Quasi-periodic signal mode extraction method based on adaptive truncation calibration

Abstract

The invention discloses a quasi-periodic signal mode extraction method based on adaptive truncation calibration, and the method comprises the specific steps: S1, extracting a main frequency spectrum peak, and calculating an average period T as a subsequent period truncation reference; s2, intercepting a T-length signal as a first period S1; intercepting signal segments point by point and determining a normalization coefficient e (d); s3, calculating the value of e (d), and selecting a maximum value point as an optimal interception point; calculating an amplitude correction coefficient a1, and normalizing an initial periodic signal taking T as a reference; and S4, repeating the steps S1 to S3, iteratively extracting multiple sections of calibrated periodic signals, and performing weighted average on the periodic signals based on the signal-to-noise ratio of each period to obtain a stable periodic mode. According to the quasi-periodic signal mode extraction method based on adaptive truncation calibration disclosed by the invention, the problem that a traditional time domain averaging method in the prior art is excessively dependent on a strict period is solved.

Description

Technical Field

[0001] This invention belongs to the technical field of signal processing and mechanical vibration monitoring methods, specifically relating to a method for extracting quasi-periodic signal patterns based on adaptive truncation calibration. Background Technology

[0002] Rotating and reciprocating machinery are key equipment in core industrial sectors such as energy, rail transportation, and intelligent manufacturing. Their operating status directly determines production efficiency and system safety. During operation, core components such as bearings generate vibration signals. Accurately extracting feature patterns from these signals is a crucial prerequisite for equipment fault diagnosis and has significant engineering implications for ensuring the stable operation of industrial systems. Currently, the mainstream technique for pattern extraction from these signals in industry is the time-domain averaging method. This method, under the premise of strictly consistent signal periods, determines the period length through synchronous rotational speed, truncates the long time-series signal into n segments according to the period, and then averages each segment point by point after phase alignment to suppress random noise and highlight periodic patterns. However, the effectiveness of this method is fundamentally limited. Its core relies entirely on the premise of strictly consistent signal periods, and the accurate alignment of each periodic segment also depends on a constant period. In other words, the time-domain averaging method is only applicable to strictly periodic signals and has zero tolerance for periodic fluctuations. This is the core reason why it cannot adapt to actual working conditions. In real-world industrial scenarios, the rotational speed of rotating and reciprocating machinery inevitably fluctuates due to factors such as load changes and energy fluctuations, directly causing the period of the bearing vibration signal to change dynamically over time. In this situation, the traditional time-domain averaging method fails due to its rigid dependence on a strict period: when truncating to a fixed period, over-truncation or under-truncation deviations occur, leading to distortion of information in individual period segments; more seriously, deviation signals from the previous period are mistakenly incorporated into the next period, causing phase misalignment accumulation, ultimately resulting in complete failure of phase alignment across all segments. Consequently, the fault features extracted based on the average signal are completely distorted. This is the core problem that the traditional time-domain averaging method cannot solve due to its reliance on a strict period.

[0003] Besides time-domain averaging, industry has also attempted to extract patterns using Fourier transform and short-time Fourier transform, but all have shortcomings. The Fourier transform, based on the principle of linear superposition, decomposes a time-domain signal into a series of sine and cosine functions, and is only suitable for processing stationary signals whose statistical characteristics do not change over time. When processing quasi-periodic signals, because these signals have time-varying periodic characteristics, the Fourier transform compresses the time dimension information, making it unable to effectively capture the periodic dynamic changes of the signal, thus producing frequency ambiguity. This ambiguity causes the representations of different frequency components in the frequency domain to overlap, making it difficult to accurately identify the true pattern of the signal. The short-time Fourier transform, by introducing a sliding window function, achieves time-frequency analysis of signals to a certain extent, providing a new approach for non-stationary signal processing. However, its time-frequency resolution is strictly limited by the characteristics of the window function. When faced with quasi-periodic signals exhibiting large periodic fluctuations, choosing a fixed window length presents a dilemma: a short window length, while quickly capturing periodic changes, reduces the signal's frequency domain resolution, making it difficult to preserve waveform details; conversely, a long window length, while improving frequency domain resolution and preserving waveform features, results in a delayed response to periodic changes, failing to track the signal's dynamic evolution in a timely manner. This inherent contradiction leads to significant limitations in short-time Fourier transform pattern extraction for quasi-periodic signals, hindering high-precision pattern recognition. Consequently, existing techniques suffer from insufficient frequency resolution and poor time-frequency convergence in signal extraction. Summary of the Invention

[0004] The purpose of this invention is to provide a method for extracting quasi-periodic signal patterns based on adaptive truncation calibration, which solves the problem that the traditional time-domain averaging method in the prior art is too dependent on strict periodicity.

[0005] The technical solution adopted in this invention is a method for extracting quasi-periodic signal patterns based on adaptive truncation calibration, which specifically includes the following steps: S1, perform fast Fourier transform on the vibration signal, extract the main frequency spectrum peak, and calculate the average period T as the reference for subsequent period truncation; S2, extract a signal of length T as the first period S1; since the starting point of subsequent periods deviates from the theoretical position nT, set the search range, extract signal segments point by point and determine the normalization coefficient e(d); S3, calculate the value of e(d), select the maximum value point as the optimal intercept point; calculate the amplitude correction coefficient a1, and normalize the initial periodic signal based on T; S4. Repeat steps S1-S3 to iteratively extract multiple calibrated periodic signals. Based on the signal-to-noise ratio of each period, perform a weighted average of the periodic signals to obtain a stable periodic pattern.

[0006] The invention is further characterized in that, The vibration signal acquisition process in S1 is as follows: for the bearings of rotating and reciprocating machinery, the vibration signal is acquired by synchronously monitoring the real-time speed of the machinery, adaptively adjusting the vibration signal truncation parameters, and aligning the phase of the signal segments.

[0007] The specific process for extracting the main frequency spectral peak in S1 is as follows: In the frequency domain spectrum, identify and extract the spectral peak corresponding to the main frequency.

[0008] The specific steps for identifying and extracting the spectral peak corresponding to the dominant frequency are as follows: In the frequency domain amplitude spectrum, select the spectral peak corresponding to the maximum amplitude as the candidate dominant frequency; if there are multiple spectral peaks with similar amplitudes, and the difference between their amplitudes and the maximum amplitude is less than 10% of the maximum amplitude, then calculate the center frequency of each candidate spectral peak. The frequency band with the highest energy percentage is selected as the main frequency. If the vibration signal has frequency drift, the main frequency sequence after judgment is smoothed by moving average with a sliding window length of no less than 3 periods to reduce the impact of instantaneous fluctuations on period estimation. Based on the aforementioned main frequency f, the average period T of the signal is calculated using the formula T=1 / f, and used as the initial reference for subsequent period interception.

[0009] S2 specifically involves: in the original vibration signal, sliding a window of length T backward from the first sampling point, calculating the normalized cross-correlation coefficient e(d) between each window and the immediately following window. When the coefficients of at least three consecutive windows are greater than the preset threshold p, the signal is determined to have entered the stable periodic region, and the starting point of the first window that meets this condition is taken as the effective starting point. If the length L of the start transition segment is known, the segment is skipped first and then the sliding detection is performed. From this effective starting point, continuous sampling points of length T are taken as the first period S1 and the normalized coefficient e(d) is determined.

[0010] S3 specifically refers to the following: In the subsequent vibration signal, due to periodic fluctuations, measurement noise, or interference in the quasi-periodic signal, the starting point of the subsequent period will not be strictly located at the theoretical starting position nT. Therefore, based on the maximum phase drift corresponding to ±2% of the measured rotational speed fluctuation, the starting point search range is set to 0.2T before and after the theoretical starting point nT. Within this range, continuous sampling points of equal length T to the first period S1 are extracted as candidate periods S2. The normalized cross-correlation coefficient e(d) is calculated within the search range, and e(d) quantifies the waveform similarity between the candidate period S2 and the first period S1, with a value range of []. [1,1], where 1 indicates complete correlation; the point that makes e(d) reach its maximum value is selected as the optimal intercept point, and the amplitude correction coefficient a1 is calculated simultaneously to perform amplitude normalization on S2, thereby realizing the alignment and amplitude calibration of the time-domain waveform.

[0011] e(d) is represented as: (1); in, This is the amplitude correction coefficient, used to eliminate the interference of amplitude differences on similarity evaluation; The first period determined in S2, Candidate periods are selected within the search range; the symbol " "" indicates the inner product of the signal.

[0012] The amplitude correction factor a1 is expressed as: (2); in," " indicates the inner product of the signal; Candidate period energy, For the first cycle The energy. The candidate period is determined by the energy ratio of the two periods. Amplitude scale calibration to match the first cycle Consistency is maintained to avoid the similarity of waveforms being obscured by amplitude differences, thereby improving the accuracy of the difference evaluation function.

[0013] Compared with the prior art, the beneficial effects of the present invention are: (1) This invention provides a method for extracting quasi-periodic signal patterns based on adaptive truncation calibration, which is applicable to the extraction and noise reduction of periodic patterns of vibration signals, pressure signals and acoustic signals in rotating and reciprocating machinery. Its core lies in the adaptive processing of signal period fluctuations and amplitude changes, which achieves noise suppression while efficiently extracting periodic patterns, significantly improving the signal-to-noise ratio and feature extraction accuracy of signal analysis. Compared with the traditional time-domain averaging method, this invention effectively overcomes its dependence on strictly periodic signals. Even under complex working conditions with speed fluctuations, load disturbances and noise interference, it can still accurately extract signal patterns, and the final extraction results are more stable and clear, providing reliable support for mechanical fault diagnosis and operating status analysis. At the same time, it has a wide range of applications and can be further extended to the processing scenarios of various quasi-periodic signals such as vibration, pressure and acoustic signals. Attached Figure Description

[0014] Figure 1 This is a flowchart illustrating the quasi-periodic signal pattern extraction method based on adaptive truncation calibration of the present invention. Figure 2 This is a diagram of the original periodic vibration signal. Detailed Implementation

[0015] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0016] This invention provides a method for extracting quasi-periodic signal patterns based on adaptive truncation calibration, such as... Figure 1 As shown, the specific steps include: S1, perform fast Fourier transform on the vibration signal, extract the main frequency spectrum peak, and calculate the average period T as the reference for subsequent period truncation; The vibration signal acquisition process in S1 is as follows: for the bearings of rotating and reciprocating machinery, the vibration signal is acquired by synchronously monitoring the real-time speed of the machinery, adaptively adjusting the vibration signal truncation parameters, and aligning the phase of the signal segments.

[0017] The specific process for extracting the main frequency spectral peak in S1 is as follows: In the frequency domain spectrum, identify and extract the spectral peak corresponding to the main frequency.

[0018] The specific steps for identifying and extracting the spectral peak corresponding to the dominant frequency are as follows: In the frequency domain amplitude spectrum, select the spectral peak corresponding to the maximum amplitude as the candidate dominant frequency; if there are multiple spectral peaks with similar amplitudes, and the difference between their amplitudes and the maximum amplitude is less than 10% of the maximum amplitude, then calculate the center frequency of each candidate spectral peak. The frequency band with the highest energy percentage is selected as the main frequency. If the vibration signal has frequency drift, the main frequency sequence after judgment is smoothed by moving average with a sliding window length of no less than 3 periods to reduce the impact of instantaneous fluctuations on period estimation. Based on the aforementioned main frequency f, the average period T of the signal is calculated using the formula T=1 / f, and used as the initial reference for subsequent period interception.

[0019] S2, extract a signal of length T as the first period S1; since the starting point of subsequent periods deviates from the theoretical position nT, set the search range, extract signal segments point by point and determine the normalization coefficient e(d); S2 specifically involves: in the original vibration signal, sliding a window of length T backward from the first sampling point, calculating the normalized cross-correlation coefficient e(d) between each window and the immediately following window. When the coefficients of at least three consecutive windows are greater than the preset threshold p, the signal is determined to have entered the stable periodic region, and the starting point of the first window that meets this condition is taken as the effective starting point. If the length L of the start transition segment is known, the segment is skipped first and then the sliding detection is performed. From this effective starting point, continuous sampling points of length T are taken as the first period S1 and the normalized coefficient e(d) is determined.

[0020] S3, calculate the value of e(d), select the maximum value point as the optimal intercept point; calculate the amplitude correction coefficient a1, and normalize the initial periodic signal based on T; S3 specifically refers to the following: In the subsequent vibration signal, due to periodic fluctuations, measurement noise, or interference in the quasi-periodic signal, the starting point of the subsequent period will not be strictly located at the theoretical starting position nT. Therefore, based on the maximum phase drift corresponding to ±2% of the measured rotational speed fluctuation, the starting point search range is set to 0.2T before and after the theoretical starting point nT. Within this range, continuous sampling points of equal length T to the first period S1 are extracted as candidate periods S2. The normalized cross-correlation coefficient e(d) is calculated within the search range, and e(d) quantifies the waveform similarity between the candidate period S2 and the first period S1, with a value range of []. [1,1], where 1 indicates complete correlation; the point that makes e(d) reach its maximum value is selected as the optimal intercept point, and the amplitude correction coefficient a1 is calculated simultaneously to perform amplitude normalization on S2, thereby realizing the alignment and amplitude calibration of the time-domain waveform.

[0021] e(d) is represented as: (1); in, This is the amplitude correction coefficient, used to eliminate the interference of amplitude differences on similarity evaluation; The first period determined in S2, Candidate periods are selected within the search range; the symbol " "" indicates the inner product of the signal.

[0022] The amplitude correction factor a1 is expressed as: (2); in," " indicates the inner product of the signal; Candidate period energy, For the first cycle The energy. The candidate period is determined by the energy ratio of the two periods. Amplitude scale calibration to match the first cycle Consistency is maintained to avoid the similarity of waveforms being obscured by amplitude differences, thereby improving the accuracy of the difference evaluation function.

[0023] S4. Repeat steps S1-S3 to iteratively extract multiple calibrated periodic signals. Based on the signal-to-noise ratio of each period, perform a weighted average of the periodic signals to obtain a stable periodic pattern.

[0024] Example 1 This embodiment provides a method for extracting quasi-periodic signal patterns based on adaptive truncation calibration, such as... Figure 1 As shown, the specific steps are as follows: Step 1: For the bearings of rotating and reciprocating machinery, the vibration signals are collected by synchronously monitoring the real-time rotational speed of the machinery, adaptively adjusting the vibration signal truncation parameters, and achieving phase alignment of the signal segments.

[0025] Step 2: Perform a fast Fourier transform on the original vibration signal to convert the signal from the time domain to the frequency domain; in the frequency domain spectrum, identify and extract the spectral peak corresponding to the main frequency, which reflects the core periodicity of the signal. Step 3: In the original vibration signal, slide a window of length T backward from the first sampling point, calculate the normalized cross-correlation coefficient between each window and the immediately following window. When the coefficients of at least 3 consecutive windows are all greater than the preset threshold p, the signal is determined to have entered the stable periodic region, and the starting point of the first window that meets this condition is taken as the "effective starting point". If the length L of the start transition segment is known, skip the segment first and then perform sliding detection. From this effective starting point, continuous sampling points of length T are taken as the first period S1.

[0026] Step 4: In subsequent vibration signals, due to periodic fluctuations, measurement noise, or interference in the quasi-periodic signal, the starting point of subsequent cycles will not be strictly located at the theoretical starting position nT of the subsequent cycles. Therefore, based on the maximum phase drift corresponding to ±2% of the measured rotational speed fluctuation, the starting point search range is set to 0.2T before and after the theoretical starting point nT (i.e., the sampling point range [N]). [0.2T, N+0.2T]), and within this range, continuously sampled points with the same length as the first period S1 and all of the same length T are selected as candidate periods S2, so as to balance computational efficiency and optimal starting point capture probability, while ensuring that subsequent inner product calculations have a unified benchmark.

[0027] Step 5: Calculate the normalized cross-correlation coefficient e(d) within the search range. e(d) quantifies the waveform similarity between candidate period S2 and the first period S1, with a value range of [...]. 1, 1], where 1 indicates complete correlation; select the point that makes e(d) reach its maximum value as the optimal intercept point, and simultaneously calculate the amplitude correction coefficient a1 to perform amplitude normalization processing on S2, thereby realizing the alignment and amplitude calibration of the time domain waveform.

[0028] Step 6: Iterate through steps 4 and 5 until the remaining signal is less than 0.8T or e(d) < 0.7 for 5 consecutive times; extract the period for each segment. ,by With the current weighted template Energy ratio calculation Those below 5 dB were discarded, and... A stable pattern is obtained by weighted averaging all qualified cycles.

[0029] Example 2 Based on Example 1, the main frequency is determined according to the following rules: (1) In the frequency domain amplitude spectrum, select the spectral peak corresponding to the maximum amplitude as the candidate main frequency; (2) If there are multiple spectral peaks with similar amplitudes, and the difference between their amplitudes and the maximum amplitude is less than 10% of the maximum amplitude, then calculate the proportion of energy of each candidate spectral peak in its own center frequency 5% band to the total energy, and select the one with the highest energy proportion as the main frequency. (3) If the signal has frequency drift, the main frequency sequence after judgment is smoothed by moving average with a sliding window length of not less than 3 periods to reduce the impact of instantaneous fluctuations on period estimation. Based on the aforementioned main frequency f, the average period T of the signal is calculated using the formula T = 1 / f, and used as the initial reference for subsequent period interception.

[0030] Example 3 Based on Example 2, the normalized cross-correlation coefficient e(d) is expressed as: (1); in, This is the amplitude correction coefficient, used to eliminate the interference of amplitude differences on similarity evaluation; The first period determined in S2, Candidate periods are selected within the search range; the symbol " "" indicates the inner product of the signal.

[0031] It should be noted that, due to periodic fluctuation errors, measurement system errors, or environmental noise interference in the quasi-periodic signal, the candidate period... The actual starting point will not be exactly equal to the ending point of the previous period. Therefore, it is necessary to set the starting search range. Where T represents step T. A defined average period, d, is the allowable deviation, to ensure coverage of actual possible starting point offsets.

[0032] Example 4 Based on Example 3, amplitude correction coefficient This is used to eliminate amplitude fluctuations between signals of different periods, specifically expressed as: (2); in," " indicates the inner product of the signal; Candidate period energy, Template cycle The energy. The candidate period is determined by the energy ratio of the two periods. Amplitude scale calibration to match the first cycle Consistency is crucial to avoid the similarity of waveforms being masked by amplitude differences, thereby improving the accuracy of the difference evaluation function. In practical applications, to ensure that the truncated signal corresponds to each visually perceived fluctuation process, it is necessary to manually select a suitable first-cycle signal as the template period.

[0033] Example 5 This embodiment takes the vibration signal of a compressor casing as an example, specifically: Based on Nyquist's theorem, the sampling frequency is set at 20kHz to cover the compressor's main frequency and harmonic components, avoiding aliasing. A piezoelectric accelerometer is selected, paired with a 24-bit AD acquisition card, to meet the signal acquisition requirements under both normal and abnormal operating conditions. The acquisition time is 10 seconds, and uncompressed storage ensures signal integrity. Noise reduction and other preprocessing are performed after acquisition. Perform an FFT on the preprocessed signal to obtain the amplitude spectrum, search for the maximum value within the sampling frequency, determine a main frequency, corresponding to 400 sampling points.

[0034] Cut off 400 points from the beginning of the preprocessed signal. Verify that the peak value is within the range and there are no abnormal pulses; if invalid, continue truncation. Record. Mean, variance, and peak value are used for subsequent calculations.

[0035] Theoretically, the starting point of the second cycle is 401 points. Considering phase shift and noise, the search range is set to [380, 420], and 400-point signal segments are extracted point by point. The normalized cross-correlation coefficient is used as the evaluation function. The formula is as follows: Select the starting point corresponding to the maximum value based on the calculated value; Calculate the amplitude correction factor The formula is as follows: This completes signal normalization; Using the previous cycle correction signal as a reference, the search range for subsequent cycles is narrowed. The total number of sampling points corresponds to 500 cycle segments. After removing abnormal segments, 485 valid segments are retained.

[0036] By performing the above steps on all periodic signal segments, a stable periodic pattern is finally obtained.

[0037] Example 6 This embodiment uses the wind turbine gearbox housing as an example. Specifically, the input shaft speed of the wind turbine gearbox fluctuates between 22 Hz ± 1.5 Hz, corresponding to a period T ≈ 45.45 ms. When sampling at fs = 25 kHz, T corresponds to exactly 1136 sampling points. Based on the limit offset of ±1.5 Hz, the maximum period deviation can be calculated to be ±3.1 ms, approximately 6.8% of T. To balance computational load and acquisition probability, the search half-width is compromised to 0.15 T, i.e., 6.8 ms ≈ 170 points.

[0038] The first cycle S1 is fixed at point 1136; the theoretical starting point of the next cycle is n = 1137, and the search interval is defined as [n]. 170, n + 170] = [967, 1307]; At each of the 341 starting positions, segments of 1136 points are extracted to form candidate S2; the normalized inner product (S2·S1) / |S1|² is calculated, and the starting point corresponding to the maximum value is taken as the optimal match; If the maximum inner product is less than 0.90, the sudden change in rotational speed is considered too large. The current cycle is abandoned and the search half-width is expanded to 0.25 T in the next cycle. The above process is then repeated.

[0039] Experiments show that under these parameters, the total time for processing 10 seconds of data (≈220 cycles) from a single channel is 0.28 seconds, and the final extracted average period waveform has an error of <0.3% compared to the key phase reference, meeting the real-time and accuracy requirements for online monitoring.

[0040] Example 7 This embodiment uses a high-speed spindle dynamic balancing test bench as an example. Specifically, it is an online dynamic balancing test bench for high-speed spindles with a target speed of 12,000 r / min (200 Hz) and a measured short-term fluctuation of ±0.5%. The sampling rate is fs = 204.8 kHz, and the theoretical period is T = 1 / 200 Hz = 5 ms, corresponding to 1024 points. Based on a ±0.5% adjustment, the maximum period drift is approximately ±0.025 ms ≈ ±5 points; to leave a 50% margin, the search half-width is taken as 0.0075 T, i.e., 8 points.

[0041] Take the first 1024 points of the stable segment as template S1; The theoretical starting point for the next cycle is n = 1025, and the search window is

[1025] . [8, 1025+8] = [1017, 1033]; 1024 points are intercepted at each of these 17 starting points to obtain 17 candidate S2s; calculate the amplitude correction coefficient a1 = ‖S2‖² / ‖S1‖², and find the difference evaluation function e(d)=a1·(S2·S1); The candidate with the largest e(d) and a signal-to-noise ratio ≥25 dB is selected as the starting point for this cycle, and the iteration continues.

[0042] Tests show that the single-cycle processing time is 42 µs, accounting for 0.8% of the real-time cycle (5 ms), successfully capturing a tiny rotational speed fluctuation of 0.5%, maintaining a constant segment length of 1024 points, ensuring consistency of the inner product reference, and the error between the extracted waveform and the bond phase reference is <0.1 °.

[0043] like Figure 2 As shown, a typical original periodic vibration signal is presented. It is evident that both the period length and amplitude exhibit slight drift, visually demonstrating the characteristics of a quasi-periodic signal and explaining why traditional fixed truncation methods are prone to phase misalignment and waveform distortion. The quasi-periodic signal pattern extraction method based on adaptive truncation calibration provided in this invention is applied to the extraction and noise reduction of periodic patterns from vibration, pressure, or acoustic signals of rotating and reciprocating machinery. This method adaptively processes periodic fluctuations and amplitude variations, effectively extracting periodic patterns while suppressing noise, significantly improving the signal-to-noise ratio and feature extraction in signal analysis.

Claims

1. A method for extracting quasi-periodic signal patterns based on adaptive truncation calibration, characterized in that, Specifically, the following steps are included: S1, perform fast Fourier transform on the vibration signal, extract the main frequency spectrum peak, and calculate the average period T as the reference for subsequent period truncation; S2, extract a signal of length T as the first period S1; since the starting point of subsequent periods deviates from the theoretical position nT, set the search range, extract signal segments point by point and determine the normalization coefficient e(d); S3, calculate the value of e(d), select the maximum value point as the optimal intercept point; calculate the amplitude correction coefficient a1, and normalize the initial periodic signal based on T; S4. Repeat steps S1-S3 to iteratively extract multiple calibrated periodic signals. Based on the signal-to-noise ratio of each period, perform a weighted average of the periodic signals to obtain a stable periodic pattern.

2. The method for extracting quasi-periodic signal patterns based on adaptive truncation calibration according to claim 1, characterized in that, The vibration signal acquisition process described in S1 is as follows: for the bearings of rotating and reciprocating machinery, the vibration signal is acquired by synchronously monitoring the real-time rotational speed of the machinery, adaptively adjusting the vibration signal truncation parameters, and achieving phase alignment of the signal segments.

3. The method for extracting quasi-periodic signal patterns based on adaptive truncation calibration according to claim 1, characterized in that, The specific process for extracting the main frequency spectral peak in S1 is as follows: In the frequency domain spectrum, identify and extract the spectral peak corresponding to the main frequency.

4. The method for extracting quasi-periodic signal patterns based on adaptive truncation calibration according to claim 3, characterized in that, The specific steps for identifying and extracting the spectral peaks corresponding to the main frequency are as follows: In the frequency domain amplitude spectrum, select the spectral peak corresponding to the maximum amplitude as the candidate main frequency; if there are multiple spectral peaks with similar amplitudes, and the difference between their amplitudes and the maximum amplitude is less than 10% of the maximum amplitude, then calculate the center frequency of each candidate spectral peak. The frequency band with the highest energy percentage is selected as the main frequency. If the vibration signal has frequency drift, the main frequency sequence after judgment is smoothed by moving average with a sliding window length of no less than 3 periods to reduce the impact of instantaneous fluctuations on period estimation. Based on the aforementioned main frequency f, the average period T of the signal is calculated using the formula T=1 / f, and used as the initial reference for subsequent period interception.

5. The method for extracting quasi-periodic signal patterns based on adaptive truncation calibration according to claim 1, characterized in that, Specifically, S2 involves sliding a window of length T backward from the first sampling point in the original vibration signal, calculating the normalized cross-correlation coefficient e(d) between each window and the immediately following window, and determining that the signal has entered the stable periodic region when the coefficients of at least three consecutive windows are greater than the preset threshold p. The starting point of the first window that meets this condition is taken as the effective starting point. If the length L of the start transition segment is known, the segment is skipped first and then the sliding detection is performed. From this effective starting point, a continuous sampling point of length T is taken as the first period S1 and the normalized coefficient e(d) is determined.

6. The method for extracting quasi-periodic signal patterns based on adaptive truncation calibration according to claim 1, characterized in that, Specifically, S3 involves the following: In subsequent vibration signals, due to periodic fluctuations, measurement noise, or interference in the quasi-periodic signal, the starting point of the subsequent period will not be strictly located at the theoretical starting position nT. Therefore, based on the maximum phase drift corresponding to ±2% of the measured rotational speed fluctuation, the starting point search range is set to 0.2T before and after the theoretical starting point nT. Within this range, consecutive sampling points of equal length T to the first period S1 are extracted as candidate periods S2. The normalized cross-correlation coefficient e(d) is calculated within the search range, and e(d) quantifies the waveform similarity between the candidate period S2 and the first period S1, with a value range of []. [1,1], where 1 indicates complete correlation; the point that makes e(d) reach its maximum value is selected as the optimal intercept point, and the amplitude correction coefficient a1 is calculated simultaneously to perform amplitude normalization on S2, thereby realizing the alignment and amplitude calibration of the time-domain waveform.

7. The method for extracting quasi-periodic signal patterns based on adaptive truncation calibration according to claim 6, characterized in that, The term e(d) is represented as: （1）； in, This is the amplitude correction coefficient, used to eliminate the interference of amplitude differences on similarity evaluation; The first period determined in S2, Candidate periods are selected within the search range; the symbol " "" indicates the inner product of the signal.

8. The method for extracting quasi-periodic signal patterns based on adaptive truncation calibration according to claim 7, characterized in that, The amplitude correction coefficient a1 is expressed as: ） （2）； in," " indicates the inner product of the signal; Candidate period energy, For the first cycle The energy, through the energy ratio of two cycles, determines the candidate cycle. Amplitude scale calibration to match the first cycle Consistency is maintained to avoid the similarity of waveforms being obscured by amplitude differences, thereby improving the accuracy of the difference evaluation function.

Images