A multi-source heterogeneous signal fusion acquisition method based on timestamp alignment

Abstract

The application belongs to the technical field of signal processing and industrial monitoring, and particularly relates to a multi-source heterogeneous signal fusion acquisition method based on timestamp alignment, which comprises the following steps: S1, obtaining high-speed service signals and low-speed state signals of a system to be monitored, performing up-sampling processing on the low-speed state signals, aligning the time resolution of the low-speed state signals with the characteristic window step of the high-speed service signals, and obtaining a time-base normalized to-be-calibrated sequence; S2, calculating the local standard deviation of the high-speed service signals in a current calculation window and the first-order difference absolute value, combining a preset normalized amplification coefficient, and constructing a fluctuation energy density index reflecting the fluctuation characteristics of the signals. The application effectively prevents the misjudgment of Gaussian white noise of a sensor as a correlation signal, and can adapt to the response time drift caused by equipment aging, improve the accuracy of fault analysis from a second level to a millisecond level, and effectively solve the problems of heterogeneous signal causality inversion and fusion difficulty.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of signal processing and industrial monitoring, and particularly relates to a multi-source heterogeneous signal fusion acquisition method based on timestamp alignment. BACKGROUND

[0002] In large scientific devices such as heavy ion accelerators, synchrotron radiation sources and complex industrial scenes such as precision semiconductor manufacturing, the control system usually needs to process heterogeneous signals with different sources and greatly different sampling rates. Taking the low-level radio frequency control system of the accelerator as an example, the system mainly acquires two types of key data: one is high-speed business signals such as sampling signals of radio frequency cavities, which have extremely high sampling rates and are transmitted through high-speed buses, carrying high-precision hardware timestamps; the other is low-speed state signals such as cavity vacuum degree or cooling water temperature, which have relatively low sampling rates and are transmitted through slow control channels, usually only carrying software timestamps at the second level.

[0003] However, the prior art has significant defects in processing the above multi-source heterogeneous signals. Due to the thermal inertia of the sensor itself or the gas diffusion time, the physical response often has a lag of hundreds of milliseconds or even seconds, which in turn causes the phenomenon of cause and effect inversion. For example, when a radio frequency cavity has a spark failure, the high-speed signal drops instantaneously, while the vacuum deterioration data appears with a lag. The existing data alignment method mainly relies on absolute hardware timestamps, ignoring the existence of physical transmission lag, resulting in that the radio frequency data at the fault time is incorrectly associated with the seemingly normal vacuum data, and the real abnormal data is classified into irrelevant time periods, so that the system cannot accurately determine the fault cause.

[0004] In addition, high-speed signals focus on microsecond-level waveform distortion, while low-speed signals focus on second-level trend changes, and the feature dimensions of the two are completely split, making it difficult to directly construct a unified decision formula for heterogeneous data fusion. The existing monitoring system can usually only analyze each type of signal independently, lacks effective cooperative diagnosis mechanism, and is easily disturbed by noise of a single signal source, resulting in false positives or false negatives. Therefore, there is an urgent need for a method that does not rely on absolute hardware timestamps, but is based on signal features for adaptive alignment and fusion. SUMMARY

[0005] The application provides a multi-source heterogeneous signal fusion acquisition method based on timestamp alignment to solve the technical problems in the prior art that the timestamp alignment of heterogeneous signals fails due to physical response lag, it is difficult to accurately determine the fault cause, and heterogeneous data fusion is difficult.

[0006] In a first aspect, the application provides a multi-source heterogeneous signal fusion acquisition method based on timestamp alignment, comprising the following steps:

[0007] S1. Acquire the high-speed service signal and low-speed status signal of the system to be monitored. Upsample the low-speed status signal to align its time resolution with the feature window step size of the high-speed service signal to obtain the time-base normalized calibration sequence.

[0008] S2 calculates the local standard deviation and first-order difference absolute value of the high-speed service signal within the current calculation window, and constructs a fluctuation energy density index that reflects the fluctuation characteristics of the signal by combining it with the preset normalized amplification coefficient.

[0009] S3, based on the fluctuation energy density index and the sequence to be calibrated, constructs a weighted response coupling function containing a rate of change gating term, traverses the lag time variables within a preset search range, and marks the lag time that makes the weighted response coupling function reach its maximum value as the optimal physical lag time;

[0010] S4 utilizes the optimal physical lag time to perform time compensation on the sequence to be calibrated, and calculates the multi-source collaborative anomaly index in combination with the fluctuation energy density index. Based on the multi-source collaborative anomaly index, it realizes the fusion fault diagnosis of multi-source heterogeneous signals.

[0011] Furthermore, the high-speed service signals include amplitude monitoring signals of the accelerator radio frequency cavity, and the low-speed status signals include vacuum monitoring signals of the cavity.

[0012] Furthermore, the fluctuation energy density index The calculation formula is:

[0013] ;

[0014] In the formula, Indicates the first The amplitude of high-speed service signals at each sampling point; This represents the local mean of the high-speed service signal within the current calculation window; Indicates the length of the sliding window; It is a natural constant; This is the normalized amplification factor.

[0015] Furthermore, the weighted response coupling function The calculation formula is:

[0016] ;

[0017] In the formula, The lag time variable to be searched; This is the total length of the sample used to calculate the coupling degree; For the first The fluctuation energy density index at each point; Interpolated data for the low-speed state signal after time shifting; is a steady-state reference value of the low-speed state signal; is an attenuation base; is a rate of change sensitivity coefficient.

[0018] Further, the optimal physical lag time is determined as follows: within a preset search range , the value that maximizes is selected as .

[0019] Further, the calculation formula of the multi-source collaborative anomaly index is as follows:

[0020] ;

[0021] In the formula, is a basic weight coefficient of the high-speed service signal; is a weight coefficient of the collaborative effect; is an amplitude normalization factor; is an optimal physical lag time; is the low-speed state signal data after time compensation; is a steady-state reference value of the low-speed state signal.

[0022] Further, the specific process of the fusion fault diagnosis is as follows: when the multi-source collaborative anomaly index exceeds a preset threshold, it is determined that a physical correlation fault occurs in the system; when the multi-source collaborative anomaly index does not exceed the preset threshold but the fluctuation energy density index is high, it is determined that a non-physical fault occurs in the system or there is electrical noise.

[0023] Further, in S1, the low-speed state signal sequence is up-sampled by using a cubic spline interpolation method.

[0024] Further, in S3, the preset search range is 0-2000 ms.

[0025] Further, the normalization amplification coefficient is set to 1000.

[0026] ​The beneficial effects are as follows: The core innovation of this invention lies in proposing an adaptive soft alignment method based on signal fluctuation fingerprint characteristics. Compared with existing technologies, this method does not rely on absolute hardware timestamps. Instead, it automatically searches for and compensates for physical transmission lag time by calculating the coupling relationship between the energy dispersion of high-speed signals and the rate of change response of low-speed signals. In particular, it uniquely introduces a rate of change gating mechanism, effectively preventing the misjudgment of Gaussian white noise from sensors as correlated signals. This method can adapt to the response time drift caused by equipment aging, improving the accuracy of fault analysis from the second level to the millisecond level, and effectively solving the problems of causal inversion and fusion difficulties of heterogeneous signals. Attached Figure Description

[0027] Figure 1 This is a flowchart of a multi-source heterogeneous signal fusion acquisition method based on timestamp alignment.

[0028] Figure 2 This is a schematic diagram illustrating the time misalignment of multi-source signals under current technological conditions.

[0029] Figure 3 This is a schematic diagram illustrating the automatic timing alignment and fusion achieved after applying the method of this invention. Detailed Implementation

[0030] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0031] An embodiment of the multi-source heterogeneous signal fusion acquisition method based on timestamp alignment provided by this invention:

[0032] like Figure 1 As shown, the multi-source heterogeneous signal fusion acquisition method based on timestamp alignment includes the following steps:

[0033] S1: Acquire the high-speed service signal and low-speed status signal of the system to be monitored. Upsample the low-speed status signal to align its time resolution with the feature window step size of the high-speed service signal, and obtain the time-base normalized calibration sequence.

[0034] In this embodiment, the heterogeneous signal to be analyzed mainly originates from the accelerator's low-level radio frequency control system. The high-speed service signal sequence is defined as... For example, the amplitude monitoring signal of the radio frequency cavity, its sampling rate Set to 100MHz, define the low-speed state signal sequence as follows: For example, the vacuum monitoring signal of the cavity, its sampling rate Set to 10Hz. Because... Much larger To perform calculations on the same logical dimension, the temporal resolution of low-speed state signals needs to be improved. Specifically, cubic spline interpolation is used to... Upsampling is performed to logically align the sampling point density with the feature window step size of the high-speed service signal, resulting in the calibration sequence. For example, if the step size for feature extraction of high-speed service signals corresponds to 1ms, then the low-speed state signal will be interpolated to a frequency of 1kHz. It should be noted that at this time... Although the number of data points corresponds to high-speed service signals, there is still an unknown physical lag time in the timing of physical events. .

[0035] By normalizing the time base of the dataset, the obstacle of huge differences in sampling rates between different signal sources is eliminated, making subsequent feature fusion calculations possible.

[0036] S2 calculates the local standard deviation and first-order difference absolute value of the high-speed service signal within the current calculation window, and constructs a fluctuation energy density index that reflects the fluctuation characteristics of the signal by combining it with the preset normalized amplification coefficient.

[0037] In this embodiment, in order to extract fingerprint features characterizing the occurrence of faults from the originally stable high-speed radio frequency signal, a fluctuation energy density index is constructed. The calculation formula is as follows:

[0038]

[0039] In the formula, Indicates the first The amplitude of the high-speed service signal at each sampling point is directly acquired by the ADC. This represents the local mean of the high-speed service signal within the current calculation window, obtained by taking the first value. Before The arithmetic mean of the data points is used to eliminate the DC component. The length of the sliding window is represented, for example, 100 points, which is determined by the periodicity of the signal; the square root part of the first term is actually the calculation of the local standard deviation within the window. When sparking occurs, the energy dispersion increases instantaneously. It is the first-order absolute difference, used to capture instantaneous changes in a signal; natural constant , approximately equal to 2.718, is used to ensure that the minimum value of the logarithmic term is not less than 1; This is the normalization amplification factor, for example, 1000, used to map small difference values ​​to the sensitive region of the logarithmic function.

[0040] For ease of understanding, a specific calculation example is given: suppose the current window length , the local mean , the current sampling point , the previous sampling point .

[0041] First, calculate the local standard deviation part, assuming that the variance of the data in the window is about 5 after square root; then calculate the first-order difference part: ; the logarithmic term calculation: ; the final fluctuation energy density index: .

[0042] When the system is running normally, the signal is smooth, the standard deviation is close to 0, and the difference is also close to 0, at this time value is very small, while when the spark occurs, the index will increase significantly.

[0043] By constructing the fluctuation energy density index, the microsecond-level waveform distortion can be converted into a high signal-to-noise ratio characteristic value, providing a clear fingerprint basis for subsequent alignment.

[0044] S3, based on the fluctuation energy density index and the sequence to be calibrated, construct a weighted response coupling degree function containing a change rate gating term, and traverse the lag time variable within the preset search range, mark the lag time that makes the weighted response coupling degree function reach the maximum value as the best physical lag time.

[0045] In this embodiment, in order to realize adaptive lag search and prevent background noise interference, the weighted response coupling degree function is constructed, and the calculation formula is as follows:

[0046]

[0047] In the formula, is the lag time variable to be searched, and the search range , for example, is set to 0ms to 2000ms; is the total length of the sample; is the interpolated data of the low-speed state signal after translation; is the steady-state reference value of the low-speed signal, for example, the average value when the system is idle; is the absolute value of the deviation degree term; is the change rate gating term; is the decay base, for example, take 0.5; is the change rate sensitivity coefficient, used to adapt the signal dimension and prevent the index from approaching 1 due to too small value.

[0048] The logic of the change rate gating in this formula is: when the low-speed signal changes slowly, the exponential term The entire gating item becomes At this point, even if the radio frequency signal fluctuates, the coupling degree is not calculated, thus filtering out false correlations.

[0049] The calculation example is as follows: Suppose at a certain moment... Fluctuation energy density index .

[0050] Assuming attempt delay time The corresponding low-speed signal value at this time benchmark value The deviation is Furthermore, at this moment, the low-speed signal underwent a sudden change, with the change amount... .

[0051] set up , The index part: Gated items: Single-point coupling contribution value: .

[0052] If the low-speed signal does not change at this time, the gating term is 0, and the contribution value is 0. This is achieved by traversing all possible... Choose the option that makes the sum of the sums ... The largest As the optimal physical lag time .

[0053] By introducing an adaptive search with a rate-of-change gating mechanism, the interference of static drift of environmental parameters can be eliminated, and the inherent physical transmission lag time of the system can be accurately locked.

[0054] S4 utilizes the optimal physical lag time to perform time compensation on the sequence to be calibrated, and calculates the multi-source collaborative anomaly index in combination with the fluctuation energy density index. Based on the multi-source collaborative anomaly index, it realizes the fusion fault diagnosis of multi-source heterogeneous signals.

[0055] In this embodiment, the calculated Final time compensation is performed on the low-speed signal, and a multi-source collaborative anomaly index is constructed. :

[0056]

[0057] In the formula, This is the basic weighting coefficient for high-speed service signals, for example, 1; The synergy effect weighting coefficient, for example, 10, is used to highlight concurrent failures; This is the magnitude normalization factor, for example, 100, used to adjust the order of magnitude of the difference to fit the nonlinear interval of the logarithmic function; The best lag time is automatically searched.

[0058] The calculation example is as follows:

[0059] Case A, false firing: However, the low-speed signal after compensation is close to the reference value, and the deviation is 0.

[0060] The index is low.

[0061] Case B, true firing / true vacuum deterioration: The low-speed signal after compensation deviates from the reference value by 0.5.

[0062] .

[0063] It can be seen that the index of the real physical correlation fault is much higher than that of single interference, and the system can automatically determine the fault type according to the set threshold.

[0064] In order to intuitively show the technical effect of the present application, a group of comparison effect diagrams are obtained through system data processing:

[0065] Figure 2 The existing technology state is shown, and the time dislocation of the multi-source signal exists, from Figure 2 It can be clearly observed that at 3.0 seconds on the time axis, the fluctuation characteristics of the high-speed radio frequency signal have occurred a sharp peak representing a fault, however, due to the physical transmission lag, the environmental monitoring data does not begin to show a clear upward trend until about 3.8 seconds. This about 0.8 second physical transmission lag causes the two curves to be completely inconsistent at the moment of fault occurrence, and it is easy to draw a wrong conclusion.

[0066] Figure 3 The effect after applying the method of the present application is shown, and the fluctuation trend of the environmental data curve is pulled back to 3.0 seconds after automatic compensation of the best lag time, and the radio frequency signal fluctuation characteristics are perfectly logically aligned on the time axis. The filled area shows the multi-source collaborative anomaly index calculated by the present application, and only at the moment of fault occurrence, due to the high coincidence of the radio frequency fluctuation and the vacuum deterioration characteristics after alignment, the anomaly index instantaneously soars to the peak value, which intuitively proves that the present application can accurately lock the real moment of fault occurrence and its cause.

[0067] Through multi-source collaborative anomaly judgment, single-source noise and multi-source concurrent fault can be effectively distinguished, and the hidden physical cause and effect relationship is clearly presented, which significantly improves the accuracy of fault diagnosis.

[0068] The above are preferred embodiments of the present application, and do not limit the protection scope of the present application, so: any equivalent changes made according to the structure, shape, principle of the present application should be covered within the protection scope of the present application.

Claims

1. A method for fusion and acquisition of multi-source heterogeneous signals based on timestamp alignment, characterized in that, Includes the following steps: S1. Acquire the high-speed service signal and low-speed status signal of the system to be monitored. Upsample the low-speed status signal to align its time resolution with the feature window step size of the high-speed service signal to obtain the time-base normalized calibration sequence. S2 calculates the local standard deviation and first-order absolute difference of the high-speed service signal within the current calculation window. Combined with a preset normalized amplification factor, it constructs a fluctuation energy density index reflecting the signal fluctuation characteristics. satisfy: ； Indicates the first The amplitude of high-speed service signals at each sampling point; This represents the local mean of high-speed service signals within the current calculation window; Indicates the length of the sliding window; It is a natural constant; This is the normalized amplification factor; S3, based on the fluctuation energy density index and the sequence to be calibrated, constructs a weighted response coupling function containing a rate of change gating term, traverses the lag time variables within a preset search range, and marks the lag time that makes the weighted response coupling function reach its maximum value as the optimal physical lag time; Weighted response coupling function satisfy: ； The lag time variable to be searched; This is the total length of the sample used to calculate the coupling degree; For the first The fluctuation energy density index at each point; Interpolated data for the low-speed state signal after time shifting; This is the steady-state reference value for the low-speed signal; The attenuation base; This is the sensitivity coefficient for the rate of change; S4 utilizes the optimal physical lag time to perform time compensation on the sequence to be calibrated, and calculates the multi-source collaborative anomaly index in combination with the fluctuation energy density index. Based on the multi-source collaborative anomaly index, it realizes the fusion fault diagnosis of multi-source heterogeneous signals.

2. The multi-source heterogeneous signal fusion acquisition method based on timestamp alignment according to claim 1, characterized in that, High-speed service signals include amplitude monitoring signals of the accelerator radio frequency cavity, while low-speed status signals include vacuum monitoring signals of the cavity.

3. The multi-source heterogeneous signal fusion acquisition method based on timestamp alignment according to claim 1, characterized in that, Optimal physical lag time The method for determining it is: within the preset search range Inside, select to make The largest value As .

4. The multi-source heterogeneous signal fusion acquisition method based on timestamp alignment according to claim 1, characterized in that, Multi-source collaborative anomaly index The calculation formula is: ； In the formula, These are the basic weighting coefficients for high-speed service signals; These are the weighting coefficients for the synergistic effect; This is the amplitude normalization factor; The optimal physical lag time; This is the time-compensated low-speed state signal data; This is the steady-state reference value for the low-speed signal.

5. The multi-source heterogeneous signal fusion acquisition method based on timestamp alignment according to claim 4, characterized in that, The specific process of fusion fault diagnosis is as follows: when the multi-source collaborative anomaly index When the threshold is exceeded, a physical correlation fault is determined to have occurred in the system; when the multi-source collaborative anomaly index... The energy density index did not exceed the preset threshold but fluctuated. When the value is high, it is determined that the system has experienced a non-physical fault or that there is electrical noise.

6. The multi-source heterogeneous signal fusion acquisition method based on timestamp alignment according to claim 1, characterized in that, In S1, cubic spline interpolation is used to upsample the low-speed state signal sequence.

7. The multi-source heterogeneous signal fusion acquisition method based on timestamp alignment according to claim 1, characterized in that, In S3, the preset search range is 0-2000ms.

8. The multi-source heterogeneous signal fusion acquisition method based on timestamp alignment according to claim 1, characterized in that, Set the normalized amplification factor to 1000.

Images