Variable pitch bearing impact pulse detection method and system based on voiceprint measurement

Abstract

The invention relates to a variable pitch bearing impact pulse detection method and system based on voiceprint measurement, and the method comprises the steps: obtaining an original voiceprint signal during the operation of a variable pitch bearing, and enabling the frequency band of the original voiceprint signal to cover a set range; performing dynamic noise reduction and band-pass filtering processing on the original voiceprint signal to obtain a clean signal; extracting a multi-dimensional feature vector at least comprising an impact strength index ISI, a pulse repetition period PRP and a frequency band energy ratio BER from the cleaning signal; and the multi-dimensional feature vector outputs a continuous damage coefficient representing the damage degree of the bearing based on the prediction model, and the fault type is identified according to the damage coefficient and the numerical value of each feature in the multi-dimensional feature vector. A broadband MEMS microphone array is adopted to collect sound signals, and wind noise is dynamically suppressed through a time-frequency mask; extracting multi-dimensional characteristics of impact strength, pulse period and frequency band energy ratio; and a support vector regression model is constructed, and multi-feature quantification damage degree and fault type identification are fused.

Description

Technical Field

[0001] This invention relates to the field of wind power generation technology, specifically to a method and system for detecting impact pulses in pitch bearings based on acoustic signature measurement. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] As a critical transmission component of wind turbine generators, the health of pitch bearings directly affects the overall operational safety, stability, and power generation efficiency of the turbine. During wind turbine operation, pitch bearings are subjected to complex alternating loads and harsh environmental conditions over extended periods, making them prone to localized damage such as pitting, cracking, and spalling. Failure to detect early faults in a timely manner can lead to bearing failure or even turbine shutdown, resulting in significant economic losses.

[0004] Currently, monitoring solutions based on vibration and temperature sensors are widely used in industry. Vibration sensors determine the condition of bearings by detecting vibration signals during operation, while temperature sensors indirectly reflect faults by monitoring abnormal temperature rises. However, existing technologies have the following significant shortcomings: Frequency band coverage has limitations. Traditional vibration sensors are mostly designed based on a single frequency band, which cannot effectively capture both high-frequency impact pulses (such as transient impacts caused by early pitting corrosion) and low-frequency vibration characteristics (such as modulation phenomena caused by peeling), resulting in the easy omission of early and weak fault characteristics.

[0005] Sensitive to environmental interference. Wind noise interference is significant in windy environments, especially traditional acoustic sensors which have a low signal-to-noise ratio when collecting sound signals, often obscuring effective fault features and making reliable extraction difficult.

[0006] Installation and maintenance are complex. Existing sensors typically require drilling and bolting for installation, which not only compromises the integrity of the bearing structure but also increases installation costs and maintenance difficulty, hindering large-scale application.

[0007] Diagnostic algorithms are limited. Existing methods mostly rely on simple threshold judgments or single indicators, lacking the ability to model multi-feature fusion and nonlinear relationships, making it difficult to distinguish between different types of faults (such as pitting and cracks), and resulting in a high false alarm rate.

[0008] To overcome the above-mentioned shortcomings, existing technologies have proposed some methods, such as using broadband sensors and combining frequency domain analysis algorithms. However, these methods still suffer from poor real-time performance, weak adaptability, and high computational complexity. In particular, there is a lack of systematic solutions for the operating characteristics of pitch bearings and the special environment of wind farms. Summary of the Invention

[0009] To address the technical problems mentioned above, this invention provides a method and system for detecting impact pulses in pitch bearings based on acoustic signature measurement. The method employs a wideband MEMS microphone array to acquire acoustic signals from 100Hz to 80kHz, dynamically suppressing wind noise through time-frequency masking. It extracts multi-dimensional features including impact intensity, pulse period, and frequency band energy ratio, and constructs a support vector regression model to quantify damage levels and identify fault types by integrating multiple features. Combined with magnetically mounted hardware, it achieves non-invasive, high signal-to-noise ratio acquisition and real-time diagnosis, significantly improving the early damage detection capability and wind farm environment adaptability of pitch bearings.

[0010] To achieve the above objectives, the present invention adopts the following technical solution: The first aspect of the present invention provides a method for detecting impact pulses in pitch bearings based on acoustic signature measurement, comprising the following steps: Acquire the original acoustic signature signal during the operation of the pitch bearing, with the frequency band coverage of the original acoustic signature signal within a set range; Dynamic noise reduction and bandpass filtering are performed on the original voiceprint signal to obtain a clean signal; Extract a multi-dimensional feature vector from the clean signal, including at least the impact intensity index (ISI), pulse repetition period (PRP), and frequency band energy ratio (BER). The multidimensional feature vector outputs a continuous damage coefficient representing the degree of bearing damage based on the prediction model. The fault type is identified based on the damage coefficient and the values ​​of each feature in the multidimensional feature vector.

[0011] Furthermore, dynamic noise reduction specifically includes: The original voiceprint signal is subjected to a short-time Fourier transform (STFT) to obtain the time spectrum; Generating a time-frequency mask matrix based on the spectral characteristics of wind noise; The time-frequency spectrum is suppressed by multiplication using a time-frequency mask matrix to obtain the denoised time-frequency spectrum; The inverse short-time Fourier transform is performed on the denoised time spectrum to obtain the denoised time-domain signal.

[0012] Furthermore, a multi-dimensional feature vector is extracted from the clean signal, including at least the impact intensity index (ISI), pulse repetition period (PRP), and frequency band energy ratio (BER), specifically: The formula for calculating the Impact Strength Index (ISI) is as follows: ; The formula for calculating the pulse repetition period (PRP) is as follows: ; The formula for calculating the band power ratio (BER) is: ; in, Represents the peak value of a signal sequence. Represents the root mean square value of the signal sequence; Indicates the first The occurrence time of each pulse peak, where median is the median operator; Indicates high-frequency energy. Indicates low-frequency energy. This represents the smoothing factor.

[0013] Furthermore, the prediction model is a pre-trained Support Vector Regression (SVR) model, and the kernel function of the SVR model is the Radial Basis Function (RBF) kernel, as shown in the following equation: ; in, Indicates the kernel width parameter. This represents the square of the Euclidean norm.

[0014] Furthermore, the fault type is identified, specifically: If the damage coefficient D exceeds the threshold If the impact strength index (ISI) is the dominant characteristic, then the fault type is identified as pitting corrosion. If the damage coefficient D exceeds the threshold If the pulse repetition period (PRP) exhibits abnormal periodicity, then the fault type is identified as spalling. If the damage coefficient D exceeds the threshold If the frequency band energy ratio (BER) is significantly greater than 1, then the fault type is identified as a crack.

[0015] Furthermore, threshold For adaptive thresholding, the formula is: To achieve adaptive adjustment, among which, As the baseline threshold, To adjust the coefficient, This represents the bearing speed.

[0016] A second aspect of the present invention provides a pitch bearing impact pulse detection system based on acoustic signature measurement, comprising: MEMS microphone array is used to acquire the raw acoustic signature signal during the operation of the pitch bearing, and the frequency band coverage of the raw acoustic signature signal is set. The processor performs dynamic noise reduction and bandpass filtering on the original acoustic signal to obtain a clean signal; it extracts a multi-dimensional feature vector from the clean signal, including at least the impact intensity index (ISI), pulse repetition period (PRP), and frequency band energy ratio (BER); the multi-dimensional feature vector outputs a continuous damage coefficient representing the degree of bearing damage based on the prediction model; and the fault type is identified based on the damage coefficient and the values ​​of each feature in the multi-dimensional feature vector.

[0017] A third aspect of the present invention provides a computer program product including computer-readable instructions that, when executed on an electronic device, cause the electronic device to implement the above-described pitch bearing impact pulse detection method based on acoustic signature measurement.

[0018] A fourth aspect of the present invention provides an electronic device including at least one processor and a memory connected to the processor, the memory being used to store a computer program; the processor being used to execute the computer program, enabling the electronic device to implement the above-described pitch bearing impact pulse detection method based on acoustic signature measurement.

[0019] A fifth aspect of the present invention provides a computer storage medium carrying one or more computer programs, which, when executed by an electronic device, enable the electronic device to implement the above-described pitch bearing impact pulse detection method based on acoustic signature measurement.

[0020] Compared with existing technologies, one or more of the above technical solutions have the following beneficial effects: 1. By employing a wideband acoustic signature sensor (100 Hz to 80 kHz) combined with a dynamic time-frequency masking noise reduction algorithm, high-frequency impact pulse signals are effectively captured and wind noise interference is significantly suppressed, resulting in a substantial improvement in the signal-to-noise ratio. This enables the detection of early-stage micro-damage with a minimum size of 0.5 mm. Compared to traditional vibration monitoring methods, the warning time can be advanced by tens of hours, achieving true early fault detection.

[0021] 2. Multi-dimensional and complementary feature vectors, including the Impact Intensity Index (ISI), Pulse Repetition Period (PRP), and Frequency Band Energy Ratio (BER), were extracted. A Support Vector Regression (SVR) model was then used for multi-feature fusion and nonlinear regression, enabling accurate quantification of damage severity and effective differentiation of different fault types such as pitting, cracking, and spalling. In high-noise environments (e.g., wind speed > 8 m / s), the accuracy rate reached 89.6%, with a false alarm rate of less than 5%, significantly reducing the risk of misjudgment.

[0022] 3. In terms of hardware, it adopts a MEMS microphone array and magnetic mounting design, eliminating the need for drilling holes in the bearing. This enables non-invasive and rapid installation and maintenance. It has strong vibration resistance (>10g) and an IP67 protection rating, allowing it to operate stably in harsh wind environments ranging from -40°C to 85°C. This solves the pain points of traditional sensors, such as complex installation and damage to the bearing structure.

[0023] 4. This solution optimizes algorithm efficiency through a collaborative computing architecture of STM32 microcontroller and FPGA, supporting embedded real-time inference of SVR models with a computation latency of less than 0.2 seconds, meeting the real-time requirements of online monitoring. Simultaneously, the overall solution boasts low power consumption (<1W), supports PoE power supply, and facilitates on-site deployment and long-term operation. Attached Figure Description

[0024] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0025] Figure 1 This is a schematic diagram of the overall process of the pitch bearing impact pulse detection method based on acoustic signature measurement provided by one or more embodiments of the present invention. Detailed Implementation

[0026] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0027] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0028] As introduced in the background section, existing bearing detection solutions suffer from limited frequency band coverage, sensitivity to environmental interference, complex installation and maintenance, and a single diagnostic algorithm. Some existing technologies attempt to solve these problems, but they suffer from poor real-time performance, weak adaptability, and high computational complexity. In particular, they lack systematic solutions for the operating characteristics of pitch bearings and the special environment of wind farms.

[0029] The following embodiments present a method and system for detecting impact pulses in pitch bearings based on acoustic signature measurements. A wideband MEMS microphone array is used to acquire acoustic signals from 100Hz to 80kHz, and wind noise is dynamically suppressed using time-frequency masks. Multi-dimensional features, including impact intensity, pulse period, and frequency band energy ratio, are extracted. A support vector regression model is constructed to quantify the damage level and identify the fault type by fusing multiple features. Combined with magnetically mounted hardware, non-invasive, high signal-to-noise ratio acquisition and real-time diagnosis are achieved, significantly improving the early damage detection capability and wind field adaptability of pitch bearings.

[0030] Example 1: like Figure 1 As shown, the method for detecting impact pulses in pitch bearings based on acoustic signature measurement includes the following steps: Acquire the original acoustic signature signal during the operation of the pitch bearing, with the frequency band coverage of the original acoustic signature signal within a set range; Dynamic noise reduction and bandpass filtering are performed on the original voiceprint signal to obtain a clean signal; Extract a multi-dimensional feature vector from the clean signal, including at least the impact intensity index (ISI), pulse repetition period (PRP), and frequency band energy ratio (BER). The multidimensional feature vector outputs a continuous damage coefficient representing the degree of bearing damage based on the prediction model. The fault type is identified based on the damage coefficient and the values ​​of each feature in the multidimensional feature vector.

[0031] (1) Acquisition and preprocessing of voiceprint signals.

[0032] Objective: To capture the impact pulse signal of the pitch bearing using a wideband acoustic signature sensor and dynamically suppress wind noise interference to ensure signal purity.

[0033] Step 1.1 Signal Acquisition. Acquire the raw voiceprint signal at a sampling frequency of not less than 160 kHz (satisfying the Nyquist sampling theorem). The signal frequency band covers 100Hz to 80kHz; the original acoustic signature signal includes a mixture of bearing impact sound emission, wind noise, mechanical noise and other original time-domain signals.

[0034] Step 1.2 Dynamic noise reduction algorithm. Apply time-frequency mask. Suppressing wind noise and obtaining a clean signal: ; in, This indicates a noisy voiceprint signal. This represents a time-frequency mask matrix, generated based on wind noise spectral characteristics (such as low-frequency dominance), typically estimated using a threshold or neural network. The mask values ​​range from [0,1], where values ​​close to 1 represent noise-dominant regions. Noise suppression is achieved through mask subtraction.

[0035] This formula can be derived as follows: assuming the noise is additive interference. ,Right now However, in the time-frequency domain, the masking method is approximately a multiplication suppression, and in practical implementation, the short-time Fourier transform (STFT) spectrum can be used for operation.

[0036] Step 1.2.1 Perform a short-time Fourier transform (STFT) on the signal to obtain the time spectrum. Estimated mask For noise spectrum estimation, Avoid dividing by zero for small positive numbers; Step 1.2.2 Reconstruct the time-domain signal using the inverse STFT to obtain a clean time-domain signal.

[0037] Step 1.3 Bandpass filter design.

[0038] A 4th-order Butterworth bandpass filter is used, with a passband range of [missing information]. to The transfer function is: ; in, The central angular frequency is calculated using the geometric mean. For example, , hour, ,but ; The quality factor is used to control the filter bandwidth (typical value). To achieve the maximum flat response).

[0039] The transfer function is the standard form of a 4th-order Butterworth filter, composed of two cascaded 2nd-order filters. The transfer function of each 2nd-order unit is... The cascaded denominator is squared to obtain the overall fourth-order response.

[0040] During the design process, the s-domain can be converted to the z-domain using a bilinear transform to realize a digital filter. Furthermore, this can be combined with a low-pass filter (cutoff frequency). (used to suppress low-frequency background noise), the transfer function is similar to a low-order Butterworth form.

[0041] (2) Impact pulse feature extraction.

[0042] Objective: To extract key features from the denoised signal for subsequent fault diagnosis, thereby improving the robustness and discriminative power of the features.

[0043] This step targets the high-frequency impact pulse signal of the pitch bearing, quantifying its amplitude, periodicity, and spectral distribution characteristics to ensure characteristic stability under noise interference.

[0044] Step 2.1 Impact Intensity Index (ISI): This step quantifies the ratio of pulse amplitude to background noise by calculating the peak and root mean square values ​​of the signal within a time window. A logarithmic scale is used to enhance sensitivity to weak signals. ; in, The peak value of the signal is typically represented by scanning a time-domain signal sequence. Find the maximum absolute value, that is: ; The root mean square value is represented by the following formula: ; in, The number of sampling points. This index reflects the relative intensity of the impact pulse. A logarithmic transformation (with a coefficient of 20, derived from the acoustic definition of decibels) can convert the linear ratio to the decibel level, facilitating threshold setting and fault classification (e.g., ISI > 15 dB indicates potential pitting).

[0045] In practical calculations, the signal is processed using a window function (such as the Hanning window) to reduce edge effects. The typical window length is 10% to 20% of the signal length. This formula is derived from the standard definition of impulse intensity and improves resolution in low signal-to-noise ratio environments by logarithmically compressing the dynamic range.

[0046] This step quantifies the relative intensity of the impact signal; a larger value indicates a more severe impact.

[0047] Step 2.2 Pulse Repetition Period (PRP): The signal is smoothed and peaks are detected to find the occurrence times of a series of impulse pulses, which are used to reflect the periodicity of the fault and quantify the interval distribution between pulse events, as shown in the following formula: ; The above formula calculates the time interval between consecutive pulses, reflecting the periodicity of the fault. Its reciprocal corresponds to the fault characteristic frequency, which is related to the bearing speed and geometric dimensions. Among these, Indicates the first The occurrence time of each pulse peak is used to identify the pulse position sequence through a peak detection algorithm. ( (This refers to the number of pulses); the median is used to suppress the influence of outliers and ensure robustness. This indicator is suitable for periodic failures such as bearing spalling, and its value is typically related to the bearing speed. Related.

[0048] Step 2.2.1 Apply Gaussian filtering to smooth the signal to assist in peak localization, with the threshold set to 3 to 5 times RMS; Step 2.2.2 If the number of pulses m < 5, then PRP is set to the default value (e.g., signal duration / 2) to avoid invalid calculations. This formula avoids the bias of traditional frequency estimation by using the median of the statistical interval.

[0049] Step 2.3: Bandwidth Energy Ratio (BER), calculate the Welch power spectral density of the signal, evaluate the high and low frequency energy distribution, and quantify the spectral shift of the impulse pulse. ; in, High-frequency energy is typically defined as the energy in the frequency spectrum. (The spectrum S(f) is calculated using FFT). Indicates low-frequency energy. This represents a smoothing factor to prevent the denominator from being zero and to suppress numerical instability. This index reflects the fault-induced spectral shift (high-frequency dominance indicates sharp impact, such as a crack; low-frequency dominance indicates background vibration).

[0050] Step 2.3.1 Perform Welch power spectral density estimation on the signal to reduce variance, with a window function overlap of 50%.

[0051] Step 2.3.2 It can be dynamically adjusted to 2-3 times the main frequency of the signal. The formula design avoids direct division overflow, ensures numerical stability through ε, and facilitates comparison with the threshold (for example, BER>1 indicates high-frequency anomaly, triggering an early warning).

[0052] (3) Bearing damage assessment model.

[0053] Objective: Based on extracted impact pulse features, this study aims to quantify bearing damage levels using a support vector regression (SVR) model, and to achieve fault type identification and early warning. The model employs multi-feature fusion input and outputs damage coefficients to achieve nonlinear regression prediction, thereby improving the accuracy and robustness of diagnosis.

[0054] Step 3.1 SVR model construction.

[0055] Step 3.1.1 Input feature vector Output damage coefficient (Where 0 represents healthy, and 1 represents severe damage): ; in, Represents the Lagrange multipliers. Represents the kernel function. Indicates the bias term. This represents the number of support vectors. The formula originates from the dual form of SVR and is used to regress the degree of bearing damage. In actual solutions, it is optimized using quadratic programming tools (such as the LIBSVM library). And calculate using KKT conditions Model hyperparameters (such as penalty factors) and insensitive loss The mean squared error (MSE) is minimized through grid search and cross-validation.

[0056] The model uses kernel functions (such as the RBF kernel) to map features to a high-dimensional space for nonlinear regression calculations, outputting a continuous damage coefficient. D=0 represents health, D=1 represents severe damage, and intermediate values ​​represent different degrees of degeneration.

[0057] Step 3.1.2 Kernel Function Selection: A radial basis function (RBF) kernel is used to handle the nonlinearity of the feature space. ; in This represents the kernel width parameter (typical values ​​0.1-10, optimized through cross-validation to control the smoothness and generalization ability of the model). The square of the Euclidean norm is expressed by the following formula: (j (Corresponding to ISI, PRP, BER). This kernel function maps low-dimensional features to an infinite-dimensional Hilbert space, improving SVR's adaptability to noisy data. It can be dynamically adjusted to the reciprocal of the characteristic variance to balance overfitting and underfitting.

[0058] Step 3.1.3 Fault type identification, based on prediction Threshold fusion is used to identify fault types (such as pitting). And ISI is the dominant force; peeling: Furthermore, PRP is abnormal; cracks: comprehensive. And BER offset). Threshold adaptive adjustment: ; in This represents the baseline threshold (typically 0.5, determined from historical data). This indicates the adjustment factor (typical value 0.01-0.05, set according to noise level). This indicates the bearing rotational speed (in rad / s). This formula ensures that the threshold varies with operating conditions, improving recognition accuracy.

[0059] This solution enables quantitative assessment of bearing damage, adapts to nonlinear data and noise interference, and improves early warning capabilities. In environments with wind speeds >10m / s, prediction accuracy is improved by 20% (MSE reduced to 0.05), with a false alarm rate <5%; the minimum damage detection size reaches 0.5mm. The model is specifically optimized for pitch bearings, and the RBF kernel ensures high robustness, supporting embedded real-time computation.

[0060] experiment.

[0061] The detection performance of this solution in the experimental scenario is as follows: Damage identification: The smallest successfully detected damage area was 0.3 mm. 2 The surface damage from the immersion water occurred 42 hours earlier than traditional vibration monitoring methods. This result was obtained through surface measurement of the immersion water, and the quantitative index is the damage area A, calculated using the following formula: ; Where S represents the boundary of the damaged region, which is obtained by integration using an image processing algorithm.

[0062] Resistance: In an environment with a wind speed of 8 m / s, the identification accuracy rate reaches 89.6% (the undamaged positioning accuracy rate is 52.3%). Accuracy P is defined as:

[0063] TP, TN, FP, and FN represent true positive, true negative, false positive, and false negative, respectively.

[0064] These results demonstrate that, under conditions of strong noise and dynamic wind fields, this scheme exhibits excellent early damage detection capability and robustness, significantly reducing the risk of misjudgment compared to existing technologies.

[0065] Comparative experiment.

[0066] To further highlight the advantages of this scheme, comparative experiments were conducted with traditional vibration and acoustic emission methods. Experimental conditions were standardized: impact energy 5J, noise level SNR = 20dB, and 200 test samples. Key performance indicators are compared in the table below. Table 1 Comparison of Key Performance Indicators

[0067] Minimum time-sensitive damage size: This invention can detect 0.2mm. 2 The damage is far superior to the 2.5mm of traditional methods. 2 1.0mm of harmonic emission method 2 This indicator is calculated through threshold comparison, representing the minimum time-related damage size. satisfy: ; in Indicates the size of the damage. This represents the detection probability for a given size.

[0068] Detection time: The average detection time of this method is 0.18s, significantly lower than the 1.0s of the traditional vibration method. This time T is obtained through end-to-end timing. T = tcollection + tprocessing + tprediction Each component represents the duration of signal acquisition, noise reduction, and model prediction, respectively. MathType is equivalent to: T equals the summation, with three t subscripts; rendered as a time decomposition formula, supporting data from the implementation examples.

[0069] False alarm rate: The false alarm rate of this scheme is 4.1%, far lower than the 18.7% of the traditional vibration method and the 9.5% of the acoustic emission method. The false alarm rate (FPR) is calculated as follows: ; The comparative results show that this proposed method outperforms existing methods in terms of detection sensitivity, response speed, and accuracy, making it particularly suitable for engineering scenarios with scarce data and high noise levels. These experimental data have fully validated this proposed method, supporting its practical application in the field of composite material health monitoring.

[0070] This solution employs a wideband acoustic signature sensor (100 Hz to 80 kHz) combined with a dynamic time-frequency masking noise reduction algorithm to effectively capture high-frequency impact pulse signals and significantly suppress wind noise interference, resulting in a substantial improvement in the signal-to-noise ratio. This enables the detection of early-stage micro-damage with a minimum size of 0.5 mm, providing an early warning time tens of hours earlier than traditional vibration monitoring methods, thus achieving true early fault detection.

[0071] This scheme extracts multi-dimensional and complementary feature vectors such as the impact intensity index (ISI), pulse repetition period (PRP), and frequency band energy ratio (BER), and uses a support vector regression (SVR) model for multi-feature fusion and nonlinear regression. This enables precise quantification of damage severity and effective differentiation of different fault types such as pitting, cracking, and spalling. In high-noise environments (e.g., wind speed > 8 m / s), the accuracy reaches 89.6%, with a false alarm rate of less than 5%, significantly reducing the risk of misjudgment.

[0072] Example 2: A pitch bearing impact pulse detection system based on acoustic signature measurement includes: A microphone array is used to acquire the raw acoustic signature signal during the operation of the pitch bearing, with the frequency band of the raw acoustic signature signal covering a set range. The processor performs dynamic noise reduction and bandpass filtering on the original acoustic signal to obtain a clean signal; it extracts a multi-dimensional feature vector from the clean signal, including at least the impact intensity index (ISI), pulse repetition period (PRP), and frequency band energy ratio (BER); the multi-dimensional feature vector outputs a continuous damage coefficient representing the degree of bearing damage based on the prediction model; and the fault type is identified based on the damage coefficient and the values ​​of each feature in the multi-dimensional feature vector.

[0073] Hardware design: Sensing unit: MEMS microphone array (signal-to-noise ratio >70dB, frequency response 20Hz-20kHz); Processing unit: STM32H743 microcontroller combined with FPGA for collaborative processing, supporting real-time SVR inference (computation latency <0.2s). Storage units: NOR Flash (4MB) for firmware storage, SDRAM (64MB) for feature cache; Communication interface: PoE power supply and data transmission, supports Ethernet or wireless protocols, power consumption <1W in low power mode.

[0074] Installation plan: Location: 3 o'clock position on the outer ring of the pitch bearing (the area of ​​maximum stress), which facilitates the capture of radial impact signals; Fixing method: Double fixation with magnetic base and industrial glue, vibration resistance >10g, no drilling required for installation; Protection design: IP67 protection rating, temperature resistance -40°C to 85°C, adaptable to harsh wind field environments.

[0075] This solution employs a MEMS microphone array and magnetic mounting design, eliminating the need for drilling holes in the bearing. This enables non-invasive, rapid installation and maintenance. It boasts strong vibration resistance (>10g) and an IP67 protection rating, allowing stable operation in harsh wind environments ranging from -40°C to 85°C. This solution addresses the pain points of traditional sensors, such as complex installation and damage to the bearing structure.

[0076] This solution optimizes algorithm efficiency through a collaborative computing architecture of STM32 microcontroller and FPGA, supports embedded real-time inference of SVR models, and has a computation latency of less than 0.2 seconds, meeting the real-time requirements of online monitoring. Meanwhile, the overall solution has low power consumption (<1W), supports PoE power supply, and is easy to deploy in the field and operate for extended periods.

[0077] Example 3: A computer program product includes computer-readable instructions that, when executed on an electronic device, cause the electronic device to implement the aforementioned pitch bearing impact pulse detection method based on acoustic signature measurement.

[0078] Example 4: An electronic device includes at least one processor and a memory connected to the processor, the memory being used to store a computer program; the processor is used to execute the computer program, enabling the electronic device to implement the above-described pitch bearing impact pulse detection method based on acoustic signature measurement.

[0079] Example 5: A computer storage medium carrying one or more computer programs, which, when executed by an electronic device, enable the electronic device to implement the above-mentioned pitch bearing impact pulse detection method based on acoustic signature measurement, including the following steps.

[0080] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for detecting impact pulses in pitch bearings based on acoustic signature measurement, characterized in that, Includes the following steps: Acquire the original acoustic signature signal during the operation of the pitch bearing, wherein the frequency band of the original acoustic signature signal covers a set range; The original voiceprint signal is subjected to dynamic noise reduction and bandpass filtering to obtain a clean signal; Extract a multi-dimensional feature vector from the cleaning signal, including at least the impact intensity index (ISI), pulse repetition period (PRP), and frequency band energy ratio (BER). The multi-dimensional feature vector outputs a continuous damage coefficient representing the degree of bearing damage based on the prediction model. The fault type is identified based on the damage coefficient and the values ​​of each feature in the multi-dimensional feature vector.

2. The method for detecting impact pulses in pitch bearings based on acoustic signature measurement as described in claim 1, characterized in that, Dynamic noise reduction, specifically: The original voiceprint signal is subjected to a short-time Fourier transform (STFT) to obtain the time spectrum; Generating a time-frequency mask matrix based on the spectral characteristics of wind noise; The time-frequency mask matrix is ​​used to perform multiplication suppression on the time-frequency spectrum to obtain the noise-reduced time-frequency spectrum. The denoised time-domain signal is obtained by performing an inverse short-time Fourier transform on the denoised time-domain spectrum.

3. The method for detecting impact pulses in pitch bearings based on acoustic signature measurement as described in claim 1, characterized in that, Extract a multi-dimensional feature vector from the clean signal, including at least the impact intensity index (ISI), pulse repetition period (PRP), and frequency band energy ratio (BER). Specifically: The formula for calculating the impact strength index (ISI) is as follows: ; The formula for calculating the pulse repetition period (PRP) is as follows: ; The formula for calculating the frequency band energy ratio (BER) is as follows: ; in, Represents the peak value of a signal sequence. Represents the root mean square value of the signal sequence; Indicates the first The occurrence time of each pulse peak, where median is the median operator; Indicates high-frequency energy. Indicates low-frequency energy. This represents the smoothing factor.

4. The method for detecting impact pulses in pitch bearings based on acoustic signature measurement as described in claim 1, characterized in that, The prediction model is a pre-trained Support Vector Regression (SVR) model, and the kernel function of the SVR model is the Radial Basis Function (RBF) kernel, as shown in the following equation: ； in, Indicates the kernel width parameter. This represents the square of the Euclidean norm.

5. The method for detecting impact pulses in pitch bearings based on acoustic signature measurement as described in claim 1, characterized in that, Identify the fault type, specifically: If the damage coefficient D exceeds the threshold If the impact strength index (ISI) is the dominant feature, then the fault type is identified as pitting corrosion. If the damage coefficient D exceeds the threshold If the pulse repetition period (PRP) exhibits abnormal periodicity, then the fault type is identified as peeling. If the damage coefficient D exceeds the threshold If the frequency band energy ratio (BER) is significantly greater than 1, then the fault type is identified as a crack.

6. The method for detecting impact pulses in pitch bearings based on acoustic signature measurement as described in claim 5, characterized in that, The threshold For adaptive thresholding, the formula is: To achieve adaptive adjustment, among which, As the baseline threshold, To adjust the coefficient, This represents the bearing speed.

7. A pitch bearing impact pulse detection system based on acoustic signature measurement, characterized in that, include: MEMS microphone array is used to acquire the raw acoustic signature signal during the operation of the pitch bearing, and the frequency band coverage of the raw acoustic signature signal is set. The processor performs dynamic noise reduction and bandpass filtering on the original acoustic signal to obtain a clean signal; it extracts a multi-dimensional feature vector from the clean signal, including at least the impact intensity index (ISI), pulse repetition period (PRP), and frequency band energy ratio (BER); the multi-dimensional feature vector outputs a continuous damage coefficient representing the degree of bearing damage based on the prediction model; and the fault type is identified based on the damage coefficient and the values ​​of each feature in the multi-dimensional feature vector.

8. A computer storage medium, characterized in that, The storage medium carries one or more computer programs that, when executed by an electronic device, enable the electronic device to perform the steps of the pitch bearing impact pulse detection method based on acoustic signature measurement as described in any one of claims 1-6.

9. A computer program product, characterized in that, Includes computer-readable instructions that, when executed on an electronic device, cause the electronic device to perform the steps in the pitch bearing impact pulse detection method based on acoustic signature measurement as described in any one of claims 1-6.

10. An electronic device, characterized in that, It includes at least one processor and a memory connected to the processor, the memory being used to store a computer program; the processor is used to execute the computer program, enabling the electronic device to perform the steps in the pitch bearing impact pulse detection method based on acoustic signature measurement as described in any one of claims 1-6.

Images