion integration. It improves the fault extraction method, extracts weak fault features from vibration data, and achieves effective and accurate identification of fault location and fault degree in the early stage of fault, which is conducive to early detection and timely handling of faults.

[0010] The technical solution of this method is as follows: Figure 1 As shown, its six modules function as follows:

[0011] The data filtering module takes vibration and rotational speed data from the data acquisition as input to determine the anomalies in the sensor-acquired data and the validity of the data for fault diagnosis. Data without anomalies and valid can be input into the feature extraction module; otherwise, the fault diagnosis module is directly invoked to output a conclusion.

[0012] The feature extraction module can extract the effective components of data based on the fault feature intervals of the detected location, thereby extracting fault features using relevant algorithms. This algorithm improves data utilization and can output standardized fault features. This module can integrate the fault features as input to the fault pre-detection module, and the extracted fault features can be used as input to the feature enhancement module.

[0013] The fault pre-detection module can determine whether a fault exists in the detected location by enhancing the time-domain characteristics of the data based on the frequency domain and time-domain features. Data indicating a fault can be used as input to the feature enhancement module; otherwise, the fault diagnosis module is directly invoked to output the conclusion.

[0014] The feature enhancement module can optimize the algorithm parameters based on the fault features extracted by the feature extraction module, and then perform time-domain feature enhancement and weak signal detection on the data, using the results as input to the fault identification module.

[0015] The fault identification module can combine the results of the feature enhancement module with... Figure 2 The fault identification strategy shown determines the fault location and severity, and uses the results as input to the fault identification module. If the results of the data fault feature enhancement do not point to a fault in a certain part, the detected part is considered normal, and the data is returned to the feature enhancement module to re-optimize parameters based on the fault characteristics of other detected parts, perform fault enhancement and weak signal detection, and then input it back to the fault identification module until all parts have been detected.

[0016] The fault diagnosis module summarizes and outputs the conclusions from the data filtering module, fault pre-detection module, and fault identification module. The final output includes: data anomalies, invalid data detection, normal detection location, localized damage to gears, uniform wear on gears, and damage to bearings. For results indicating a fault, the module will output the degree of fault. Attached Figure Description

[0017] Figure 1 System technical solution flowchart

[0018] Figure 2 Fault identification strategy Detailed Implementation

[0019] In the data filtering module, anomaly detection is achieved by calculating the total valid value and kurtosis of the data. If the total valid value is less than 100%, the anomaly detection is performed. or kurtosis greater than Then the data is considered to be abnormal, among which

[0020] This represents an abnormal parameter with valid values, determined by the characteristic differences between abnormal and normal signals.

[0021] This indicates the effective value level, which is determined by the sensor sensitivity.

[0022] This represents the kurtosis anomaly parameter, which is determined by the characteristic differences between the anomalous signal and the normal signal.

[0023] This indicates the kurtosis level, and a kurtosis level of 3 is selected.

[0024] The data validity determination in the data filtering module is achieved by calculating the rotational speed and its variation. The rotational speed should not be lower than 80% of the full-load speed, and the fluctuation of the rotational speed should be less than 20%. Otherwise, the data is considered not valid for fault detection.

[0025] The effective component extraction model for the detection site in the feature extraction module is as follows:

[0026]

[0027] In the formula

[0028] Represents frequency amplitude

[0029] Indicates the cutoff frequency of the effective component in gear detection;

[0030] Indicates the cutoff frequency of the effective component in bearing detection;

[0031] Indicates the spindle speed;

[0032] This indicates the meshing frequency band parameters, which are set according to the target extracted components;

[0033] This represents the feature frequency parameter, which is set according to the target extracted components.

[0034] This represents the characteristic frequency band parameter, which is set according to the target extracted components;

[0035] This indicates the frequency conversion parameter, which is set according to the target extracted components.

[0036] Indicates the filter order;

[0037] Indicates the characteristic frequency of bearing inner ring failure;

[0038] Indicates the transmission ratio;

[0039] Indicates the number of teeth.

[0040] The fault feature extraction model in the feature extraction module is as follows:

[0041]

[0042] In the formula

[0043] This represents the periodicity parameter of the algorithm. , These are the periodic parameter ranges for gears and bearings, respectively.

[0044] Indicates period The characteristic parameters at that time;

[0045] Represents a data sequence;

[0046] Indicates the number of sampling points;

[0047] Indicates the total number of sampling points;

[0048] express The calculation range parameter can be set to adjust the calculation range.

[0049] Indicates the order of the characteristic frequency;

[0050] Indicates the characteristic frequency of bearing failure;

[0051] Indicates the sampling frequency.

[0052] Fault characteristic thresholds in the fault pre-detection module Based on extensive engineering practice and experimental simulations, if This indicates a fault exists; the specific fault and its severity require further analysis, and the periodic parameters need to be determined. It is used as input to the feature enhancement module; otherwise, it is assumed that there is no fault.

[0053] The algorithm parameter optimization in the feature enhancement module is based on the feature parameters extracted in the feature extraction module. and its periodic parameters Optimize algorithm parameters in temporal feature enhancement and weak signal detection.

[0054] The model for temporal feature enhancement in the feature enhancement module is as follows:

[0055]

[0056] In the formula

[0057] Represents the original vibration signal;

[0058] This represents the filtered signal;

[0059] Represents the filter coefficients;

[0060] This represents the number of iterations, with an upper limit of 1. The level of fault enhancement can be changed by setting it; the default value is 30.

[0061] Indicates the first Replace the original signal weighting coefficients;

[0062] Indicates the first Alternate filtering signal weighting coefficients;

[0063] Indicates the periodic parameter;

[0064] This represents the displacement factor, with an upper limit of 1. The accuracy of the periodic feature enhancement can be changed by setting the default value to 1;

[0065] This represents the filter parameters. Setting these parameters can change the accuracy of fault enhancement. The default value is 100.

[0066] Indicates the number of sampling points.

[0067] The model for weak signal detection in the feature enhancement module is as follows:

[0068]

[0069] In the formula

[0070] Output for weak signal detection;

[0071] The damping coefficient;

[0072] This refers to the amplitude of the internal driving force.

[0073] The envelope signal represents the time-domain feature enhancement data.

[0074] In the fault identification module, fault location determination is achieved through the results of time-domain feature enhancement and weak signal detection. If time-domain features exist and a weak sinusoidal signal with the corresponding characteristic frequency is detected, the fault location can be determined. The fault identification strategy in the fault identification module is as follows:

[0075] Strategy 1: In gear inspection, if meshing frequency impact and characteristic frequency impact are detected after time-domain feature enhancement, but the weak sinusoidal signal of the characteristic frequency is not detected, then the gear is considered to have uniform wear.

[0076] Strategy 2: In gear inspection, if meshing frequency impact and characteristic frequency impact are detected after time-domain feature enhancement, and a weak sinusoidal signal of the characteristic frequency is detected, then the gear is considered to have local damage, and the location of the damage can be determined by the characteristic frequency.

[0077] Strategy 3: In bearing inspection, if a characteristic frequency impact is detected after time-domain feature enhancement, and a weak sinusoidal signal of shaft rotation frequency or cage frequency is detected, the bearing is considered to be damaged, and the damaged location can be determined by the characteristic frequency.

[0078] Strategy 4: If no periodic impact is detected after time-domain feature enhancement during detection, the detected area is considered normal.

[0079] The fault severity determination in the fault identification module is based on the weak signal detection results. The value is determined by the magnitude of the fault; the closer the value is to 0, the milder the fault; the larger the value, the more severe the fault.

Claims

1. An early fault detection system for a wind turbine, characterized in that The system calls the data screening module to screen data without exception and detect effective vibration data; The system calls the feature extraction module to extract data effective components and fault features; The system calls the fault pre-detection module to preliminarily detect faults according to original signals and time-domain feature enhanced signals; The system calls the feature enhancement module to optimize algorithm parameters by using fault features extracted by the feature extraction module, so as to realize time-domain feature enhancement of data and weak signal detection; The system calls the fault identification module to detect fault positions and degrees based on results of the feature enhancement module and fault identification strategies, and returns the feature enhancement module to enhance features of other detection parts after detection, until all parts are detected; The system calls the fault diagnosis module to summarize conclusions in the data screening module, the fault pre-detection module and the fault identification module, and outputs the detection results. The fault feature extraction model in the feature extraction module of the system is as follows: wherein: periodic parameters of the representation algorithm, , are the periodic parameter ranges of the gear, bearing, respectively; representative period characteristic parameters at times represents a data sequence; represents the number of sampling points; N represents the total number of samples; representing a calculation range parameter of the calculation range, the parameter being set to adjust the calculation range; denotes the characteristic frequency order; represents a bearing fault characteristic frequency; denotes the sampling frequency. The time-domain feature enhancement model in the feature enhancement module of the system is as follows: In the formula represents the original vibration signal; represents the filtered signal; denotes filter coefficients; denotes the number of iterations, which has an upper limit of The degree of fault enhancement can be varied by setting, with a default value of 30. represents the first generation original signal weight coefficients; representing the filter signal weight coefficients; denotes a periodic parameter; denotes a displacement factor, which has an upper limit of , the accuracy of the periodicity feature enhancement can be changed by setting, and the default value is 1; represents a filter parameter, which sets the precision of the fault enhancement that can be changed by setting a default value of 100; represents the number of sampling points.

2. A wind turbine early fault detection system according to claim 1, characterized by feature extraction The effective component extraction model in the module is as follows: wherein representative frequency amplitude of the representative frequency represents the gear detection active ingredient cutoff frequency; represents the bearing detection active ingredient cutoff frequency; denotes the spindle rotation frequency; represents an engagement band parameter, set in accordance with the target extraction component; representing a characteristic frequency parameter, set in accordance with the target extraction component; representing a characteristic frequency band parameter, set in accordance with the target extraction component; represents a frequency conversion parameter, set in accordance with the target extraction component; denotes the filter order; represents the bearing inner ring fault characteristic frequency; denotes the transmission ratio; represents the number of teeth.

3. A wind turbine early fault detection system according to claim 1, characterized by feature enhancement The weak signal detection model in the module is as follows: wherein: Output for weak signal detection; is the damping coefficient; is the internal actuation force amplitude; an envelope signal representing the time-domain feature enhancement data.

4. A wind turbine early fault detection system according to claim 1, characterised in that The fault identification strategy in the fault identification module is as follows: Strategy one: in gear detection, if meshing frequency impact and characteristic frequency impact are detected after time-domain feature enhancement, but no weak sinusoidal signal of the characteristic frequency is detected, it is considered that the gear has uniform wear; Strategy two: in gear detection, if meshing frequency impact and characteristic frequency impact are detected after time-domain feature enhancement, and a weak sinusoidal signal of the characteristic frequency is detected, it is considered that the gear has local damage, and the damage position can be determined by the characteristic frequency; Strategy three: in bearing detection, if characteristic frequency impact is detected after time-domain feature enhancement, and a weak sinusoidal signal of shaft rotation frequency or cage frequency is detected, it is considered that the bearing has damage, and the damage position can be determined by the characteristic frequency; Strategy four: in detection, if no periodic impact is detected after time-domain feature enhancement, it is considered that the detection part is normal.

Images