Detection data processing method and system of helium leak detector for diaphragm compressor

Abstract

The invention relates to the technical field of electronic data processing, and discloses a detection data processing method and system for a helium leak detector for a diaphragm compressor, and the method comprises the steps: obtaining an original helium concentration time sequence signal and a periodic frequency parameter of the stroke of the diaphragm compressor; constructing a notch filter bank, and filtering the original helium concentration time sequence signal to obtain a preliminary signal; performing wavelet transform decomposition on the initial signal, performing threshold processing on a wavelet coefficient, and then performing reconstruction to obtain a de-noised signal; iterative fitting is carried out on the de-noised signal, a dynamic background baseline is estimated, and a background correction signal is obtained; calculating a first-order time derivative sequence, a local information entropy sequence and morphological correlation; and when the amplitude of the first-order time derivative sequence is continuously higher than a first threshold value, the local information entropy sequence is lower than a second threshold value, and the morphological correlation degree is higher than a third threshold value, judging that leakage occurs. According to the invention, multi-dimensional and high-confidence identification can be carried out on the leakage event, and the detection sensitivity, the anti-interference capability and the reliability of tiny leakage are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of electronic data processing, and in particular to a method and system for processing detection data of a helium leak detector for a diaphragm compressor. Background Art

[0002] Due to their structural characteristics, diaphragm compressors offer unique advantages in transporting high-purity, flammable, explosive, or toxic gases. They are used in the chemical, nuclear, semiconductor, food, and pharmaceutical industries. To ensure operational safety and media purity, high-precision airtightness testing of the compressor and its components is crucial. Helium mass spectrometry is currently recognized as one of the most sensitive industrial leak detection techniques. Using helium as a tracer gas, a helium leak detector detects trace amounts of helium escaping or infiltrating, thereby determining the presence of a leak. However, when performing helium leak detection on diaphragm compressors online or under simulated operating conditions, the raw helium concentration signal collected by the leak detector often has an extremely low signal-to-noise ratio. This is primarily due to the signal not only containing random environmental noise and the instrument's own thermal noise, but also, more seriously, the cyclical reciprocating motion of the compressor's crankshaft-connecting rod mechanism generates strong periodic interference, whose frequency is highly correlated with the compressor's stroke frequency and its harmonic components. This interference often overwhelms the slowly varying helium concentration signal caused by actual, small leaks, posing a significant challenge to accurate leak identification.

[0003] To extract weak leakage signals from a strong interference background, existing data processing methods typically employ simple filtering or thresholding. For example, moving average or low-pass filters are used to smooth the signal to suppress high-frequency random noise. However, while these linear filters suppress noise, they also tend to blur the edges of the actual leakage signal, reducing the instantaneous detection response and making it difficult to effectively remove periodic interference at specific frequencies directly related to the compressor's operating status. Another approach is to set a fixed concentration threshold, declaring a leak when the signal exceeds it. However, due to factors such as ambient temperature fluctuations and sensor aging, the background signal (baseline) of a helium leak detector slowly drifts. Fixed threshold strategies cannot adapt to this dynamically changing baseline, easily leading to false positives or missed negatives. Furthermore, some transient disturbances that are not leaks can cause brief, sudden changes in the signal, making it difficult to distinguish them from true exponentially increasing leakage patterns based solely on amplitude. Consequently, existing technologies for processing leak detection signals from dynamic equipment such as diaphragm compressors suffer from poor robustness, low sensitivity, and insufficient automated identification capabilities, making them difficult to meet the requirements of high-reliability continuous monitoring. Summary of the Invention

[0004] The present invention provides a method and system for processing detection data of a helium leak detector for a diaphragm compressor to solve the problems of poor robustness, low sensitivity and insufficient automatic recognition capability in the above-mentioned prior art, which are difficult to meet the requirements of high-reliability continuous monitoring.

[0005] In a first aspect, a method for processing detection data of a helium leak detector for a diaphragm compressor of the present invention comprises the following steps: The method obtains a raw helium concentration time series signal collected by a helium leak detector at a preset sampling frequency, and simultaneously obtains the periodic frequency parameters of the diaphragm compressor stroke. Based on the periodic frequency parameters, a notch filter bank is constructed for the fundamental frequency and harmonic frequencies of the periodic frequency parameters to filter the raw helium concentration time series signal to obtain a preliminary signal. The preliminary signal is decomposed by wavelet transform, and the wavelet coefficient thresholds of each decomposition scale layer are calculated. The wavelet coefficients are thresholded and reconstructed to obtain a denoised signal. The denoised signal is iteratively fitted to estimate the dynamic background baseline of the denoised signal, and the dynamic background baseline is subtracted from the denoised signal to obtain a background correction signal. Based on the background correction signal, its first-order time derivative sequence and local information entropy sequence are respectively calculated within a sliding time window. The background correction signal is simultaneously subjected to a normalized cross-correlation operation with a template signal of a preset typical leakage exponential rise model to obtain a morphological correlation. A leak is determined to have occurred when the amplitude of the first-order time derivative sequence is continuously higher than a first threshold, the local information entropy sequence is lower than a second threshold, and the morphological correlation is higher than a third threshold within a preset time.

[0006] Preferably, the construction of a notch filter group for the fundamental frequency and harmonic frequencies of the periodic frequency parameter includes: taking the periodic frequency parameter as the fundamental frequency, and selecting the fundamental frequency and multiple harmonic frequencies of its preset order as notch center frequencies; for each notch center frequency, designing a digital notch filter with a preset quality factor; and cascading the digital notch filters to form the notch filter group.

[0007] Preferably, the preliminary signal is subjected to wavelet transform decomposition, and the wavelet coefficient thresholds of each decomposition scale layer are calculated, and the wavelet coefficients are threshold processed and then reconstructed to obtain the denoised signal, including: selecting a predetermined wavelet basis function, performing multi-layer discrete wavelet transform decomposition on the preliminary signal to obtain detail coefficients of each layer and approximate coefficients of the highest layer; based on the Stein unbiased risk estimation criterion, independently calculating the denoising threshold of the detail coefficients of each layer; using a preset threshold function to process the detail coefficients; and performing inverse wavelet transform on the processed detail coefficients and the unprocessed approximate coefficients of the highest layer to reconstruct the denoised signal.

[0008] Preferably, the iterative fitting of the denoised signal to estimate the dynamic background baseline of the denoised signal includes: iteratively fitting the denoised signal using an asymmetric least squares smoothing algorithm, setting a smoothness parameter and an asymmetry parameter for the asymmetric least squares smoothing algorithm; fitting the denoised signal in an iterative weighted manner, stopping the iteration when the baselines calculated from two adjacent iterations meet a preset convergence criterion, and using the final fitting result as the dynamic background baseline.

[0009] Preferably, the background correction signal is based on the first-order time derivative sequence and the local information entropy sequence thereof, which are respectively calculated within the sliding time window, and the background correction signal is subjected to normalized cross-correlation operation with a template signal of a preset typical leakage exponential rising model to obtain the morphological correlation, including: within the sliding time window, using a numerical differentiation method to calculate the first-order time derivative sequence of the background correction signal; quantizing the background correction signal within the sliding time window, and calculating its information entropy sequence as the local information entropy sequence; and performing normalized cross-correlation operation on the background correction signal within the sliding time window and a preset template signal representing a typical leakage feature to obtain the morphological correlation.

[0010] Preferably, the determination of leakage includes: when the amplitude of the first-order time derivative sequence is continuously higher than a first preset threshold value for a preset duration, and within the same duration, the local information entropy sequence is lower than a second preset threshold value, and the morphological correlation is higher than a third preset threshold value, it is determined that leakage has occurred.

[0011] Preferably, the digital notch filter is a second-order IIR digital notch filter.

[0012] Preferably, the selecting a predetermined wavelet basis function to perform multi-layer discrete wavelet transform decomposition on the preliminary signal includes: selecting Daubechies 4 wavelet basis to perform 5-layer discrete wavelet transform decomposition on the preliminary signal.

[0013] Preferably, the information entropy is Shannon information entropy.

[0014] In a second aspect, the helium leak detector data processing system for a diaphragm compressor of the present invention includes a memory and a processor. The memory stores computer instructions. When the processor executes the computer instructions, the above-mentioned helium leak detector data processing method for a diaphragm compressor is implemented.

[0015] The beneficial effects of the present invention are as follows: the present invention can accurately eliminate the strong periodic interference generated by the mechanical movement of the equipment itself by specially constructing a notch filter group for the compressor stroke frequency and its harmonics, providing a pure signal basis for subsequent processing. On this basis, combined with multi-scale wavelet transform and threshold processing based on specific criteria, it can effectively filter out random noise while retaining the edge details of the real leakage signal to the maximum extent, avoiding the blurring of signal characteristics. Furthermore, an asymmetric least squares smoothing algorithm is used to estimate and deduct the background baseline of the signal, which effectively overcomes the baseline drift problem caused by environmental changes or instrument aging, and avoids the misjudgment caused by the fixed threshold method. By comprehensively examining the rate of change of the first-order derivative of the signal, the signal complexity reflected by the local information entropy, and the high morphological correlation with the typical leakage model for joint judgment, multi-dimensional and high-confidence recognition of real leakage events is achieved, which significantly improves the detection sensitivity, anti-interference ability and accuracy and reliability of small leaks. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A schematic flow chart of a method for processing detection data of a helium leak detector for a diaphragm compressor provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0017] The embodiments of the present invention are described in detail below, and examples of the embodiments are shown in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and intended to be used to explain the present invention, but should not be understood as limiting the present invention.

[0018] like Figure 1 As shown, an embodiment of the method for processing detection data of a helium leak detector for a diaphragm compressor provided by the present invention specifically includes the following steps: S1. Obtain the original helium concentration time series signal collected by the helium leak detector at a preset sampling frequency, and synchronously obtain the periodic frequency parameters of the diaphragm compressor stroke.

[0019] Specifically, the helium concentration data is collected at a sampling frequency of 100 Hz by the built-in data acquisition module of the helium leak detector to form the original helium concentration time series signal. At the same time, the operating speed of the diaphragm compressor is obtained by the speed sensor installed on the crankshaft, for example, 300 revolutions per minute, and the periodic frequency parameter f0 of the stroke is converted to 5 Hz.

[0020] S2. Based on the periodic frequency parameter, directly construct a notch filter bank for the fundamental frequency and at least the first five harmonic frequencies of the periodic frequency parameter, and filter the original helium concentration time series signal to suppress periodic interference and obtain a preliminary signal.

[0021] Specifically, using the acquired periodic frequency parameter f0 as the fundamental frequency, the interference frequencies to be suppressed are determined to be f0, 2f0, 3f0, 4f0, 5f0, and 6f0. For each interference frequency, a second-order IIR digital notch filter is designed. These individual notch filters are cascaded to form a notch filter bank. The raw helium concentration time series signal is sequentially passed through the notch filter bank to obtain a preliminary signal.

[0022] S3. Decompose the preliminary signal by multi-scale discrete wavelet transform, calculate the wavelet coefficient threshold of each decomposition scale layer based on Stein's unbiased risk estimation criterion, perform threshold processing on the wavelet coefficients and reconstruct them to obtain a denoised signal.

[0023] Specifically, the preliminary signal is decomposed into five layers of discrete wavelet transforms using the Daubechies 4 wavelet basis, yielding high-frequency detail coefficients and low-frequency approximation coefficients at each layer. For each layer of high-frequency detail coefficients, an optimal data-driven threshold is calculated using the SURE criterion. A soft threshold function is then used to shrink the high-frequency coefficients, while preserving the low-frequency approximation coefficients. The signal is then reconstructed using an inverse discrete wavelet transform to obtain the denoised signal.

[0024] S4. Iteratively fitting the denoised signal using an asymmetric least squares smoothing algorithm to estimate the dynamic background baseline of the denoised signal, and subtracting the dynamic background baseline from the denoised signal to obtain a background correction signal.

[0025] Specifically, an asymmetric least squares smoothing algorithm is employed, with an asymmetry factor p (ranging from 0.001 to 0.1) and a smoothness factor lambda. During the iteration process, signal points above the current fitted baseline are assigned smaller weights, while those below the baseline are assigned larger weights. Through multiple iterations, a weighted penalized least squares problem is solved to obtain a dynamic background baseline that closely matches the signal base. The background-corrected signal is then subtracted from the dynamic background baseline from the denoised signal.

[0026] S5. Based on the background correction signal, its first-order time derivative sequence and local information entropy sequence are calculated respectively within the sliding time window, and at the same time, the background correction signal is subjected to normalized cross-correlation operation with the template signal of a preset typical leakage exponential rise model to obtain morphological correlation.

[0027] Specifically, a sliding time window of length N is set, illustratively, N is equal to 50 data points. Within each window, the first-order derivative of the background correction signal is calculated using the difference method. At the same time, the amplitude of the signal within the window is histogram-generated, its probability distribution is calculated, and the local information entropy is calculated according to the Shannon entropy formula. In addition, an exponentially rising function of the form A multiplied by 1 minus e minus t divided by τ is pre-generated as a template signal, and a standardized cross-correlation calculation is performed with the background correction signal within the window to obtain a morphological correlation value ranging from -1 to 1.

[0028] S6. When the amplitude of the first-order time derivative sequence is continuously higher than a first threshold within a preset time, the local information entropy sequence is lower than a second threshold, and the morphological correlation is higher than a third threshold, it is determined that leakage occurs.

[0029] Specifically, the first threshold is set to 0.1, the second threshold is set to 2.5, and the third threshold is set to 0.85. The preset time is set to 0.5 seconds. During the data processing process, the three indicators are continuously monitored. If the amplitude of the first-order time derivative series is greater than 0.1 for 0.5 seconds, and the value of the local information entropy series is less than 2.5 during the same time period, and the value of the morphological correlation is also greater than 0.85, then a leak alarm is triggered.

[0030] In an optional embodiment, a notch filter group is directly constructed for the fundamental frequency and at least the first five harmonic frequencies of the periodic frequency parameter, including: taking the periodic frequency parameter as the fundamental frequency, and selecting the fundamental frequency and multiple harmonic frequencies of its preset order as the notch center frequencies; for each notch center frequency, designing a digital notch filter with a preset quality factor; and cascading the digital notch filters to form the notch filter group.

[0031] Specifically, in industrial environments, pipeline vibration signals are often severely affected by 50 Hz power frequency interference and its harmonics. Therefore, this embodiment selects 50 Hz as the periodic frequency parameter, i.e., the fundamental frequency. In order to completely eliminate this interference, the fundamental frequency and its second, third, fourth, and fifth harmonics are filtered out, that is, the notch center frequencies are set to 50 Hz, 100 Hz, 150 Hz, 200 Hz, and 250 Hz. The main power frequency interference bands can be fully covered. For each notch center frequency determined above, a second-order IIR digital notch filter is designed. In order to accurately filter out the interference frequency while retaining the effective signals in the adjacent frequency bands to the maximum extent, the quality factor Q value of the filter is set to a higher value, such as 35. Subsequently, these five digital notch filters designed for 50 Hz, 100 Hz, 150 Hz, 200 Hz, and 250 Hz, respectively, are cascaded. After the signal passes through the notch filter bank, the power frequency and its harmonic components are effectively suppressed, thus obtaining a pure preliminary signal.

[0032] In an optional embodiment, the preliminary signal is decomposed by multi-scale discrete wavelet transform, and the wavelet coefficient threshold of each decomposition scale layer is calculated based on the Stein unbiased risk estimation criterion, and the wavelet coefficient is threshold-processed and then reconstructed to obtain a denoised signal, including: selecting a predetermined wavelet basis function, performing multi-layer discrete wavelet transform decomposition on the preliminary signal to obtain detail coefficients of each layer and approximate coefficients of the highest layer; based on the Stein unbiased risk estimation criterion, independently calculating the denoising threshold of the detail coefficient of each layer; using a preset threshold function to process the detail coefficient; and performing an inverse wavelet transform on the processed detail coefficient and the unprocessed approximate coefficient of the highest layer to reconstruct the denoised signal.

[0033] Specifically, to effectively separate noise from the preliminary signal, this embodiment uses the Daubechies 5 wavelet basis function, or db5 wavelet, which has excellent time-frequency localization properties. The notch-filtered preliminary signal undergoes a five-layer discrete wavelet transform decomposition. This decomposition produces five layers of detail coefficients, each representing signal detail at different frequency scales, and a top-level approximation coefficient representing the signal's low-frequency profile. For example, the first layer of detail coefficients corresponds to the highest frequency portion of the signal, while the fifth layer of approximation coefficients corresponds to the signal's smoothest components.

[0034] Specifically, after decomposition, rather than using a unified threshold, the Stein unbiased risk estimation criterion is applied to independently calculate a denoising threshold (also known as a wavelet coefficient threshold) for the detail coefficients at each decomposition scale. For example, the threshold for the detail coefficients at the first level is calculated to be 0.8, the threshold for the second level is 0.5, and so on. Subsequently, a soft threshold function is applied to the detail coefficients at each level. This function sets coefficients below the threshold to zero and shrinks coefficients above the threshold toward zero, resulting in a smoother reconstructed signal than a hard threshold function. Finally, the thresholded detail coefficients of the five levels are combined with the unprocessed approximate coefficients of the highest level to reconstruct the signal using an inverse wavelet transform, resulting in a denoised signal that effectively removes random noise.

[0035] In an optional embodiment, an asymmetric least squares smoothing algorithm is used to iteratively fit the denoised signal to estimate the dynamic background baseline of the denoised signal, including: setting a smoothness parameter and an asymmetry parameter for the asymmetric least squares smoothing algorithm; fitting the denoised signal by an iterative weighted manner, when the baselines calculated by two adjacent iterations meet a preset convergence criterion, stopping the iteration, and using the final fitting result as the dynamic background baseline.

[0036] Specifically, to accurately extract the slowly changing dynamic background baseline in the signal while ignoring upward pulses caused by leakage, this embodiment sets two key parameters for the asymmetric least squares smoothing algorithm. The smoothness parameter lambda is set to 10 to the power of 6. A large value ensures that the fitted baseline is sufficiently smooth. The asymmetry parameter p is set to 0.01. A value much smaller than 0.5 causes the asymmetric least squares smoothing algorithm to weight signal points above the current baseline much less than signal points below the baseline during fitting, resulting in the fitting result tending to be closer to the signal's lower envelope.

[0037] Baseline fitting is performed iteratively. In each iteration, the weights are dynamically adjusted using the asymmetry parameter p based on the relative position of the current data point to the previous baseline, and the weighted least squares fit curve is recalculated. This process is repeated until the root mean square error between the baselines calculated in two consecutive iterations is less than a preset convergence criterion, such as less than 10 to the power of -6. When this criterion is met, the iterations cease, and the resulting final fitting curve is considered a dynamic background baseline that accurately reflects the changing operating conditions.

[0038] In an optional embodiment, based on the background correction signal, its first-order time derivative sequence and local information entropy sequence are respectively calculated within a sliding time window, and the background correction signal is subjected to normalized cross-correlation operation with a template signal of a preset typical leakage exponential rising model to obtain morphological correlation, including: within the sliding time window, using a numerical differentiation method to calculate the first-order time derivative sequence of the background correction signal; quantizing the background correction signal within the sliding time window, and calculating its information entropy sequence as the local information entropy sequence; and performing normalized cross-correlation operation on the background correction signal within the sliding time window and a preset template signal representing a typical leakage feature to obtain the morphological correlation.

[0039] Specifically, a sliding time window of 2 seconds is used to analyze the background-corrected signal. Within each time window, the central difference method is used to calculate the first-order time derivative of the signal. The magnitude of the derivative reflects the rate of change of the signal amplitude, and the derivative value increases sharply when a leak occurs. Simultaneously, the signal amplitude range within the sliding time window is evenly divided into 64 quantization levels, and the Shannon information entropy of the signal is calculated based on this. Under normal operating conditions, the signal is highly random and has a high information entropy. However, when a leak occurs, the signal morphology becomes more regular, and the information entropy decreases significantly.

[0040] To assess the similarity between the background correction signal's morphology and the leak signature, a template signal representing a typical exponentially rising leak model is pre-constructed. This template signal can be obtained by processing monitoring data from a known, typical, minor leak event. Within a sliding time window, a normalized cross-correlation is performed between the background correction signal within that window and the preset template signal. The result, known as the morphological correlation, ranges between -1 and 1. A value close to 1, such as 0.9, indicates that the waveform of the current background correction signal is highly consistent with that of the typical leak signature.

[0041] In an optional embodiment, determining whether leakage occurs includes: when the amplitude of the first-order time derivative sequence is continuously higher than a first preset threshold value for a preset duration, and within the same duration, the local information entropy sequence is lower than a second preset threshold value, and the morphological correlation is higher than a third preset threshold value, determining that leakage occurs.

[0042] Specifically, in order to ensure the accuracy and reliability of leakage judgment, this embodiment sets three independent criteria and a duration condition. The first preset threshold is for the first-order time derivative, set to 5.0 units per second, which is used to capture the sudden change of the signal. The second preset threshold is for the local information entropy, set to 3.5, which is used to identify the transition of the signal from random to regular. The third preset threshold is for the morphological correlation, set to 0.85, which is used to confirm the matching degree between the signal waveform and the leakage feature. These thresholds are obtained through statistical analysis of a large amount of historical normal operation data and simulated leakage data. Only when the above three conditions are met at the same time can the leakage alarm be triggered. This multi-feature fusion and duration constraint judgment strategy can effectively filter out false alarms caused by single impacts or short-term interferences. Only when the signal shows a continuous behavior that meets all the key characteristics of the leakage will it be finally determined that a leakage has occurred.

[0043] The implementation principle of the data processing method for the helium leak detector for diaphragm compressors of the embodiment of the present invention is as follows: by constructing a notch filter group for the compressor stroke frequency and its harmonics, it is possible to accurately eliminate the strong periodic interference generated by the mechanical movement of the equipment itself, providing a pure signal basis for subsequent processing. Moreover, by combining multi-scale wavelet transform and threshold processing, it is possible to effectively filter out random noise while retaining the edge details of the real leakage signal to the maximum extent, avoiding the blurring of signal features. In addition, the asymmetric least squares smoothing algorithm is used to estimate and deduct the background baseline of the signal, effectively overcoming the baseline drift problem caused by environmental changes or instrument aging, and avoiding the misjudgment caused by the fixed threshold method. Moreover, by comprehensively examining the rate of change of the first-order derivative of the signal, the signal complexity reflected by the local information entropy, and the high morphological correlation with the typical leakage model for joint judgment, it is possible to perform multi-dimensional and high-confidence identification of real leakage events, significantly improving the detection sensitivity, anti-interference ability, and accuracy and reliability of small leaks.

[0044] An embodiment of a helium leak detector data processing system for a diaphragm compressor provided by the present invention includes a memory and a processor. The memory stores computer instructions. When the processor executes the computer instructions, the helium leak detector data processing method for a diaphragm compressor provided in the above embodiment is implemented.

[0045] Although the embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and are not to be construed as limitations on the present invention. A person skilled in the art may change, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. A method for processing detection data of a helium leak detector for a diaphragm compressor, characterized in that: The method comprises the following steps: obtaining an original helium concentration time series signal collected by a helium leak detector at a preset sampling frequency, and synchronously obtaining a periodic frequency parameter of a diaphragm compressor stroke; constructing a notch filter group for the fundamental frequency and harmonic frequency of the periodic frequency parameter according to the periodic frequency parameter, filtering the original helium concentration time series signal to obtain a preliminary signal; performing wavelet transform decomposition on the preliminary signal, calculating the wavelet coefficient threshold of each decomposition scale layer, performing threshold processing on the wavelet coefficient, and then reconstructing the signal to obtain a denoised signal; performing iterative fitting on the denoised signal, estimating a dynamic background baseline of the denoised signal, and subtracting the dynamic background baseline from the denoised signal to obtain a background correction signal; Based on the background correction signal, its first-order time derivative sequence and local information entropy sequence are calculated respectively within the sliding time window. At the same time, the background correction signal is subjected to normalized cross-correlation operation with the template signal of a pre-set typical leakage exponential rising model to obtain the morphological correlation. When the amplitude of the first-order time derivative sequence is continuously higher than the first threshold within the preset time, the local information entropy sequence is lower than the second threshold, and the morphological correlation is higher than the third threshold, leakage is determined to have occurred.

2. The method for processing detection data of a helium leak detector for a diaphragm compressor according to claim 1, characterized in that: The method of constructing a notch filter group for the fundamental frequency and harmonic frequencies of the periodic frequency parameter includes: taking the periodic frequency parameter as the fundamental frequency, and selecting the fundamental frequency and multiple harmonic frequencies of the fundamental frequency and preset orders as notch center frequencies; designing a digital notch filter with a preset quality factor for each notch center frequency; and cascading the digital notch filters to form the notch filter group.

3. The method for processing detection data of a helium leak detector for a diaphragm compressor according to claim 1, characterized in that: The method of performing wavelet transform decomposition on the preliminary signal, calculating the wavelet coefficient threshold of each decomposition scale layer, and reconstructing the wavelet coefficient after threshold processing to obtain the denoised signal includes: selecting a predetermined wavelet basis function, performing multi-layer discrete wavelet transform decomposition on the preliminary signal to obtain detail coefficients of each layer and approximate coefficients of the highest layer; independently calculating the denoising threshold of the detail coefficients of each layer based on the Stein unbiased risk estimation criterion; processing the detail coefficients using a preset threshold function; and performing wavelet inverse transform on the processed detail coefficients and the unprocessed approximate coefficients of the highest layer to reconstruct the denoised signal.

4. The method for processing detection data of a helium leak detector for a diaphragm compressor according to claim 1, characterized in that: The denoised signal is iteratively fitted to estimate the dynamic background baseline of the denoised signal, including: iteratively fitting the denoised signal using an asymmetric least squares smoothing algorithm, setting a smoothness parameter and an asymmetry parameter for the asymmetric least squares smoothing algorithm; fitting the denoised signal in an iterative weighted manner, stopping the iteration when the baselines calculated from two adjacent iterations meet a preset convergence criterion, and using the final fitting result as the dynamic background baseline.

5. The method for processing detection data of a helium leak detector for a diaphragm compressor according to claim 1, characterized in that: Based on the background correction signal, its first-order time derivative sequence and local information entropy sequence are calculated respectively within the sliding time window, and the background correction signal is subjected to normalized cross-correlation operation with a template signal of a preset typical leakage exponential rising model to obtain morphological correlation, including: within the sliding time window, using a numerical differentiation method to calculate the first-order time derivative sequence of the background correction signal; quantizing the background correction signal within the sliding time window and calculating its information entropy sequence as the local information entropy sequence; and performing normalized cross-correlation operation on the background correction signal within the sliding time window and a preset template signal representing a typical leakage feature to obtain the morphological correlation.

6. The method for processing detection data of a helium leak detector for a diaphragm compressor according to claim 1, characterized in that: The determination of leakage includes: when the amplitude of the first-order time derivative sequence is continuously higher than a first preset threshold value within a preset duration, and within the same duration, the local information entropy sequence is lower than a second preset threshold value, and the morphological correlation is higher than a third preset threshold value, it is determined that leakage has occurred.

7. The method for processing detection data of a helium leak detector for a diaphragm compressor according to claim 2, characterized in that: The digital notch filter is a second-order IIR digital notch filter.

8. The method for processing detection data of a helium leak detector for a diaphragm compressor according to claim 3, characterized in that: The step of selecting a predetermined wavelet basis function and performing multi-layer discrete wavelet transform decomposition on the preliminary signal includes: selecting Daubechies 4 wavelet basis to perform 5-layer discrete wavelet transform decomposition on the preliminary signal.

9. The method for processing detection data of a helium leak detector for a diaphragm compressor according to claim 5, characterized in that: The information entropy is Shannon information entropy.

10. A helium leak detector data processing system for a diaphragm compressor, characterized in that: The method comprises a memory and a processor, wherein the memory stores computer instructions, and when the processor executes the computer instructions, the method for processing detection data of a helium leak detector for a diaphragm compressor according to any one of claims 1 to 9 is implemented.

Images