Periodic abnormal sound adaptive identification method

Abstract

The invention relates to the technical field of signal processing, and discloses a periodic abnormal sound adaptive identification method, which comprises the steps of dividing a time domain detection signal of equipment to be detected based on a frequency domain and a time domain, obtaining corresponding sub-signals in candidate frequency bands, calculating a comprehensive score based on various frequency spectrum characteristics of the sub-signals, and obtaining a periodic abnormal sound adaptive identification result. Obtaining a final score of each candidate frequency band, and taking the candidate frequency band with the highest final score as a target frequency band; and constructing a self-adaptive threshold value based on the median of the sliding window and the absolute deviation of the median, and judging whether the periodic abnormal sound exists or not based on the energy curve and the self-adaptive threshold value through an equal interval condition that the interval range is not greater than a preset tolerance and a persistence condition that the number of peak values is not less than a preset minimum peak number. According to the method, the target frequency band is accurately obtained by integrating multiple frequency spectrum features, the missing detection rate and the false detection rate are reduced from the source, the detection robustness is guaranteed through the self-adaptive threshold value, and meanwhile accurate recognition of the periodic abnormal sound is achieved based on the equal-interval and continuous conditions.

Description

Technical Field

[0001] This invention relates to the field of signal processing technology, and in particular to an adaptive method for periodic abnormal sound recognition. Background Technology

[0002] During the operation of rotating machinery, gear transmission devices, fans, and motors, periodic impact noises frequently occur. These signals often originate from bearing defects, poor gear meshing, rotor imbalance, or other repetitive mechanical impact processes. They typically manifest as regularly spaced pulses or narrow-band high-energy components in acoustic or vibration signals. Periodic impact noises not only severely affect the noise quality of equipment and user experience, but more importantly, they are often a significant indicator of early mechanical failures. Therefore, timely and accurate detection of periodic impact noises has important engineering and economic significance.

[0003] Currently, in engineering practice, the detection of periodic impact noises mostly relies on manual listening or observation of time-frequency spectra. A common approach is to collect sound signals using a microphone or vibration signals using an accelerometer, then calculate the short-time Fourier transform, and have the tester observe the spectrum for regular spikes or energy stripes. However, this method is highly subjective, depends on human experience, and is difficult to automate for batch testing.

[0004] Some studies have attempted to identify periodic signals using energy curves and peak detection. For example, integrating energy within a specific frequency band and detecting local peaks, or using fixed thresholds to identify impact events; however, the selection of the target frequency band often relies on prior knowledge or manual observation, lacking adaptability and prone to missed or false detections. Commonly used fixed threshold methods are sensitive to changes in environmental noise and lack stability under different operating conditions. Furthermore, the determination of periodicity mostly relies on checking the number of peaks or simple intervals, lacking quantitative consistency indicators, thus easily leading to misjudgments in complex noise environments.

[0005] In summary, existing methods rely on manual selection of target frequency bands, which is inefficient and difficult to apply to automated production lines. Furthermore, the lack of standardized detection methods due to subjective biases makes them prone to missed detections and false detections when dealing with different equipment or complex and variable operating conditions. At the same time, fixed peak detection thresholds cannot guarantee the stability of identification under different operating conditions and have poor environmental robustness. Moreover, the periodicity confirmation mostly only examines the number of peaks or performs simple interval checks, which makes it difficult to distinguish between real periodic impacts and randomly occurring noise pulses in complex industrial noise backgrounds, and is prone to misjudgment. Summary of the Invention

[0006] Therefore, the technical problem to be solved by the present invention is to overcome the problems of low efficiency, low accuracy and poor robustness of periodic abnormal sound recognition caused by the reliance on manual acquisition of target frequency bands, fixed detection thresholds and simple periodic judgment logic in the prior art.

[0007] To address the aforementioned technical problems, this invention provides a method for adaptive recognition of periodic abnormal sounds, comprising: Acquire the time-domain detection signal of the device under test; The entire frequency range is divided into multiple candidate frequency bands. Based on the candidate frequency bands, the time-domain detection signal is filtered to obtain the candidate signal corresponding to each candidate frequency band. The candidate signals corresponding to each candidate frequency band are divided into multiple sub-signals using a preset sliding window. For each candidate frequency band, extract multiple spectral features of its corresponding sub-signals and perform weighted summation to obtain the comprehensive score of each sub-signal of each candidate frequency band; for each candidate frequency band, obtain the median of the comprehensive scores of all its sub-signals as the final score; and select the candidate frequency band with the highest final score as the target frequency band. Perform a short-time Fourier transform on the candidate signal corresponding to the target frequency band to obtain the time spectrum; calculate the power spectral density of the time spectrum, sum the power spectral density in the frequency direction to obtain the energy curve of the target frequency band; By using a preset sliding window to slide on the energy curve, the median and absolute deviation of the energy value within each sliding window are obtained, and an adaptive threshold is constructed. The moments when the energy value in the energy curve of the target frequency band exceeds the adaptive threshold are marked as candidate peaks and a candidate peak set is formed. The time interval between adjacent candidate peaks in the candidate peak set is calculated, and candidate peaks whose time interval with the two adjacent candidate peaks on the left and right is less than the preset interval are discarded to obtain an optimized candidate peak set. If the difference between the maximum and minimum time intervals between all adjacent candidate peaks in the optimized candidate peak set is not greater than the preset tolerance, and the number of candidate peaks in the optimized candidate peak set is not less than the preset minimum number of peaks, then there is a periodic abnormal sound in the candidate signal corresponding to the target frequency band.

[0008] Preferably, the time-domain detection signal of the device under test is an acoustic signal collected by a microphone or a vibration signal collected by an accelerometer.

[0009] Preferably, after acquiring the time-domain detection signal of the device under test, the process includes preprocessing the time-domain detection signal; the preprocessing includes DC removal, amplitude normalization, and anti-aliasing filtering.

[0010] Preferably, the entire frequency range is divided into multiple candidate frequency bands based on a preset center frequency and a preset bandwidth.

[0011] Preferably, the sub-signal has multiple spectral characteristics, including spectral kurtosis, envelope periodicity signal-to-noise ratio, modulation spectral kurtosis, and spectral entropy.

[0012] Preferably, a short-time Fourier transform is performed on the candidate signal corresponding to the target frequency band to obtain the time spectrum, which is expressed as: ; in, Indicates the first Frequency in a time window spectral components at the location; , This indicates the number of points in the short-time Fourier transform. Indicates and Distance is Candidate signals for each time window, frame shift This represents the number of sample points between two adjacent time windows. Represents the window function. It represents the imaginary unit.

[0013] Preferably, the time-frequency spectrum is scaled based on the time window length to obtain the power spectral density, which is expressed as: ; in, Indicates the first Frequency in a time window The power spectral density estimate at that location, express, This represents the square of the L2 norm of the window function.

[0014] Preferably, the median and absolute deviation of the energy values ​​within each sliding window are obtained to construct an adaptive threshold, including: Get the The median of energy values ​​within a sliding window ; Get the The median absolute deviation of energy values ​​within a sliding window ; Based on the The median and absolute deviation of the energy values ​​within the nth sliding window are used to obtain the nth... Adaptive thresholds corresponding to each sliding window ; in, In the energy curve, the first... The energy value within a sliding window, This indicates the preset coefficient.

[0015] Preferably, after determining that there is a periodic abnormal sound in the candidate signal corresponding to the target frequency band, the process includes: A periodic consistency index is constructed based on the ratio of the standard deviation to the mean of the time interval between all adjacent peaks in the optimized candidate peak set. If the periodic consistency index is not greater than the preset consistency threshold, then the periodic abnormal sound is determined to be periodically stable.

[0016] Preferably, after determining that there is a periodic abnormal sound in the candidate signal corresponding to the target frequency band, the process includes: Energy acquisition curve in delay autocorrelation function after ; If the energy value at the delay point Higher than the corresponding adaptive threshold, and delay If the average time interval of all adjacent candidate peaks in the preferred candidate peak set is the same, then the periodic abnormal sound is determined to be periodically stable. in, This represents the average energy value. This prevents extremely small positive numbers with a denominator of 0. The time index represents the energy curve.

[0017] Compared with the prior art, the above-described technical solution of the present invention has the following advantages:

[0018] The periodic abnormal sound adaptive identification method described in this invention divides the time-domain detection signal of the device under test into frequency and time domains, obtains the corresponding sub-signals in each candidate frequency band, and calculates a comprehensive score based on multiple spectral features of each sub-signal to obtain the final score of each candidate frequency band. The frequency band with the highest final score is selected as the target frequency band. The selection of the target frequency band eliminates the problem of inconsistent judgment criteria caused by differences in experience among different personnel, improves the objectivity and consistency of detection, and significantly improves detection efficiency and automation level. By comprehensively summing multiple spectral features to obtain a comprehensive score, the frequency band that truly contains fault characteristics can be more accurately located, thereby reducing the rate of missed detection and false detection from the source.

[0019] This invention constructs an adaptive threshold based on the absolute deviation between the median and the mean value in the sliding window. This allows the threshold to accurately track the level of background noise without being affected by the peak value itself. Furthermore, the threshold can be dynamically adjusted according to changes in the local noise level of the signal, ensuring excellent detection performance under different operating loads and environmental noise backgrounds. It has strong versatility and stability, and improves the robustness of abnormal sound recognition.

[0020] When determining the presence of periodic abnormal sounds based on energy curves and adaptive thresholds, this invention performs preliminary screening by using an equal-interval condition where the interval range is no greater than a preset tolerance, and a persistence condition where the number of peaks is no less than a preset minimum number of peaks. The equal-interval condition ensures that the selected peak sequence has a basic regularity in its temporal distribution, while the persistence condition ensures that the regularity pattern has lasted for a sufficiently long time rather than appearing by chance, thereby achieving accurate identification of periodic abnormal sounds.

[0021] This invention further introduces a periodic consistency index to accurately quantify the volatility of the interval, and uses an autocorrelation function for cross-validation. Based on a multi-index joint judgment mechanism, it strictly distinguishes between true periodic shocks and random shocks, effectively avoiding misjudgments in strong noise environments and further ensuring the accuracy of periodic abnormal sound identification. Attached Figure Description

[0022] To make the content of this invention easier to understand, the invention will be further described in detail below with reference to specific embodiments and accompanying drawings, wherein: Figure 1 This is a flowchart of the periodic heterophony adaptive recognition method of the present invention; Figure 2 This is a graph showing the results of determining forward and reverse rotation under conditions of high background noise and no looping sound; Figure 3 This is a graph showing the results of determining forward and reverse rotation under low background noise and no looping sound conditions; Figure 4 This is a diagram showing the results of determining forward and reverse rotation under conditions of no interference but with recurring abnormal noise; Figure 5 This is a diagram showing the results of determining forward and reverse rotation under conditions of interference and recurring abnormal noise; Figure 6 This is a schematic diagram of the judgment results corresponding to various detection environments. Detailed Implementation

[0023] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments described are not intended to limit the present invention.

[0024] Reference Figure 1 The flowchart shown illustrates the steps of the periodic heterophony adaptive recognition method of the present invention, which specifically include: S101: Acquire the time-domain detection signal of the device under test; S102: Divide the full frequency range into multiple candidate frequency bands, filter the time-domain detection signal based on the candidate frequency bands, and obtain the candidate signal corresponding to each candidate frequency band; S103: Divide the candidate signals corresponding to each candidate frequency band into multiple sub-signals using a preset sliding window; S104: For each candidate frequency band, extract multiple spectral features of its corresponding sub-signals and perform weighted summation to obtain the comprehensive score of each sub-signal of each candidate frequency band; for each candidate frequency band, obtain the median of the comprehensive scores of all its sub-signals as the final score; obtain the candidate frequency band with the highest final score as the target frequency band. S105: Perform a short-time Fourier transform on the candidate signal corresponding to the target frequency band to obtain the time spectrum; calculate the power spectral density of the time spectrum, sum the power spectral density in the frequency direction, and obtain the energy curve of the target frequency band; Wherein, the time spectrum is represented as: ; in, Indicates the first Frequency in a time window spectral components at the location; , This indicates the number of points in the short-time Fourier transform. Indicates and Distance is Candidate signals for each time window, frame shift This represents the number of sample points between two adjacent time windows. Represents the window function. Represents the imaginary unit; Power spectral density, expressed as: ; in, Indicates the first Frequency in a time window The power spectral density estimate at that location, express, This represents the square of the L2 norm of the window function; S106: Using a preset sliding window to slide on the energy curve, obtain the median and absolute deviation of the energy values ​​within each sliding window, and construct an adaptive threshold, including: Get the The median of energy values ​​within a sliding window ; Get the The median absolute deviation of energy values ​​within a sliding window ; Based on the The median and absolute deviation of the energy values ​​within the nth sliding window are used to obtain the nth... Adaptive thresholds corresponding to each sliding window ; in, In the energy curve, the first... The energy value within a sliding window, Indicates the preset coefficient; S107: Mark the moments when the energy value in the energy curve of the target frequency band exceeds the adaptive threshold as candidate peaks and form a candidate peak set; calculate the time interval between adjacent candidate peaks in the candidate peak set, and discard candidate peaks whose time interval with the two adjacent candidate peaks on the left and right is less than the preset interval to obtain an optimized candidate peak set. S108: If the difference between the maximum and minimum time intervals between all adjacent candidate peaks in the optimized candidate peak set is not greater than the preset tolerance, and the number of candidate peaks in the optimized candidate peak set is not less than the preset minimum number of peaks, then there is a periodic abnormal sound in the candidate signal corresponding to the target frequency band.

[0025] In this embodiment, the time-domain detection signal of the device under test is an acoustic signal collected by a microphone or a vibration signal collected by an accelerometer. The sensor can be flexibly configured to collect the corresponding time-domain detection signal according to the type of device under test. After collecting the time-domain detection signal of the device under test, this embodiment includes preprocessing the time-domain detection signal; the preprocessing includes DC removal, amplitude normalization, and anti-aliasing filtering.

[0026] In this embodiment, the entire frequency range is divided into multiple candidate frequency bands based on a preset center frequency and a preset bandwidth. For each sub-signal in the candidate frequency band, there are multiple spectral characteristics, including spectral kurtosis, envelope periodicity signal-to-noise ratio, modulation spectral kurtosis, and spectral entropy.

[0027] The periodic abnormal noise adaptive identification method described in this invention divides the time-domain detection signal of the device under test into frequency and time domains, obtains the corresponding sub-signals in each candidate frequency band, and calculates a comprehensive score based on multiple spectral features of each sub-signal to obtain the final score of each candidate frequency band. The frequency band with the highest final score is selected as the target frequency band. The selection of the target frequency band eliminates the problem of inconsistent judgment criteria caused by differences in experience among different personnel, improves the objectivity and consistency of detection, and significantly improves detection efficiency and automation level. By comprehensively summing multiple spectral features to obtain a comprehensive score, it can more accurately locate the frequency band that truly contains fault characteristics, thereby reducing the missed detection and false detection rates from the source. This invention constructs an adaptive threshold based on the absolute deviation between the median and the median in the sliding window, so that the threshold can accurately track the background noise level without being affected by the impact peak itself. Moreover, the threshold can be dynamically adjusted with changes in the local noise level of the signal, ensuring that it can maintain excellent detection performance under different operating loads and different environmental noise backgrounds. It has extremely strong versatility and stability, and improves the robustness of abnormal noise identification. When determining the presence of periodic abnormal sounds based on energy curves and adaptive thresholds, this invention performs preliminary screening by using an equal-interval condition where the interval range is no greater than a preset tolerance, and a persistence condition where the number of peaks is no less than a preset minimum number of peaks. The equal-interval condition ensures that the selected peak sequence has a basic regularity in its temporal distribution, while the persistence condition ensures that the regularity pattern has lasted for a sufficiently long time rather than appearing by chance, thereby achieving accurate identification of periodic abnormal sounds.

[0028] Specifically, after determining that there are periodic abnormal sounds in the candidate signals corresponding to the target frequency band, the process includes: constructing a periodic consistency index based on the ratio of the standard deviation to the average value of the time interval between all adjacent peaks in the optimized candidate peak set; if the periodic consistency index is not greater than a preset consistency threshold, then the periodic abnormal sound is determined to be periodically stable.

[0029] Specifically, after determining that there is a periodic abnormal sound in the candidate signal corresponding to the target frequency band, the process includes: acquiring the energy curve during the delay. autocorrelation function after If the energy value at the delay point Higher than the corresponding adaptive threshold, and delay If the time interval between all adjacent candidate peaks in the preferred candidate peak set is the same, then the periodicity of the abnormal sound is determined to be stable; where, This represents the average energy value. This prevents extremely small positive numbers with a denominator of 0. The time index represents the energy curve.

[0030] This invention further introduces a periodic consistency index to accurately quantify the volatility of the interval, and uses an autocorrelation function for cross-validation. Based on a multi-index joint judgment mechanism, it strictly distinguishes between true periodic shocks and random shocks, effectively avoiding misjudgments in strong noise environments and further ensuring the accuracy of periodic abnormal sound identification.

[0031] Based on the above embodiments, in this embodiment of the invention, a detection device is provided to achieve periodic impact noise detection, including: The sensor, depending on the device being tested, selects a microphone to collect sound signals or an accelerometer to collect vibration signals; The data acquisition card is used to perform high-precision sampling of the analog signals output by the sensor. The sampling rate is set to 51200Hz to ensure accurate recording of the impact signals in the 20kHz frequency range. The acquisition and processing system includes a computer or an embedded signal processing unit, which is used to acquire acquisition signals and run the periodic abnormal sound adaptive recognition method provided by the present invention, perform on-site analysis and judgment on the acquisition data, and output detection results.

[0032] Specifically, the adaptive identification method for periodic abnormal sounds of this invention mainly includes three parts: signal acquisition and preprocessing, automatic frequency band positioning and energy curve construction, and adaptive thresholding and periodicity determination. Signal acquisition and preprocessing involves acquiring acoustic or vibration signals through sensors, and performing DC removal, amplitude normalization, and anti-aliasing filtering on the acquired signals to ensure input signal quality. Automatic frequency band positioning and energy curve construction involves calculating spectral characteristics across the entire frequency range, automatically selecting the frequency band where periodic abnormal sounds occur as the target frequency band based on indicators such as spectral kurtosis, envelope modulation characteristics, and spectral entropy, and constructing a time-varying energy curve within the target frequency band. Adaptive thresholding and periodicity determination involves dynamically determining the threshold based on the median and absolute deviation of the median of a sliding window, extracting candidate peak values, and jointly determining them using equal spacing conditions and periodic consistency indicators. If necessary, autocorrelation analysis is used for verification to ultimately confirm the existence of periodic impact abnormal sounds and their characteristic parameters.

[0033] Based on the above description, in this embodiment, the adaptive identification method for periodic abnormal sounds based on cyclic peak determination specifically includes:

[0034] S201: Signal Acquisition and Preprocessing Acoustic or vibration signals are collected by sensors, and the collected signals are subjected to DC removal, amplitude normalization, and anti-aliasing filtering to ensure the accuracy of dynamic range and frequency components and guarantee the quality of input signals.

[0035] S202: Automatic frequency band positioning and energy curve construction: Calculates spectral characteristics across the entire frequency range, automatically selects the target frequency band based on indicators such as spectral kurtosis, envelope modulation characteristics, and spectral entropy, and constructs an energy curve that changes over time within the target frequency band; S202-1: Use an automatic frequency band locator to determine frequency bands that may contain periodic abnormal sounds; A series of center frequencies and bandwidth combinations are set across the entire frequency range to obtain multiple candidate frequency bands; then, bandpass filtering is performed on the signal in each candidate frequency band to obtain candidate signals. ; Within each candidate frequency band, a set of feature metrics are extracted, including spectral kurtosis (SK) and envelope periodicity signal-to-noise ratio (SNR). Modulation spectral kurtosis (MSK) and spectral entropy (H) are used to measure whether there are periodic abnormal sounds in the frequency band; ① Spectral kurtosis is used to characterize the sharpness of energy distribution and is defined as: ; in, It is the normalized distribution of power spectral density within the frequency band. and These are its mean and variance, respectively. Indicates the center frequency of the candidate frequency band. Parameters to ensure the denominator is not zero; ② In order to capture the amplitude modulation characteristics, the signal is obtained through Hilbert transform. envelope and in the modulation frequency range Internal calculation of Welch envelope spectrum When the spectral peak is significantly higher than the background, the envelope periodicity signal-to-noise ratio is defined as: ; ③ The frequency corresponding to the maximum spectral peak is the main modulation frequency. To further reflect the prominent properties of the envelope spectrum, it is normalized to and the modulation spectrum kurtosis is calculated, expressed as: ; in, and These are the mean and variance of the modulation spectrum, respectively. ④ The concentration of energy distribution is measured by spectral entropy: ; in, This refers to the number of frequency points within the band. By weighting and combining the above indicators, a comprehensive score for this frequency band is obtained, expressed as: ; in, It is the normalized modulation signal-to-noise ratio; The median score across multiple time windows is used as the final score, and the frequency band with the highest score is selected as the target frequency band. ; This invention utilizes multi-dimensional feature indicators such as spectral kurtosis, envelope modulation characteristics, and spectral entropy to scan and evaluate the entire frequency range, enabling automatic determination of the target frequency band without human intervention. Compared to traditional methods that rely on testers observing spectrum diagrams or selecting frequency bands based on experience, this invention achieves objectivity and automation in frequency band selection, significantly reducing human bias and improving detection efficiency and consistency.

[0036] S202-2: After determining the target frequency band, the bandpass signal will undergo a short-time Fourier transform: ; Based on this, the power spectral density is calculated and expressed as: ; in, The original signal, For length is Window function, R represents the FFT length of each frame, and R represents the frame shift (hop size). Represents the imaginary unit. Indicates frequency index, Indicates the sampling rate; S202-3: Summing the power spectral density along the frequency direction to construct the energy curve.

[0037] S203: Adaptive Threshold and Periodicity Determination: The threshold is dynamically determined based on the median and absolute deviation of the median in a sliding window. Candidate peak values ​​are extracted and jointly judged by equal interval conditions and periodic consistency index. If necessary, autocorrelation analysis is combined for verification to finally confirm the existence of periodic impact noise and its characteristic parameters. S203-1: In order to identify significant peaks from the energy curve, this invention uses the median and absolute deviation of the median of a sliding window to define an adaptive threshold, including: Median of the sliding window ; absolute deviation of the median of the sliding window ; Obtain the adaptive detour threshold: ; In the candidate peak extraction stage, this invention adopts an adaptive threshold method based on the absolute deviation of the median and the median in the sliding window. This method can dynamically adjust the judgment threshold according to the changes in signal energy level and background noise, thereby effectively avoiding the problem of failure of traditional fixed threshold methods when the noise is large or the environmental conditions change. Through this adaptive mechanism, this invention can maintain the stability and robustness of detection and judgment under different working conditions and different noise backgrounds.

[0038] S203-2: When the energy curve exceeds this adaptive threshold, the corresponding moment is marked as a candidate peak; After the candidate peaks are identified, the present invention further confirms the existence of periodic abnormal sounds through interval consistency; the time positions of the candidate peaks are recorded sequentially as follows: The time interval between adjacent peaks is: Take an interval greater than A continuous peak sequence, if there exists a peak whose intervals with both adjacent peaks are less than [a certain value]... If the peak is not found in a given sequence of peaks, then discard the peak; if the peak is found in a given sequence of peaks, then discard the peak. satisfy: And the number of peaks in the sequence is not less than If so, it can be determined that the signal segment contains periodic abnormal sounds; among them, Preset tolerance; S203-3: To further quantify regularity, a periodic consistency indicator is introduced: When this indicator is less than the corresponding consistency threshold, it indicates that the periodicity is very significant. and These represent the standard deviation and mean of the interval time matrix composed of the intervals of all peaks in the candidate peak set, respectively. S203-4: Meanwhile, the autocorrelation function of the energy curve: ; In delay If a significant peak exists nearby, and the delay corresponding to that peak is equal to the mean of the candidate interval... If the results are consistent, the existence of periodic abnormal sounds is further confirmed.

[0039] In the periodic confirmation stage, this invention not only examines whether the time interval between adjacent peaks meets the equal spacing condition, but also introduces a periodic consistency index. To quantify the stability of the interval. When A lower value indicates a significant regularity in the peak interval, thus enhancing the reliability of the periodicity determination. Furthermore, this invention incorporates autocorrelation analysis of the energy curve as an auxiliary verification method. When the autocorrelation function shows a significant peak at a position consistent with the average interval, the existence of periodic impact abnormalities can be further confirmed. Through the combined determination of equal interval conditions, periodic consistency indicators, and autocorrelation analysis, this invention effectively avoids misjudgments and omissions caused by single-indicator determinations, making the detection results more robust.

[0040] Reference Figure 2 The image shown is a graph illustrating the results of determining forward and reverse rotation under conditions of high background noise and no looping sound. Figure 2 The first column shows the time-frequency plot and energy envelope curve of the STFT in forward rotation, and the second column shows the time-frequency plot and energy envelope curve in reverse rotation; in the figures, triangular markers indicate candidate peaks. (Refer to...) Figure 3 The image shown is a graph illustrating the results of determining clockwise and counterclockwise rotation under low background noise and no looping sound conditions; refer to... Figure 4 The image shown is a diagram illustrating the results of determining forward and reverse rotation under conditions of interference-free operation with recurring abnormal noise; refer to... Figure 5 The image shown is a diagram illustrating the results of determining forward and reverse rotation under conditions of interference and recurring abnormal noise; refer to... Figure 6 The diagram shown illustrates the judgment results for various detection environments.

[0041] In summary, the periodic abnormal noise adaptive identification method proposed in this invention integrates automatic frequency band localization, energy curve construction, adaptive threshold peak screening, and periodic determination technology based on equal spacing and consistency indices. By extracting frequency domain features from acoustic or vibration signals and robustly screening candidate peaks using adaptive thresholds and statistical indices, the method further utilizes equal spacing conditions and periodic consistency indices to confirm periodic impact characteristics, thereby achieving automated detection of impact abnormal noises in complex noise environments. This method avoids reliance on manual observation of the spectrum, significantly improving the accuracy and robustness of detection. In verification, the method of this invention can accurately identify periodic impact abnormal noises in typical mechanical equipment, and the detection results have high stability and reliability, demonstrating good engineering application value. Furthermore, this invention can be applied to the detection of periodic impact abnormal noises in the acoustic performance of products in industrial production lines. Because this invention effectively reduces manual intervention through automatic frequency band localization, energy curve construction, and transfer feature extraction, and uses equal spacing and consistency indices for robust determination, it enables rapid screening of large batches of products on fully automated inspection lines, possessing significant engineering application prospects and promotional value.

[0042] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0043] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0044] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0045] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0046] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A periodicity hetero-tone adaptive recognition method, characterized in that, The method comprises the following steps: Collecting a time-domain detection signal of a device to be detected; Dividing the full frequency range into multiple candidate frequency bands, filtering the time-domain detection signal based on the candidate frequency bands, and obtaining a candidate signal corresponding to each candidate frequency band; Dividing the candidate signal corresponding to each candidate frequency band into multiple sub-signals using a preset sliding window; For each candidate frequency band, extracting multiple spectral features of each sub-signal corresponding thereto, and performing weighted summation to obtain a comprehensive score of each sub-signal of the candidate frequency band; for each candidate frequency band, obtaining the median of the comprehensive scores of all sub-signals thereof as a final score; Obtaining the candidate frequency band with the highest final score as a target frequency band; Performing short-time Fourier transform on the candidate signal corresponding to the target frequency band to obtain a time-frequency spectrum; calculating the power spectral density of the time-frequency spectrum, and summing the power spectral density in the frequency direction to obtain an energy curve of the target frequency band; Sliding the preset sliding window on the energy curve to obtain the median and median absolute deviation of the energy value in each sliding window, and constructing an adaptive threshold; Marking the time when the energy value of the energy curve of the target frequency band exceeds the adaptive threshold as a candidate peak to form a candidate peak set; calculating the time interval between adjacent candidate peaks in the candidate peak set, discarding the candidate peaks with a time interval smaller than a preset interval with the left and right adjacent candidate peaks, and obtaining an optimized candidate peak set; If the difference between the maximum and minimum values of the time interval between all adjacent candidate peaks in the optimized candidate peak set is not greater than a preset tolerance, and the number of candidate peaks in the optimized candidate peak set is not less than a preset minimum peak number, then the candidate signal corresponding to the target frequency band contains periodic abnormal sound.

2. The method of claim 1, wherein, The time-domain detection signal of the device to be detected is an acoustic signal collected by a microphone or a vibration signal collected by an acceleration sensor.

3. The method of claim 1, wherein, After collecting the time-domain detection signal of the device to be detected, the time-domain detection signal is preprocessed, which includes removing direct current, amplitude normalization, and anti-aliasing filtering.

4. The method of claim 1, wherein, The full frequency range is divided into multiple candidate frequency bands based on a preset center frequency and a preset bandwidth.

5. The method of claim 1, wherein, The multiple spectral features of the sub-signals include spectral kurtosis, envelope periodicity signal-to-noise ratio, modulation spectrum kurtosis, and spectral entropy.

6. The method of claim 1, wherein, The short-time Fourier transform is performed on the candidate signal corresponding to the target frequency band to obtain a time-frequency spectrum, which is represented as: ； wherein denotes the frequency spectrum component at the frequency in the jth time window; denotes the number of points of the short-time Fourier transform, denotes the distance of the candidate signal to the jth time window, denotes the frame shift of the candidate signal, denotes the number of sample points between two adjacent time windows, denotes the window function, denotes the imaginary unit.​​ 7. The method of claim 6, wherein, The power spectral density is obtained by scaling the time-frequency spectrum based on the time window length, which is represented as: ； in, Indicates the first Frequency in a time window The power spectral density estimate at that location, express, This represents the square of the L2 norm of the window function.

8. The method of claim 1, wherein, The median and median absolute deviation of the energy value in each sliding window are obtained to construct an adaptive threshold, which includes: obtaining a median of energy values within a first sliding window ; Acquiring a median absolute deviation of energy values in a first sliding window ; Based on the The median and absolute deviation of the energy values ​​within the nth sliding window are used to obtain the nth... Adaptive thresholds corresponding to each sliding window ; wherein, represents an energy value in a sliding window of the energy curve, represents a preset coefficient.​ 9. The method of claim 1, wherein, After determining that the candidate signal corresponding to the target frequency band contains periodic abnormal sound, the following steps are included: A periodic consistency index is constructed based on the ratio of the standard deviation to the average value of the time interval between all adjacent peaks in the optimized candidate peak set; If the periodic consistency index is not greater than a preset consistency threshold, it is determined that the periodicity of the periodic abnormal sound is stable.

10. The method of claim 1, wherein, After determining that the candidate signal corresponding to the target frequency band contains periodic abnormal sound, the following steps are included: acquiring an energy curve in the autocorrelation function after a delay ;​ if the energy value at the delay is higher than the corresponding adaptive threshold, and the delay is the same as the average of the time intervals of all adjacent candidate peaks in the set of preferred candidate peaks, then the periodicity of the periodicity artifact is determined. wherein, represents the average value of the energy values, represents a very small positive number to prevent the denominator from being zero, represents the time index of the energy curve.

Images