A circuit signal data processing method and system

Abstract

The application discloses a circuit signal data processing method and system, which comprises the following steps: dividing a noise frequency band by performing spectrum analysis on circuit signal data; generating compensation parameters to identify interference mode types based on the amplitude difference of the frequency band; generating a signal gain correction table; identifying a high-frequency interference section according to the correction table and tracing back to the source of the interference to establish an associated path; generating an interference classification table according to the path level depth to construct an adaptive filtering configuration; performing transient pulse detection on the filtered signal to identify abnormal peak sections, analyzing the attenuation characteristics to generate weight coefficients for feature enhancement; identifying signal mutation points by using a key feature set, performing time attenuation weighted evaluation on time sequence density for sparse sampling, constructing a fragmentation storage strategy according to data distribution and dynamically adjusting the storage priority, and generating structured signal feature data. The application can grade and inhibit different sources of interference, effectively extract transient features and optimize data storage access efficiency.

Description

Technical Field

[0001] This invention relates to the field of signal processing technology, and in particular to a method and system for processing circuit signal data. Background Technology

[0002] During the operation of industrial equipment, the electrical signals continuously collected by sensors are crucial for assessing the equipment's health status. However, the electromagnetic environment in industrial settings is complex, and the collected signals are subject to various interference components during transmission, including power line coupling clutter, high-frequency pulses from switching devices, and random background noise. These interference components originate from different sources and have varying frequency characteristics, intertwining with the useful signal in the frequency domain and causing a decrease in the signal-to-noise ratio of the original data.

[0003] Existing signal preprocessing methods typically employ fixed-parameter filters to uniformly process acquired data, failing to distinguish the sources and propagation characteristics of different interference components, thus hindering targeted suppression. Furthermore, abnormal equipment conditions are often accompanied by short-duration impact signals. These transient components exhibit rapid amplitude changes and short durations, making it difficult for conventional feature extraction methods to effectively identify and utilize their attenuation patterns. In addition, the feature data generated by monitoring systems is unevenly distributed temporally, with dense data in some periods and sparse data in others. Fixed storage allocation methods easily lead to concentrated access hotspots, impacting data retrieval efficiency. Summary of the Invention

[0004] This invention discloses a circuit signal data processing method and system, which aims to identify interference frequency bands and trace interference sources through spectrum analysis to form a hierarchical compensation strategy. It combines transient detection and attenuation characteristic analysis to extract impulse response features and implement weight enhancement. At the same time, it constructs a segmented storage strategy based on the temporal density of feature data and dynamically adjusts the storage priority to form structured signal feature data, providing data support for industrial equipment condition monitoring.

[0005] The first aspect of this invention provides a circuit signal data processing method, comprising the following steps: Collect circuit signal data from industrial equipment sensors, perform noise spectrum analysis on the circuit signal data to identify interference mode types, and extract noise compensation coefficients based on the interference mode types to generate a signal gain correction table; Based on the signal gain correction table, the circuit signal data is gain-corrected to identify high-frequency interference segments. The high-frequency interference segments are grouped and sorted according to interference intensity to generate an interference classification table. An adaptive filtering configuration is constructed using the interference classification table to extract the characteristics of the filtered signal. Transient pulse detection is performed on the filtered signal features to obtain transient pulse response. Abnormal peak segments are identified from the transient pulse response to generate peak weight coefficients. The peak weight coefficients are then used to perform feature enhancement processing on the filtered signal features to generate a key feature set. The key feature set is used to identify signal mutation points, and the signal mutation points are sparsely sampled according to the temporal distribution to generate a feature label sequence. A segmented storage strategy is constructed based on the feature label sequence. Based on the sharded storage strategy, storage load monitoring is performed to obtain load fluctuation status. The load fluctuation status is used to generate dynamic storage priority. Based on the dynamic storage priority, the feature tag sequence is converted into structured signal feature data.

[0006] A second aspect of the present invention provides a circuit signal data processing system, comprising: The data acquisition module is used to acquire circuit signal data from industrial equipment sensors, perform noise spectrum analysis on the circuit signal data to identify interference mode types, and extract noise compensation coefficients based on the interference mode types to generate a signal gain correction table. The filtering module is used to perform gain correction on the circuit signal data according to the signal gain correction table, identify high-frequency interference segments, group and sort the high-frequency interference segments according to interference intensity to generate an interference classification table, and construct an adaptive filtering configuration through the interference classification table to extract the features of the filtered signal. The feature extraction module is used to perform transient pulse detection on the filtered signal features to obtain transient pulse response, identify abnormal peak segments from the transient pulse response to generate peak weight coefficients, and use the peak weight coefficients to perform feature enhancement processing on the filtered signal features to generate a key feature set. The data encoding module is used to identify signal mutation points using the key feature set, sparsely sample the signal mutation points according to the time-series distribution to generate a feature label sequence, and construct a segmented storage strategy based on the feature label sequence. The data storage module is used to monitor the storage load based on the sharded storage strategy to obtain the load fluctuation status, generate a dynamic storage priority using the load fluctuation status, and convert the feature tag sequence into structured signal feature data according to the dynamic storage priority.

[0007] The beneficial effects of this invention are reflected in the following points: First, by using spectrum analysis, the interference components of the circuit signal are divided into high-noise and low-noise frequency bands. Based on the amplitude differences of the frequency bands, notch depth is calculated to generate compensation parameters. On this basis, high-frequency interference segments are decomposed into strong and weak interference components. The stable frequency components in the weak interference components are used to trace the interference source and establish harmonic correlation paths from the source frequency to each interference segment. The interference is then graded according to the hierarchical depth of the correlation paths, enabling the filtering strategy to distinguish the source and propagation level of the interference, thus exhibiting stronger targeting compared to fixed-parameter filters. Second, wavelet transform is used to detect transient pulses in the filtered signal and identify abnormal peak segments. By exponentially fitting the amplitude sequence after the peak position within the peak segment, attenuation rate parameters are extracted. The matching degree between the attenuation rate parameters and typical fault impact characteristics is used to generate peak weight coefficients. Peaks with attenuation rates within the typical fault range receive high weights, while peaks deviating from the normal range receive low weights. This method enables the feature enhancement process to distinguish transient components with different attenuation characteristics, highlighting impact responses with fault indication significance. Finally, after identifying signal mutation points, a time-decay weighted method is used for density evaluation. Recently occurring mutation points are given higher evaluation weights, while the weights of distant mutation points decay over time. Sparse sampling and fragmentation are performed based on the weighted density distribution. At the same time, the real-time load of each storage block is monitored, and the storage priority is dynamically adjusted according to the timeliness requirements of data access. Data that can be accessed late is prioritized to smooth out load peaks. This method enables the storage strategy to adapt to the uneven distribution of feature data in time and avoids access bottlenecks caused by the concentration of hot data.

[0008] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0009] The accompanying drawings illustrate specific examples of the technical solutions described in this invention and, together with the detailed embodiments, form part of the specification, serving to explain the technical solutions, principles, and effects of this invention.

[0010] Unless otherwise specified, the same reference numerals in different figures represent the same or similar technical features, and different reference numerals may be used to represent the same or similar technical features.

[0011] Figure 1 This is a schematic flowchart of a circuit signal data processing method according to the present invention.

[0012] Figure 2 This is a structural block diagram of a circuit signal data processing system according to the present invention. Detailed Implementation

[0013] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0014] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0015] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0016] The technical solutions of the embodiments of this application will be described below.

[0017] like Figure 1 As shown, this embodiment of the invention provides a circuit signal data processing method, including the following steps S110-S150: Step S110: Collect circuit signal data from industrial equipment sensors, perform noise spectrum analysis on the circuit signal data to identify interference mode types, and extract noise compensation coefficients based on the interference mode types to generate a signal gain correction table.

[0018] Specifically, circuit signal data from industrial equipment sensors is acquired. These sensors include temperature sensors, pressure sensors, vibration sensors, current sensors, and voltage sensors. They monitor the equipment's operating status in real time and convert physical quantities into electrical signals. Circuit signal data is acquired through an analog front-end circuit. The bandwidth of the acquisition channel is set to 0 to 100 kHz to cover the typical operating frequency band of industrial equipment, and the sampling rate is set to 250 kHz to meet the requirements of the Nyquist sampling theorem, ensuring that high-frequency components do not experience aliasing distortion. The circuit signal data is quantized using a 16-bit analog-to-digital converter, with a minimum resolvable voltage of approximately 0.15 mV, a level of accuracy capable of capturing subtle changes in sensor signals. During transmission, the circuit signals from industrial equipment sensors are susceptible to electromagnetic interference, ground noise, and power supply ripple. These interference sources manifest in the spectrum as peaks at specific frequencies or a rise in the broadband noise floor. Immediately after acquisition, the circuit signal data undergoes digital filtering preprocessing. The filter employs a dual notch design at 50 Hz and 60 Hz to suppress power frequency interference and its harmonic components, with a notch depth greater than 40 dB to ensure effective attenuation of power frequency interference. The acquisition time for circuit signal data is set to 10 consecutive seconds to obtain sufficient frequency domain resolution. The frequency resolution after Fast Fourier Transform (FFT) of 10 seconds of data is approximately 0.1Hz, which is sufficient to distinguish adjacent interference frequency components. The amplitude range of the circuit signal output by the sensor is typically between ±5V, and the input impedance of the analog front-end circuit is set to 1MΩ to reduce the load effect on the sensor output.

[0019] In some embodiments, the step of performing noise spectrum analysis on the circuit signal data to identify the interference mode type includes: constructing a frequency domain amplitude distribution map using the circuit signal data; performing peak-valley segmentation from the frequency domain amplitude distribution map to generate a low-noise frequency band and a high-noise frequency band; performing frequency domain compensation based on the low-noise frequency band and the high-noise frequency band to form compensation parameters; and forming the interference mode type based on the frequency difference characteristics of the compensation parameters.

[0020] A frequency domain amplitude distribution map was constructed using circuit signal data. The total number of sampling points for the circuit signal data reached 2.5 million within a 10-second acquisition time. The large data volume necessitated segmented processing to improve the stability of the spectrum estimation. The segment length was set to 4096 sampling points, corresponding to a time window of approximately 16.4 ms. A 50% overlap rate was used between adjacent segments to increase spectral smoothness. Fast Fourier Transform (FFT) transformed the circuit signal data from the time domain to the frequency domain. The transformed spectrum displays the frequency component distribution of the signal with frequency on the horizontal axis and amplitude on the vertical axis. The amplitude at each frequency point reflects the intensity of that frequency component in the original signal. A Hanning window function was applied before the Fourier transform of the circuit signal data to reduce spectral leakage. The smoothing properties of the window function reduced the trade-off between the main lobe width and side lobe amplitude. The frequency axis of the frequency domain amplitude distribution map covers the range of 0 to 125 kHz, corresponding to half the sampling rate, and all frequency components of the circuit signal data are characterized within this range. Periodic interference, such as switching frequency interference from a power supply, exhibits sharp peaks at the corresponding frequencies and harmonic positions in the frequency domain amplitude distribution plot, with peak amplitudes significantly higher than the background noise floor. Broadband noise, such as thermal noise and quantization noise, appears as a relatively flat noise floor in the frequency domain amplitude distribution plot; the floor height reflects the power spectral density of the noise. After constructing the frequency domain amplitude distribution plot, a logarithmic coordinate transformation is performed to convert the amplitudes to dB units, facilitating the observation of frequency components at different orders of magnitude. The logarithmic coordinate system allows weak signals and strong interference signals to be clearly presented in the same view.

[0021] Peak-valley segmentation is performed on the frequency domain amplitude distribution map to generate low-noise and high-noise frequency bands. Peak-valley segmentation is based on the significant amplitude difference between the interference region and the background noise region on the spectral curve. A typical segmentation boundary is the 35dB difference between the low-noise segment (-20dB amplitude) and the high-noise segment (15dB amplitude). Peak-valley segmentation binarizes the spectral curve by setting a dynamic threshold. The threshold is set to the noise floor mean plus three standard deviations. Frequency segments above the threshold are marked as peak segments, and segments below the threshold are marked as valley segments. The noise floor mean is estimated by averaging the lowest 20% of frequency points in the frequency domain amplitude distribution map. This estimation method assumes that most frequency points are not affected by strong interference. After thresholding, the peak segments identified in the frequency domain amplitude distribution map are filtered based on peak amplitude and bandwidth. Narrow-band peaks with peak amplitudes below 10dB of the noise floor are considered false detections and are removed, retaining only peak segments with significant amplitudes and reasonable bandwidths. Low-noise bands are defined as continuous frequency segments with amplitudes below a threshold, while high-noise bands are defined as continuous frequency segments with amplitudes above a threshold. The bandwidth ratio of low-noise and high-noise bands reflects the overall noise pollution level of the circuit signal; a high-noise band ratio exceeding 30% indicates a severe interference environment. After the low-noise and high-noise band sets are formed, the center frequency and average amplitude of each band are recorded. The average amplitude of low-noise bands is typically between -25dB and -10dB, while the average amplitude of high-noise bands is typically between 5dB and 30dB.

[0022] Compensation parameters are formed based on frequency domain compensation of low-noise and high-noise frequency bands. The core of these parameters is configuring a corresponding notch filter for each high-noise frequency band. An interference peak with a center frequency of 150Hz and an average amplitude of 22dB has a 40dB amplitude difference compared to the background noise with an average amplitude of -18dB; this difference is the basis for setting the notch depth. The compensation parameters employ an adaptive notch filter structure. For each high-noise frequency band, the required notch center frequency and notch depth are calculated. The notch center frequency is set to the center frequency of the high-noise frequency band. The notch depth calculation formula is D_notch = A_high - A_low, where D_notch is the notch depth, A_high is the average amplitude of the high-noise frequency band, and A_low is the average amplitude of the low-noise frequency band. The average amplitude of the low-noise frequency band is obtained by weighted averaging of the amplitudes of all low-noise frequency bands. The weighting coefficient is proportional to the bandwidth of each low-noise frequency band, with the broadband region contributing more to the average value within the low-noise frequency band. The compensation parameters contain configuration information for multiple notch filters, each corresponding to a high-noise frequency band. The number of notch filters in the compensation parameters equals the number of high-noise frequency bands. If a high-noise frequency band contains multiple discrete interference peaks, it is split into multiple sub-bands for separate compensation. Splitting is based on the frequency interval between peaks; splitting is performed when the interval is greater than three times the half-width at half-maximum (WHM) of the peak. Each of the split sub-bands generates an independent compensation parameter entry. The notch filters in the compensation parameters are sorted from high to low according to their notch depth. Filters with a notch depth greater than 30dB are marked as strong compensation items, those with a notch depth between 15dB and 30dB are marked as medium compensation items, and those with a notch depth less than 15dB are marked as weak compensation items. The total number of compensation parameter entries reflects the complexity of the interference received by the circuit signal. When the number of entries exceeds 10, it indicates a complex interference environment requiring multi-level compensation processing.

[0023] Interference mode types are formed based on the frequency difference characteristics of the compensation parameters. The identification of interference mode types is based on the distribution pattern of notch frequencies in the compensation parameters. A 50Hz evenly spaced distribution of notch frequencies (50Hz for the first, 100Hz for the second, and 150Hz for the third) points to a power frequency harmonic interference source. Interference mode types are classified according to the distribution pattern of the notch center frequency in the compensation parameters. If the notch frequencies are evenly spaced and the intervals correspond to integer multiples of a fundamental frequency, it is classified as a harmonic interference mode, typically generated by the periodic operation of switching power supplies or motors. If there is an isolated single notch frequency in the compensation parameters that is close to the power frequency of 50Hz or 60Hz, the interference mode type is classified as a power frequency interference mode, originating from inductive coupling of power lines. When the notch frequency distribution in the compensation parameters exhibits a dense multi-peak characteristic within a certain frequency band, the interference mode type is classified as a broadband interference mode, which may be generated by arc discharge or high-frequency switching devices. Interference mode types are further categorized based on the distribution characteristics of notch depth in the compensation parameters. The proportion of strong, medium, and weak compensation terms in the compensation parameters is used to determine the uniformity of interference. Notch depth differences of less than 5 dB are marked as uniform interference, while differences of more than 20 dB are marked as non-uniform interference. The fundamental frequency in the harmonic interference mode is extracted by calculating the greatest common divisor of the notch frequency sequence in the compensation parameters. The interference intensity level is recorded simultaneously with the identification results of the interference mode type: an average notch depth greater than 30 dB indicates a strong interference level, between 15 dB and 30 dB indicates a medium interference level, and less than 15 dB indicates a weak interference level.

[0024] Noise compensation coefficients are extracted based on interference mode type to generate a signal gain correction table. When the interference mode type is harmonic interference and the fundamental frequency is 50Hz, the noise compensation coefficients are set to the corresponding notch depth values ​​at harmonic frequencies such as 50Hz, 100Hz, and 150Hz, forming a discrete sequence of compensation coefficients. The calculation of noise compensation coefficients prioritizes the frequency points marked as strong compensation terms in the compensation parameters. The noise compensation coefficient corresponding to a strong compensation term is taken as the linear value of its notch depth. The noise compensation coefficient is defined as follows: D_notch represents the notch depth in dB. For harmonic interference modes, the coefficient is set as the ratio of the notch depth at each harmonic frequency to the fundamental frequency notch depth. The type of interference mode determines how this ratio sequence is calculated. For power frequency interference, the noise compensation coefficient is directly taken as a linear value of the power frequency notch depth, supplemented with the notch depth of the power frequency harmonics to form a noise compensation coefficient vector. For broadband interference, the noise compensation coefficient is obtained by fitting a continuous curve to the notch depth within the interference frequency band using cubic spline interpolation. The signal gain correction table is constructed based on the noise compensation coefficient. The table records the gain correction factor corresponding to each frequency point, indexed by frequency. The gain correction factor is equal to the reciprocal of the noise compensation coefficient. The interference intensity level affects the adjustment range of the gain correction factor in the signal gain correction table. The gain correction factor for frequencies corresponding to strong interference levels is closer to 0.1, while the gain correction factor for frequencies corresponding to weak interference levels is closer to 1.0. When the interference mode type is marked as uniform interference, the gain correction factor at each interference frequency point in the signal gain correction table adopts a uniform correction strategy; when marked as non-uniform interference, the gain correction factor at each interference frequency point is calculated independently point by point according to the notch depth. The resolution of the signal gain correction table on the frequency axis is set to 1Hz, covering the effective frequency band from 0 to 100kHz. The signal gain correction table contains gain correction factors for a total of 100,000 frequency points. The value range of the gain correction factor is limited to between 0.1 and 1.0. After the signal gain correction table is generated, it is smoothed by moving average, and the window length is set to 5 frequency points.

[0025] Step S120: Based on the signal gain correction table, the circuit signal data is gain corrected to identify high-frequency interference sections. The high-frequency interference sections are grouped and sorted according to interference intensity to generate an interference classification table. An adaptive filtering configuration is constructed using the interference classification table to extract the characteristics of the filtered signal.

[0026] Specifically, gain correction is performed on the circuit signal data using a signal gain correction table to identify high-frequency interference zones. In the signal gain correction table, the gain correction factor at frequency 35kHz is 0.15, and at frequency 72kHz it is 0.22. These correction factors below 0.3 indicate concentrated areas of high-frequency interference. After the circuit signal data is converted to the frequency domain using a Fast Fourier Transform, the amplitude at each frequency point is multiplied by the corresponding gain correction factor in the signal gain correction table to complete the gain correction. The corrected amplitude is equal to the original amplitude multiplied by the gain correction factor. Again, the gain correction factor at 35kHz and 72kHz in the signal gain correction table is 0.22. These correction factors below 0.3 indicate concentrated areas of high-frequency interference. Within the 100kHz frequency band covered by the signal gain correction table, frequency points with gain correction factors below 0.5 are marked as potential interference points, and the energy of the circuit signal data at these frequency points needs to be suppressed. High-frequency interference segments are defined as a continuous sequence of potential interference points. When the frequency interval between adjacent interference points is less than 100Hz, they are merged into the same segment; when the interval is greater than 100Hz, they are divided into different segments. After gain correction, the identification of high-frequency interference segments is accomplished by scanning the distribution of gain correction factors in the signal gain correction table, proceeding point by point from low frequency to high frequency. The minimum value of the gain correction factor within the high-frequency interference segment is recorded; this minimum value reflects the peak interference intensity of that segment. The number of high-frequency interference segments above 20kHz in the signal gain correction table is significantly greater than that in the low-frequency band.

[0027] In some embodiments, the step of grouping and sorting the high-frequency interference segments according to interference intensity to generate an interference classification table includes: decomposing the high-frequency interference segments into strong interference components and weak interference components; identifying interference source characteristics from the weak interference components to generate source tracing parameters; using the source tracing parameters in combination with the spectral characteristics of the strong interference components to trace and locate the source to form an interference association path; and generating an interference classification table based on the hierarchical depth of the interference association path.

[0028] The high-frequency interference segment is decomposed into strong interference components and weak interference components. The division between strong and weak components is based on the degree of difference in the minimum gain correction factor of each segment. The intensity difference between a segment with a minimum gain correction factor of 0.12 and a segment with a minimum of 0.38 is more than three times, and this difference constitutes the quantitative basis for component division. Strong interference components include segments in the high-frequency interference segment with a minimum gain correction factor below 0.2, while weak interference components include segments with a minimum gain correction factor between 0.2 and 0.5. The total number of high-frequency interference segments in a typical industrial environment is approximately 15 to 30, with strong interference components accounting for approximately 30% to 40% and weak interference components accounting for approximately 60% to 70%. Although the weak interference components have lower interference intensity, they are more numerous, and the cumulative energy of weak interference components in the high-frequency interference segment may approach or exceed that of strong interference components. The frequency distribution of strong interference components exhibits a clustering characteristic, with the 35kHz to 45kHz frequency band being a typical clustering region in the high-frequency interference segment. These segments typically correspond to the operating frequency of the switching power supply and its harmonics. The frequency distribution of weak interference components is relatively dispersed. In high-frequency interference segments, weak interference components are often scattered across the entire high-frequency range. This dispersed characteristic helps in tracing the propagation path of interference. Strong and weak interference components each form an independent set of segments, and each segment records two parameters: the center frequency and the minimum gain correction factor.

[0029] For example, the step of identifying interference source features from the weak interference components to generate source tracking parameters includes: obtaining frequency stability key points in the weak interference components; detecting signal delay features at the frequency stability key points to generate delay tracking parameters; using the delay tracking parameters to perform reverse localization to generate a source candidate set; and generating source tracking parameters based on the source candidate set and the delay tracking parameters.

[0030] Key points for frequency stability in weak interference components are identified. The selection of key points is based on the time-series fluctuation amplitude of the segment's center frequency. Segments with frequency fluctuations of only 0.2Hz have higher tracking value than segments with fluctuations reaching 3.5Hz; the former typically corresponds to a fixed-frequency interference source driven by a crystal oscillator. Key points for frequency stability are selected from segments in the weak interference component with frequency fluctuation amplitudes less than 0.5Hz, whose center frequencies hardly change over time. The frequency stability of weak interference components is quantified by calculating the time-series standard deviation of the center frequencies of each segment; a smaller standard deviation indicates higher stability. The number of key points for frequency stability typically accounts for 20% to 30% of the total number of segments in the weak interference component, and these key points carry core information for tracing the interference source. Segments in the weak interference component that simultaneously possess high frequency stability and high amplitude stability are marked as preferred key points in the key points for frequency stability. Amplitude stability is measured by the time-series coefficient of variation of the minimum gain correction factor of the segment. Key points for frequency stability record the center frequency of the segments; segments in the weak interference component that meet the selection criteria are written into the set of key points for frequency stability one by one. In servo driver power supply scenarios, the switching frequency of the DC-DC converter inside the driver is controlled by a crystal oscillator. The generated interference manifests as a highly stable frequency segment in the weak interference component, and this segment is preferentially included in the key points of frequency stability.

[0031] Delay tracking parameters are generated by detecting signal delay characteristics at key frequency stability points. Segments with harmonic relationships within these key frequency stability points exhibit a fixed phase difference, which is directly related to the delay time of signal propagation from the interference source to the acquisition point. Delay tracking parameters are extracted by analyzing the phase information of the signal at these key frequency stability points. Multiple harmonics generated by the same interference source exhibit a fixed phase correlation. The generation of delay tracking parameters prioritizes segments marked as preferred key points within the key frequency stability points, as these have higher phase information quality, resulting in more accurate delay estimates. Phase extraction at key frequency stability points employs a short-time Fourier transform method, with a transform window length set to 1024 sampling points to obtain sufficient phase resolution. The delay tracking parameters include the absolute phase of each key frequency stability point and the calculated delay estimate, typically ranging from 0.1 μs to 100 μs. Segments with similar delay estimates among key frequency stability points are grouped into the same delay group in the delay tracking parameters; segments within the same group may originate from the same or adjacent interference sources. The delay tracking parameters calculate the mean delay and standard deviation of delay within each delay group. Delay groups with a standard deviation of less than 1 μs among the frequency stability key points have high positioning accuracy. The delay tracking parameters ultimately include a list of delay groups, with each delay group recording the key point number and mean delay within the group.

[0032] A source candidate set is generated using delay tracking parameters for reverse localization. The source candidate set is generated based on the difference in the mean delay of each delay group in the delay tracking parameters. The group with a mean delay of 15.8 μs differs from the group with a mean delay of 47.5 μs by approximately 31.7 μs; this time difference indicates that the two groups originate from interference sources at different locations. The source candidate set is generated based on the mean delay of each delay group in the delay tracking parameters. Each delay group corresponds to one candidate source; a smaller mean delay indicates that the interference source is closer to the acquisition point. The delay group with the smallest mean delay in the delay tracking parameters corresponds to the interference source closest to the acquisition point, and the source candidate set marks this group as a near-end source candidate. Each candidate in the source candidate set records its mean delay and the associated frequency stability key point number. The standard deviation of the delay groups in the delay tracking parameters is converted into location confidence and written into the source candidate set. The delay difference between different delay groups in the delay tracking parameters reflects the relative positional relationship between each interference source; the source candidate set uses this information to construct a relative position sequence of sources. The number of candidates in the source candidate set is equal to the number of delay groups in the delay tracking parameters, typically ranging from 3 to 8. Each delay group in the delay tracking parameters maps to one candidate in the source candidate set. After the source candidate set is formed, it is sorted in ascending order of mean delay, and the sorting result reflects the spatial distribution of each interference source from near to far.

[0033] Source tracing parameters are generated based on the source candidate set and delay tracing parameters. The proximal candidate with a localization confidence of 0.92 in the source candidate set corresponds to the delay group with the smallest mean delay in the delay tracing parameters; this candidate is written as the primary source in the source tracing parameters. The source tracing parameters integrate candidate information from the source candidate set and delay feature information from the delay tracing parameters to form a complete description of the interference source. The frequency stability key point numbers associated with each delay group in the delay tracing parameters are transcribed into the source tracing parameters to establish the correspondence between the source and the interference frequency. Each source entry in the source tracing parameters includes two fields: a suspected fundamental frequency and a confidence score. The suspected fundamental frequency is obtained by extracting the lowest frequency from the frequency stability key points associated with the source candidate set, and the confidence score is inherited from the localization confidence of the source candidate set. Multiple candidates with similar localization confidence in the source candidate set are recorded independently in the source tracing parameters; only when the delay difference between corresponding delay groups in the delay tracing parameters is less than 2 μs are they considered for merging. The source tracing parameters ultimately include a list of sources. Information from the source candidate set and delay tracing parameters is fused in the source tracing parameters. For each source, a suspected fundamental frequency and confidence score are recorded.

[0034] Interference correlation paths are formed by tracing the source using source tracking parameters combined with the spectral characteristics of strong interference components. These paths are organized in a tree structure, with the suspected fundamental frequency in the source tracking parameters as the root node, and segments of strong and weak interference components related to that fundamental frequency as child nodes. The root node and child nodes are associated through harmonic multiples. When an interference source operates, it generates fundamental frequency interference along with multiple harmonic interferences at integer multiples of the frequency. These harmonic interferences are scattered across different positions in the spectrum, forming independent interference segments. The tree structure groups these dispersed segments back to a common source, forming a hierarchical correlation description. The center frequency of the strong interference component is matched with integer multiples of the suspected fundamental frequency in the source tracking parameters; a matching error of less than 1% confirms the correlation. Each path in the interference correlation path records the complete harmonic chain from the source frequency to the terminal interference segment, with the path length equal to the harmonic order, i.e., the hierarchical depth. Suspected fundamental frequencies with a confidence score higher than 0.8 in the source tracking parameters are prioritized for strong interference component matching, and high-confidence correlation chains are prioritized for construction in the interference correlation path. If any segment in the strong interference component cannot be associated with any suspected fundamental frequency in the source tracing parameters, the interference association path marks these segments as unknown source interference, with a level depth of 0. After the interference association path is constructed, the number of strong interference component segments associated with each source frequency is counted, and the suspected fundamental frequency with the most associated segments in the source tracing parameters is determined as the main interference source.

[0035] An interference classification table is generated based on the hierarchical depth of the interference association path. The classification is based on the hierarchical depth of each segment within the interference association path. For example, the 36kHz segment associated with the 12kHz main interference source has a depth of 3, the 48kHz segment has a depth of 4, and the 60kHz segment has a depth of 5. These depth values ​​directly map to the interference level. The table groups all segments in the interference association path according to hierarchical depth: depths 1 to 2 are classified as Level 1 interference, depths 3 to 4 as Level 2 interference, and depths 5 and above as Level 3 interference. Segments with shallower hierarchical depths in the interference association path typically correspond to stronger interference intensity; the average intensity of Level 1 interference in the table is significantly higher than that of Level 3 interference. The table records the frequency range of each interference level and the associated source frequency. The interference association path shows that Level 1 interference is mainly concentrated in the low-frequency band, while Level 3 interference is mainly distributed in the high-frequency band. Segments marked as having unknown sources in the interference association path are separately classified as independent interference levels in the table; the source of this level of interference cannot be traced through harmonic relationships. The interference classification table is sorted from high to low according to the interference level, with Level 1 interference listed first to indicate the interference components that need to be dealt with first. The hierarchical depth information of each segment in the interference association path is fully mapped to the level field of the interference classification table.

[0036] An adaptive filtering configuration is constructed using an interference classification table to extract the characteristics of the filtered signal. The frequency band corresponding to Level 1 interference in the interference classification table is suppressed using a high-order notch filter. The filter order is set to 8th order to obtain a steep stopband characteristic, and the notch depth is adaptively adjusted according to the interference intensity recorded in the interference classification table. Level 2 and Level 3 interference in the interference classification table are processed using 6th and 4th order notch filters, respectively. This differentiated configuration of filter order achieves a reasonable allocation of computational resources. The adaptive filtering configuration sets the center frequency and bandwidth of the filters according to the frequency range of each level of interference in the interference classification table. The filtering bandwidth for Level 1 interference is set to 2% of the center frequency, and the filtering bandwidths for Level 2 and Level 3 interference are set to 3% and 5%, respectively. The segments of independent interference levels in the interference classification table adopt a broadband suppression strategy, with the filtering bandwidth extended to 10% of the center frequency to cover the possible frequency drift range. The adaptive filtering configuration integrates the filter parameters corresponding to each level of interference in the interference classification table into filter banks. The filter banks are cascaded according to the interference level, with the Level 1 interference filter located at the front end of the signal processing chain. After the circuit signal data is processed by the filter bank configured with adaptive filtering, the energy of each interference frequency point is effectively suppressed, and the filtered signal features are extracted from the processed time-domain waveform. The filtered signal features include time-domain statistical parameters and frequency-domain distribution parameters. The time-domain statistical parameters include mean, variance, peak-to-peak value, and peak factor, while the frequency-domain distribution parameters include the spectral centroid and spectral bandwidth. The calculation of each parameter in the filtered signal features is based on a complete 10-second data segment after filtering, and the adaptive filtering guided by the interference classification table ensures the signal-to-noise ratio of the extracted features.

[0037] Step S130: Transient pulse detection is performed on the filtered signal features to obtain transient pulse response. Abnormal peak segments are identified from the transient pulse response to generate peak weight coefficients. The peak weight coefficients are used to perform feature enhancement processing on the filtered signal features to generate a key feature set.

[0038] Specifically, transient pulse detection is performed on the filtered signal features to obtain the transient pulse response. When the centroid of the spectrum in the filtered signal features is 32kHz and the bandwidth is 18kHz, the signal energy is mainly concentrated in the mid-to-high frequency band, which is a typical distribution area for transient pulses. Transient pulse detection uses wavelet transform to decompose the time-domain waveform corresponding to the filtered signal features into multiple scales. The db4 wavelet is selected as the wavelet basis function, and the decomposition level is set to 6 levels to cover the complete frequency band from low to high frequencies. The peak-to-peak value parameter in the filtered signal features indicates the dynamic range of the signal; a larger peak-to-peak value indicates the possible presence of transient components with abrupt amplitude changes. The peak factor parameter in the filtered signal features is used to predict the likelihood of transient pulses; a peak factor greater than 5 indicates the presence of significant spike components in the signal, and the search range for transient pulse detection is correspondingly expanded. The transient pulse response is obtained by detecting abrupt changes in the wavelet decomposition coefficients. An abrupt change is defined as the position where the difference in wavelet coefficients between adjacent sampling points exceeds three times the standard deviation of the local mean. The variance parameter of the filtered signal features reflects the overall fluctuation of the signal. When the variance is large, the sensitivity threshold for transient pulse detection is correspondingly increased to avoid false detections. The transient pulse response records the time position and amplitude of each abrupt change point, and the mean parameter of the filtered signal features is used as the reference for amplitude normalization and written into the transient pulse response. In the scenario of motor bearing fault monitoring, each time the bearing rolling element passes through the damage point, it will generate periodic transient pulses in the waveform corresponding to the filtered signal features. The transient pulse response captures the time position and amplitude characteristics of these pulses.

[0039] In some embodiments, the step of identifying abnormal peak segments from the transient impulse response and generating peak weight coefficients includes: identifying amplitude abrupt changes in the transient impulse response to generate abrupt change distribution regions; extracting segments exceeding a threshold from the abrupt change distribution regions to generate abnormal peak segments; performing peak attenuation analysis based on the abnormal peak segments to form attenuation rate parameters; and applying the attenuation rate parameters to perform weight transformation to generate peak weight coefficients.

[0040] Amplitude abrupt change identification is performed on the transient impulse response to generate abrupt change distribution regions. Amplitude abrupt changes are determined based on whether the amplitude difference between adjacent sampling points exceeds a threshold. When the amplitude at the 127th sampling point is 2.8V and jumps to 8.5V at the 128th sampling point, the amplitude difference of 5.7V constitutes a typical abrupt change characteristic. Amplitude abrupt change identification calculates the amplitude difference between adjacent sampling points in the transient impulse response point by point. Before the amplitude difference calculation, the amplitude of each sampling point is divided by the normalized benchmark recorded in the transient impulse response for normalization. Points where the normalized amplitude difference exceeds twice the overall amplitude standard deviation of the transient impulse response are marked as abrupt change points. The distribution of abrupt change points in the transient impulse response exhibits a clustering characteristic; multiple adjacent abrupt change points constitute a abrupt change region, while isolated single abrupt change points form independent regions. The abrupt change distribution regions cluster the abrupt change points in the transient impulse response according to spatial proximity. Adjacent abrupt change points with a time interval of less than 1ms are grouped into the same abrupt change distribution region, while those with a time interval greater than 1ms are segmented into different abrupt change distribution regions. The transient impulse response spans 10 seconds, and the number of abrupt change distribution regions typically ranges from 20 to 100, closely related to the equipment's operating status. The start and end times, maximum amplitude difference, and peak position of each abrupt change distribution region are recorded. The location of the abrupt change point with the largest amplitude difference in the transient impulse response is marked as the peak position of the abrupt change distribution region. The abrupt change distribution regions are sorted from highest to lowest according to their maximum amplitude difference; this sorting result initially reflects the degree of anomaly in each region.

[0041] Abnormal peak segments are generated by extracting segments exceeding a threshold from the mutation distribution area. The selection of abnormal peak segments requires distinguishing between normal fluctuations and true abnormal peaks. Regions with a maximum amplitude difference of 6.2V show a significant difference in abnormality compared to regions with a maximum amplitude difference of 1.1V; the threshold is used to quantify this distinction. The extraction threshold for abnormal peak segments is determined based on the statistical distribution of the maximum amplitude difference in the mutation distribution area. The threshold is set as the mean amplitude difference plus two standard deviations. Mutation distribution areas exceeding this threshold are extracted as abnormal peak segments. Approximately 15% to 25% of the regions in the mutation distribution area meet the threshold condition and are extracted as abnormal peak segments. These segments correspond to the time periods in the signal where amplitude changes are most drastic. Abnormal peak segments inherit the start and end times and peak position information of the mutation distribution area and add a peak amplitude field to record the maximum signal amplitude within the segment. Regions in the mutation distribution area that do not reach the threshold are not recorded in the abnormal peak segments; the amplitude changes in these regions fall within the normal signal fluctuation range. The duration of abnormal peak segments is typically between 0.1 ms and 5 ms. Areas with excessively long durations within the abrupt change distribution zone require further verification even if the amplitude difference exceeds the threshold. Durations exceeding 10 ms may indicate a step signal rather than a transient peak. After the abnormal peak segments are formed, they are sorted by time, and the sorting results demonstrate the temporal distribution pattern of the abnormal peaks within the monitoring period.

[0042] Attenuation rate parameters are derived by analyzing the attenuation of peaks in anomalous peak segments. The signal amplitude following the peak position within an anomalous peak segment exhibits an exponential decay trend, and the attenuation rate reflects the speed of energy dissipation and the damping characteristics of the signal propagation path. The attenuation rate parameter is obtained by exponentially fitting the amplitude sequence following the peak position within the anomalous peak segment. The fitting model uses an exponential decay function, and the attenuation rate parameter λ characterizes the rate of amplitude decay. Segments with larger attenuation rate parameters within the anomalous peak segment correspond to rapidly decaying peaks, typically generated by nearby impact sources; segments with smaller attenuation rate parameters correspond to slowly decaying peaks, which may have propagated over longer paths. The typical range for the attenuation rate parameter is 100 to 5000 ms. Peaks with attenuation rate parameters below 100 ms in an anomalous peak segment may contain low-frequency oscillation components. The attenuation rate parameter is calculated independently for each segment within the anomalous peak segment, using the data segment from the peak position to when the amplitude decays to 10% of the peak amplitude. The decay rate parameter also records the goodness-of-fit index. Sections with a goodness-of-fit below 0.8 in the abnormal peak region are marked as atypical decay, and the decay process in these sections may be affected by multiple reflections or superimposed interference. In the gearbox fault diagnosis scenario, the impact signal generated by pitting on the tooth surface exhibits a medium-speed decay characteristic with a decay rate parameter of approximately 800 to 1500 per second in the abnormal peak region.

[0043] The attenuation rate parameter is used to perform a weight transformation to generate peak weight coefficients. The design principle of the weight transformation is to assign weights based on the fault indication significance of the attenuation rate parameter. A fast attenuation peak with an attenuation rate of 2500 ms differs in fault characteristics from a slow attenuation peak with an attenuation rate of 300 ms, and the weight allocation needs to reflect this difference. The peak weight coefficients are generated by performing a nonlinear mapping on the attenuation rate parameter, with the mapping function being... Where W is the peak weight coefficient, λ is the decay rate parameter, λ_mid is the median of the optimal decay rate (1250 ms), and k is the curve steepness coefficient, determined according to the desired effective decay rate range, typically set to 0.004 ms. Abnormal peak segments with decay rate parameters in the range of 500 to 2000 ms receive high weight coefficients of 0.8 to 1.0, corresponding to the attenuation characteristics of impact signals generated by typical mechanical faults. The peak weight coefficient ranges from 0.1 to 1.0; abnormal peak segments with decay rate parameters exceeding the normal range receive low weights to reduce their contribution to feature enhancement. Atypical decay segments with a goodness of fit below 0.8 in the decay rate parameter are uniformly assigned a value of 0.5 in the peak weight coefficient to avoid excessive influence of abnormal decay patterns on weight allocation. Each abnormal peak segment corresponds one-to-one with a peak weight coefficient, with each segment receiving a weight value between 0.1 and 1.0. After the peak weight coefficient is formed, it is bound to the time location information of the abnormal peak segment to form a time-weight mapping relationship.

[0044] A key feature set is generated by using peak weighting coefficients to enhance the features of the filtered signal. The time segment corresponding to the abnormal peak segment with a peak weighting coefficient of 0.95 receives the highest amplification during feature enhancement, while the segment with a peak weighting coefficient of 0.2 receives only slight enhancement. The feature enhancement process segments the time-domain waveform corresponding to the filtered signal features, and the enhancement coefficient for each segment is determined based on the maximum value of the peak weighting coefficient within that time segment. Statistical parameters such as peak-to-peak value and variance in the filtered signal features are recalculated after feature enhancement; the signal amplitude in segments with higher peak weighting coefficients is amplified, significantly increasing the values ​​of these statistical parameters. The key feature set includes the enhanced time-domain statistical features and frequency-domain distribution features. The time-domain features include the enhanced mean, enhanced variance, enhanced peak-to-peak value, and enhanced peak factor; the frequency-domain features include the enhanced spectral centroid and enhanced spectral bandwidth. The role of the peak weighting coefficient in feature enhancement is to highlight the contribution of abnormal peak segments; the time segments marked with high weight by the peak weighting coefficient in the filtered signal features have a greater impact on the key feature set. The key feature set also records statistical information on the peak weighting coefficients, including the weight mean, weight variance, and the proportion of high-weight segments. This information reflects the overall distribution characteristics of abnormal peaks in the signal. After the filtered signal features are weighted and enhanced by the peak weighting coefficients, the feature parameters in the key feature set are more sensitive to abnormal equipment conditions.

[0045] Step S140: Identify signal mutation points using key feature sets, sparsely sample the signal mutation points according to their temporal distribution to generate feature label sequences, and construct a segmented storage strategy based on the feature label sequences.

[0046] Specifically, signal abrupt change points are identified using a key feature set. When the enhanced peak factor in the key feature set suddenly increases from the normal value of 3.5 to 8.2, this change indicates the appearance of a significant spike in the signal waveform, and the corresponding moment is marked as a signal abrupt change point. The identification of signal abrupt change points involves analyzing the rate of change of the time series of each feature parameter in the key feature set. Moments with a rate of change exceeding a set threshold are extracted as candidate abrupt change points. The rate of change is calculated by dividing the difference between adjacent sampling points by the sampling interval. The confidence level of the signal abrupt change point is highest when both the enhanced variance and the enhanced peak-to-peak value in the key feature set show abrupt changes simultaneously; the confidence level decreases accordingly when only one parameter changes. Abrupt changes in the enhanced mean in the key feature set usually indicate a shift in the overall signal level, while abrupt changes in the enhanced spectral bandwidth reflect a change in the signal frequency distribution range. Analysis of these two parameters in conjunction with the enhanced spectral centroid can identify the overall migration of frequency domain characteristics. Signal abrupt change points record the timestamp of the abrupt change, the name of the feature parameter triggering the abrupt change, and the magnitude of the abrupt change. The magnitude of the abrupt change is defined as the ratio of the feature parameter values ​​before and after the change. Sudden changes in the weighted mean parameter of the key feature set are often accompanied by linked changes in multiple other feature parameters. When a signal mutation point detects a sudden change in the weighted mean, it simultaneously checks the changes in related parameters. The weight variance and the proportion of high-weight segments in the key feature set are used to determine the priority of the signal mutation point. When the weight variance is large and the proportion of high-weight segments exceeds 40%, the signal mutation point identified in the corresponding time period is marked as a high-priority mutation. In the scenario of CNC machine tool spindle monitoring, the moment the tool breaks, it will cause a sudden change in the enhanced peak factor and the enhanced spectral centroid in the key feature set. The signal mutation point captures this linked change and marks it as a high-priority mutation. The number of signal mutation points is related to the operating status of the equipment. During normal operation, signal mutation points are sparsely distributed, while during abnormal conditions, signal mutation points appear densely. The feature sensitivity of the key feature set ensures the effective identification of mutation points.

[0047] In some embodiments, the step of sparsely sampling the signal mutation points according to their temporal distribution to generate a feature label sequence includes: performing a temporal density statistical evaluation on the signal mutation points to generate a temporal density distribution map; identifying dense mutation periods from the temporal density distribution map to generate dense period identifiers; performing downsampling processing on the dense period identifiers to obtain a sparsification parameter set; and performing sparse reconstruction based on the sparsification parameter set to generate a feature label sequence.

[0048] For example, the step of performing time-series density statistical evaluation on the signal abrupt change points to generate a time-series density distribution map includes: assigning time-series weights to the signal abrupt change points to generate an initial weight sequence; forming a decay weight coefficient based on the time decay characteristics of the initial weight sequence; using the decay weight coefficient to perform weighted statistics on the signal abrupt change points to generate density data; and arranging the density data in time sequence to generate a time-series density distribution map.

[0049] An initial weight sequence is generated by assigning time-series weights to signal mutation points. The contribution of signal mutation points with different amplitudes to density statistics needs to be differentiated; a high-amplitude mutation (amplitude 5.2) should receive a larger statistical weight than a low-amplitude mutation (amplitude 1.3). The initial weight sequence assigns a weight value to each signal mutation point, calculated based on both mutation amplitude and priority; a larger amplitude and higher priority result in a larger weight value. The mutation amplitude of each signal mutation point is normalized to the range of 0 to 1 and used as the base component for weight calculation, accounting for 60% of the weight values ​​in the initial weight sequence. High-priority mutations receive a priority bonus of 0.4 in the initial weight sequence, while ordinary mutations receive a priority bonus of 0. The weight values ​​in the initial weight sequence range from 0.3 to 1.0, with the smallest mutation amplitude and ordinary priority point receiving the lowest weight of 0.3. The initial weight sequence and signal mutation points correspond one-to-one in time sequence, with the sequence length equal to the number of signal mutation points, forming a complete time-weight mapping relationship. The timestamp information of the signal mutation point is preserved in the initial weight sequence, and each sequence element contains two fields: timestamp and weight value.

[0050] The decay weight coefficients are formed based on the time decay characteristics of the initial weight sequence. The introduction of time decay characteristics causes earlier abrupt changes to gradually decrease their contribution to the density assessment at the current time step, while later abrupt changes maintain a higher contribution. The decay weight coefficients are calculated using an exponential decay model, with the following formula: Where W_decay is the attenuation weight coefficient, Δt is the time difference between the timestamp of the signal mutation point and the current evaluation time, and τ is the attenuation time constant, set to 2 seconds. The attenuation time constant of the attenuation weight coefficient can be adjusted according to the monitoring scenario. A smaller time constant is used for rapidly changing signals in the initial weight sequence to improve response speed, while a larger time constant is used for slowly changing signals to enhance smoothness. The attenuation weight coefficient is related to the temporal position of each mutation point in the initial weight sequence. The same mutation point obtains different attenuation weight coefficients at different evaluation times; the closer to the evaluation time, the closer the attenuation weight coefficient is to 1. The effective weight is obtained by multiplying the weight value of the initial weight sequence by the attenuation weight coefficient. The effective weight reflects the actual contribution of each mutation point to the density evaluation at the current time. The attenuation weight coefficient ensures that recent mutations in the initial weight sequence dominate the density evaluation, while the influence of distant mutations gradually decreases as the time difference increases. The attenuation weight coefficient ranges from 0 to 1. When the time difference is 0, the attenuation weight coefficient is 1, indicating no attenuation; when the time difference approaches infinity, the attenuation weight coefficient approaches 0, indicating complete attenuation.

[0051] A density data set is generated by weighting signal abrupt changes using an attenuation weighting coefficient. During the weighted statistical process, recent abrupt changes contribute significantly due to their attenuation weighting coefficient being close to 1, while distant abrupt changes contribute only slightly due to their coefficient approaching 0, thus reflecting the time-sensitive nature of density assessment. Density data is calculated once at each assessment time by summing the effective weights of all signal abrupt changes within a specified time window prior to that time, with the time window width set to 3 seconds. The effective weight of a signal abrupt change is equal to the corresponding weight value in the initial weight sequence multiplied by the attenuation weighting coefficient. The density data is then aggregated to obtain the density assessment value by summing all effective weights within the time window. The density data is sampled uniformly at 0.1-second intervals within the monitoring period, generating 100 assessment time points in a 10-second monitoring period, forming a complete density time series. The attenuation weighting coefficient ensures that points closer to the assessment time contribute more to the density data, achieving temporal locality in density assessment. The density values ​​at each assessment time constitute the density time series, with periods of concentrated signal abrupt changes corresponding to high-value regions in the series.

[0052] The density data is arranged chronologically to generate a time-series density distribution map. The density values ​​fluctuate significantly over time, peaking at 12.6 at the 3.5-second evaluation time and dropping to only 2.1 at the 7.2-second evaluation time. This fluctuation pattern forms a complete density change curve after chronological arrangement. The time-series density distribution map is plotted with the evaluation time on the horizontal axis and the density values ​​of the density data on the vertical axis. The horizontal axis covers the entire monitoring period from 0 to 10 seconds, while the vertical axis range is adaptively set based on the maximum density value. Density values ​​at adjacent evaluation times are connected using linear interpolation, resulting in a continuous curve shape for easy observation of density change trends. The density data is smoothed to eliminate evaluation noise. The smoothing window width is set to 5 evaluation time points, and moving average filtering is used as the smoothing method. The maximum density values ​​in the density data are marked as peak points in the time-series density distribution map. These peak points correspond to the times when signal abrupt changes are most concentrated, and both the time location and density value of each peak point are recorded. The time-series density distribution map also labels the statistical characteristics of the density data, including the density mean, density maximum, and density standard deviation. These statistical values ​​are used to determine the identification threshold for dense periods. After the time-series density distribution map is generated, it stores two versions of the density data: the original version for precise analysis and the smoothed version for trend observation.

[0053] Dense period identifiers are generated by identifying abrupt changes in the temporal density distribution map. The difference between high-density windows and low-density windows forms the basis for dividing dense periods. Windows with a density value of 28 points / second need to be identified as dense periods for subsequent targeted processing. The generation of dense period identifiers involves thresholding the density values ​​of each window in the temporal density distribution map. The threshold is set to the mean density value plus 1.5 times the standard deviation; windows exceeding the threshold are identified as dense periods. A smoothed numerical version of the temporal density distribution map is used for thresholding to avoid misjudgments caused by noise, while the original numerical version is used for precise location of dense period boundaries. Multiple adjacent high-density windows in the temporal density distribution map are merged into a single continuous dense period identifier. The merging condition is that the density values ​​of adjacent windows all exceed 80% of the threshold, forming a complete dense period interval. The dense period identifier records the start and end times of the period, the peak density, and the total number of signal abrupt change points within the period. The period corresponding to the maximum density value in the temporal density distribution map is marked as the main dense period, and the time position of the peak point is used to determine the core area of ​​the main dense period. The density gradient information from the time-series density distribution map is used to determine the boundary accuracy of dense period markers. The location with the largest absolute gradient value is set as the period boundary to ensure accurate positioning of the start and end points of dense periods. Dense period markers typically occupy 10% to 30% of the monitoring period, and the time-series density distribution map shows that dense period markers contain more than 70% of signal abrupt change points. After the dense period markers are formed, they are sorted from high to low peak density, and the sorting result indicates the processing priority of each dense period.

[0054] Downsampling is performed on dense time period identifiers to obtain a sparsity parameter set. The necessity of downsampling stems from the data redundancy caused by the excessive number of abrupt changes within dense time periods. Retaining all data from a period with a peak density of 35 points / second would significantly increase storage and computational overhead. The sparsity parameter set calculates a downsampling ratio for each dense time period identifier, determined by the ratio of peak density to target density, set to 8 points / second. Periods with short start-end time spans but a large number of abrupt changes within the dense time period identifier receive higher downsampling ratios, ranging from 4:1 to 6:1 in the sparsity parameter set. The sparsity parameter set also includes a sampling strategy field, with two strategies: uniform sampling and priority sampling. Uniform sampling retains signal abrupt changes at fixed intervals, while priority sampling prioritizes the retention of high-priority abrupt changes. The sampling strategy for each time period in the dense time period identifier is determined based on the proportion of high-priority mutation points within that time period. This proportion is calculated by dividing the number of high-priority mutation points by the total number of mutation points recorded in the dense time period identifier. When the proportion exceeds 30%, priority sampling is used to ensure that important mutation points are not missed; when the proportion is below 30%, uniform sampling is used to simplify the processing flow. A sparsity parameter set corresponds one-to-one with the dense time period identifier. Each dense time period identifier receives a set of parameters containing the downsampling ratio and sampling strategy. For time periods marked as the main dense time periods in the dense time period identifier, a more conservative downsampling ratio is set in the sparsity parameter set to ensure that key mutation information is not lost.

[0055] Feature marker sequences are generated through sparse reconstruction based on the sparsity parameter set. Sparse reconstruction filters abrupt changes within dense time periods according to the configuration of the sparsity parameter set; when the downsampling ratio is 4:1, only 5 representative abrupt changes are retained out of the original 20. The feature marker sequence is generated by filtering and recombining signal abrupt changes according to the configuration of the sparsity parameter set. Abrupt changes within dense time periods are downsampled according to the strategy specified by the sparsity parameter set, while abrupt changes within non-dense time periods are all retained without filtering. In time periods where the sampling strategy in the sparsity parameter set is priority sampling, the feature marker sequence prioritizes high-priority abrupt changes, and the remaining quota is evenly distributed to ordinary abrupt changes over time to ensure that high-value information is preserved. Each marker in the feature marker sequence inherits the timestamp, feature parameter name, and abrupt change amplitude information of the original signal abrupt change, and adds a sparse marker field to record whether the point has undergone downsampling processing. The downsampling ratio of the sparsity parameter set directly determines the data volume of the feature marker sequence; typically, the data volume of the feature marker sequence is 40% to 60% of the number of original signal abrupt changes. After sparse reconstruction, the feature marker sequence undergoes a continuity check to examine whether the downsampling ratio configured in the sparsification parameter group led to the loss of key mutation information. If the time interval between adjacent retained points exceeds 0.5 seconds, a point-filling mechanism is triggered. The feature marker sequence is finally arranged in chronological order, and the downsampling results of each dense period and the complete retained results of non-dense periods in the sparsification parameter group are merged to form a unified sequence structure.

[0056] In some embodiments, constructing a sharded storage strategy based on the feature marker sequence includes: performing data density location and identification on the feature marker sequence to generate a data-dense region; performing dispersion assessment on the data-dense region to form a distributed storage index; using the distributed storage index to fragment the data-dense region to generate a distributed storage region; and using the distributed storage region to implement cross-region allocation to generate a sharded storage strategy.

[0057] Data-dense regions are identified by locating data density in the feature-labeled sequence. The distribution of label points along the time axis exhibits significant density variations: 28 label points are clustered between 2.0 and 3.5 seconds, while only 6 are clustered between 6.0 and 7.5 seconds. This uneven distribution necessitates identification through data-dense region location. Data-dense region identification involves dividing the feature-labeled sequence into time windows with a width of 0.5 seconds, consistent with the window width of the time-series density distribution map. Windows containing more than twice the average density of the sequence are identified as candidate dense regions. Adjacent candidate dense regions in the feature-labeled sequence are merged into a single, continuous data-dense region. The start and end times and the number of label points contained within the merged region are recorded. The identification threshold for data-dense regions is adaptively adjusted based on the overall density of the feature-labeled sequence; the threshold is increased when the sequence density is high to prevent excessive areas from being identified as dense regions. Label points in the feature-labeled sequence that have undergone sparse reconstruction recover their original density information during data-dense region identification. The sparse label field indicates that the point originated from downsampling, and the original density is used to accurately assess storage requirements. Data-dense regions are directly related to the distribution pattern of marker points in the feature marker sequence. Periods with dense signal abrupt changes still exhibit data-dense regions after sparse reconstruction. The number and total duration of data-dense regions reflect the storage requirements of the feature marker sequence; a higher proportion of dense regions indicates greater storage pressure.

[0058] Dispersion assessment is performed on data-intensive areas to generate a distributed storage metric. The dispersion assessment quantifies the distribution characteristics of the time intervals between marked points within the data-intensive area. A uniform interval distribution is suitable for sequential storage, while drastic interval fluctuations require distributed storage to optimize access efficiency. The distributed storage metric is obtained by calculating the coefficient of variation (COP) of the time intervals between marked points within the data-intensive area. The COP equals the standard deviation of the interval divided by the mean of the intervals; a higher COP indicates higher dispersion. When the time intervals in the data-intensive area range from 0.02 seconds to 0.15 seconds, the COP calculated by the distributed storage metric is approximately 0.6, indicating a moderate level of distributed storage requirement. The distributed storage metric is calculated independently for each data-intensive area. The calculation considers the number of marked points recorded in the data-intensive area to determine if the statistical sample size is sufficient. For data-intensive areas with fewer than 5 marked points, the distributed storage metric is set to a default value of 0.3. Areas within the data-intensive area with a distributed storage metric value higher than 0.5 are marked as high-dispersion areas, areas with a distributed storage metric value lower than 0.3 are marked as low-dispersion areas, and areas in between are marked as medium-dispersion areas. The distributed storage metric also calculates the interval range ratio within data-intensive areas. The range ratio is equal to the ratio of the maximum interval to the minimum interval. A larger range ratio indicates a more uneven interval distribution, and a range ratio exceeding 5 requires special handling. Once the distributed storage metric is generated, it is bound to the data-intensive areas, forming a region-metric mapping relationship.

[0059] Distributed storage regions are generated by fragmenting data-intensive areas using distributed storage metrics. Fragmentation breaks down highly dispersed data-intensive areas into multiple smaller storage units. A region with a distributed storage metric value of 0.72 can be divided into 4 to 6 independent distributed storage regions. Distributed storage regions are generated by dividing the data-intensive area according to the granularity guided by the distributed storage metric. The granularity of the division is positively correlated with the distributed storage metric value; the higher the metric value, the finer the division. Data-intensive areas marked as highly dispersed by the distributed storage metric are divided into 4 to 6 distributed storage regions, those marked as medium dispersed are divided into 2 to 3 distributed storage regions, and those marked as low dispersed are divided into 1 to 2 distributed storage regions or remain undivided. The division boundaries of distributed storage regions are selected at locations within the data-intensive area with larger time intervals. Dividing at larger intervals reduces cross-region data access and improves the locality of data retrieval. The range ratio parameter in the distributed storage metric helps determine the number of divisions; when the range ratio exceeds 5, the number of divisions is increased to isolate abnormal interval segments. The distributed storage area inherits the start and end time range of the data-intensive area it belongs to, and further refines it into multiple sub-time periods, each of which constitutes an independent distributed storage area. The distributed storage area record contains the number of marker points and the estimated storage capacity. The coefficient of variation in the distributed storage metric is used to estimate the access frequency weight of each distributed storage area.

[0060] A sharding storage strategy is implemented using distributed storage areas to generate cross-area allocation. Cross-area allocation plans storage locations based on the capacity and access characteristics of each distributed storage area. Small blocks with an estimated capacity of 2.5KB and large blocks of 18KB are placed using differentiated strategies to achieve load balancing. The sharding storage strategy allocates storage locations based on the capacity and access frequency weight of the distributed storage areas, prioritizing larger capacity and higher access frequency areas to high-speed storage regions. Multiple blocks from the same original data-intensive area within a distributed storage area are allocated to different physical storage partitions in the sharding storage strategy to avoid access bottlenecks caused by the concentration of hot data. The sharding storage strategy generates storage addresses and backup strategies for each distributed storage area. The storage addresses specify the primary and backup storage locations, and the backup strategies determine the number of backups based on the importance level of the distributed storage area. Blocks with higher access frequency weights within a distributed storage area are configured with more backup copies in the sharding storage strategy to ensure the availability and fault tolerance of frequently accessed data. The sharding storage strategy also includes an inter-distributed storage area association index, which records the storage locations of each distributed storage area originally belonging to the same data-intensive area, supporting data integrity reconstruction. After the sharded storage strategy is generated, a storage mapping table is formed. The mapping table is indexed by the distributed storage area number and records the storage address, capacity, access weight and backup configuration of each block.

[0061] Step S150: Based on the sharded storage strategy, monitor the storage load to obtain the load fluctuation status, use the load fluctuation status to generate dynamic storage priority, and convert the feature tag sequence into structured signal feature data according to the dynamic storage priority.

[0062] Specifically, storage load monitoring is performed based on the sharded storage strategy to obtain load fluctuation status. The basic data for storage load monitoring comes from the storage mapping table in the sharded storage strategy. The distributed storage area numbers in the mapping table range from DS-001 to DS-023, and the access frequency weight of each block varies from 0.2 to 0.95. These parameters provide monitoring targets for load statistics. Storage load monitoring performs real-time access volume statistics for each storage location specified by the sharded storage strategy, with the statistical period set to 100 milliseconds to capture short-term load fluctuations. The load of sequential storage blocks and distributed storage areas in the sharded storage strategy is statistically analyzed separately. The load of sequential storage blocks is usually relatively stable, while the load of distributed storage areas fluctuates more drastically. The load fluctuation status records the number of accesses, read / write ratio, and average response time of each storage block within the monitoring period. Blocks with higher access frequency weights in the sharded storage strategy usually show higher accesses during load fluctuation. The associated indexes of the sharded storage strategy are reflected in the load fluctuation status as two components: index access load and data access load. The index access load corresponds to the query overhead of the associated index, and the data access load corresponds to the actual data read / write overhead. During load fluctuations, the load data for each storage block is organized in a time series, with the series length equal to the number of monitoring periods. A 10-second monitoring period generates 100 load sampling points at 100-millisecond statistical intervals. In the sharded storage strategy, access to backup replicas is statistically analyzed separately during load fluctuations. The load distribution between the primary and backup replicas reflects the redundancy utilization of data access.

[0063] In some embodiments, generating dynamic storage priorities using the load fluctuation state includes: converting the load fluctuation state into a resource utilization distribution; identifying high-load periods from the resource utilization distribution to generate load peak markers; performing load balancing analysis on the load peak markers to form latency tolerance parameters; and reconstructing priorities based on the latency tolerance parameters to generate dynamic storage priorities.

[0064] The load fluctuation state is converted into a resource utilization distribution. The purpose of the resource utilization distribution is to normalize the load data of different storage blocks. A block accessed 1200 times per second differs by orders of magnitude in resource consumption from a block accessed 150 times per second; normalization makes this difference comparable. The resource utilization distribution normalizes the access load of each storage block in the load fluctuation state, using the design capacity and bandwidth limit of each block as the normalization benchmark. Utilization rate equals the actual load divided by the capacity limit. The read / write ratio parameter in the load fluctuation state affects the calculation weight of the resource utilization distribution. Write operations typically consume 1.5 to 2 times the resource consumption of read operations; the resource utilization distribution weights and amplifies the write operation load. The resource utilization distribution displays the resource usage status of each storage block with time as the horizontal axis and utilization percentage as the vertical axis. Monitoring period data in the load fluctuation state is mapped point by point to the resource utilization distribution curve. The average response time parameter in the load fluctuation state is converted into a response latency indicator in the resource utilization distribution. When the response time exceeds a set threshold, the corresponding utilization rate is marked as overloaded. The resource utilization distribution generates independent utilization curves for each storage block during load fluctuations, with the number of curves equal to the total number of storage blocks. The resource utilization distribution also calculates a global utilization metric, which is a weighted average of the utilization rates of each block, with the weights derived from the access frequency weights during load fluctuations.

[0065] High-load periods are identified from the resource utilization distribution to generate load peak markers. The purpose of identifying high-load periods is to provide a basis for resource scheduling decisions. Periods with 85% utilization differ significantly in storage pressure from periods with only 20% utilization; the former requires priority attention and handling. Load peak marker generation involves threshold detection of the utilization value at each moment in the resource utilization distribution, with a threshold set at 70%. Periods with utilization exceeding the threshold are identified as high-load periods. Multiple adjacent high-utilization moments in the resource utilization distribution are merged into a single continuous load peak marker, with the merging condition being that the utilization rate at adjacent moments remains above 90% of the threshold. The load peak marker records the start and end times, peak utilization, and duration of the high-load period. The period with the highest peak utilization in the resource utilization distribution is marked as the primary peak. Peak detection is performed on the global utilization curve and the utilization curves of each block in the resource utilization distribution, and the load peak marker distinguishes between global peaks and local peaks. In multi-channel parallel signal acquisition scenarios, when data from each channel is written to storage simultaneously, synchronous load peaks are formed in the resource utilization distribution. Load peak markers capture this characteristic of synchronous multi-channel writing. After the load peak markers are formed, they are sorted from high to low according to the peak utilization rate, and the sorting results indicate the processing urgency of each high-load period.

[0066] Load balancing analysis is performed on load peak markers to generate latency tolerance parameters. The core task of load balancing analysis is to assess the distributability and delayability of each load peak. The main peak with a peak occupancy of 92% and a duration of 2.3 seconds has the greatest impact on storage performance and needs to be prioritized for balancing. Latency tolerance parameters are generated by analyzing the timeliness requirements of data access within the corresponding time period of the load peak marker. Data access with low timeliness requirements can be delayed until the off-peak period. Data access within each high-load period of the load peak marker is categorized by source: access from real-time monitoring has high timeliness requirements, while access from historical queries has low timeliness requirements. The latency tolerance parameter calculates the proportion of delayable access for each load peak marker. The proportion of delayable access equals the number of low-timeliness accesses divided by the total number of accesses. The higher the proportion, the easier it is to mitigate the peak through latency strategies. Short-duration peaks in the load peak markers usually correspond to bursty accesses, and the latency tolerance parameter sets a lower tolerance for these peaks to avoid latency accumulation. The latency tolerance parameter ranges from 0 to 1. A value of 0 indicates that all accesses during that period cannot be delayed, while a value of 1 indicates that all accesses can be delayed. The latency tolerance parameter also records the recommended latency duration, which is determined based on the peak-to-valley interval of the load peak marker and the amount of accesses that can be delayed.

[0067] Dynamic storage priorities are generated based on latency tolerance parameters. The principle of priority refactoring is to lower the priority of deferred access to smooth the load curve. During high-load periods with a latency tolerance parameter of 0.65, 65% of data accesses can be delayed, and these accesses become the primary targets for priority adjustment. Dynamic storage priorities redistribute access requests for each storage block based on a comprehensive assessment of the latency tolerance parameter and the current load status. During periods with a high proportion of deferred accesses in the latency tolerance parameter, dynamic storage priorities lower the priority of low-time-sensitivity accesses by 1 to 2 levels from the default value. Dynamic storage priorities use a 5-level priority system, with level 1 being the highest priority for accesses with the highest real-time requirements and level 5 being the lowest priority for accesses that can be significantly delayed. The recommended latency duration information in the latency tolerance parameter is written into the latency configuration field of the dynamic storage priorities to guide the actual delayed execution of access requests. Dynamic storage priorities generate independent priority configurations for each storage block, including access quotas for each priority level and adjustment strategies corresponding to the latency tolerance parameter. In continuous monitoring scenarios for industrial equipment, the timeliness requirements for signal data during steady-state operation are relatively low. Dynamic storage priority lowers the storage priority of this type of data, reserving a high-priority channel for potentially abnormal signal data. After the dynamic storage priority is formed, it interfaces with the storage scheduling module to achieve adaptive load balancing based on latency tolerance parameters.

[0068] The feature tag sequence is converted into structured signal feature data based on dynamic storage priority. The execution order of the structured conversion is determined by the priority mapping relationship in the dynamic storage priority. Storage blocks with priority level 1 correspond to high-priority mutation points, and blocks with priority level 4 correspond to ordinary mutation points. High-priority data is converted and stored first. The generation of structured signal feature data follows the format conversion of the feature tag sequence according to the order specified by the dynamic storage priority. The timestamp, feature parameter name, mutation amplitude, and other fields of each tag point in the feature tag sequence are reorganized into a hierarchical structure in the structured signal feature data. The first layer is segmented by time, the second layer is grouped by feature type, and the third layer contains specific values. The delay configuration field of the dynamic storage priority affects the timing of the generation of structured signal feature data. The structured conversion of low-priority data can be delayed until the off-peak period for batch execution. The sparse tag fields in the feature tag sequence are converted into data integrity identifiers in the structured signal feature data, indicating whether the data point is the original data or representative data after downsampling. Structured signal feature data establishes a multi-dimensional index for feature-labeled sequences, supporting data retrieval by time range, feature type, and abrupt change magnitude, among other methods. The priority configuration of each storage block in dynamic storage priority is reflected in the structured signal feature data as storage location tags, indicating whether the data is stored in a high-speed or conventional area. Once the structured signal feature data is generated, a complete signal feature archive is formed, containing all information about the feature-labeled sequences and a description of the storage layout guided by dynamic storage priority.

[0069] To implement the circuit signal data processing method corresponding to the above method embodiments, in order to achieve the corresponding functions and technical effects. See also Figure 2 , Figure 2 This paper shows a structural block diagram of a circuit signal data processing system 200 provided in an embodiment of the present application, including: The data acquisition module 201 is used to acquire circuit signal data of industrial equipment sensors, perform noise spectrum analysis on the circuit signal data to identify interference mode types, and extract noise compensation coefficients based on the interference mode types to generate a signal gain correction table. The filtering module 202 is used to perform gain correction on the circuit signal data according to the signal gain correction table, identify high-frequency interference segments, group and sort the high-frequency interference segments according to interference intensity to generate an interference classification table, and construct an adaptive filtering configuration through the interference classification table to extract the characteristics of the filtered signal. Feature extraction module 203 is used to perform transient pulse detection on the filtered signal features to obtain transient pulse response, identify abnormal peak segments from the transient pulse response to generate peak weight coefficients, and use the peak weight coefficients to perform feature enhancement processing on the filtered signal features to generate a key feature set. Data encoding module 204 is used to identify signal mutation points using the key feature set, sparsely sample the signal mutation points according to the time-series distribution to generate a feature label sequence, and construct a segmented storage strategy based on the feature label sequence; The data storage module 205 is used to monitor the storage load based on the sharded storage strategy to obtain the load fluctuation status, generate a dynamic storage priority using the load fluctuation status, and convert the feature tag sequence into structured signal feature data according to the dynamic storage priority.

[0070] The circuit signal data processing system 200 described above can implement one of the circuit signal data processing methods of the above method embodiments. The options in the above method embodiments are also applicable to this embodiment, and will not be detailed here. The remaining content of this application embodiment can be referred to the content of the above method embodiments, and will not be repeated in this embodiment.

[0071] The above embodiments are not an exhaustive list based on the present invention, and there may be many other embodiments not listed. Any substitutions and improvements made without departing from the concept of the present invention are within the protection scope of the present invention.

Claims

1. A method for processing circuit signal data, characterized in that, include: Collect circuit signal data from industrial equipment sensors, perform noise spectrum analysis on the circuit signal data to identify interference mode types, and extract noise compensation coefficients based on the interference mode types to generate a signal gain correction table; The circuit signal data is gain-corrected and high-frequency interference segments are identified based on the signal gain correction table. These high-frequency interference segments are then grouped and sorted according to interference intensity to generate an interference classification table. This process includes: decomposing the high-frequency interference segments into strong interference components and weak interference components; identifying interference source characteristics from the weak interference components to generate source tracing parameters; using the source tracing parameters in conjunction with the spectral characteristics of the strong interference components to trace the source and establish interference correlation paths; generating an interference classification table based on the hierarchical depth of the interference correlation paths; and constructing an adaptive filtering configuration using the interference classification table to extract the filtered signal features. Transient pulse detection is performed on the filtered signal features to obtain transient pulse response. Abnormal peak segments are identified from the transient pulse response to generate peak weight coefficients. The peak weight coefficients are then used to perform feature enhancement processing on the filtered signal features to generate a key feature set. The key feature set is used to identify signal mutation points, and the signal mutation points are sparsely sampled according to the temporal distribution to generate a feature label sequence. A segmented storage strategy is constructed based on the feature label sequence. Based on the sharded storage strategy, storage load monitoring is performed to obtain load fluctuation status. The load fluctuation status is used to generate dynamic storage priority. Based on the dynamic storage priority, the feature tag sequence is converted into structured signal feature data.

2. The method according to claim 1, characterized in that, The step of performing noise spectrum analysis on the circuit signal data to identify interference mode types includes: A frequency domain amplitude distribution diagram is constructed using the circuit signal data; The frequency domain amplitude distribution map is segmented into peaks and valleys to generate low-noise and high-noise frequency bands. Compensation parameters are formed by frequency domain compensation based on the low-noise frequency band and the high-noise frequency band. Interference mode types are formed based on the frequency difference characteristics of the compensation parameters.

3. The method according to claim 1, characterized in that, The step of identifying abnormal peak segments from the transient impulse response and generating peak weighting coefficients includes: The transient impulse response is subjected to amplitude mutation identification to generate a mutation distribution region; From the mutation distribution area, segments exceeding the threshold are extracted to generate abnormal spike segments; Based on the abnormal peak segments, peak attenuation analysis is performed to generate attenuation rate parameters; The attenuation rate parameter is used to perform a weight transformation to generate peak weight coefficients.

4. The method according to claim 1, characterized in that, The step of sparsely sampling the signal abrupt change points according to their temporal distribution to generate a feature marker sequence includes: A time-density statistical evaluation is performed on the signal abrupt change points to generate a time-density distribution map; Dense period identifiers are generated by identifying abrupt and dense periodic changes from the temporal density distribution map. Perform downsampling processing on the dense time period identifiers to obtain a sparsity parameter set; The feature label sequence is generated by sparse reconstruction based on the sparsification parameter set.

5. The method according to claim 1, characterized in that, The step of constructing a sharding storage strategy based on the feature tag sequence includes: Data density localization and identification are performed on the feature marker sequence to generate data dense regions; For the data-intensive areas, a dispersion assessment is performed to generate distributed storage metrics; The data-intensive area is fragmented using the aforementioned distributed storage metrics to generate distributed storage areas; The distributed storage area is used to implement a cross-region allocation strategy to generate fragmented storage.

6. The method according to claim 1, characterized in that, The process of generating dynamic storage priorities using the load fluctuation state includes: Convert the load fluctuation state into a resource utilization distribution; High-load periods are identified from the resource utilization distribution to generate load peak markers; Perform load balancing analysis on the load peak markers to generate latency tolerance parameters; Dynamic storage priorities are generated by prioritizing based on the latency tolerance parameter.

7. The method according to claim 1, characterized in that, The step of identifying interference source features from the weak interference components to generate source tracing parameters includes: Obtain the key points of frequency stability in the weak interference components; At the key points of frequency stability, signal delay characteristics are detected to generate delay tracking parameters; The aforementioned delay tracking parameters are used to perform reverse localization to generate a candidate set of sources; Source tracing parameters are generated based on the source candidate set and the delay tracing parameters.

8. The method according to claim 4, characterized in that, The step of performing time-series density statistical evaluation on the signal abrupt change points to generate a time-series density distribution map includes: An initial weight sequence is generated by performing time-series weight allocation on the signal abrupt change points; The decay weight coefficients are formed based on the time decay characteristics of the initial weight sequence; The attenuation weighting coefficient is used to perform weighted statistics on the signal abrupt points to generate density data; The density data is arranged in time sequence to generate a time-series density distribution map.

9. A circuit signal data processing system, characterized in that, include: The data acquisition module is used to acquire circuit signal data from industrial equipment sensors, perform noise spectrum analysis on the circuit signal data to identify interference mode types, and extract noise compensation coefficients based on the interference mode types to generate a signal gain correction table. The filtering module is used to perform gain correction on the circuit signal data according to the signal gain correction table, identify high-frequency interference segments, and group and sort the high-frequency interference segments according to interference intensity to generate an interference classification table. This includes: decomposing the high-frequency interference segments into strong interference components and weak interference components; identifying interference source characteristics from the weak interference components to generate source tracing parameters; using the source tracing parameters in combination with the spectral characteristics of the strong interference components to trace the source and form an interference association path; generating an interference classification table based on the hierarchical depth of the interference association path; and constructing an adaptive filtering configuration using the interference classification table to extract the filtered signal features. The feature extraction module is used to perform transient pulse detection on the filtered signal features to obtain transient pulse response, identify abnormal peak segments from the transient pulse response to generate peak weight coefficients, and use the peak weight coefficients to perform feature enhancement processing on the filtered signal features to generate a key feature set. The data encoding module is used to identify signal mutation points using the key feature set, sparsely sample the signal mutation points according to the time-series distribution to generate a feature label sequence, and construct a segmented storage strategy based on the feature label sequence. The data storage module is used to monitor the storage load based on the sharded storage strategy to obtain the load fluctuation status, generate a dynamic storage priority using the load fluctuation status, and convert the feature tag sequence into structured signal feature data according to the dynamic storage priority.

Images