A compressor whole-machine seal detection method based on pressure decay characteristics

Abstract

This invention relates to the field of sealing detection technology, and particularly to a method for detecting the overall sealing of a compressor based on pressure decay characteristics. The method includes: acquiring a time-series signal during the compressor's overall sealing detection process and preprocessing it to obtain a preprocessed pressure sequence; constructing a sliding window for the pressure sequence and calculating a trend direction index, extracting the corresponding time period as the decay monitoring stage signal; determining the adaptive leakage characteristic frequency boundary of the decay monitoring stage signal and performing variational mode decomposition under an initial mode decomposition number; statistically analyzing the number of effective modes whose low-frequency energy concentration meets set conditions, and adaptively iteratively adjusting the mode decomposition number until the number of effective modes converges to a target value to determine the optimal mode decomposition number; and calculating the total pressure drop of the leakage signal under the optimal mode decomposition number and determining the sealing status. This invention improves the accuracy of detecting abnormal sealing conditions in the entire compressor.

Description

Technical Field

[0001] This invention relates to the field of sealing testing technology, and in particular to a method for testing the sealing of a compressor based on pressure decay characteristics. Background Technology

[0002] Compressors are core power equipment in petrochemical, natural gas transmission, and refrigeration and air conditioning industries, and their overall sealing performance directly affects the system's operating efficiency and safety stability. Media leakage caused by seal failure not only results in energy loss but can also lead to safety accidents and environmental pollution in severe cases. The pressure decay method, which involves charging the system to its rated pressure and continuously monitoring the pressure change over time to determine the sealing status, is a non-contact, highly sensitive whole-system testing method. However, during compressor operation, the pressure decay signal is affected by multiple mechanical factors such as piston reciprocating motion, speed fluctuations, and valve opening and closing. This results in a complex, non-stationary pressure decay signal with multiple superimposed components. The original pressure curve alone is insufficient to accurately identify subtle leakage characteristics, necessitating a high-precision sealing detection method capable of adaptively analyzing complex pressure signals.

[0003] Chinese patent document CN117572300B discloses a method for detecting inter-turn short-circuit faults in motors based on variational mode decomposition fused with deep learning, including the following steps: establishing a fault model of a permanent magnet synchronous motor; extracting fault features using a variational mode decomposition algorithm; optimizing variational mode decomposition parameters using a sparrow search algorithm; establishing a deep pyramid pooling residual convolutional neural network model; and using SPP (Self-Protected Programming) technology. Training of ResCNN network; Real-time diagnosis of inter-turn short circuit faults in permanent magnet synchronous motors.

[0004] Variational mode decomposition (VMD) is an adaptive signal decomposition method based on a variational optimization framework proposed in recent years. It uses an alternating direction multiplier method for iterative solution to decompose a composite signal into a finite number of band-limited eigenmode functions. Compared with traditional empirical mode decomposition, it has a significant advantage in resisting mode aliasing, and each mode has a clear physical meaning and a highly interpretable mathematical framework, making it promising for applications in mechanical signal feature extraction. However, VMD has a key limitation in compressor seal detection: the pressure decay process in compressor seal detection exhibits significant stages, evolving continuously in terms of the number and frequency distribution of effective components from pressurization, pressure stabilization, to natural decay monitoring. During the pressurization stage, mechanical excitation is strong and signal components are numerous, while during the decay monitoring stage, low-frequency leakage characteristics gradually dominate and the number of components decreases significantly. The core hyperparameter of VMD is the mode decomposition number. It must be pre-set before the algorithm executes, and cannot perceive the dynamic changes in the signal component structure between the above stages. Fixed The value mechanism leads to frequent under-decomposition of the decomposition results when the number of signal components evolves with the stage. Effective leakage features are merged into the same mode and cannot be extracted independently, or over-decomposition occurs, generating spurious modes without physical meaning and introducing noise interference, which seriously restricts the accuracy of leakage feature extraction and the reliability of seal detection. Summary of the Invention

[0005] To address the technical problem of difficulty in accurately extracting weak leakage characteristics when processing complex non-stationary pressure decay signals of compressors, where fixed signal decomposition parameters lead to under- or over-decomposition of modes, this invention provides a compressor whole-machine seal detection method based on pressure decay characteristics.

[0006] This invention provides a method for detecting the overall seal of a compressor based on pressure decay characteristics, employing the following technical solution: A compressor whole-unit seal testing method based on pressure decay characteristics includes the following steps: S1: Acquire the timing signal during the compressor's overall sealing test process and preprocess it to obtain the preprocessed pressure sequence; S2: Construct a sliding window for the pressure sequence and calculate the trend direction index. When the trend direction index of consecutive adjacent windows meets the preset threshold condition, extract the corresponding time period as the attenuation monitoring stage signal. S3: Determine the adaptive leakage characteristic frequency boundary of the signal during the attenuation monitoring stage, and perform variational mode decomposition under the initial mode decomposition number to calculate the low-frequency energy concentration of each mode component under the constraint of the adaptive leakage characteristic frequency boundary. S4: Count the number of effective modes that meet the set conditions for low-frequency energy concentration, and adaptively iterate and adjust the number of mode decompositions based on the number of effective modes until the number of effective modes converges to the target value, so as to determine the optimal number of mode decompositions; S5: Under the optimal modal decomposition number, select the modal component with the highest low-frequency energy concentration to reconstruct the leakage signal, calculate the total pressure drop of the leakage signal, and determine the sealing status.

[0007] The technical advantages of this invention are as follows: Addressing the problem that existing variational mode decomposition algorithms, which require manual fixing of the mode decomposition number, are prone to over- or under-decomposition under complex pressure signals, this invention accurately identifies the attenuation monitoring stage by constructing a trend direction index, avoiding the misleading influence of complex mechanical interference during the pressurization stage. It adaptively determines the leakage characteristic frequency boundary and low-frequency energy concentration, achieving automatic iterative optimization and optimal matching of the mode decomposition number. This enables the invention to accurately remove background noise in complex real-world industrial scenarios and reconstruct extremely weak real leakage signals with high fidelity, thereby improving the accuracy of detecting abnormal sealing conditions in the compressor as a whole.

[0008] Preferably, the steps for acquiring the timing signal during the compressor's overall seal testing process and preprocessing it to obtain the preprocessed pressure sequence are as follows: The system acquires the pressure and temperature signals of the inner wall of the cavity, which are synchronously collected by the sensors. It calculates the pressure change between adjacent sampling points step by step, identifies jump points that exceed the estimated range as outliers, and replaces the outliers with the average of the adjacent normal sampling points before and after the point. Based on the ideal gas law, it converts the measured pressure values ​​of the inner wall of the cavity at each moment into equivalent standard pressures at the reference temperature to complete temperature compensation. It then uses a sliding window of a set length to perform mean filtering on the compensated equivalent standard pressure sequence to obtain the preprocessed pressure sequence.

[0009] The technical advantages of this invention are as follows: By synchronously acquiring the pressure and temperature signals of the inner wall of the cavity, the measured pressure is converted into the equivalent standard pressure at the reference temperature using the ideal gas law, effectively eliminating the systematic deviation of the pressure reading caused by the temperature drift during the test. At the same time, combined with the elimination of abnormal jump points and the moving average filtering, the random fluctuation noise of the sensor is fully suppressed while preserving the low-frequency leakage trend characteristics, thus providing a high-quality basic sequence with a high signal-to-noise ratio for subsequent data analysis.

[0010] Preferably, when the trend direction index of consecutive adjacent windows meets the preset threshold condition, the step of extracting the corresponding time period as the attenuation monitoring stage signal is as follows: construct a sliding window sequence for the pressure sequence using the number of sampling points corresponding to the set duration as the window length and step size; extract the pressure value at the starting sampling point of the window, the pressure value at the ending sampling point of the window, the maximum pressure value within the window, and the minimum pressure value within the window for each sliding window; calculate the difference between the pressure value at the starting sampling point of the window and the pressure value at the ending sampling point of the window to obtain the net pressure decrease; calculate the difference between the maximum pressure value within the window and the minimum pressure value within the window to obtain the total pressure change amplitude; when the trend direction index of a consecutive set number of adjacent windows all exceed the preset threshold, it is determined that the corresponding time period is in the attenuation monitoring stage, and the pressure signal of that time period is extracted as the attenuation monitoring stage signal.

[0011] The technical advantages of this invention are as follows: By using a set window to calculate the ratio of the net pressure drop to the total pressure change to construct a trend direction index, it can intuitively and quantitatively reflect the monotonically decreasing intensity of pressure changes. Compared with existing methods that rely on manual experience or a single amplitude threshold to determine the signal period, this invention effectively overcomes the interference of short-term pressure fluctuations and background noise, accurately distinguishes the pure natural decay monitoring stage from the pressurization and stabilization stages, and cuts off the risk of subsequent algorithm decomposition failure caused by irrelevant high-frequency mechanical excitation from the data input source.

[0012] Preferably, the step of determining the adaptive leakage characteristic frequency boundary is as follows: Apply a Fast Fourier Transform to the signal during the attenuation monitoring phase to calculate the power spectral density sequence of the entire signal. Locate the frequency point corresponding to the maximum value in the power spectral density sequence and record it as the main peak frequency. Scan point by point from the main peak frequency in the direction of increasing frequency and calculate the difference value of the power spectral density between each frequency point and the adjacent frequency point. Use the difference value as the difference value of the corresponding frequency point. When the sign of the difference value changes from negative to positive, define the frequency point corresponding to the difference value as the adaptive leakage characteristic frequency boundary.

[0013] The technical advantages of this invention are as follows: by calculating the signal power spectral density through fast Fourier transform, and accurately locating the first spectral valley that turns from negative to positive as the adaptive leakage characteristic frequency boundary by scanning point by point from the main peak frequency, compared with the practice of manually presetting a single fixed boundary, the boundary positioning mechanism of this invention can automatically migrate and adapt according to the differences in spectral structure caused by the cavity volume, working pressure and leakage aperture of the specific model, realizing the accurate delineation of cross-device characteristic frequency bands without manual intervention, and enhancing the versatility and ease of implementation of the detection method on diverse compressors.

[0014] Preferably, the steps for calculating the low-frequency energy concentration of each modal component are as follows: Variational mode decomposition is performed on the signal during the attenuation monitoring stage using the initial mode decomposition number to obtain multiple mode components, and the power spectral density sequence of each mode component is calculated. Using the adaptive leakage characteristic frequency boundary as the dividing line, the power spectral density sequence of each modal component is divided into two intervals: low frequency band and full frequency band. The cumulative power spectral density of each modal component at each frequency point in the low-frequency range is calculated, as well as the total cumulative power spectral density at each frequency point in the full frequency range. The ratio obtained by dividing the cumulative power spectral density of the low-frequency band by the total cumulative power spectral density of the entire frequency band is defined as the low-frequency energy concentration of the corresponding modal component.

[0015] The technical effect of this invention is as follows: by calculating the ratio of the cumulative power spectral density of the modal components in the low-frequency band to the total cumulative value of the entire frequency band, a low-frequency energy concentration index is constructed. Since compressor leakage is essentially a low-frequency slowly changing physical process, this index can objectively reveal the proportion of the actual leakage component in each decomposition mode.

[0016] Preferably, the steps for adaptively and iteratively adjusting the number of mode decompositions based on the number of effective modes until the number of effective modes converges to the target value, to determine the optimal number of mode decompositions, are as follows: The low-frequency energy concentration of each modal component is compared with a preset energy concentration threshold. The number of modal components with a low-frequency energy concentration greater than or equal to the threshold is counted and taken as the effective number of modes. When the effective number of modes is greater than 1, it is determined that over-decomposition has occurred, and the current mode decomposition number is reduced by 1 before re-entering the variational mode decomposition step. When the effective number of modes is equal to 0, it is determined that under-decomposition has occurred, and the current mode decomposition number is increased by 1 before re-entering the variational mode decomposition step. When the effective number of modes is equal to 1, it is determined that the current mode decomposition number matches the actual component structure of the signal, and the current value is determined as the optimal mode decomposition number and the iteration is terminated.

[0017] The technical effect of this invention is as follows: using an effective number of modes equal to 1 as the core criterion for complete adaptation between the modal decomposition number and the actual component structure of the current signal, the invention guides the iterative increase or decrease of the modal decomposition number in both directions by automatically determining over-decomposition when the number is greater than 1 or under-decomposition when the number is equal to 0. This ensures that a single and pure physical leakage mode can be adaptively locked and retained under various unknown and complex working conditions, thereby reducing the false alarm rate and false negative rate of on-site engineering detection.

[0018] Preferably, the iteration process in step S4 further includes a step to prevent oscillations: Maintain a historical access table. During each iteration, save the current modal decomposition number and the corresponding number of valid modes as records in the historical access table. Before each parameter adjustment, check whether the new modal decomposition number that needs to be updated already exists in the historical access table. If the new modal decomposition number already exists in the historical access table, the iteration is determined to have entered an oscillation state. At this time, calculate the absolute value of the difference between the number of valid modes in all records of the historical access table and 1, select the modal decomposition number corresponding to the record with the smallest absolute value, and forcibly terminate the iteration with the decomposition result of its corresponding round.

[0019] The technical advantages of this invention are as follows: By maintaining a historical access table during the adaptive iterative optimization process, the update trajectory of the modal decomposition number in each round is recorded and monitored in real time. When a new parameter is detected to already exist in the past record, the algorithm is promptly determined to be in an oscillation state and forcibly terminated. At the same time, the parameter combination with the closest number of effective modes in the historical record to the theoretical target value is selected. This effectively overcomes the technical problem that the algorithm parameters are prone to repeated jumps and cannot converge when decomposing non-stationary signals, and ensures high computing efficiency and stable output of the detection process in real industrial edge computing devices.

[0020] Preferably, the steps to prevent oscillation also include: When selecting the record with the smallest absolute difference, if there are multiple records with the smallest absolute difference, the record with the smallest number of effective modes is selected from the records with a number of effective modes greater than 1, and its corresponding modality decomposition number is used as the criterion for terminating the iteration. An adjustment upper limit is set for the modality decomposition number. When the modality decomposition number increases to the adjustment upper limit during the iteration process, the iteration is forcibly terminated to prevent infinite loop.

[0021] The technical effect of this invention is as follows: when dealing with forced termination of oscillation, the historical records with the smallest effective mode number are selected first to approximate the optimal decomposition state to the greatest extent. The upper limit of the modal decomposition number is strictly set to cut off the dead loop path, which effectively avoids the exhaustion of the monitoring system's computing power or the suspension and lag caused by sudden strong electromagnetic or mechanical interference on site, and ensures the continuous and uninterrupted operation of the online monitoring system of the equipment.

[0022] Preferably, under the optimal modal decomposition number, the step of selecting the modal component with the highest low-frequency energy concentration to reconstruct the leakage signal, calculating the total pressure drop of the leakage signal, and determining the sealing status is as follows: During the attenuation monitoring phase, the signal values ​​at the start and end times of the reconstructed leakage signal are extracted, and the two are subtracted to obtain the total pressure drop; if the total pressure drop is less than or equal to the preset sealing qualification threshold, the judgment result that the sealing performance of the whole machine is qualified is output; if the total pressure drop is greater than the sealing qualification threshold, the sealing status is determined to be abnormal, and the leakage rate is estimated based on the pressure drop rate during the start to end period; when the leakage rate is lower than a preset multiple of the historical typical leakage rate, it is determined to be a minor leakage, and when it reaches or exceeds this multiple, it is determined to be a serious leakage.

[0023] The technical advantages of this invention are as follows: After obtaining a high-fidelity reconstructed leakage signal, the total pressure drop at the beginning and end of the signal is strictly compared with the historical qualified threshold to complete the basic seal compliance judgment. After identifying anomalies, the pressure drop rate is further calculated to perform a step-by-step rating of the leakage degree. Compared with the binary judgment method of simply outputting qualified or unqualified, this invention can not only effectively alarm, but also output fine-grained equipment health status profiles such as minor leakage and serious leakage. This provides on-site maintenance engineers with a scientific and intuitive quantitative decision-making basis for quickly locating the hidden danger level and formulating differentiated maintenance strategies.

[0024] Preferably, the specific steps for applying mean filtering to the compensated equivalent standard pressure sequence using a sliding window of a set length are as follows: obtaining the sampling frequency when acquiring the pressure value of the inner wall of the acquisition cavity; dynamically determining the length of the sliding window based on the sampling frequency, wherein the length of the sliding window is positively correlated with the sampling frequency to maintain the smooth effect of the mean filtering on a fixed time scale; and performing sliding mean filtering on the equivalent standard pressure sequence using a sliding window of dynamically determined length.

[0025] This invention offers the following technical advantages: Addressing the engineering challenges of significant environmental interference and difficulty in accurately extracting subtle leakage characteristics during compressor sealing testing, this invention proposes a fully adaptive analytical method based on variational mode decomposition. Unlike traditional methods that rely on fixed manual parameters, leading to over- or under-decomposition, this invention constructs a trend direction index to accurately isolate pure attenuated signal segments and dynamically captures characteristic frequency boundaries using spectral characteristics. Furthermore, it uses low-frequency energy concentration to guide the automatic optimization of the decomposition count, eliminating reliance on expert experience. This achieves high-fidelity reconstruction of real leakage signals under conditions of strong interference and cross-model operation, improving the intelligence level of early warning of minor leaks in power equipment and the reliability of on-site testing. Attached Figure Description

[0026] Figure 1 This is a flowchart of the compressor whole machine seal detection method based on pressure decay characteristics according to the present invention.

[0027] Figure 2 This invention is fixed Value and Adaptation A diagram showing the comparison of VMD decomposition results.

[0028] Figure 3 This is the present invention. A schematic diagram of the value iteration convergence process.

[0029] Figure 4 This is a schematic diagram comparing the pressure decay curves of the present invention under sealed and leaking conditions. 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] This invention discloses a method for detecting the overall seal of a compressor based on pressure decay characteristics, referring to... Figure 1 This includes the following steps: Step S1: Signal acquisition and preprocessing.

[0032] The compressor's overall sealing test is implemented based on the pressure decay method. During the test, both pressure and temperature time-series signals need to be collected simultaneously. Specifically, a piezoresistive digital pressure sensor with a matched range is installed on both the compressor's inlet and outlet sides. The sensors are sealed to the tested chambers via pressure guide tubes, and sampling frequencies are used to measure the pressure. The pressure value of the inner wall of the chamber is acquired in real time, and the acquisition time covers the complete detection cycle from pressurization to the end of decay monitoring. Before the test, the compressor's inlet and outlet pipes and all connecting flanges are sealed to ensure that the chamber under test is in a sealed state. A platinum resistance temperature sensor is installed on the outer wall of the chamber under test, and the chamber wall temperature is recorded synchronously with the pressure sensor at the same sampling frequency for subsequent pressure compensation. The output signals of the two types of sensors are collected by a data acquisition card and stored as synchronous dual-channel digital time-series data in a timestamp-aligned manner.

[0033] The collected data underwent the following preprocessing steps, including outlier removal: Pressure changes between adjacent sampling points were calculated progressively; for jump points exceeding the physically reasonable range estimated based on the cavity volume and working pressure, the average of the two normal sampling points before and after that point was used as the replacement; Temperature compensation: Based on the ideal gas law, the measured pressure values ​​at each moment were converted to the equivalent standard pressure at the reference temperature to eliminate the systematic influence of temperature drift on the pressure readings; Moving average filtering: A length of... A sliding window of sampling points performs mean filtering on the compensated pressure sequence, suppressing sensor random noise while preserving the low-frequency trend characteristics of the signal. The pressure sequence after the above preprocessing is used as the input signal for subsequent steps.

[0034] Step S2: Construction of the trend direction index and identification of the decay monitoring stage.

[0035] In the overall sealing inspection of compressors, the attenuation monitoring stage exhibits objective characteristics distinct from other stages: the cavity pressure shows a continuous, unidirectional decreasing trend over time. Specifically, when a sealing defect exists, gas continuously escapes through the defect channel. According to the ideal gas law and the principle of mass conservation, under approximately isothermal conditions, the cavity pressure is directly proportional to the amount of gas within the cavity. The unidirectional gas escape leads to a monotonically decreasing pressure. Even if multiple defects exist in the cavity, and the escape direction from each channel is consistent, the cumulative effect still maintains an overall monotonically decreasing trend.

[0036] During the attenuation monitoring phase, within any time window, the pressure value at the end of the window is lower than the pressure value at the beginning, and the net pressure drop increases with the increase of leakage. The net pressure change is positive during the pressurization phase and approaches zero during the stabilization phase; the directions of these two phases are fundamentally different from those of the attenuation phase.

[0037] If the attenuation monitoring stage is not accurately identified, and the pressurization stage signal is mistakenly sent to variational mode decomposition, the pressurization stage signal has a large number of components, corresponding to... The value is much larger than the actual number of components in the decay stage, resulting in severe over-decomposition of the decay stage signal, making it impossible to extract leakage characteristics. Taking a compressor as an example: the signal in the charging stage contains frequency and multiple harmonics, with more than 5 effective components; in the decay stage, when the compressor stops, the effective components are only 1 to 2. If the charging stage signal is used... Directly applying the values ​​to the decay phase can lead to spurious modes generated by over-decomposition, which will mask the true leakage characteristics and result in inaccurate detection conclusions. Therefore, it is necessary to construct an index that can accurately measure the directional intensity of pressure changes within a time window to achieve precise identification during the decay monitoring phase.

[0038] For the preprocessed pressure sequence, with Each sampling point represents a window length of 10 seconds. This duration, under typical leakage rate conditions, can generate a significant net pressure change within the window, causing the trend signal to exceed the sensor's background noise level. A sliding window sequence is constructed using 1000 sampling points as the step size. Four features are extracted from each window: the pressure value at the initial sampling point of the window. Pressure value at the sampling point at the end of the window Maximum pressure within the window Minimum pressure within the window .

[0039] Pressure value at the start of the window Subtract the pressure value at the end of the window The net pressure decrease within the window is obtained; the maximum pressure within the window is used as the criterion. Subtract the minimum pressure within the window The total pressure change within the window is obtained; the net pressure decrease is divided by the total pressure change, and the resulting ratio is defined as the trend direction index of the window. The trend direction index is calculated sequentially for each sliding window of the pressure sequence, forming a trend direction index sequence. When the trend direction index of three consecutive adjacent windows exceeds a threshold... When the corresponding time period is determined to be in the attenuation monitoring stage, the pressure signal of that time period is extracted as the object of subsequent analysis.

[0040] The greater the leakage in the cavity, the greater the net pressure drop per unit time. The short-term total pressure change caused by sensor background noise changes less with increasing leakage; therefore, the larger the trend direction index, the closer it is to the theoretical upper limit of 1. During the pressure stabilization phase, the net pressure drop approaches zero, and the trend direction index approaches zero. During the pressurization phase, the initial pressure is lower than the final pressure, resulting in a negative net pressure drop and a negative index. The trend direction index values ​​for the three phases do not overlap, and the threshold... It can effectively distinguish the attenuation monitoring stage from other stages, and the identified attenuation monitoring stage signal provides an accurate signal range for subsequent variational mode decomposition, thus avoiding the problems caused by stage misjudgment at the signal source level. The problem of mismatch between the value and the actual component structure of the signal.

[0041] Step S3: Determination of adaptive leakage characteristic frequency boundary and construction of low-frequency energy concentration.

[0042] In the scenario of compressor whole-unit seal inspection, the equivalent opening area of ​​the sealing defect is extremely small, and the escape of gas through this tiny opening is a slow and continuous process. The volume of the gas cavity being measured is several orders of magnitude larger than the equivalent opening area of ​​the defect, and the change in cavity pressure due to leakage can only be observed on a longer time scale. This difference in physical scale determines that the pressure drop signal caused by leakage is a slowly varying trend component with an extremely low frequency. The larger the cavity volume and the smaller the defect opening, the slower the corresponding pressure drop change, and the lower the frequency of the leakage component. The physical process of sensor circuit response to external environmental mechanical vibration is much faster than the change in cavity pressure due to leakage, corresponding to a signal component distributed in a relatively high frequency range.

[0043] The aforementioned physical mechanism generates an objectively existing spectral structure on the power spectrum: the slowly varying leakage component forms a concentrated main peak in the low-frequency band, while sensor noise and environmental vibration components are distributed in a relatively high-frequency range. Between these two types of components lies a spectral valley region with a significantly lower power spectral density than those on either side. This spectral valley is not artificially set, but is jointly determined by the physical characteristics of the leakage process and the noise source, and its position adaptively shifts with changes in cavity volume, operating pressure, and defect area. Taking a large-volume low-pressure model as an example, the leakage main peak frequency is extremely low, and the spectral valley appears at a lower frequency; taking a small-volume high-pressure model as an example, the leakage main peak frequency is relatively high, and the spectral valley shifts to the right accordingly. Regardless of changes in the model parameters, the spectral valley always objectively exists between the leakage main peak and the noise distribution area, forming the common physical basis for cross-model adaptive frequency boundary positioning.

[0044] If a fixed frequency value is used as the upper limit of the leakage characteristic frequency band, the relative relationship between the leakage main peak position and the fixed boundary varies systematically across different models. Furthermore, the fixed boundary lacks universality among models with significant differences in cavity volume or operating pressure. Therefore, the frequency position of the first spectral valley to the right of the leakage main peak in the signal power spectrum is used as the adaptive leakage characteristic frequency boundary. This boundary is determined by the spectral structure of the signal itself and automatically migrates with changes in the physical characteristics of leakage. There is no need to preset frequency parameters for different models, thus achieving cross-model adaptive applicability at the frequency boundary level.

[0045] Apply a Fast Fourier Transform to the signal identified in step S2 during the attenuation monitoring phase, and calculate the power spectral density sequence of the entire signal with a frequency resolution of [missing information]. ,in This represents the total number of signal sampling points during the attenuation monitoring phase. Locate the frequency point corresponding to the maximum value in the power spectral density sequence, and denote it as the main peak frequency. .from The system scans point by point in the direction of increasing frequency, calculating the difference in power spectral density between each frequency point and its adjacent frequency points. This difference is then used as the difference value for the corresponding frequency point. When the sign of the difference value changes from negative to positive (i.e., the power spectral density changes from a continuous decrease to an increase for the first time), the frequency point corresponding to this difference value is defined as the adaptive leakage characteristic frequency boundary. Using the initial mode decomposition number Variational mode decomposition is performed on the signal during the attenuation monitoring phase to obtain an initial set of mode components. A fast Fourier transform is then applied to each mode component to calculate its power spectral density sequence. To define the boundaries, the power spectral density sequences of each modal component are divided into low-frequency bands. With full frequency band Two intervals are used to calculate the cumulative power spectral density of all frequency points in each interval.

[0046] For each modal component, its power spectral density in the low-frequency band The accumulated value at each frequency point within the range, divided by the modal component across the entire frequency band. The ratio of the sum of the power spectral densities at all frequency points within the range is defined as the low-frequency energy concentration of that modal component. The value ranges from 0 to 1. Threshold: Greater than or equal to The modal components are determined to be effective leakage modes, below which... The modal components are identified as noise modes, and the effective mode count is recorded as... The number of noise modes is denoted as ,satisfy .

[0047] The greater the cavity leakage, the more concentrated the low-frequency main peak energy caused by the leakage, and the closer the low-frequency energy concentration of the effective leakage mode is to 1; the sensor background noise and environmental vibration components are distributed in a relatively high frequency range, and the low-frequency energy concentration of the corresponding mode remains at a low level. The spectral valley positioning method enables... It consistently falls at the natural frequency boundary between the two types of components. The difference in low-frequency energy concentration between the effective leakage mode and the noise mode remains significant across different models. Its applicability is not affected by the model parameters. When When the value is too large, the leakage component is split, and the low-frequency energy concentration of each split sub-mode is higher than that of the noise mode. In addition, the number of modes that meet the threshold increases, and the effective mode count increases accordingly.

[0048] Step S4: Adaptive adjustment of mode decomposition number.

[0049] In the attenuation monitoring phase of compressor overall seal testing, the pressure signal to be analyzed is dominated by a single physical process: gas escapes unidirectionally through sealing defects, causing a continuous decrease in cavity pressure. Regardless of the number of sealing defects in the cavity, the combined leakage effect of each channel is macroscopically manifested as a single, monotonically decreasing pressure curve, and in the frequency domain, it is represented by a single dominant low-frequency component. The number and spatial distribution of defects only affect the attenuation rate and do not generate additional independent low-frequency components. Therefore, in the attenuation monitoring phase, the number of effective low-frequency components has a clear a priori constraint, theoretically being only one.

[0050] This prior constraint determines the variational mode decomposition. The value's reasonableness provides direct evidence. When When the value is appropriate, the decomposition result should contain exactly one low-frequency energy concentration that reaches a certain level. Effective leakage mode; when When the value is too large, the variational mode decomposition algorithm forcibly splits the single leakage low-frequency component into multiple artificial sub-components, resulting in the number of effective modes exceeding 1; when When the value is too small, the leakage signal is forced to alias with noise in the same mode, and the low-frequency energy concentration of each mode is difficult to achieve. The number of effective modes is 0. Let's illustrate this with a specific detection example: Performing variational mode decomposition, three out of the five modes showed a low-frequency energy concentration exceeding 0.6, indicating that the leakage component was excessively decomposed; After adjusting to 2, there is only one effective leakage mode, and the other mode is a noise mode. The decomposition result matches the physical structure of the signal, and the number of effective modes converges from 3 to the target value of 1. The direction and degree of deviation of the number of effective modes from the target value of 1 directly indicate the... The direction and magnitude of value adjustment provide criteria for constructing an iterative adjustment mechanism.

[0051] Arrange the low-frequency energy concentration of each modal component in step S3 from largest to smallest, and confirm that the low-frequency energy concentration reaches a certain level. The set of modal components and their corresponding quantities.

[0052] when If the value is greater than 1, the current mode decomposition number will be... Decrease by 1, return to step S3 with the updated The value is recalculated by performing variational mode decomposition and low-frequency energy concentration calculation; when When the value is 0, the current mode decomposition number is... Increase by 1, and return to step S3 to perform a re-decomposition; when When it is exactly equal to 1, determine the current state. The value matches the actual component structure of the signal during the attenuation monitoring stage, the iteration terminates, and the current decomposition result is used to proceed to step S5. To prevent extreme cases... The value continues to increase without stopping, setting The adjustment limit is ,when Increase to When the current decomposition result is reached, the iteration is forcibly terminated.

[0053] when When the value is large, the more sub-modes the leaked low-frequency component is split into, the larger the effective mode count, and the greater the deviation from the target value of 1; the larger the effective mode count, the stronger the current... The more severe the over-decomposition of the value of the signal, the more... The reduction operation continues until the effective mode count converges to 1. This metric directly transforms the prior knowledge of the single low-frequency component of the signal during the attenuation monitoring phase into... The iterative adjustment criterion for the value converges the mode decomposition number to a value that matches the physical structure of the signal within a finite number of steps, fundamentally solving the problem of variational mode decomposition being fixed. The problem of over-decomposition and under-decomposition caused by the value.

[0054] Step S5: Adaptive variational mode decomposition execution and sealing status determination.

[0055] The trend direction index, low-frequency energy concentration, and effective mode count constructed in steps S2 to S4 are integrated into a complete iterative execution process.

[0056] Using the pressure sequence preprocessed in step S1 as input, the method described in step S2 is applied... A sliding window of sampling points is scanned step by step, and the trend direction index is calculated for each window. When three consecutive adjacent windows When all values ​​exceed 0.5, the corresponding time period is identified as the attenuation monitoring stage, and the pressure signal for that time period is extracted.

[0057] With initial mode decomposition number Variational mode decomposition is performed on the signal during the attenuation monitoring phase. The low-frequency energy concentration is calculated for each modal component using the fast Fourier transform according to the method described in step S3. ,by Statistical counting of effective modes for thresholds .

[0058] Perform according to the rules described in step S4 Value adjustment: If Then Updated to: And when lower bound 2 is triggered, it will be forcibly terminated according to step S4; if Then Updated to ;like The iteration terminates. Each adjustment is followed by an updated... The variational mode decomposition and low-frequency energy concentration calculation are re-executed until... or Reaching the upper limit .

[0059] At the same time, to avoid the occurrence To mitigate the fluctuations, maintain a historical access table to record each iteration. Value and corresponding Before each adjustment, first check the items that need to be updated in the current iteration. Check if the value already exists in the historical access table: If not, update. The value is then re-executed for variational mode decomposition; if it already exists in the historical access table, the iteration is determined to have entered an oscillation phase, and the value is retrieved from all records in the historical access table. The smallest record corresponds to The value is used to force the termination of the iteration based on the decomposition result of this round; if | If there are equal choices, take priority. The minimum value in , When convergence terminates directly, achieve The current result will be forcibly terminated, and both will take precedence over oscillation detection.

[0060] For example, for a certain determination of the best During the process of setting values, the historical record table is initially empty. .

[0061] During the first round of iterations: After decomposition The historical access table has been updated to: ,express , ;because Therefore, it needs to be updated. Value No history of visits The records will be updated normally. ; During the second round of iterations: After decomposition The historical access table has been updated to: ;because Therefore, it needs to be updated. Value No history of visits Then it will update normally. ; During the third iteration: , after decomposition , a decomposition has occurred; at this time, the historical access table becomes: ; because is greater than 1, the to-be-updated value is , and already exists in the historical access table, concussion detection is triggered, wherein the 3 records in the historical access table all have a value of 1, there is a parallel condition, and the priority is to take the corresponding to the minimum value , select , corresponding decomposition result is taken as the optimal decomposition value, the mode with the highest low-frequency energy concentration is taken as the main leakage representative mode, and the iteration is forcibly terminated.

[0062] Since variational mode decomposition has the monotonicity that components are gradually refined with the increase of for the same signal, that is, the effective mode count is 0 when is too small, 1 when appropriate, and exceeds 1 when too large, the above iteration process can converge to a termination state within a limited number of steps. After the iteration is terminated, the modal component with the highest low-frequency energy concentration is confirmed as the effective leakage modal, and the remaining modals are used as noise modals and do not participate in subsequent analysis.

[0063] After the iteration is terminated, the leakage signal is reconstructed with the time-domain waveform of the effective leakage mode. For the reconstructed leakage signal, take the signal value at the start time during the attenuation monitoring phase and the signal value at the end time , calculate the total pressure drop in this time period .

[0064] Based on the historical qualified product detection data of compressors of the same model or the same batch, count the upper limit of pressure drop within the same detection duration under normal sealing condition, and take this upper limit as the sealing qualification judgment threshold .

[0065] If , it is determined that the overall sealing performance of this compressor is qualified; if , it is determined that the sealing state is abnormal, the leakage rate is further estimated by the pressure drop rate within the time period from the start time to the end time of the leakage signal, the pressure drop rate is the pressure drop per unit time, the leakage degree is divided into slight leakage and severe leakage according to the magnitude of the leakage rate, the sealing state judgment result is output, and the start and end times of the attenuation monitoring stage, the total pressure drop, the leakage rate and each iteration's value and Complete data archives are maintained for subsequent quality traceability. For example, a minor leak is defined as a leak rate less than 0.5 times the historical typical leak rate, while a severe leak is defined as a leak rate greater than 0.5 times the historical typical leak rate.

[0066] The technical effects of this invention can also be illustrated in conjunction with the accompanying drawings. Figure 2 For fixed Value and Adaptation Comparison chart of VMD decomposition results Figure 2 Using the same attenuation monitoring phase signal as input, compare with a fixed Value and Adaptation The variational mode decomposition results of the value. The upper side shows the fixed value. When the value is too large, the over-decomposition result shows that the leaked low-frequency component is split into multiple sub-modes with similar center frequencies; the lower side shows the result after adaptive iteration convergence. The decomposition results show that the effective leakage mode is unique and the low-frequency energy concentration is significantly higher than that of the noise mode.

[0067] Figure 3 for A schematic diagram of the value iteration convergence process. Figure 3 This illustrates the iterative process of adaptive variational mode decomposition. Value and The relationship between iteration rounds and the number of iterations is shown on the horizontal axis, with the number of iterations on the left and the number of iterations on the right on the vertical axis. Value, right side is The image shows... The value is determined by the initial value. Starting from this point, adjustments will be made gradually based on joint criteria. Follow The value changes monotonically and converges to 1, and the iteration terminates within a finite number of rounds.

[0068] Figure 4 This is a comparison chart of pressure decay curves under sealed and leaking conditions. Figure 4 The diagram shows the reconstructed representative mode signals of the leak from the sealed prototype and the leaking prototype during the attenuation monitoring phase, extracted by adaptive variational mode decomposition. The horizontal axis represents time, and the vertical axis represents the equivalent standard pressure. The total pressure drop of the reconstructed signal from the sealed prototype is also shown. Below the threshold Leaked prototype Exceed The two are clearly distinguishable, verifying the effectiveness of the conclusion on the sealing status.

[0069] The above are all preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Therefore, all equivalent changes made in accordance with the structure, shape and principle of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for detecting the overall seal of a compressor based on pressure decay characteristics, characterized in that, Includes the following steps: S1: Acquire the timing signal during the compressor's overall sealing test process and preprocess it to obtain the preprocessed pressure sequence; S2: Construct a sliding window for the pressure sequence and calculate the trend direction index. When the trend direction index of consecutive adjacent windows meets the preset threshold condition, extract the corresponding time period as the attenuation monitoring stage signal. S3: Determine the adaptive leakage characteristic frequency boundary of the signal during the attenuation monitoring stage, and perform variational mode decomposition under the initial mode decomposition number to calculate the low-frequency energy concentration of each mode component under the constraint of the adaptive leakage characteristic frequency boundary; the steps for determining the adaptive leakage characteristic frequency boundary are as follows: Apply a Fast Fourier Transform to the signal during the attenuation monitoring phase to calculate the power spectral density sequence of the entire signal; locate the frequency point corresponding to the maximum value in the power spectral density sequence and record it as the main peak frequency; scan point by point from the main peak frequency in the direction of frequency increase and calculate the difference value of the power spectral density between each frequency point and the adjacent frequency point. Use the difference value as the difference value of the corresponding frequency point. When the sign of the difference value changes from negative to positive, define the frequency point corresponding to the difference value as the adaptive leakage characteristic frequency boundary. S4: Count the number of effective modes that meet the set conditions for low-frequency energy concentration, and adaptively iterate and adjust the number of mode decompositions based on the number of effective modes until the number of effective modes converges to the target value, so as to determine the optimal number of mode decompositions; S5: Under the optimal modal decomposition number, select the modal component with the highest low-frequency energy concentration to reconstruct the leakage signal, calculate the total pressure drop of the leakage signal, and determine the sealing status.

2. The compressor whole-machine seal detection method based on pressure decay characteristics according to claim 1, characterized in that, The steps for acquiring the timing signal during the compressor's overall seal testing process and preprocessing it to obtain the preprocessed pressure sequence are as follows: Acquire the pressure value and temperature signal of the inner wall of the cavity synchronously collected by the sensor; The pressure change between adjacent sampling points is calculated step by step. Points that jump beyond the estimated range are identified as outliers and replaced with the average of the adjacent normal sampling points before and after the outlier. Based on the ideal gas law, the measured pressure values ​​on the inner wall of the cavity at each moment are converted into equivalent standard pressures at the reference temperature to complete temperature compensation. The equivalent standard pressure sequence after compensation is filtered by mean using a sliding window of a set length to obtain the preprocessed pressure sequence.

3. The compressor whole-machine seal detection method based on pressure decay characteristics according to claim 1, characterized in that, When the trend direction index of consecutive adjacent windows meets the preset threshold condition, the steps to extract the corresponding time period as the signal of the attenuation monitoring stage are as follows: construct a sliding window sequence for the pressure sequence with the number of sampling points corresponding to the set duration as the window length and step size; extract the pressure value of the starting sampling point, the pressure value of the ending sampling point, the maximum pressure value and the minimum pressure value within the window for each sliding window. The net pressure drop is obtained by subtracting the pressure value at the last sampling point from the pressure value at the beginning of the window; the total pressure change is obtained by subtracting the minimum pressure value from the maximum pressure value within the window. When the trend direction index of a set number of consecutive adjacent windows all exceed a preset threshold, the corresponding time period is determined to be in the attenuation monitoring stage, and the pressure signal of that time period is extracted as the attenuation monitoring stage signal.

4. The compressor whole-machine seal detection method based on pressure decay characteristics according to claim 1, characterized in that, The steps for calculating the low-frequency energy concentration of each modal component are as follows: Variational mode decomposition is performed on the signal during the attenuation monitoring stage using the initial mode decomposition number to obtain multiple mode components, and the power spectral density sequence of each mode component is calculated. Using the adaptive leakage characteristic frequency boundary as the dividing line, the power spectral density sequence of each modal component is divided into two intervals: low frequency band and full frequency band. The cumulative power spectral density of each modal component at each frequency point in the low-frequency range is calculated, as well as the total cumulative power spectral density at each frequency point in the full frequency range. The ratio obtained by dividing the cumulative power spectral density of the low-frequency band by the total cumulative power spectral density of the entire frequency band is defined as the low-frequency energy concentration of the corresponding modal component.

5. The compressor whole-machine seal detection method based on pressure decay characteristics according to claim 1, characterized in that, The steps to adaptively and iteratively adjust the number of mode decompositions based on the number of effective modes until the number of effective modes converges to the target value are as follows: The low-frequency energy concentration of each modal component is compared with a preset energy concentration threshold, and the number of modal components with a low-frequency energy concentration greater than or equal to the threshold is counted and taken as the number of effective modes. When the number of effective modes is greater than 1, it is determined that over-decomposition has occurred, and the current mode decomposition number is reduced by 1 before re-entering the variational mode decomposition step; when the number of effective modes is equal to 0, it is determined that under-decomposition has occurred, and the current mode decomposition number is increased by 1 before re-entering the variational mode decomposition step; when the number of effective modes is equal to 1, it is determined that the current mode decomposition number matches the actual component structure of the signal, and the current value is determined as the optimal mode decomposition number and the iteration is terminated.

6. The compressor whole-machine seal detection method based on pressure decay characteristics according to claim 5, characterized in that, The iteration process in step S4 also includes steps to prevent oscillations: Maintain a historical access table. During each iteration, save the current modal decomposition number and the corresponding number of valid modes as records in the historical access table. Before each parameter adjustment, check whether the new modal decomposition number that needs to be updated already exists in the historical access table. If the new modal decomposition number already exists in the historical access table, the iteration is determined to have entered an oscillation state. At this point, calculate the absolute value of the difference between the number of valid modes in all records of the historical access table and 1, select the mode decomposition number corresponding to the record with the smallest absolute value, and force the iteration to terminate with the decomposition result of its corresponding round.

7. The compressor whole-machine seal detection method based on pressure decay characteristics according to claim 6, characterized in that, Steps to prevent oscillations also include: When selecting the record with the smallest absolute difference, if there are multiple records with the smallest absolute difference, the record with the smallest number of effective modes is selected from the records with a number of effective modes greater than 1, and its corresponding modality decomposition number is used as the criterion for terminating the iteration. An adjustment upper limit is set for the modality decomposition number. When the modality decomposition number increases to the adjustment upper limit during the iteration process, the iteration is forcibly terminated to prevent infinite loop.

8. The compressor whole-machine seal detection method based on pressure decay characteristics according to claim 1, characterized in that, Under the optimal mode decomposition number, the mode component with the highest low-frequency energy concentration is selected to reconstruct the leakage signal, calculate the total pressure drop of the leakage signal, and determine the sealing status. The steps are as follows: During the attenuation monitoring stage, the signal values ​​at the start and end times of the reconstructed leakage signal are extracted, and the two are subtracted to obtain the total pressure drop. If the total pressure drop is less than or equal to the preset sealing qualification threshold, the result of the whole machine sealing performance qualification will be output. If the total pressure drop exceeds the sealing qualification threshold, the sealing condition is deemed abnormal, and the leakage rate is estimated based on the pressure drop rate during the start to end period. When the leakage rate is lower than a preset multiple of the historical typical leakage rate, it is determined to be a minor leak; when it reaches or exceeds that multiple, it is determined to be a serious leak.

9. The compressor whole-machine seal detection method based on pressure decay characteristics according to claim 2, characterized in that, The specific steps for applying mean filtering to the compensated equivalent standard pressure sequence using a sliding window of a set length are as follows: obtain the sampling frequency when collecting the pressure value of the inner wall of the acquisition cavity; dynamically determine the length of the sliding window based on the sampling frequency, wherein the length of the sliding window is positively correlated with the sampling frequency to maintain the smooth effect of the mean filtering on a fixed time scale; and perform moving mean filtering on the equivalent standard pressure sequence using a sliding window of dynamically determined length.

Images