A fan bearing fault online detection method and system

Abstract

The present application relates to the technical field of fan bearing fault diagnosis, in particular to a fan bearing fault online detection method and system, adopting a weighted combination envelope spectrum technology, quantifying the correlation of fault characteristics and resonance frequency bands through two-dimensional spectrum coherence, combining the frequency domain signal-to-noise ratio to adaptively assign weights, aggregating and enhancing the fault information of multiple resonance frequency bands, effectively offsetting the attenuation effect in the signal transmission process, solving the problem of signal attenuation distortion caused by magnetic attraction installation through the machine shell under the limitation of fan structure size. The order tracking and constant angle increment sampling technology is introduced, the corresponding relationship between the vibration signal and the bearing rotation angle is established, the time domain signal is converted into the angle domain signal, the fault characteristic frequency and the rotating speed are decoupled, the theoretical fault frequency of the inner ring, the outer ring, the roller and other components is accurately matched with the detection result, the detection adaptability under variable working conditions is greatly improved, compared with the fixed frequency domain analysis of the traditional acceleration envelope method, the misjudgment caused by frequency offset is effectively avoided.

Description

Technical Field

[0001] This invention relates to the field of wind turbine bearing fault diagnosis technology, specifically to an online detection method and system for wind turbine bearing faults. Background Technology

[0002] In industrial sectors such as semiconductor manufacturing and power generation, fans serve as critical fluid transport and heat dissipation equipment. Their long-term, continuous, and stable operation directly determines the reliability of the production line and the overall operation and maintenance costs. Fan bearings, as core transmission components, are prone to failure under high-speed rotation conditions due to wear, pitting, and fatigue spalling. Failure to detect these failures in a timely manner can quickly lead to bearing jamming, shaft bending, or even complete machine shutdown, resulting in significant economic losses. Therefore, research into online detection technology for fan bearing failures, enabling accurate early identification and severity assessment of faults, has become one of the core requirements for intelligent operation and maintenance of industrial equipment.

[0003] In existing technologies, vibration signal analysis is the mainstream approach for wind turbine bearing fault detection. Among these, acceleration envelope analysis is widely used in practical engineering due to its ability to effectively extract the transient impact characteristics generated by faults. Its core detection principle is as follows: When a local defect occurs in a wind turbine bearing, the periodic collision between the roller and the defective area during rotation generates a transient impact signal. This impact signal excites the inherent resonance of the bearing and the wind turbine casing, forming a high-frequency vibration signal containing fault characteristics. After acquiring this vibration acceleration signal using an accelerometer, the signal within the resonance frequency band is filtered out by a high-pass filter. Then, envelope detection and low-pass filtering are performed to remove the high-frequency carrier and extract the envelope signal. Finally, the envelope signal is subjected to spectral analysis. If a significant peak appears at the theoretical fault characteristic frequency of the bearing's inner ring, outer ring, roller, or cage, the corresponding component is determined to be faulty.

[0004] Meanwhile, in specific practical scenarios, due to the limitations of the wind turbine's structural dimensions, after the wind turbine is encapsulated and installed, the sensor cannot be directly installed near the bearing housing and can only be installed on the outer surface of the casing. In practical applications, a flat boss is usually glued at the connection between the casing and the stator blades, and the vibration sensor is installed on the boss by magnetic attraction to detect the vibration signal of the bearing. However, this installation method will cause the vibration signal to attenuate and be distorted during transmission, further increasing the difficulty of fault detection.

[0005] Considering the actual operating scenarios, installation limitations, and application practices of existing technologies, traditional acceleration envelope detection methods still have many insurmountable shortcomings in adapting to variable operating conditions of wind turbines, extracting early and subtle faults, and accurately diagnosing faults. They cannot meet the preventative maintenance needs of high-reliability equipment such as wind turbines in the semiconductor industry. Firstly, the rotational speed of wind turbines fluctuates slightly during actual operation. Traditional acceleration envelope methods directly perform frequency domain analysis on time-domain vibration signals without establishing a correlation between the vibration signal and the bearing rotation angle. This leads to a "frequency ambiguity" phenomenon caused by the coupling of fault characteristic frequencies with rotational speed, making it impossible to accurately match the theoretical fault frequencies of various bearing components. This results in extremely poor adaptability to detection under variable operating conditions.

[0006] Secondly, the transient impact signal generated by early weak bearing faults has extremely low energy, and it is attenuated even more severely after being transmitted through the casing. It is easily masked by the background noise generated by the vibration of the fan casing and pipeline. The traditional acceleration envelope method only performs integration processing on a single resonant frequency band, without considering the distribution differences of fault information in different resonant frequency bands. It cannot aggregate and enhance the weak fault features in multiple frequency bands, making it difficult to achieve effective identification of early faults.

[0007] Secondly, existing detection methods mostly rely on fixed thresholds for alarms based on a single envelope feature, only achieving a qualitative judgment of "fault presence / absence." They cannot accurately locate specific faulty components such as the inner or outer rings, nor can they quantitatively assess the severity of the fault. Furthermore, environmental interference and signal transmission distortion easily lead to frequent threshold triggering and false alarms, while early, weak faults are masked because their signal amplitude does not reach the threshold, resulting in missed detections and severely impacting the reliability of the detection results. Finally, traditional methods only employ shallow signal processing techniques such as filtering and envelope extraction, failing to uncover the multidimensional temporal features contained in the wind turbine bearing vibration signal. They lack the ability to identify complex bearing faults and cannot predict fault development trends by combining historical detection data. Summary of the Invention

[0008] The purpose of this invention is to provide an online detection method and system for wind turbine bearing faults, addressing the aforementioned problems.

[0009] The technical solution adopted in this invention is as follows: an online detection method for wind turbine bearing faults, comprising the following steps:

[0010] Signal acquisition involves magnetically mounting an accelerometer on the fan casing to monitor the vibration of the fan bearing and acquire the time-domain signal of the vibration acceleration. Simultaneously, a Hall sensor is set near the fan bearing to acquire the key phase signal of the fan bearing rotation speed.

[0011] Signal preprocessing establishes a time-bearing rotation angle correspondence using the rotational speed key phase signal, converting the vibration acceleration time-domain signal into a vibration acceleration angular-domain signal.

[0012] Feature extraction is performed to estimate the two-dimensional spectral coherence of the vibration acceleration angular domain signal. An envelope spectrum slice weighting function is constructed, and a weighted combined envelope spectrum is calculated and fused with time-domain features to form a comprehensive feature set.

[0013] Fault diagnosis involves inputting a comprehensive feature set into a hybrid model, outputting alarm prediction values ​​and inversely normalizing them, and combining the FFT spectrum with the weighted combined envelope spectrum to quantify the severity of the fault.

[0014] Online alarms are set according to weighted probability, and alarm data is added to the historical sample database for online optimization.

[0015] Furthermore, in the signal acquisition, the vibration acceleration time-domain signal includes a time-domain curve, kurtosis index, and envelope RMS value;

[0016] The time-domain curve is used to reflect the time-domain waveform of the original vibration acceleration;

[0017] The kurtosis index is a statistical measure used to reflect the impulse characteristics of a signal;

[0018] The effective value of the envelope is used to reflect the location where bearing vibration occurs;

[0019] The rotational speed key phase signal is used to record the relationship between the bearing rotation angle and time.

[0020] Furthermore, in the signal preprocessing, the vibration acceleration time-domain signal is converted into a vibration acceleration angular-domain signal by inversion, which serves as the direct input for spectral correlation and spectral coherence.

[0021] Furthermore, the feature extraction includes the following sub-steps:

[0022] Estimate the two-dimensional spectral coherence of angular domain vibrational acceleration signals and quantize the correlation between cyclic frequency and spectral frequency;

[0023] The fault information content at each spectral frequency is evaluated by using a frequency domain signal-to-noise ratio metric, and an adaptive threshold is set to construct an envelope spectrum slice weighting function.

[0024] The weighted combined envelope spectrum is calculated by integrating the two-dimensional spectral coherence according to the weighting function.

[0025] The weighted combination envelope spectrum features are integrated with time-domain features, kurtosis index and envelope effective value to form a comprehensive feature set for fault diagnosis.

[0026] Furthermore, the hybrid model in the fault diagnosis is a hybrid neural network consisting of CNN, LSTM, and attention layer.

[0027] Furthermore, the fault diagnosis includes structural calculation of the input comprehensive feature set using a hybrid neural network of CNN+LSTM+attention layer and bearing fault probability quantification.

[0028] Furthermore, the structural calculation of the CNN+LSTM+attention layer hybrid neural network on the input comprehensive feature set includes the following sub-steps:

[0029] Input tensor construction expands the comprehensive feature set of the input into a temporal feature tensor adapted to wind turbine vibration;

[0030] In the CNN layer, spatial features are extracted from the temporal feature tensor for local spatial correlation.

[0031] In the LSTM layer, the fault development trend in the temporal feature tensor is captured;

[0032] At the attention layer, the features at critical moments of the fault are given high weights through the attention mechanism, and the original probability of the wind turbine bearing failure is output.

[0033] Furthermore, the bearing failure probability quantification includes the following sub-steps:

[0034] Generate the order FFT spectrum by performing an order tracking FFT on the vibration acceleration angular domain signal to obtain the order FFT spectrum;

[0035] The failure probability of each component of the wind turbine bearing is calculated. For the inner ring, outer ring, rollers and cage of the wind turbine bearing, the failure probability is calculated by order FFT spectrum and weighted combined envelope spectrum respectively, and then weighted and fused.

[0036] The fault diagnosis and hybrid model are integrated. The cyclic frequency sequence of the weighted combined envelope spectrum, the amplitude characteristics of the order FFT spectrum, and the fault probability of each component are used as inputs to the CNN+LSTM+attention layer hybrid neural network. The hybrid neural network outputs the confidence level of the fault type.

[0037] The present invention also provides an online detection system for wind turbine bearing faults, which includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the above-described online detection method for wind turbine bearing faults.

[0038] The beneficial effects of the present invention include at least one of the following;

[0039] 1. A method for online detection of wind turbine bearing faults is provided. The method adopts weighted combination envelope spectrum technology, which uses two-dimensional spectrum coherent quantization to determine the correlation between fault features and resonant frequency bands. Combined with frequency domain signal-to-noise ratio adaptive weight allocation, the fault information of multiple resonant frequency bands is aggregated and enhanced, effectively offsetting the attenuation effect during signal transmission. This solves the problem that the sensor cannot be directly installed in the bearing chamber due to the structural size limitation of the wind turbine and needs to be magnetically installed through the casing boss, which leads to signal attenuation and distortion.

[0040] 2. By introducing order tracking and constant angle incremental sampling technology, the correspondence between vibration signal and bearing rotation angle is established, and the time domain signal is converted into angular domain signal, so that the fault characteristic frequency is decoupled from the rotation speed. This ensures that the theoretical fault frequency of components such as inner ring, outer ring, and roller is accurately matched with the detection results, which greatly improves the detection adaptability under varying working conditions. Compared with the fixed frequency domain analysis of the traditional acceleration envelope method, it effectively avoids misjudgment caused by frequency shift.

[0041] 3. By mining multi-dimensional time-series deep features through a hybrid neural network of CNN+LSTM+attention layer, and combining the fault probability fusion model of FFT spectrum and weighted combined envelope spectrum, it can not only accurately locate specific faulty components, but also quantitatively assess the severity of faults through weighted fault probability, providing maintenance personnel with clear repair directions and priority criteria.

[0042] 4. The system adopts a lightweight aluminum alloy cabinet design, supports quick-connect interfaces and multi-channel expansion, and is suitable for installation and maintenance scenarios in industrial sites; at the same time, it outputs structured alarm information, reducing the professional threshold for operation and maintenance personnel. Compared with the traditional technology that relies on experts to interpret data, it greatly improves operation and maintenance efficiency and reduces unplanned downtime. Attached Figure Description

[0043] Figure 1 This is an overall outline diagram for online detection of wind turbine bearing faults.

[0044] Figure 2 This is a schematic diagram of the structure of an online fault detection system for wind turbine bearings.

[0045] Figure 3 Flowchart of online detection method for wind turbine bearing faults;

[0046] Figure 4 Flowchart for structural calculation of the input comprehensive feature set;

[0047] Figure 5 Flowchart for quantifying bearing failure probability;

[0048] Figure 6 This is a schematic diagram of the structure of an electronic device. Detailed Implementation

[0049] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described in the accompanying drawings can generally be arranged and designed in various different configurations.

[0050] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0051] It should be noted that, unless otherwise specified, the embodiments and features described in this invention can be combined with each other.

[0052] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0053] like Figure 3 As shown, an online detection method for fan bearing faults is applicable to the detection of continuously operating fans in semiconductor processing, and includes the following steps:

[0054] Signal acquisition involves magnetically mounting an accelerometer on the fan casing to monitor the vibration of the fan bearing and acquire the time-domain signal of the vibration acceleration. Simultaneously, a Hall sensor is set near the fan bearing to acquire the key phase signal of the fan bearing rotation speed.

[0055] Signal preprocessing establishes a time-bearing rotation angle correspondence using the rotational speed key phase signal, converting the vibration acceleration time-domain signal into a vibration acceleration angular-domain signal.

[0056] Feature extraction is performed to estimate the two-dimensional spectral coherence of the vibration acceleration angular domain signal. An envelope spectrum slice weighting function is constructed, and a weighted combined envelope spectrum is calculated and fused with time-domain features to form a comprehensive feature set.

[0057] Fault diagnosis involves inputting a comprehensive feature set into a hybrid model, outputting alarm prediction values ​​and inversely normalizing them, and combining the FFT spectrum with the weighted combined envelope spectrum to quantify the severity of the fault.

[0058] Online alarms are set according to weighted probability, and alarm data is added to the historical sample database for online optimization.

[0059] The purpose of this design is to provide an online detection method for wind turbine bearing faults. It adopts weighted combination envelope spectrum technology, which uses two-dimensional spectrum coherent quantification to correlate fault features with resonant frequency bands. Combined with frequency domain signal-to-noise ratio adaptive weight allocation, it aggregates and enhances fault information from multiple resonant frequency bands, effectively offsetting the attenuation effect during signal transmission. This solves the problem that sensors cannot be directly installed in the bearing housing due to wind turbine structural size limitations and need to be magnetically installed through the casing boss, which leads to signal attenuation and distortion.

[0060] Meanwhile, by introducing order tracking and constant angle incremental sampling technology, the correspondence between vibration signal and bearing rotation angle is established, and the time domain signal is converted into angular domain signal, so that the fault characteristic frequency is decoupled from the rotation speed. This ensures that the theoretical fault frequency of components such as inner ring, outer ring, and roller is accurately matched with the detection results, which greatly improves the detection adaptability under varying working conditions. Compared with the fixed frequency domain analysis of the traditional acceleration envelope method, it effectively avoids misjudgment caused by frequency shift.

[0061] like Figure 1 and Figure 2 As shown, in this embodiment, an AC146178 AC fan is used as the monitoring object for real-time monitoring and analysis of the bearing operating status. Due to the structural size limitations of the fan, the accelerometer cannot be directly installed near the bearing housing and can only be installed on the outer surface of the casing. In this embodiment, a flat boss is glued to the connection between the casing and the stator blades, and the vibration sensor is installed on the boss by magnetic attraction to detect the bearing vibration signal.

[0062] In this embodiment, the vibration acceleration time-domain signal acquired during signal acquisition includes the time-domain curve, kurtosis index, and envelope RMS value.

[0063] The time-domain curve is used to reflect the time-domain waveform of the original vibration acceleration;

[0064] The kurtosis index is a statistical measure used to reflect the impulse characteristics of a signal;

[0065] The effective value of the envelope is used to reflect the location where bearing vibration occurs;

[0066] The rotational speed key phase signal is used to record the relationship between the bearing rotation angle and time.

[0067] It should be noted that, in this embodiment, a specific implementation method is provided, and the kurtosis index is expressed by the following formula:

[0068] ;

[0069] Where K is the kurtosis index, and x is the sampled value of the vibration acceleration time-domain signal. For mathematical expectation operators, The mean of the signal. Let V be the signal variance.

[0070] The effective value of the envelope is expressed by the following formula:

[0071] ;

[0072] Among them, E env Here, N represents the effective value of the envelope signal, and N is the number of sampling points for the envelope signal. Let be the sampled value of the i-th envelope signal.

[0073] The rotational speed key phase signal is used to record the relationship between the bearing's rotational angle and time, expressed by the following formula:

[0074] ;

[0075] in, Let t be the bearing rotation angle, and t be time. This is a mapping function between rotational speed and angle.

[0076] Meanwhile, in the signal preprocessing, the vibration acceleration time-domain signal is converted into a vibration acceleration angular-domain signal by inversion, which serves as the direct input for spectral correlation and spectral coherence.

[0077] The transformation relationship is as follows:

[0078] ;

[0079] in, x is the angular domain vibration acceleration signal, and x(·) is the time domain vibration acceleration signal. for The inverse function is used to deduce the corresponding time by rotating the angle.

[0080] The purpose of this design is to focus on acquiring effective signals under the constraints of the wind turbine structure during the signal acquisition phase. Vibration acceleration signals, including time-domain curves, kurtosis indices, and envelope RMS values, are acquired through magnetically attached accelerometers. The kurtosis index reflects the transient characteristics of the fault by quantifying the signal impact characteristics, while the envelope RMS value locates the vibration location through time-domain energy statistics. Simultaneously acquired speed key phase signals establish a mapping relationship between time and bearing rotation angle. Then, the time-domain signal is converted into an angular domain signal through inverse function transformation, eliminating the interference of speed fluctuations on frequency domain analysis. This design adapts to the structural size constraints of the wind turbine. By magnetically attaching the sensor through the casing boss, effective signals are acquired without altering the core structure of the wind turbine. Furthermore, the angular domain signal is converted into a stable input for subsequent spectral coherence calculations and order FFT analysis, solving the frequency ambiguity problem under varying operating conditions and laying the foundation for accurate matching of fault characteristic frequencies.

[0081] In this embodiment, the two-dimensional spectral coherence of the angular domain vibration acceleration signal is estimated, and the correlation between the cyclic frequency and the spectral frequency is quantified. The calculation formula is as follows:

[0082] ;

[0083] in, The two-dimensional spectral coherence coefficient, The cyclic frequency corresponds to the fault characteristic frequency. The frequency of the spectrum corresponds to the frequency of the resonant band. The cross-power spectral density of the angular domain signal. for The self-power spectral density at that location, for The self-power spectral density at that location.

[0084] The fault information content at each spectral frequency is evaluated using a frequency domain signal-to-noise ratio (SNR) metric. An adaptive threshold is set, and an envelope spectral slice weighting function is constructed. The frequency domain SNR calculation formula is as follows:

[0085] ;

[0086] Among them, SNR(f s )for The frequency domain signal-to-noise ratio at that location. For the bandwidth, in this embodiment, the value is set to 50~100Hz. For frequency Spectral energy at the location, for Noise energy at that location.

[0087] Meanwhile, the formula for calculating the weighting function is:

[0088] ;

[0089] Among them, w(f s )for The weight function at the point, The signal-to-noise ratio threshold. , These represent the upper and lower limits of the resonant frequency band.

[0090] Furthermore, the weighted combined envelope spectrum is calculated by integrating the two-dimensional spectral coherence using a weighting function. The calculation formula is as follows:

[0091] ;

[0092] in, for The weighted combination envelope spectrum value, for and Spectral energy at the intersection, df s For the integral infinitesimal element of the spectral frequency;

[0093] Finally, the weighted combined envelope spectrum features are fused with time-domain features, kurtosis indices, and effective envelope values ​​to form a comprehensive feature set for fault diagnosis, expressed as:

[0094] ;

[0095] in, ... These are the characteristic frequencies of faults in various components of the bearing. This is the effective value of vibration acceleration. This represents the effective value of the vibration velocity.

[0096] The purpose of this design is that the core of feature extraction is to coherently quantify the correlation between fault feature frequencies and resonant frequency bands in a two-dimensional spectrum, and then construct an adaptive weighting function based on the frequency domain signal-to-noise ratio measure to selectively enhance fault information in multiple resonant frequency bands, thereby assigning high weights to frequency bands rich in fault information and setting noise-dominated frequency bands to zero. Finally, a weighted combined envelope spectrum is obtained through integration. At the same time, time-domain features are integrated to form a comprehensive feature set containing spatial, energy, and statistical multi-dimensional information. This offsets the signal attenuation and distortion caused by indirect sensor installation, effectively extracts high-frequency impact features of early weak faults, and solves the defect of traditional single-band envelope spectra that cannot capture information in multiple resonant frequency bands. The comprehensive feature set provides comprehensive input data for subsequent fault diagnosis, avoids the diagnostic bias caused by single features, and improves the robustness of fault identification.

[0097] like Figure 4 As shown, this embodiment provides a fault diagnosis neutralization model based on a CNN+LSTM+attention layer hybrid neural network. Fault diagnosis includes structural calculation of the CNN+LSTM+attention layer hybrid neural network on the input comprehensive feature set and bearing fault probability quantification. The structural calculation of the CNN+LSTM+attention layer hybrid neural network on the input comprehensive feature set includes the following sub-steps:

[0098] Input tensor construction: The comprehensive feature set of the input is expanded into a time-series feature tensor adapted to wind turbine vibration, and its expression is:

[0099] ;

[0100] Among them, F in The temporal feature tensor is used as the input to the hybrid neural network. Let be the comprehensive feature set at the t-th sampling time, seq_len be the temporal window length, and the tensor dimension be (batch, seq_len, m), where m is the feature dimension and batch is the batch size.

[0101] In the CNN layer, spatial features are extracted from the temporal feature tensor for local spatial correlation. The calculation formula is as follows:

[0102] ;

[0103] in, The output features of the CNN layer Convolution kernel weights (dimensions) ), k is the kernel size, which can be selected from 3 to 5. For CNN feature dimensions, For bias terms, ReLU activation function , It is the local feature sequence from time t to t+k in the temporal feature tensor;

[0104] In the LSTM layer, the fault development trend in the temporal feature tensor is captured, and its core calculation is as follows:

[0105] ;

[0106] in, , , These are the input gate, forget gate, and output gate of the LSTM layer at time t, respectively. Let be the cell state at time t. Let represent the candidate cell state at time t. Let be the hidden state at time t. , , , These are the weight matrices for the input gate, forget gate, output gate, and cell state, respectively. , , , These are the bias terms for the input gate, forget gate, output gate, and cell state, respectively, and ⊙ represents the Hadamard product;

[0107] At the attention layer, features at critical moments of failure are assigned high weights through an attention mechanism, and the original probability of wind turbine bearing failure is output. The calculation formula is as follows:

[0108] ;

[0109] Among them, To score attention, , Here is the attention weight matrix. For attention layer bias terms, To normalize attention weights, For global feature vectors, For the weights of the fully connected layer, For bias terms of fully connected layers, This represents the original failure probability.

[0110] At the same time, such as Figure 5 As shown in the figure, this embodiment provides a specific step for quantifying the probability of bearing failure, which includes the following sub-steps:

[0111] To generate the order FFT spectrum, perform an order tracking FFT on the vibration acceleration angular domain signal to obtain the order FFT spectrum. The calculation formula is as follows:

[0112] ;

[0113] in, The magnitude of the FFT spectrum is given by order N, where N is the number of sampling points. Let be the rotation angle of the nth sampling point. Here, j represents the order frequency, and j is the imaginary unit.

[0114] The failure probability of each component of the wind turbine bearing is calculated. For the inner ring, outer ring, rollers, and cage of the wind turbine bearing, the failure probability is calculated using the order-wise FFT spectrum and the weighted combined envelope spectrum, and then weighted and fused. The formula for calculating the failure probability of the FFT spectrum is as follows:

[0115] ;

[0116] in, For components FFT failure probability, For components Failure frequency Spectral energy within the range, Normal frequency band energy, Sum all fault types, where k is the index variable for the fault type;

[0117] The formula for calculating the fault probability of the weighted combined envelope spectrum is:

[0118] ;

[0119] in, For components WCES failure probability, This is the WCES reference value for the normal frequency band. The weighted combined envelope spectrum value corresponding to the fault of component c. Sum all fault types, where k is the index variable for the fault type;

[0120] The weighted fusion formula is:

[0121] ;

[0122] in, To weight the failure probability, in this embodiment, the following settings are respectively set and As weighting coefficients, set them respectively. =0.6、 =0.4.

[0123] Fault diagnosis is integrated with a hybrid model. The cyclic frequency sequence of the weighted combined envelope spectrum, the amplitude characteristics of the order FFT spectrum, and the fault probability of each component are used as inputs to a hybrid neural network consisting of a CNN, LSTM, and an attention layer. The hybrid neural network outputs the fault type confidence score, where the inverse normalization formula is as follows:

[0124] ;

[0125] in, This is the alarm prediction value after inverse normalization. , These are the maximum and minimum values ​​of the features in the training set.

[0126] The purpose of this design is to construct a hybrid neural network using temporal feature tensors, extract local spatial correlations of features through CNN layers, capture the temporal trend of fault development through LSTM layers, and strengthen the feature weights at critical moments of the fault through attention layers. Finally, the original fault probability is output through softmax. Fault probability quantization calculates the fault probabilities of the bearing inner ring, outer ring, rollers, and cage respectively through order FFT spectrum and weighted combined envelope spectrum. The probabilities are then fused with a weight of 0.6:0.4 to adapt to the detection needs of mid-to-late stage and early stage faults. This is then fused with the neural network output to obtain the fault type confidence. At the same time, the predicted values ​​of actual physical quantities are restored through inverse normalization. By mining deep temporal features through hybrid neural networks, the problem of traditional shallow signal processing being unable to identify complex faults is solved. Fault probability quantization enables accurate location of faulty components and quantitative assessment of severity, breaking through the limitation of traditional single thresholds that can only make qualitative judgments. Inverse normalization ensures that the predicted values ​​are consistent with the actual working conditions, improving the engineering practicality of the diagnostic results.

[0127] This embodiment provides an online detection system for wind turbine bearing faults based on an online detection method for wind turbine bearing faults. The system includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. It also includes a data acquisition module and an interaction module.

[0128] The acquisition module includes an accelerometer, a Hall sensor, and a data acquisition card. The accelerometer is used to monitor the vibration of the wind turbine bearing and acquire the vibration acceleration time-domain signal. The data acquisition card is used to receive the signals acquired by the accelerometer and the Hall sensor.

[0129] The interaction module includes an HMI screen and a connecting cable. The HMI screen is used to display the processing results of the processor, and the connecting cable is used to connect the HMI screen and the processor.

[0130] The processor is equipped with an edge computing unit for performing two-dimensional spectral coherence estimation, weighted combined envelope spectrum generation, and hybrid model inference.

[0131] The memory contains historical sample data.

[0132] This system uses a processor as its core. The acquisition module achieves synchronous signal acquisition and transmission through accelerometers, Hall sensors, and data acquisition cards. The processor's built-in edge computing unit completes core calculations such as two-dimensional spectral coherence estimation, weighted combination envelope spectrum generation, and hybrid model inference. The memory stores historical sample data to support online model optimization. The interaction module displays diagnostic results intuitively through an HMI screen. The hardware modules are adapted to industrial site installation requirements, and quick-connect interfaces and multi-channel designs enhance deployment flexibility. The edge computing unit ensures real-time detection.

[0133] like Figure 6 As shown in the diagram, this application provides an electronic device structure diagram for an online wind turbine bearing fault detection system. The electronic device includes a memory and a processor. The memory stores computer-readable instructions, and the processor executes the computer-readable instructions stored in the memory to implement the data processing method for online wind turbine bearing fault detection described in any of the above embodiments.

[0134] In this embodiment of the application, the electronic device also includes a bus and a computer program stored in the memory and executable on the processor, such as a program for a data processing method for online detection of wind turbine bearing faults.

[0135] Combination Figure 6 The memory in the electronic device stores multiple computer-readable instructions to implement a data processing method for online detection of wind turbine bearing faults, and the processor can execute the multiple instructions to implement the method.

[0136] Specifically, the processor's implementation method for the above instructions can be found in the description of the relevant steps in the corresponding embodiment of the figure, and will not be repeated here.

[0137] Those skilled in the art will understand that the schematic diagram is merely an example of an electronic device and does not constitute a limitation on the electronic device. The electronic device may be a bus-type structure or a star-type structure. The electronic device may also include more or fewer other hardware or software than shown in the diagram, or different component arrangements. For example, the electronic device may also include input / output devices, network access devices, etc.

[0138] It should be noted that electronic devices are merely examples. Other existing or future electronic products that are suitable for this application should also be included within the scope of protection of this application and are incorporated herein by reference.

[0139] The memory includes at least one type of readable storage medium, which can be non-volatile or volatile. The readable storage medium includes flash memory, portable hard drives, multimedia cards, card-type memory (e.g., SD or DX memory), magnetic storage, magnetic disks, optical disks, etc. In some embodiments, the memory can be an internal storage unit of an electronic device, such as a portable hard drive. In other embodiments, the memory can be an external storage device of the electronic device, such as a plug-in portable hard drive, SmartMediaCard (SMC), Secure Digital (SD) card, or FlashCard. The memory can be used not only to store application software and various types of data installed in the electronic device, such as the code for online detection methods of wind turbine bearing faults, but also to temporarily store data that has been output or will be output.

[0140] In some embodiments, a processor can be composed of integrated circuits, such as a single packaged integrated circuit or multiple integrated circuits packaged with the same or different functions. This includes combinations of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. The processor is the control unit of the electronic device, connecting various components of the device through various interfaces and lines. It executes programs or modules stored in the memory (e.g., a program for online detection of wind turbine bearing faults) and calls data stored in the memory to perform various functions and process data within the electronic device.

[0141] The processor executes the operating system of the electronic device and various installed applications. The processor executes the applications to implement the steps in the above embodiments of the online detection method for wind turbine bearing faults, such as the steps shown in the figure.

[0142] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete this application. The one or more modules / units may be a series of computer-readable instruction segments capable of performing a specific function, which describe the execution process of the computer program in an electronic device. For example, the computer program may be divided into a receiving module, a preprocessing module, a projection module, and a determining module.

[0143] The integrated unit implemented as a software functional module described above can be stored in a computer-readable storage medium. This software functional module, stored in a storage medium, includes several instructions to cause a computer device (which may be a personal computer, computer equipment, or network device, etc.) or processor to execute portions of the online wind turbine bearing fault detection method described in the various embodiments of this application.

[0144] When modules / units integrated into an electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware devices. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above.

[0145] This application provides a computer-readable storage medium storing computer-readable instructions. These instructions are executed by a processor in an electronic device to implement the online detection method for wind turbine bearing faults described in any of the above embodiments. It can be applied to one or more electronic devices. An electronic device is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions. Its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0146] Electronic devices can be any electronic product that allows human-computer interaction with a customer, such as personal computers, tablets, smartphones, personal digital assistants (PDAs), game consoles, interactive network television (IPTV), smart wearable devices, etc.

[0147] Electronic devices may also include network devices and / or client devices. The network devices include, but are not limited to, a single network server, a server group consisting of multiple network servers, or a cloud based on cloud computing consisting of a large number of hosts or network servers.

[0148] The networks in which electronic devices are located include, but are not limited to, the Internet, wide area networks, metropolitan area networks, local area networks, and virtual private networks (VPNs).

[0149] The computer program includes computer program code, which may be in the form of source code, object code, executable file, or some intermediate form. The computer-readable medium may include any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory, and other memory.

[0150] Furthermore, the computer-readable storage medium may primarily include a stored program area and a stored data area, wherein the stored program area may store the operating system, an application program required for at least one function, etc.; and the stored data area may store data created based on the use of blockchain nodes, etc.

[0151] The bus can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This bus can be divided into address bus, data bus, control bus, etc. For ease of illustration, only one arrow is used in the diagram, but this does not mean that there is only one bus or one type of bus. The bus is configured to implement the connection and communication between the memory and at least one processor, etc.

[0152] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0153] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0154] Furthermore, the functional modules in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0155] Furthermore, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices described in the specification may also be implemented by a single unit or device through software or hardware. Terms such as "first," "second," etc., are used to indicate names and do not indicate any specific order.

[0156] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An online detection method for fan bearing faults, applicable to the detection of continuously operating fans in semiconductor processing, characterized in that... Includes the following steps: Signal acquisition involves magnetically installing an accelerometer on the fan casing to monitor the vibration of the fan bearing and collect the time-domain signal of the vibration acceleration. Simultaneously, a Hall sensor is set up near the fan bearing to collect the key phase signal of the fan bearing rotation speed. Signal preprocessing establishes a time-bearing rotation angle correspondence using the rotational speed key phase signal, converting the vibration acceleration time-domain signal into a vibration acceleration angular-domain signal. Feature extraction is performed to estimate the two-dimensional spectral coherence of the vibration acceleration angular domain signal. An envelope spectrum slice weighting function is constructed, and a weighted combined envelope spectrum is calculated and fused with time-domain features to form a comprehensive feature set. Fault diagnosis involves inputting a comprehensive feature set into a hybrid model, outputting an alarm prediction value and inversely normalizing it. The severity of the fault is then quantified by combining the FFT spectrum and the weighted combined envelope spectrum. Online alarms are set according to weighted probability, and alarm data is added to the historical sample database for online optimization. In the signal acquisition, the vibration acceleration time-domain signal includes the time-domain curve, kurtosis index, and envelope RMS value; The time-domain curve is used to reflect the time-domain waveform of the original vibration acceleration; The kurtosis index is a statistical measure used to reflect the impulse characteristics of a signal, and it is expressed by the following formula: ； Where x is the sampled value of the vibration acceleration time-domain signal. For mathematical expectation operators, The mean of the signal. The variance of the signal; The effective envelope value is used to reflect the location of bearing vibration, and the effective envelope value is expressed by the following formula: ； Where N is the number of envelope signal sampling points, This represents the sampled value of the i-th envelope signal; The rotational speed key phase signal is used to record the relationship between the bearing rotation angle and time, which is expressed by the following formula: ； in, Let t be the bearing rotation angle, and t be time. This is a mapping function between rotational speed and angle; In the signal preprocessing, the vibration acceleration time-domain signal is converted into a vibration acceleration angular-domain signal by inversion, which serves as the direct input for spectral correlation and spectral coherence. The conversion relationship is as follows: ； in, Let x(t) be the angular domain vibration acceleration signal, and x(t) be the time domain vibration acceleration signal. for The inverse function is used to deduce the corresponding time by rotating the angle; The feature extraction includes the following sub-steps: To estimate the two-dimensional spectral coherence of the angular domain vibrational acceleration signal and quantize the correlation between the cyclic frequency and the spectral frequency, we have the following equation: ; in, The two-dimensional spectral coherence coefficient (value 0~1). The cyclic frequency corresponds to the fault characteristic frequency. The spectral frequency corresponds to its resonant frequency band. The cross-power spectral density of the angular domain signal. for The self-power spectral density at that location, for The self-power spectral density at the location; The fault information content at each spectral frequency is evaluated using a frequency domain signal-to-noise ratio (SNR) metric. An adaptive threshold is set, and an envelope spectral slice weighting function is constructed. The frequency domain SNR calculation formula is as follows: ； in, The bandwidth can be set to a value of 50~100Hz. For frequency Spectral energy at the location, for Noise energy at the location; Meanwhile, the formula for calculating the weighting function is: ； in, The signal-to-noise ratio threshold. , These are the upper and lower limits of the resonant frequency band; The weighted combined envelope spectrum is calculated by integrating the two-dimensional spectral coherence using a weighting function, and the following equation applies: ; in, The weighted combination envelope spectrum value, for and Spectral energy at the intersection; By fusing the weighted combined envelope spectrum features with temporal features, kurtosis indices, and effective envelope values, a comprehensive feature set for fault diagnosis is formed, as shown in the following formula. ; in, , These are the characteristic frequencies of bearing component failures. This is the effective value of vibration acceleration. This represents the effective value of the vibration velocity.

2. The online detection method for wind turbine bearing faults according to claim 1, characterized in that, The hybrid model used in the fault diagnosis is a hybrid neural network consisting of CNN, LSTM, and an attention layer.

3. The online detection method for wind turbine bearing faults according to claim 2, characterized in that, The fault diagnosis includes structural calculation of the input comprehensive feature set using a hybrid neural network of CNN+LSTM+attention layer and quantification of bearing fault probability.

4. The online detection method for wind turbine bearing faults according to claim 3, characterized in that, The structural calculation of the CNN+LSTM+attention layer hybrid neural network on the input comprehensive feature set includes the following sub-steps: Input tensor construction expands the comprehensive feature set of the input into a temporal feature tensor adapted to wind turbine vibration; In the CNN layer, spatial features are extracted from the temporal feature tensor for local spatial correlation. In the LSTM layer, the fault development trend in the temporal feature tensor is captured; At the attention layer, the features at critical moments of the fault are given high weights through the attention mechanism, and the original probability of the wind turbine bearing failure is output.

5. The online detection method for wind turbine bearing faults according to claim 4, characterized in that, The bearing failure probability quantification includes the following sub-steps: Generate the order FFT spectrum by performing an order tracking FFT on the vibration acceleration angular domain signal to obtain the order FFT spectrum; The failure probability of each component of the wind turbine bearing is calculated. For the inner ring, outer ring, rollers and cage of the wind turbine bearing, the failure probability is calculated by order FFT spectrum and weighted combined envelope spectrum respectively, and then weighted and fused. The fault diagnosis and hybrid model are integrated. The cyclic frequency sequence of the weighted combined envelope spectrum, the amplitude characteristics of the order FFT spectrum, and the fault probability of each component are used as inputs to the CNN+LSTM+attention layer hybrid neural network. The hybrid neural network outputs the confidence level of the fault type.

6. An online detection system for wind turbine bearing faults, characterized in that, The method includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the online detection method for wind turbine bearing faults as described in any one of claims 1 to 5.

7. The online detection system for wind turbine bearing faults according to claim 6, characterized in that, It also includes a data acquisition module and an interaction module; The acquisition module includes an accelerometer, a Hall sensor, and a data acquisition card. The accelerometer is used to monitor the vibration of the wind turbine bearing and acquire the vibration acceleration time-domain signal. The Hall sensor is used to acquire the wind turbine bearing rotation speed key phase signal. The data acquisition card is used to receive the signals acquired by the accelerometer and the Hall sensor. The interaction module includes an HMI screen and a connecting cable. The HMI screen is used to display the processing results of the processor, and the connecting cable is used to connect the HMI screen and the processor. The processor is equipped with an edge computing unit for performing two-dimensional spectral coherence estimation, weighted combined envelope spectrum generation, and hybrid model inference. The memory contains historical sample data.

Images