Motor noise detection method and system based on waveform analysis

Abstract

The application discloses a motor noise detection method and system based on waveform analysis, and relates to the related field of motor fault diagnosis.The method comprises the following steps: collecting original noise signals in the time domain when the motor is running to obtain original waveform signals, performing Fourier transform on the original waveform signals, extracting frequency spectrum characteristics, and obtaining frequency spectrum energy distribution; performing threshold segmentation on the frequency spectrum energy distribution, screening out motors that meet preset quality standards but have preset abnormal indications, and obtaining critical qualified motors; extracting original waveform signals of the critical qualified motors, extracting transient impact components from the original waveform signals, and obtaining transient characteristics; and matching the transient characteristics with early fault modes in a pre-constructed fault feature library, and outputting potential fault detection results.The application solves the technical problem that existing motor noise detection cannot identify qualified motors with potential faults at present, and achieves the technical effect of accurately identifying qualified motors with potential faults.

Description

Technical Field

[0001] This application relates to the field of motor fault diagnosis, and in particular to a method and system for detecting motor noise based on waveform analysis. Background Technology

[0002] In the field of motor manufacturing and application, motor noise detection is a crucial step in ensuring motor quality and reliability. Accurately identifying motors that appear to pass inspection but actually harbor potential faults is extremely important, as these motors often fail shortly after being put into use, severely impacting production schedules and increasing maintenance costs. Currently, the main method for solving motor fault detection problems is to use conventional noise detection technology to perform a comprehensive analysis of motor operating noise and determine whether the motor is qualified according to established standards. Qualified motors are allowed to leave the factory, while unqualified ones are returned for processing. However, this current method, relying solely on a single conventional qualification standard, lacks in-depth analysis of the subtle features in motor noise signals that reflect early potential faults. This results in an inability to effectively distinguish motors that currently pass inspection but have potential fault risks. Consequently, these motors quickly fail after entering the market, causing losses for users.

[0003] Currently, the relevant technologies for motor noise detection have a technical problem: they cannot identify motors that are currently qualified but have potential faults. Summary of the Invention

[0004] This application provides a motor noise detection method based on waveform analysis. It employs time-domain acquisition of the noise signal during motor operation to obtain the original waveform signal. Fourier transform is used to convert the time-domain signal into a frequency-domain signal, extracting spectral energy distribution features. Threshold segmentation is applied to these spectral features to classify motors into three categories: fully qualified normal motors, substandard unqualified motors, and critically qualified motors that meet basic standards but exhibit abnormal characteristics. For critically qualified motors, transient impact components are further extracted from their original waveform signals to obtain transient features. These transient features are then matched with known early fault modes. If a match is successful, the potential fault type is output; otherwise, it is marked as a state requiring monitoring. This approach solves the technical problem of existing motor noise detection methods that cannot identify motors that are currently qualified but have potential faults, achieving the technical effect of accurately identifying qualified motors with potential faults.

[0005] This application provides a motor noise detection method based on waveform analysis, comprising: acquiring the original noise signal of the motor during operation in the time domain to obtain the original waveform signal; performing Fourier transform on the original waveform signal to extract spectral features and obtain spectral energy distribution; performing threshold segmentation on the spectral energy distribution to screen out motors that meet preset quality standards but have preset abnormal indicators, thereby obtaining critically qualified motors; extracting the original waveform signal of the critically qualified motors, extracting transient impact components from the original waveform signal to obtain transient features; and matching the transient features with early fault modes in a pre-constructed fault feature library to output potential fault detection results.

[0006] In a possible implementation, the original noise signal during motor operation is acquired in the time domain to obtain the original waveform signal. A Fourier transform is then performed on the original waveform signal to extract spectral features and obtain the spectral energy distribution. The following processing is then performed: At the motor's rated speed, a time-domain noise signal is acquired using a microphone array at a preset sampling frequency; the time-domain noise signal is denoised to obtain the original waveform signal; a Fast Fourier Transform is performed on the original waveform signal to obtain a spectrum; and the amplitudes of the main frequency, bearing characteristic frequency, and electromagnetic harmonic frequency in the spectrum are calculated to generate the spectral energy distribution.

[0007] In a possible implementation, the spectral energy distribution is threshold-segmented to filter out motors that meet preset quality standards but have preset abnormal indicators, thus obtaining critically qualified motors. The following processing is then performed: The main frequency amplitude of historical qualified motor samples is dynamically acquired; based on the dynamic threshold range of the main frequency amplitude, the preset quality standard is determined; the bearing characteristic frequency amplitude of the historical qualified motor samples is dynamically acquired; based on the dynamic threshold range of the bearing characteristic frequency amplitude, qualified bearing characteristic frequency indicators are determined; the electromagnetic harmonic frequency amplitude of the historical qualified motor samples is dynamically acquired; based on the dynamic threshold range of the electromagnetic harmonic frequency amplitude, qualified electromagnetic harmonic frequency indicators are determined; amplitudes that do not meet the qualified bearing characteristic frequency indicators and the qualified electromagnetic harmonic frequency indicators are set as preset abnormal indicators; if the main frequency amplitude in the spectral energy distribution meets the preset quality standard, and the preset abnormal indicator exists in the bearing characteristic frequency amplitude or electromagnetic harmonic frequency amplitude, the corresponding motor is marked as a critically qualified motor.

[0008] In a possible implementation, the following process is performed: the historical qualified motor sample is a motor that is the same model as the motor to be tested, has been in service, and has not experienced any faults during the early fault exposure period.

[0009] In a possible implementation, transient impulse components are extracted from the original waveform signal to obtain transient features, and the following processing is performed: wavelet packet decomposition is performed on the original waveform signal to obtain a time-frequency energy distribution matrix; pulse detection is performed on the time-frequency energy distribution matrix to calculate the pulse interval and amplitude, thereby obtaining the transient features.

[0010] In a possible implementation, a fault feature library is pre-built, and the following processing is performed: dynamically collecting a first historical critical qualified motor sample, wherein the first historical critical qualified motor sample is a motor that is consistent with the model of the motor to be tested, has been in service, and has experienced a first fault mode during the early fault exposure period; obtaining the first historical transient features of the first historical critical qualified motor sample; establishing a first mapping relationship between the first historical transient features and the first fault mode, and constructing the fault feature library based on the first mapping relationship.

[0011] In a possible implementation, a fault feature library is pre-built, and the following processing is performed: Dynamically collecting second historical critically qualified motor samples, where the second historical critically qualified motor sample is a motor whose model is inconsistent with the motor to be tested but whose fault association parameters are consistent, which has been in service and exhibited a second fault mode within the early fault exposure period; the fault association parameters include bearing model, rated speed, and number of poles; obtaining the second historical transient characteristics of the second historical critically qualified motor samples; collecting first historical qualified motor samples and second historical qualified motor samples that correspond one-to-one with the first historical critically qualified motor samples and the second historical critically qualified motor samples, calculating a conversion factor based on the frequency offset ratio between the two; converting the second historical transient characteristics according to the conversion factor to obtain second converted historical transient characteristics; establishing a second mapping relationship between the second converted historical transient characteristics and the second fault mode; and adding the second mapping relationship to the fault feature library.

[0012] In a possible implementation, the fault feature library is constructed based on the first mapping relationship, and the following processing is performed: the first mapping relationship is grouped according to different fault modes to obtain multiple first mapping relationship groups; all historical transient features are extracted from each first mapping relationship group to form multiple historical transient feature libraries; the multiple historical transient feature libraries are labeled with corresponding fault modes to obtain multiple early fault modes, and the multiple early fault modes are integrated to obtain the fault feature library.

[0013] In a possible implementation, the transient feature is matched with an early fault pattern in a pre-built fault feature library to output a potential fault detection result. The following processing is performed: traversing the fault feature library to extract a first early fault pattern; calculating the similarity between the transient feature and the historical transient features contained in the first early fault pattern; if the calculation result is greater than or equal to a preset similarity threshold, the first early fault pattern is output as a potential fault detection result.

[0014] This application also provides a motor noise detection system based on waveform analysis, comprising: a spectral energy distribution acquisition module, used to acquire the original noise signal of the motor during operation in the time domain to obtain the original waveform signal, perform Fourier transform on the original waveform signal, extract spectral features, and obtain the spectral energy distribution; a critical qualified motor acquisition module, used to perform threshold segmentation on the spectral energy distribution, screen out motors that meet preset quality standards but have preset abnormal indicators, and obtain critical qualified motors; a transient feature acquisition module, used to extract the original waveform signal of the critical qualified motor, extract transient impact components from the original waveform signal, and obtain transient features; and a potential fault detection module, used to match the transient features with early fault modes in a pre-built fault feature library, and output potential fault detection results.

[0015] The proposed method and system for motor noise detection based on waveform analysis first acquires the original noise signal of the motor during operation in the time domain to obtain the original waveform signal. Then, Fourier transform is performed on the original waveform signal to extract spectral features and obtain the spectral energy distribution. Next, threshold segmentation is performed on the spectral energy distribution to filter out motors that meet preset quality standards but exhibit preset abnormal indicators, thus identifying critically qualified motors. Then, the original waveform signal of the critically qualified motor is extracted, and transient impact components are extracted from the original waveform signal to obtain transient features. Finally, the transient features are matched with early fault modes in a pre-constructed fault feature library to output the potential fault detection result. This achieves the technical effect of accurately identifying qualified motors with potential faults. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings of the embodiments of the present invention will be briefly described below. Flowcharts are used in this application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed precisely in sequence. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from these processes.

[0017] Figure 1This is a flowchart illustrating the motor noise detection method based on waveform analysis provided in an embodiment of this application.

[0018] Figure 2 This is a schematic diagram of the structure of a motor noise detection system based on waveform analysis provided in an embodiment of this application.

[0019] Figure labeling: 10 for spectrum energy distribution acquisition module, 20 for critical qualified motor acquisition module, 30 for transient characteristic acquisition module, and 40 for potential fault detection module. Detailed Implementation

[0020] The above description is merely an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, specific embodiments of this application are given below.

[0021] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description of this application will be provided in conjunction with the accompanying drawings. The described embodiments should not be considered as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] In the following description, references to "some embodiments" describe a subset of all possible embodiments. However, it is understood that "some embodiments" can be the same or different subsets of all possible embodiments and can be combined with each other without conflict. The terms "first" and "second" are used merely to distinguish similar objects and do not represent a specific ordering of objects. The terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules not explicitly listed or inherent to these processes, methods, products, or devices. Unless otherwise defined, 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 application belongs. The terminology used herein is for the purpose of describing embodiments of this application only.

[0023] This application provides a method for detecting motor noise based on waveform analysis, such as... Figure 1 As shown, the method includes:

[0024] Step S100: The original noise signal during motor operation is acquired in the time domain to obtain the original waveform signal. The original waveform signal is then subjected to Fourier transform to extract spectral features and obtain the spectral energy distribution.

[0025] Specifically, high-precision acoustic sensors, such as capacitive microphones or accelerometers, are placed at specific locations around the motor. The raw noise signal from the motor's operation is then sampled in the time domain at a certain sampling frequency to obtain the raw waveform signal. This signal is presented in digital form, namely a set of discrete amplitude values ​​and corresponding time points. Time-domain acquisition refers to the process of continuously acquiring signal samples at a certain frequency along the time axis, directly reflecting the amplitude of the signal as it changes over time, and revealing the overall trend and periodicity of the signal.

[0026] Next, the original waveform signal is processed using the Fast Fourier Transform (FFT) algorithm. FFT is a mathematical tool for converting time-domain signals into frequency-domain signals. By converting it to the frequency domain, spectral features are extracted to obtain the spectral energy distribution, which contains information about different frequency components and their corresponding energy levels.

[0027] For example, suppose that when the motor is running, the sensor collects a time-domain waveform signal, collecting a total of M data points. The horizontal axis represents time t (in seconds), and the vertical axis represents amplitude A (in Pascals or other sound pressure units). After FFT operation, a spectrum is obtained, with the horizontal axis changing to frequency f (in Hertz), and the vertical axis representing the amplitude corresponding to each frequency (reflecting the energy level of that frequency component).

[0028] In one possible implementation, the original noise signal during motor operation is acquired in the time domain to obtain an original waveform signal. A Fourier transform is then performed on the original waveform signal to extract spectral features and obtain the spectral energy distribution. Step S100 further includes step S110, where the time-domain noise signal is acquired using a microphone array at a preset sampling frequency under the motor's rated speed. Specifically, the microphone array consists of multiple microphones arranged in a certain geometric shape, such as a linear array or a circular array, capable of receiving motor noise signals from different directions and positions. The preset sampling frequency is determined based on the frequency range of the motor noise, following the Nyquist sampling theorem; the sampling frequency is at least twice the highest frequency of the signal. For example, if the highest frequency of the motor noise is 20kHz, the sampling frequency can be set to 44.1kHz or 48kHz. The signal acquired by the array is then converted to a digitized time-domain noise signal via analog-to-digital conversion (ADC).

[0029] For example, an array of eight microphones is used, evenly distributed around the motor, with each microphone spaced 10 centimeters apart. The sampling frequency is set to 48 kHz, and the noise of the motor running at its rated speed is collected for 10 seconds, resulting in eight sets of time-domain noise signal data, each containing 480,000 sampling points. These data are stored in matrix form.

[0030] Step S120 involves denoising the time-domain noise signal to obtain the original waveform signal. Specifically, spectral subtraction and wavelet thresholding are used to denoise the time-domain noise signal. Taking spectral subtraction as an example, the background noise spectrum when the motor is stationary is first obtained. Then, a short-time Fourier transform (STFT) is performed on the collected motor running noise signal to obtain the spectrum. The amplitude of the background noise spectrum is subtracted from the amplitude of each frequency component (zero is taken if the result is negative). Then, an inverse short-time Fourier transform (ISTFT) is performed to obtain the denoised original waveform signal. Wavelet thresholding involves wavelet decomposition of the time-domain signal, soft-thresholding or hard-thresholding quantization of the wavelet coefficients, removal of small coefficients representing noise, and reconstruction to obtain the denoised signal. The original waveform signal reduces environmental noise interference and can more purely reflect the noise characteristics of the motor itself.

[0031] Step S130: Perform a Fast Fourier Transform (FFT) on the original waveform signal to obtain a spectrum. Specifically, the FFT algorithm can quickly convert a time-domain signal into a frequency-domain signal. The number of FFT points is determined based on the sampling frequency and number of data points of the original waveform signal; a power of 2 can be chosen to improve computational efficiency. After the FFT operation, a complex-form spectrum is obtained. Taking the amplitude (modulus) portion yields the spectrum. The horizontal axis represents frequency, from 0 to half the sampling frequency (according to the sampling theorem), and the vertical axis represents the amplitude of each frequency component, reflecting the energy of that frequency component in the signal.

[0032] Step S140: Calculate the amplitudes of the main frequency, bearing characteristic frequency, and electromagnetic harmonic frequency in the spectrum diagram to generate the spectrum energy distribution. Specifically, the main frequency corresponds to the motor's rotational speed frequency, which can be calculated using parameters such as the number of poles and rated speed of the motor; the bearing characteristic frequency includes the characteristic frequencies of the inner ring, outer ring, and rolling elements, and its calculation formula is related to the bearing model and motor speed; the electromagnetic harmonic frequency is related to factors such as the number of poles and winding structure of the motor. By finding these specific frequency points in the spectrum diagram, reading their amplitudes, and organizing them into a spectrum energy distribution data structure, for example, storing them in dictionary form, with the key being the frequency name (main frequency, bearing characteristic frequency, etc.) and the value being the corresponding amplitude.

[0033] This implementation uses a microphone array to collect noise signals, enabling more comprehensive reception of noise from different directions and positions of the motor. This increases signal coverage and improves sampling accuracy. Through noise reduction processing, it effectively removes background environmental noise and some random noise interference, improving the signal-to-noise ratio and making the original waveform signal more clearly present the characteristics of motor noise. By using FFT to quickly obtain the spectrum diagram and calculating the amplitude of key frequency components such as the main frequency, bearing characteristic frequency, and electromagnetic harmonic frequency, it can accurately extract the spectral characteristics of different motor components (such as motor rotor, bearing, stator, etc.), enabling targeted monitoring of the operating status of various key components inside the motor. This helps to detect potential motor faults at an early stage, such as bearing wear and electromagnetic imbalance.

[0034] Step S200: Threshold segmentation is performed on the spectrum energy distribution to screen out motors that meet the preset quality standards but have preset abnormal indicators, thus obtaining critical qualified motors.

[0035] Specifically, threshold segmentation refers to a processing method that sets one or more thresholds to divide data samples into different categories or intervals, used in signal processing to distinguish between normal and abnormal signal states. Based on pre-set quality standards, i.e., determined energy threshold ranges for each frequency band, the spectral energy distribution is compared and judged point-by-point or segment-by-segment. This can be done using conditional statements and array manipulation functions in programming languages ​​to filter out motors that meet the preset quality standards (spectral energy within the acceptable range) but exhibit preset abnormal indicators (such as small energy fluctuations at certain specific frequencies). This identifies critically qualified motors, meaning motors that are basically operating normally but contain early signs of failure, existing on the borderline between qualified and unqualified. Such motors may subsequently deteriorate and fail, requiring close monitoring and attention.

[0036] In one possible implementation, the spectral energy distribution is threshold-segmented to filter out motors that meet preset quality standards but exhibit preset abnormal indicators, thus obtaining critically qualified motors. Step S200 further includes step S210, dynamically acquiring the main frequency amplitude of historical qualified motor samples, and determining the preset quality standard based on the dynamic threshold range of the main frequency amplitude. The historical qualified motor samples are motors that are consistent with the model of the motor to be tested, have been in service, and have not experienced any faults during the early fault exposure period. Specifically, through a data acquisition system, the main frequency amplitude data of motors that are consistent with the model of the motor to be tested (completely identical to the motor to be tested in terms of design, structure, and performance parameters), have been in service (have been run under actual working conditions, accumulating real operating data and fault information), and have not experienced any faults during the early fault exposure period are acquired in real time. The early fault exposure period refers to a specific time period (e.g., 6 months) after the motor starts operating. If the motor does not experience any faults during this period, it is a qualified motor; if the motor experiences a fault during this period, it is a critically qualified motor. These motors, as historical qualified samples, are stored in a database. To maintain the timeliness and representativeness of the samples, a sliding window mechanism is employed, with a fixed sample size, such as 1000 samples. When new qualified motor sample data arrives, the oldest samples are removed from the window in chronological order, ensuring that the sample is always the most up-to-date and representative dataset. Based on these main frequency amplitude data, their statistical characteristics (such as mean, standard deviation, etc.) are calculated to determine the dynamic threshold range of the main frequency amplitude, i.e., the preset quality standard. For example, the dynamic threshold range can be set to mean ± 3 times the standard deviation to cover the main frequency amplitude range of the vast majority of normally operating motors.

[0037] For example, assuming that the statistical data of the main frequency amplitude of historical qualified motor samples shows a mean of 80dB and a standard deviation of 2dB, the dynamic threshold range is set to 80dB ± 3 × 2dB, or 74dB-86dB. Only motors whose main frequency amplitude falls within this range are considered to meet the preset quality standard.

[0038] Step S220: Dynamically acquire the bearing characteristic frequency amplitude of the historical qualified motor samples. Based on the dynamic threshold range of the bearing characteristic frequency amplitude, determine the qualified bearing characteristic frequency indicators. Specifically, the same dynamic acquisition method is used to obtain the bearing characteristic frequency amplitude of the historical qualified motor samples. A sliding window is used to maintain a constant and continuously updated sample size. Based on the acquired data, its statistical characteristics are calculated to determine the dynamic threshold range of the bearing characteristic frequency amplitude, thereby clarifying the qualified bearing characteristic frequency indicators. For example, using the same mean ± 3 times standard deviation method, if the mean is 60dB and the standard deviation is 1.5dB, then the dynamic threshold range of the qualified bearing characteristic frequency amplitude is 60dB ± 3 × 1.5dB, i.e., 55.5dB-64.5dB. Bearing frequency characteristics within this range are considered normal.

[0039] Step S230 involves dynamically acquiring the electromagnetic harmonic frequency amplitude of the historical qualified motor samples and determining qualified electromagnetic harmonic frequency indicators based on the dynamic threshold range of the electromagnetic harmonic frequency amplitude. Specifically, similar to steps S210 and S220, the electromagnetic harmonic frequency amplitude of the historical qualified motor samples is dynamically acquired, and the samples are dynamically updated. The dynamic threshold range of the electromagnetic harmonic frequency amplitude is determined based on statistical characteristics and serves as the qualified electromagnetic harmonic frequency indicator.

[0040] Step S240: Amplitudes that do not conform to the qualified bearing characteristic frequency indicators and the qualified electromagnetic harmonic frequency indicators are set as preset abnormal indicators. Specifically, amplitudes that do not conform to the aforementioned qualified bearing characteristic frequency indicators and qualified electromagnetic harmonic frequency indicators, i.e., amplitudes exceeding the corresponding dynamic threshold range, are defined as preset abnormal indicators. This can be achieved by writing program code and setting conditional judgment rules to compare the collected spectral energy distribution data one by one and filter out characteristic amplitudes that do not conform to the normal range.

[0041] Step S250: If the amplitude of the dominant frequency in the spectral energy distribution meets the preset quality standard, and the amplitude of the bearing characteristic frequency or the electromagnetic harmonic frequency contains the preset abnormal indicator, the corresponding motor is marked as a critically qualified motor. Specifically, a comprehensive judgment is made on the spectral energy distribution. Program logic is written to first check whether the amplitude of the dominant frequency meets the preset quality standard (within the dynamic threshold range), and then check whether the amplitude of the bearing characteristic frequency or the electromagnetic harmonic frequency contains the preset abnormal indicator (exceeding the corresponding dynamic threshold range). If both conditions are met simultaneously, i.e., the dominant frequency is normal but the bearing or electromagnetic harmonic frequency is abnormal, the corresponding motor is marked as a critically qualified motor. Marking can be done by adding specific tags to the database or setting corresponding flag bits in the data structure.

[0042] This approach dynamically collects historical data on qualified motor samples and continuously updates the sample set, enabling timely reflection of the motor's normal characteristics over time in the actual operating environment. This makes the determination of preset quality standards and qualification indicators more accurate. Determining dynamic threshold ranges based on statistical characteristics, following data distribution patterns, better accommodates reasonable fluctuations during normal operation and avoids misjudging normal motors as abnormal due to unreasonable fixed threshold settings. By comprehensively considering the characteristics of multiple key frequency components such as the main frequency, bearing characteristic frequency, and electromagnetic harmonic frequency, it can effectively screen out motors that appear qualified but may actually show early signs of failure.

[0043] Step S300: Extract the original waveform signal of the critical qualified motor, extract the transient impact component from the original waveform signal, and obtain transient characteristics.

[0044] Specifically, for the original waveform signals of the selected critically qualified motors, wavelet transform is used to extract transient impact components. Transient impact components are short-lived, rapidly changing parts of the signal, caused by mechanical faults (such as collisions, sudden frictional changes, etc.), and appear as spikes or pulses in the waveform. Wavelet transform has good time-frequency localization properties, enabling it to focus on the transient changes in the signal. By selecting appropriate wavelet basis functions (such as db4, sym5, etc.) and decomposition scales, the original waveform signal is decomposed at multiple scales to separate the detailed signal components containing transient impact characteristics, thus obtaining transient features. These features include key parameters such as the time point, amplitude, and duration of the impact.

[0045] For example, the original waveform signal is decomposed into three levels using the db4 wavelet basis function to obtain detail coefficients at different scales, corresponding to transient impact information in different frequency bands. By reconstructing the detail signal, the waveform of the transient impact component is obtained, and information such as the time and magnitude of the impact peak are statistically analyzed as transient features.

[0046] In one possible implementation, transient impulse components are extracted from the original waveform signal to obtain transient features. Step S300 further includes step S310, performing wavelet packet decomposition on the original waveform signal to obtain a time-frequency energy distribution matrix. Specifically, wavelet packet decomposition is a multi-resolution analysis method that can decompose a signal into sub-band signals of different frequency bands. Specifically, by selecting appropriate wavelet basis functions (such as db4, sym5, etc.) and the number of decomposition levels (such as 3 levels), the original waveform signal is decomposed at multiple scales to obtain a time-frequency energy distribution matrix. The time-frequency energy distribution matrix contains energy information at different time points and in different frequency bands.

[0047] For example, suppose the original waveform signal has a sampling frequency of 48kHz and a data length of 1024 points. Using the db4 wavelet basis function, a 3-level wavelet packet decomposition is performed, resulting in 8 sub-band signals (frequency bands). Each sub-band signal corresponds to a different frequency band. The energy value of each sub-band signal is calculated, forming a time-frequency energy distribution matrix. The rows of the matrix represent time points, the columns represent frequency bands, and the matrix elements represent the energy value of the corresponding time point and frequency band.

[0048] Step S320: Pulse detection is performed on the time-frequency energy distribution matrix to calculate the pulse interval and amplitude, thereby obtaining the transient characteristics. Specifically, pulse detection is used to identify transient impulse components in the signal and calculate the pulse interval and amplitude. Specifically, an energy threshold is set based on the background noise energy level, and energy values ​​higher than this threshold are considered potential transient impulse components. The time-frequency energy distribution matrix is ​​traversed, and when the energy value at a certain time point and frequency band exceeds the set threshold, it is marked as a pulse. For each marked pulse, its amplitude (the magnitude of the energy value) and interval (the time difference between adjacent pulses) are calculated.

[0049] For example, in a time-frequency energy distribution matrix, an energy threshold is set to three standard deviations of the background noise energy. When traversing the matrix, if the energy value in the high-frequency band at a certain time point exceeds the threshold, it is marked as a pulse. The amplitude and interval of this pulse are recorded. This process is repeated to obtain the amplitude and interval of all pulses, forming a transient feature vector.

[0050] This implementation uses wavelet packet decomposition, which can decompose the signal into multiple sub-band signals of different frequency bands, capturing subtle changes in the signal at different time points and frequency bands. It can extract transient impulse components more accurately than traditional methods such as Fourier transform.

[0051] Step S400: Match the transient features with early fault modes in the pre-built fault feature library and output potential fault detection results.

[0052] Specifically, a fault feature library is established, and the transient characteristics of a large number of known fault types of motors are analyzed and classified in advance to form standard feature templates for different early fault modes. Early fault modes are relatively concealed and feature-indistinct states exhibited in the initial stage of a fault, containing typical early signs and characteristics of a specific fault type. During detection, the extracted transient feature vectors (such as vectors containing parameters like peak impact, interval time, and energy distribution) are matched one by one with each template in the fault feature library using similarity metrics such as Euclidean distance and correlation coefficient. When the similarity exceeds a set threshold, a successful match is determined, and the corresponding potential fault type is output; if no template with sufficiently high similarity is found, it is marked as requiring monitoring.

[0053] In one possible implementation, a fault feature library is pre-built. Step S400 further includes step S410, dynamically collecting a first historical critically qualified motor sample. This first historical critically qualified motor sample consists of motors of the same model as the motor to be tested, already in service, and exhibiting the first fault mode during the early fault exposure period. Specifically, motors of the same model as the motor to be tested, already in service, and exhibiting the first fault mode during the early fault exposure period are dynamically collected as samples (i.e., the first historical critically qualified motor sample). This dynamic collection also employs a sliding window mechanism, maintaining a constant sample size. When new samples arrive, a corresponding number of old samples are removed in chronological order to ensure the timeliness and representativeness of the samples. The first fault mode refers to the specific fault type that occurs in the motor during the early fault exposure period, such as bearing wear. Different motor samples may exhibit different fault modes or the same fault mode. There is a one-to-one correspondence between samples and fault modes; that is, the sample exhibits only one type of fault during the early fault exposure period. If multiple samples correspond to the same fault mode, it indicates that this fault is relatively common in that model of motor.

[0054] Step S420: Obtain the first historical transient features of the first historical critically qualified motor sample. Specifically, using the method in step S300 (such as wavelet packet decomposition), the original waveform signal of the collected first historical critically qualified motor sample is processed to extract transient features. That is, wavelet packet decomposition is first performed on the original waveform signal to obtain the time-frequency energy distribution matrix, and then pulse detection is performed to calculate transient features such as pulse interval and amplitude. These transient features are the characteristic manifestations of the motor when the first fault mode occurs, and can reflect the specific physical phenomena of the fault. For example, bearing wear may cause periodic impact pulses, and the corresponding transient features are specific pulse intervals and amplitudes.

[0055] Step S430: Establish a first mapping relationship between the first historical transient features and the first fault mode, and construct the fault feature library based on the first mapping relationship. Specifically, through data structures such as dictionaries or database tables, the acquired first historical transient features are associated with the corresponding first fault modes to establish a first mapping relationship, where the key is the transient feature (such as a combination of parameters like pulse interval and amplitude), and the value is the corresponding fault mode. Based on the above mapping relationship, a fault feature library is constructed. This library is dynamically updated, reflecting the latest and most representative fault feature information as new samples are continuously added and old samples are removed. Each entry in the fault feature library contains a specific fault mode and its corresponding transient features, used to provide a benchmark and reference for fault detection.

[0056] This approach utilizes a sample of first-historical critically qualified motors of the same model as the motor under test that failed during the early fault exposure period. The resulting fault feature library is highly targeted and practical. This mapping relationship based on actual fault data can more accurately identify potential fault types and reduce misjudgments.

[0057] In one possible implementation, a fault feature library is pre-built. Step S400 further includes step S440, dynamically collecting a second set of historically critically qualified motor samples. These second historically critically qualified motor samples are motors that are different in model from the motor under test but have the same fault association parameters, have been in service, and exhibited a second fault mode during the early fault exposure period. The fault association parameters include bearing type, rated speed, and number of poles. Specifically, the second historically critically qualified motor sample set is continuously updated using a sliding window mechanism. The second historically critically qualified motor samples have different models than the motor under test, but their bearing type, rated speed, and number of poles are the same. That is, although the overall design of the second historically critically qualified motor samples may differ from that of the motor under test, key components (such as bearings) and basic operating parameters (such as rated speed and number of poles) are consistent with the motor under test, and a specific fault mode has appeared during the early fault exposure period. Introducing second historically critically qualified motor samples supplements the fault feature library. Since these samples have the same fault association parameters as the motor under test, their fault features can be converted and used for fault detection of the motor under test, complementing the first set of historically critically qualified motor samples.

[0058] Step S450: Obtain the second historical transient features of the second historical critically qualified motor sample. Specifically, using the method in step S300 (such as wavelet packet decomposition), the original waveform signal of the acquired second historical critically qualified motor sample is processed to extract transient features. These transient features reflect the characteristic performance of the sample motor when the second fault mode occurs.

[0059] Step S460: Collect first and second historical qualified motor samples that correspond one-to-one with the first and second historical critical qualified motor samples, and calculate a conversion factor based on their frequency offset ratio. Specifically, collect first and second historical qualified motor samples that correspond one-to-one with the first and second historical critical qualified motor samples, meaning the first and second historical qualified motor samples have the same model. Furthermore, these qualified motor samples did not experience any faults during the early fault exposure period. The conversion factor is calculated based on the frequency offset ratio between the first and second historical qualified motor samples. The frequency offset ratio refers to the difference in characteristic frequencies (such as main frequency, bearing characteristic frequency, etc.) between the motor under test and the second historical qualified motor sample under the same operating conditions, reflecting the frequency difference caused by different motor models. The frequency offset ratio is calculated by comparing the characteristic frequencies of the first and second historical qualified motor samples. The conversion factor, i.e., the frequency offset ratio, is used to convert the time parameters of the second historical transient characteristics from the frequency range of the second historical qualified motor sample to the frequency range of the motor under test.

[0060] Step S470: Convert the second historical transient feature according to the conversion factor to obtain the second converted historical transient feature. Establish a second mapping relationship between the second converted historical transient feature and the second fault mode, and add the second mapping relationship to the fault feature library. Specifically, adjust the second historical transient feature using the conversion factor to obtain the second converted historical transient feature, ensuring that the second historical transient feature is consistent with the frequency range of the motor under test, making the features comparable. Associate the second converted historical transient feature with the second fault mode using a data structure (such as a dictionary or database table) to establish the second mapping relationship, where the key is the converted transient feature and the value is the corresponding fault mode. Add the second mapping relationship to the fault feature library. In this way, the fault feature library contains fault features from different models of motors (but with the same fault association parameters), enriching the content of the feature library.

[0061] This implementation reduces reliance on samples that are identical to the motor model being tested by introducing motor samples with the same fault-related parameters. By using conversion factors to adjust the transient characteristics of different motor models, the fault feature library can still perform effective fault detection even when sufficient samples of the same model are lacking.

[0062] In one possible implementation, the fault feature library is constructed based on the first mapping relationship. Step S430 further includes step S431, grouping the first mapping relationships according to different fault modes to obtain multiple first mapping relationship groups. Specifically, the first mapping relationships are grouped according to different fault modes. For example, by traversing the set of first mapping relationships, mapping relationships with the same fault mode are grouped into the same group based on the identifier of the fault mode (such as fault mode name or number). The mapping relationships within each group correspond to the same fault mode, and the fault modes between groups are different, thus obtaining multiple first mapping relationship groups.

[0063] Step S432: Extract all historical transient features from each first mapping relationship group to construct multiple historical transient feature libraries. Specifically, each first mapping relationship group contains several mapping relationships with the same failure mode, and each mapping relationship is associated with a specific historical transient feature. These features are collected to form a historical transient feature library corresponding to a specific failure mode. If there are n first mapping relationship groups, n historical transient feature libraries can be obtained.

[0064] Step S433: The multiple historical transient feature libraries are labeled with corresponding fault modes to obtain multiple early fault modes. These early fault modes are then integrated to obtain the fault feature library. Specifically, a corresponding fault mode label is added to each historical transient feature library, clearly indicating the specific fault type associated with that feature library. Then, these labeled historical transient feature libraries are integrated together to construct the fault feature library. This library uses fault modes as the classification basis, systematically organizing and storing the transient features corresponding to different fault modes.

[0065] This implementation reduces the number of samples that need to be compared one by one during fault detection by merging samples with the same fault mode. Using a grouped feature library, only a few fault modes need to be compared during detection, thus speeding up the detection process and improving efficiency.

[0066] In one possible implementation, the transient features are matched with early fault modes in a pre-built fault feature library to output potential fault detection results. Step S400 further includes step S480, traversing the fault feature library and extracting a first early fault mode. Specifically, each early fault mode in the fault feature library is accessed sequentially. The fault feature library contains multiple early fault modes, each associated with a corresponding historical transient feature library. During the traversal, the first early fault mode and its corresponding historical transient feature library are extracted first. For example, the first early fault mode in the fault feature library might be a bearing wear fault mode, and the corresponding historical transient feature library contains multiple transient features related to bearing wear.

[0067] Step S490: Calculate the similarity between the transient feature and the historical transient features contained in the first early fault mode. If the calculation result is greater than or equal to a preset similarity threshold, the first early fault mode is output as a potential fault detection result. Specifically, the extracted transient features are compared with the historical transient features contained in the first early fault mode. The similarity calculation method can use Euclidean distance, correlation coefficient, etc. A preset similarity threshold (e.g., 0.8) is set. If the calculation result is greater than or equal to the threshold, the transient feature is considered to match the early fault mode. The matched early fault mode is output as a potential fault detection result. If the similarity is lower than the threshold, the next early fault mode is traversed until a matching mode is found or all modes are traversed.

[0068] This implementation reduces the complexity of each calculation and improves overall detection efficiency by extracting and matching early failure modes one by one. Using similarity calculation to quantify and match transient features allows for an objective assessment of the similarity between features, effectively reducing false positives and improving detection accuracy.

[0069] This application employs time-domain acquisition of noise signals during motor operation to obtain the original waveform signal. Fourier transform is used to convert the time-domain signal into a frequency-domain signal, extracting spectral energy distribution features. Threshold segmentation is then applied to these spectral features to classify motors into three categories: fully qualified normal motors, substandard unqualified motors, and critically qualified motors that meet basic standards but exhibit abnormal characteristics. For critically qualified motors, transient impact components are further extracted from their original waveform signals to obtain transient features. These transient features are then matched with known early fault modes. If a match is successful, the potential fault type is output; otherwise, it is marked as a state requiring monitoring. This approach solves the technical problem of existing motor noise detection methods that cannot identify motors that are currently qualified but have potential faults, achieving the technical effect of accurately identifying qualified motors with potential faults.

[0070] In the above text, refer to Figure 1 A method for detecting motor noise based on waveform analysis according to an embodiment of the present invention has been described in detail. Next, reference will be made to... Figure 2 A waveform analysis-based motor noise detection system according to an embodiment of the present invention is described.

[0071] The waveform analysis-based motor noise detection system according to embodiments of the present invention addresses the technical problem of existing motor noise detection methods failing to identify motors that are currently qualified but have potential faults, achieving the technical effect of accurately identifying qualified motors with potential faults. The waveform analysis-based motor noise detection system includes: a spectrum energy distribution acquisition module 10, a critical qualified motor acquisition module 20, a transient feature acquisition module 30, and a potential fault detection module 40.

[0072] The spectrum energy distribution acquisition module 10 is used to acquire the original noise signal during motor operation in the time domain to obtain the original waveform signal, perform Fourier transform on the original waveform signal, extract spectrum features, and obtain the spectrum energy distribution; the critical qualified motor acquisition module 20 is used to perform threshold segmentation on the spectrum energy distribution, screen out motors that meet the preset quality standards but have preset abnormal indicators, and obtain the critical qualified motors; the transient feature acquisition module 30 is used to extract the original waveform signal of the critical qualified motor, extract the transient impact component from the original waveform signal, and obtain the transient features; the potential fault detection module 40 is used to match the transient features with the early fault modes in the pre-built fault feature library and output the potential fault detection results.

[0073] The specific configuration of the spectrum energy distribution acquisition module 10 will be described in detail below. As mentioned above, the original noise signal during motor operation is acquired in the time domain to obtain the original waveform signal. The original waveform signal is then subjected to Fourier transform to extract spectral features and obtain the spectrum energy distribution. The spectrum energy distribution acquisition module 10 may further include: a time-domain noise signal acquisition unit for acquiring the time-domain noise signal at a preset sampling frequency through a microphone array under the rated speed of the motor; a noise reduction processing unit for performing noise reduction processing on the time-domain noise signal to obtain the original waveform signal; a fast Fourier transform unit for performing a fast Fourier transform on the original waveform signal to obtain a spectrum diagram; and a spectrum energy distribution generation unit for calculating the amplitude of the main frequency, bearing characteristic frequency, and electromagnetic harmonic frequency in the spectrum diagram to generate the spectrum energy distribution.

[0074] The specific configuration of the critical qualified motor acquisition module 20 will be described in detail below. As mentioned above, the spectral energy distribution is segmented by a threshold to filter out motors that meet the preset quality standards but have preset abnormal indicators, thus obtaining critical qualified motors. The critical qualified motor acquisition module 20 may further include: a preset quality standard determination unit for dynamically collecting the main frequency amplitude of historical qualified motor samples and determining the preset quality standard based on the dynamic threshold range of the main frequency amplitude; a qualified bearing characteristic frequency indicator determination unit for dynamically collecting the bearing characteristic frequency amplitude of the historical qualified motor samples and determining the qualified bearing characteristic frequency indicator based on the dynamic threshold range of the bearing characteristic frequency amplitude; qualified bearing characteristic frequency indicator determination unit; qualified bearing characteristic frequency indicator determination unit. The electromagnetic harmonic frequency indicator determination unit is used to dynamically collect the electromagnetic harmonic frequency amplitude of the historical qualified motor samples, and determine the qualified electromagnetic harmonic frequency indicator based on the dynamic threshold range of the electromagnetic harmonic frequency amplitude; the preset abnormal indicator setting unit is used to set the amplitude of the qualified bearing characteristic frequency indicator and the qualified electromagnetic harmonic frequency indicator as the preset abnormal indicator; the critical qualified motor marking unit is used to mark the corresponding motor as a critical qualified motor if the main frequency amplitude in the spectrum energy distribution meets the preset quality standard, and the preset abnormal indicator exists in the bearing characteristic frequency amplitude or electromagnetic harmonic frequency amplitude.

[0075] The critical qualified motor acquisition module 20 may further include: the historical qualified motor sample is a motor that is consistent with the model of the motor to be tested, has been in service, and has not experienced any faults during the early fault exposure period.

[0076] The specific configuration of the transient feature acquisition module 30 will be described in detail below. As mentioned above, transient impact components are extracted from the original waveform signal to obtain transient features. The transient feature acquisition module 30 may further include: a wavelet packet decomposition unit for performing wavelet packet decomposition on the original waveform signal to obtain a time-frequency energy distribution matrix; and a pulse detection unit for performing pulse detection on the time-frequency energy distribution matrix, calculating the pulse interval and amplitude, and obtaining the transient features.

[0077] The specific configuration of the potential fault detection module 40 will be described in detail below. As mentioned above, the potential fault detection module 40 may further include: a first historical critical qualified motor sample dynamic acquisition unit for dynamically acquiring first historical critical qualified motor samples, wherein the first historical critical qualified motor samples are motors that are consistent with the model of the motor to be tested, have been in service, and have exhibited a first fault mode during the early fault exposure period; a first historical transient feature acquisition unit for acquiring the first historical transient features of the first historical critical qualified motor samples; and a fault feature library construction unit for establishing a first mapping relationship between the first historical transient features and the first fault mode, and constructing the fault feature library based on the first mapping relationship.

[0078] The pre-built fault feature library and the potential fault detection module 40 may further include: a second historical critical qualified motor sample dynamic acquisition unit for dynamically acquiring second historical critical qualified motor samples, wherein the second historical critical qualified motor sample is a motor whose model is inconsistent with the motor to be tested but whose fault association parameters are consistent, which has been in service and has exhibited a second fault mode during the early fault exposure period, and the fault association parameters include bearing model, rated speed and number of poles; a second historical transient feature acquisition unit for acquiring the second historical transient features of the second historical critical qualified motor samples; a conversion coefficient calculation unit for acquiring first historical qualified motor samples and second historical qualified motor samples that correspond one-to-one with the first historical critical qualified motor samples and the second historical critical qualified motor samples, and calculating a conversion coefficient based on the frequency offset ratio between the two; and a fault feature library expansion unit for converting the second historical transient features according to the conversion coefficient to obtain second converted historical transient features, establishing a second mapping relationship between the second converted historical transient features and the second fault mode, and adding the second mapping relationship to the fault feature library.

[0079] The fault feature library is constructed based on the first mapping relationship. The fault feature library construction unit may further include: a grouping subunit for grouping the first mapping relationship according to different fault modes to obtain multiple first mapping relationship groups; a historical transient feature library construction subunit for extracting all historical transient features from each first mapping relationship group to construct multiple historical transient feature libraries; and an early fault mode integration subunit for marking the multiple historical transient feature libraries with corresponding fault modes to obtain multiple early fault modes, and integrating the multiple early fault modes to obtain the fault feature library.

[0080] Specifically, the transient features are matched with early fault patterns in a pre-built fault feature library to output potential fault detection results. The potential fault detection module 40 may further include: a fault feature library traversal unit for traversing the fault feature library and extracting a first early fault pattern; and a potential fault detection result output unit for calculating the similarity between the transient features and the historical transient features contained in the first early fault pattern. If the calculation result is greater than or equal to a preset similarity threshold, the first early fault pattern is output as the potential fault detection result.

[0081] The waveform analysis-based motor noise detection system provided in this embodiment of the invention can execute the waveform analysis-based motor noise detection method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method.

[0082] Although this application makes various references to certain modules in the system according to the embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy distinction between each other and are not used to limit the scope of protection of this invention.

[0083] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application. In some cases, the actions or steps described in this application can be performed in a different order than that shown in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. A method for detecting motor noise based on waveform analysis, characterized in that, The method includes: The original noise signal of the motor during operation is acquired in the time domain to obtain the original waveform signal. The original waveform signal is then subjected to Fourier transform to extract the spectral features and obtain the spectral energy distribution. The spectrum energy distribution is segmented by a threshold to screen out motors that meet the preset quality standards but have preset abnormal indicators, thus obtaining critical qualified motors; Extract the original waveform signal of the critically qualified motor, and extract the transient impact component from the original waveform signal to obtain transient characteristics; The transient features are matched with early fault patterns in a pre-built fault feature library to output potential fault detection results; The original noise signal during motor operation is acquired in the time domain to obtain the original waveform signal. A Fourier transform is then performed on the original waveform signal to extract spectral features and obtain the spectral energy distribution, including: At the rated speed of the motor, time-domain noise signals are acquired by a microphone array at a preset sampling frequency. The time-domain noise signal is denoised to obtain the original waveform signal; Perform a Fast Fourier Transform on the original waveform signal to obtain a spectrum. Calculate the amplitudes of the main frequency, bearing characteristic frequency, and electromagnetic harmonic frequency in the spectrum diagram to generate the spectrum energy distribution; The spectral energy distribution is segmented by a threshold to filter out motors that meet preset quality standards but have preset abnormal indicators, thus obtaining critically qualified motors, including: The main frequency amplitude of historical qualified motor samples is dynamically collected, and the preset quality standard is determined based on the dynamic threshold range of the main frequency amplitude. The bearing characteristic frequency amplitude of the historical qualified motor samples is dynamically collected, and the qualified bearing characteristic frequency index is determined based on the dynamic threshold range of the bearing characteristic frequency amplitude. The electromagnetic harmonic frequency amplitude of the historical qualified motor samples is dynamically collected, and qualified electromagnetic harmonic frequency indicators are determined based on the dynamic threshold range of the electromagnetic harmonic frequency amplitude. The amplitude of the bearing that does not conform to the qualified bearing characteristic frequency indicator or the qualified electromagnetic harmonic frequency indicator is set as a preset abnormal indicator. If the amplitude of the main frequency in the spectrum energy distribution meets the preset quality standard, and the preset abnormal indicators exist in the amplitude of the bearing characteristic frequency or the amplitude of the electromagnetic harmonic frequency, the corresponding motor will be marked as a critically qualified motor.

2. The motor noise detection method based on waveform analysis as described in claim 1, characterized in that, The historical qualified motor samples are motors that are the same model as the motor to be tested, have been in service, and have not experienced any faults during the early fault exposure period.

3. The motor noise detection method based on waveform analysis as described in claim 1, characterized in that, The transient impact component is extracted from the original waveform signal to obtain transient characteristics, including: The original waveform signal is decomposed by wavelet packet decomposition to obtain the time-frequency energy distribution matrix; Pulse detection is performed on the time-frequency energy distribution matrix, and the pulse interval and amplitude are calculated to obtain the transient characteristics.

4. The motor noise detection method based on waveform analysis as described in claim 2, characterized in that, A pre-built fault feature library is included, including: The first historical critical qualified motor sample is dynamically collected. The first historical critical qualified motor sample is a motor with the same model as the motor to be tested, which has been in service and has shown the first fault mode during the early fault exposure period. Obtain the first historical transient features of the first historical critical qualified motor sample; Establish a first mapping relationship between the first historical transient features and the first fault mode, and construct the fault feature library based on the first mapping relationship.

5. The motor noise detection method based on waveform analysis as described in claim 4, characterized in that, The pre-built fault feature library also includes: The second historical critical qualified motor sample is dynamically collected. The second historical critical qualified motor sample is a motor that is different from the motor model to be tested but has the same fault association parameters, has been in service and has shown the second fault mode during the early fault exposure period. The fault association parameters include bearing model, rated speed and number of poles. Obtain the second historical transient features of the second historical critical qualified motor sample; Collect first and second historical qualified motor samples that correspond one-to-one with the first and second historical critical qualified motor samples, and calculate the conversion factor based on the frequency offset ratio between the two. The second historical transient feature is converted according to the conversion coefficient to obtain the second converted historical transient feature. A second mapping relationship between the second converted historical transient feature and the second fault mode is established, and the second mapping relationship is added to the fault feature library.

6. The motor noise detection method based on waveform analysis as described in claim 4, characterized in that, Constructing the fault feature library based on the first mapping relationship includes: The first mapping relationship is grouped according to different fault modes to obtain multiple groups of first mapping relationships; Extract all historical transient features from each first mapping relationship group to construct multiple historical transient feature libraries; The multiple historical transient feature libraries are labeled with corresponding fault modes to obtain multiple early fault modes, and the multiple early fault modes are integrated to obtain the fault feature library.

7. The motor noise detection method based on waveform analysis as described in claim 1, characterized in that, The transient features are matched with early fault patterns in a pre-built fault feature library to output potential fault detection results, including: Traverse the fault feature library and extract the first early fault mode; Calculate the similarity between the transient feature and the historical transient features contained in the first early fault mode. If the calculation result is greater than or equal to a preset similarity threshold, then output the first early fault mode as a potential fault detection result.

8. A motor noise detection system based on waveform analysis, characterized in that, The system is used to implement the motor noise detection method based on waveform analysis according to any one of claims 1-7, and the system comprises: The spectrum energy distribution acquisition module is used to acquire the original noise signal of the motor in the time domain to obtain the original waveform signal, perform Fourier transform on the original waveform signal, extract spectrum features, and obtain the spectrum energy distribution. The critical qualified motor acquisition module is used to perform threshold segmentation on the spectrum energy distribution, filter out motors that meet the preset quality standards but have preset abnormal indicators, and obtain critical qualified motors. The transient feature acquisition module is used to extract the original waveform signal of the critical qualified motor, extract the transient impact component from the original waveform signal, and obtain transient features; The potential fault detection module is used to match the transient features with early fault modes in a pre-built fault feature library and output potential fault detection results.

Images