A compressor efficiency evaluation method and system based on recession trend prediction

Abstract

This invention belongs to the field of compressor efficiency evaluation technology. It provides a compressor efficiency evaluation method and system based on predictive degradation trends. The method includes: determining whether high-frequency oscillations in the efficiency signal are caused by a rotating stall cluster through time-domain analysis of high-frequency periodic oscillations and frequency-domain analysis of characteristic peaks; if so, dividing the stall cluster oscillation characteristic frequency band and the low-frequency characteristic frequency band of efficiency degradation into two frequency bands; acquiring the compressor efficiency oscillation + degradation composite signal, performing wavelet packet decomposition according to the divided frequency bands, removing the decomposed oscillation characteristic frequency band signal, retaining only the efficiency degradation characteristic frequency band signal, and reconstructing a pure efficiency degradation trend curve without oscillation interference through spline interpolation and moving average; and performing linear / nonlinear degradation rate fitting calculation and analysis based on the pure efficiency degradation trend curve, combined with morphological analysis of the pure efficiency degradation trend curve, to determine the compressor efficiency degradation trend.

Description

Technical Field

[0001] This invention belongs to the field of compressor efficiency evaluation technology, specifically a compressor efficiency evaluation method and system based on the prediction of decline trend. Background Technology

[0002] In the field of compressor performance monitoring and efficiency evaluation, accurate determination of the efficiency degradation trend under near-stall conditions is crucial for the safe and stable operation of equipment and for making maintenance decisions. When a compressor enters near-stall conditions, the airflow angle of attack in the flow channel is abnormal, which easily forms a low-speed airflow cloud that propagates in a circumferential direction, namely a rotating stall cloud. This stall cloud will cause high-frequency periodic oscillations in the intake and exhaust pressure and flow signals, and the oscillation frequency is directly related to the propagation speed of the stall cloud.

[0003] Since efficiency calculations rely on the aforementioned temperature, pressure, and flow parameters, the oscillation signal is synchronously superimposed on the calculated efficiency value, causing drastic fluctuations in the efficiency signal. The core contradiction of this problem lies in the superposition of the high-frequency, large-amplitude transient efficiency oscillations caused by rotational stall with the inherent low-frequency, small-amplitude, slow-changing trend of efficiency degradation in the signal dimension, making it difficult for traditional evaluation methods to effectively separate the two.

[0004] While the time-averaging method can mitigate oscillation interference, it directly obscures the true efficiency degradation value. The instantaneous value method, on the other hand, cannot avoid oscillation interference and struggles to identify slow degradation trends from cluttered signals. Periodic oscillation signals completely mask trend changes, leading to distorted extraction of efficiency degradation trends and impacting assessment accuracy.

[0005] Therefore, the present invention provides a compressor efficiency evaluation method and system based on the prediction of decline trend. Summary of the Invention

[0006] In order to overcome the shortcomings of the prior art, at least one technical problem raised in the background art is solved.

[0007] The technical solution adopted by this invention to solve its technical problem is: a compressor efficiency evaluation method based on the prediction of decline trend, comprising the following steps:

[0008] High-frequency oscillation determination module: Collects real-time monitoring data of intake and exhaust pressure, flow rate and efficiency under near-stall state of compressor, and makes a comprehensive judgment on whether the high-frequency oscillation of efficiency signal is caused by rotating stall cluster through time domain analysis of high-frequency periodic oscillation and frequency domain analysis of characteristic peaks.

[0009] Dual-band segmentation module: If the determination is yes, the stall cluster oscillation characteristic frequency band is calculated by the real-time rotor speed of the compressor and the propagation coefficient of the rotating stall cluster, and the low-frequency characteristic frequency band of efficiency decay is calibrated to achieve dual-band segmentation;

[0010] Dual-band decomposition module: Acquires compressor efficiency oscillation + decay composite signal, performs wavelet packet decomposition according to the divided dual-band, including narrow window decomposition for the oscillation band and wide window decomposition for the decay band;

[0011] Efficiency decay reconstruction module: Removes the oscillation characteristic frequency band signal after decomposition, retains only the efficiency decay characteristic frequency band signal, and reconstructs the pure efficiency decay trend curve without oscillation interference through spline interpolation + moving average;

[0012] Efficiency degradation analysis module: Based on the pure efficiency degradation trend curve, it performs fitting calculation and analysis of linear / nonlinear degradation rates, and combines the shape analysis of the pure efficiency degradation trend curve to determine the compressor efficiency degradation trend.

[0013] As a further aspect of the present invention, the process of collecting real-time monitoring data on the inlet and outlet pressures, flow rates, and efficiency of the compressor under near-stall conditions is as follows:

[0014] Under near-stall conditions and when the anti-surge system is not triggered, data acquisition is triggered. The data format includes timestamp, intake pressure, exhaust pressure, intake flow rate, intake temperature, and exhaust temperature.

[0015] Outliers are removed using the 3σ criterion, environmental random noise is filtered out using a 5th-order Butterworth low-pass filter, and the real-time efficiency value is obtained by isentropic efficiency calculation.

[0016] As a further aspect of the present invention: the time-domain analysis process of high-frequency periodic oscillations is as follows:

[0017] The collected pressure, flow, and efficiency data are divided into data segments, and the fluctuation amplitude, fluctuation coefficient, and oscillation period consistency of each data segment are calculated. When the fluctuation amplitude and fluctuation coefficient of the pressure, flow, and efficiency signals all exceed the preset threshold and the oscillation period deviation rate is not greater than 10%, the time domain analysis is deemed to have passed. Otherwise, if any indicator does not meet the requirements, the analysis is deemed to have failed.

[0018] As a further aspect of the present invention: the frequency domain analysis process of the characteristic peaks is as follows:

[0019] Fast Fourier Transform is performed on the segmented data to extract the characteristic frequency range related to the rotor speed and the propagation coefficient of the stall cluster. If there is a significant peak in the characteristic frequency range and the peak amplitude ratio is greater than or equal to 5, and the peak frequency deviation of multiple consecutive data segments does not exceed 1Hz, then the frequency domain analysis is considered to be passed. Otherwise, if any indicator is not met, it is considered to be failed. Only when both the time domain and frequency domain judgments are passed is it determined that the high-frequency oscillation of the efficiency signal is caused by the rotating stall cluster.

[0020] As a further aspect of the present invention: the process of dual-band division is as follows:

[0021] The real-time rotor speed of the compressor is obtained, and the theoretical oscillation frequency is calculated by combining the propagation coefficient of the rotating stall cluster. The oscillation characteristic frequency band is set with the theoretical oscillation frequency as the center, and its width covers the frequency fluctuation range. At the same time, the low-frequency characteristic frequency band is calibrated according to the slow variation characteristics of efficiency decay.

[0022] As a further aspect of the present invention: the process of performing wavelet packet decomposition is as follows:

[0023] Determination of wavelet packet fundamental and decomposition level: The db4 wavelet packet is selected as the fundamental. The number of decomposition levels is calculated based on the dual-band characteristics. Wavelet packet decomposition is performed on the composite signal, and differentiated window widths are set for the two frequency bands.

[0024] The stall clique oscillation characteristic frequency band uses a narrow window width, while the efficiency degradation low-frequency characteristic frequency band uses a wide window width. After decomposition, the wavelet packet coefficients of the corresponding stall clique oscillation characteristic frequency band and efficiency degradation low-frequency characteristic frequency band are extracted to form two independent frequency band signals.

[0025] As a further aspect of the present invention: the process of reconstructing the pure efficiency degradation trend curve without oscillation interference is as follows:

[0026] Perform inverse wavelet packet transform on the wavelet packet coefficients of the retained low-frequency characteristic band of efficiency degradation to restore them to the time domain signal and obtain the degradation signal sequence;

[0027] The signal gap is processed by cubic spline interpolation algorithm. The moving average processing is performed on the interpolated and completed decay signal sequence. The smoothed decay signal sequence is sorted by timestamp to construct a pure efficiency decay trend curve with running time on the horizontal axis and efficiency value on the vertical axis.

[0028] As a further aspect of the present invention: the process for determining the compressor efficiency degradation trend is as follows:

[0029] Based on the pure efficiency decline trend curve, a standardized dataset was constructed with running time and efficiency value. The least squares method was used to fit linear, exponential and power function models respectively, and the goodness of fit of each model was calculated.

[0030] Based on the fitting results, the linear decay rate is extracted and converted into a decay rate per thousand hours, or the instantaneous rate change is analyzed by taking the derivative of the nonlinear model.

[0031] The overall shape characteristics of the pure efficiency decline trend curve are used as an auxiliary factor for judgment. By comparing the goodness of fit, quantifying the decline rate, and analyzing the curve shape, the specific category of efficiency decline can be determined as linear slow decline, nonlinear accelerated decline, or nonlinear steady decline.

[0032] As a further aspect of the present invention, the process for comprehensively determining whether efficiency decline belongs to a specific category among linear slow decline, nonlinear accelerated decline, or nonlinear steady decline is as follows:

[0033] Linear slow decay determination: The linear fitting model has the highest goodness of fit and is not lower than 0.85. The calculated decay rate per thousand hours is less than 0.1% and the fluctuation range is limited. At the same time, the pure efficiency decay trend curve is approximately straight, and the slope deviation and fitting residual of each period do not exceed the preset threshold.

[0034] Nonlinear accelerated decay determination: The goodness of fit of the exponential fitting model is the highest and not less than 0.85. The calculated average decay rate per thousand hours is not less than 0.1% and shows a continuous increasing characteristic. At the same time, the pure efficiency decay trend curve shows a concave shape.

[0035] Nonlinear steady-state decay determination: The power function fitting model has the highest goodness of fit and is not lower than 0.85. The calculated average decay rate per thousand hours gradually decreases and approaches zero. At the same time, the pure efficiency decay trend curve shows an upward convex shape and the terminal segment tends to be flat.

[0036] A compressor efficiency evaluation system based on decline trend prediction includes the following modules:

[0037] High-frequency oscillation determination module: Collects real-time monitoring data of intake and exhaust pressure, flow rate and efficiency under near-stall state of compressor, and makes a comprehensive judgment on whether the high-frequency oscillation of efficiency signal is caused by rotating stall cluster through time domain analysis of high-frequency periodic oscillation and frequency domain analysis of characteristic peaks.

[0038] Dual-band segmentation module: If the determination is yes, the stall cluster oscillation characteristic frequency band is calculated by the real-time rotor speed of the compressor and the propagation coefficient of the rotating stall cluster, and the low-frequency characteristic frequency band of efficiency decay is calibrated to achieve dual-band segmentation;

[0039] Dual-band decomposition module: Acquires compressor efficiency oscillation + decay composite signal, performs wavelet packet decomposition according to the divided dual-band, including narrow window decomposition for the oscillation band and wide window decomposition for the decay band;

[0040] Efficiency decay reconstruction module: Removes the oscillation characteristic frequency band signal after decomposition, retains only the efficiency decay characteristic frequency band signal, and reconstructs the pure efficiency decay trend curve without oscillation interference through spline interpolation + moving average;

[0041] Efficiency degradation analysis module: Based on the pure efficiency degradation trend curve, it performs fitting calculation and analysis of linear / nonlinear degradation rates, and combines the shape analysis of the pure efficiency degradation trend curve to determine the compressor efficiency degradation trend.

[0042] The beneficial effects of this invention are as follows: First, data such as compressor inlet and outlet pressure and flow rate under near-stall conditions are collected. Efficiency values ​​are calculated after preprocessing. The high-frequency oscillations in the efficiency signal are determined through a comprehensive analysis of the time and frequency domains to determine whether they are caused by a rotating stall cluster. If this is determined, the oscillation characteristic frequency band and the low-frequency efficiency decay frequency band are divided based on rotor speed and stall cluster propagation coefficient. Differential window width wavelet packet decomposition is performed on the composite signal to remove the oscillation signal. Then, the pure efficiency decay trend curve is reconstructed through interpolation and moving average. Finally, the linear or nonlinear decay trend is determined through multi-model fitting and curve morphology analysis. This scheme achieves accurate separation of oscillation and decay signals through standardized data acquisition and dual-domain determination to eliminate interference, ensuring reliable signal reconstruction. Multi-model fitting combined with quantitative indicators can accurately predict three types of decay trends, solving the problem of distorted efficiency assessment caused by near-stall oscillation interference. This provides a quantitative decision-making basis for compressor operation optimization, maintenance, and life assessment, improving assessment accuracy and engineering applicability. Attached Figure Description

[0043] The invention will now be further described with reference to the accompanying drawings.

[0044] Figure 1 This is a flowchart illustrating the steps of a compressor efficiency evaluation method based on decline trend prediction as described in an embodiment of the present invention.

[0045] Figure 2 This is a logic diagram of a compressor efficiency evaluation method based on decline trend prediction as described in an embodiment of the present invention.

[0046] Figure 3 This is a flowchart of a compressor efficiency evaluation system based on decline trend prediction, as described in an embodiment of the present invention. Detailed Implementation

[0047] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0048] Example 1: Please refer to Figures 1-2 As shown in the embodiment of the present invention, a compressor efficiency evaluation method based on decline trend prediction includes the following steps:

[0049] Step S10: Collect real-time monitoring data of intake and exhaust pressure, flow rate and efficiency of the compressor under near stall conditions, and judge whether the high-frequency oscillation of efficiency signal is caused by the rotating stall cluster through time domain analysis of high-frequency periodic oscillation and frequency domain analysis of characteristic peaks.

[0050] In step S10, the process of collecting real-time monitoring data on the inlet and outlet pressures, flow rates, and efficiency of the compressor under near-stall conditions is as follows:

[0051] A piezoelectric pressure sensor (measurement range 0-20MPa, accuracy ±0.1%FS) is deployed at the compressor inlet and outlet flanges, an electromagnetic flow sensor (measurement range 0-100kg / s, accuracy ±0.2%FS) is installed in the main flow area of ​​the inlet pipe, and a K-type thermocouple (measurement range 0-800℃, accuracy ±0.5℃) is arranged at the inlet and outlet ports and the flow channel wall.

[0052] When the compressor speed stabilizes at 60%-85% of the design speed, the pressure ratio drops to 50%-70% of the design value (typical range of near-stall conditions), and the anti-surge system is not triggered, data acquisition is triggered. The acquisition parameters are set as follows: the sampling frequency is set to 1kHz (to satisfy the Nyquist sampling theorem for stall cluster oscillation of 33-133Hz, avoiding frequency aliasing), the acquisition duration is 60s, and 60,000 sets of synchronous data are obtained (1 set every 1ms). The data format includes timestamp, intake pressure P1, exhaust pressure P2, intake flow rate qm, intake temperature T1, and exhaust temperature T2.

[0053] Outliers (data points with deviations exceeding the mean ± 3 standard deviations) are removed using the 3σ criterion. Environmental random noise is then filtered out using a 5th-order Butterworth low-pass filter (cutoff frequency 500Hz). Finally, the isentropic efficiency formula is applied. (c) p Take the specific heat capacity of air at constant pressure. ;T 2is (where η is the isentropic exhaust temperature), 60,000 real-time efficiency values ​​η were calculated.

[0054] In step S10, the time-domain analysis process of the high-frequency periodic oscillation is as follows:

[0055] The collected pressure, flow, and efficiency data are divided into data segments, specifically: each segment is divided into 10-second intervals, with 10,000 data points per segment.

[0056] The core time-domain metrics for each data segment are calculated, including:

[0057] Fluctuation amplitude A: The difference between the maximum and minimum values ​​of each data segment, i.e.: A = Xmax - Xmin (where X represents P1, P2, qm and η respectively);

[0058] Volatility coefficient (CV): The ratio of the standard deviation to the mean for each data segment;

[0059] Oscillation period consistency R: The oscillation periods of P1, P2, qm, and η are calculated through autocorrelation analysis, and the oscillation period deviation rate is taken. (TX is the pressure / flow cycle, and Tη is the efficiency cycle).

[0060] Set the time-domain analysis criteria, specifically including:

[0061] Time-domain determination condition 1: Pressure: P1 fluctuation amplitude A ≥ 0.2 MPa, fluctuation coefficient CV ≥ 2%; P2 fluctuation amplitude A ≥ 0.5 MPa, fluctuation coefficient CV ≥ 3%;

[0062] Time-domain determination condition two: Flow rate: qm fluctuation amplitude A ≥ 2 kg / s, fluctuation coefficient CV ≥ 2.5%;

[0063] Time-domain criterion three: Efficiency: η fluctuation amplitude A ≥ 2%, fluctuation coefficient CV ≥ 1.5%;

[0064] Time-domain determination condition four: Period consistency: The deviation rate R of the oscillation period between P1, P2, qm and η is ≤10%;

[0065] If all time-domain analysis criteria are met, the time-domain analysis of the high-frequency periodic oscillation is passed; otherwise, if any criterion is not met, the time-domain analysis of the high-frequency periodic oscillation is failed.

[0066] In step S10, the frequency domain analysis process of the characteristic spike is as follows:

[0067] For each data segment (divided into 10-second segments), a Fast Fourier Transform (FFT) was performed to convert it into a frequency domain amplitude-frequency curve with a frequency resolution of 0.1 Hz. Core frequency domain metrics were then extracted, including:

[0068] Characteristic frequency range F: The theoretical oscillation frequency is calculated based on the compressor rotor speed N (r / min) and the stall cluster propagation coefficient k (typical value 0.2-0.8). Set characteristic frequency range ;

[0069] Peak Amplitude A peak The maximum amplitude within the characteristic frequency range;

[0070] Peak amplitude ratio K: The ratio of the characteristic peak amplitude to the average amplitude within the adjacent 10Hz interval, i.e. , where A adj This represents the average amplitude within an adjacent 10Hz interval.

[0071] Set the frequency domain analysis criteria, specifically including:

[0072] Frequency domain determination condition 1: Existence of characteristic peaks: A peak value exists within the characteristic frequency interval F, and the peak amplitude A peak ≥0.1V (the corresponding value of the sensor output voltage, converted to a physical quantity deviation ≥80% of the time domain fluctuation amplitude);

[0073] Frequency domain judgment condition two: Peak significance: Peak amplitude ratio K≥5 (ensuring that the peak amplitude is much higher than the background noise and is not random interference);

[0074] Frequency domain judgment condition three: Peak stability: The characteristic peak frequency deviation of 6 data segments (divided into segments of 10 seconds each, for a total of 60 seconds) is ≤1Hz, and the peak amplitude ratio K is ≥5 (ensuring that the peak is not an instantaneous interference).

[0075] If all the frequency domain analysis criteria are met, the frequency domain determination of the characteristic peak is passed; otherwise, if any criterion is not met, the frequency domain determination of the characteristic peak is failed.

[0076] In step S10, the process of comprehensively determining whether the efficiency signal high-frequency oscillation is caused by the rotating stall cluster is as follows:

[0077] If both the time-domain analysis of the high-frequency periodic oscillation and the frequency-domain analysis of the characteristic peak pass, it is determined that the high-frequency oscillation of the efficiency signal is caused by the rotating stall cluster. Conversely, if either analysis fails, it is determined that the high-frequency oscillation of the efficiency signal is not caused by the rotating stall cluster, and no operation is performed.

[0078] Understandably, the significance of step S10 lies in accurately identifying the root cause of high-frequency oscillations in the efficiency signal under near-stall conditions. Through standardized data acquisition and quantified time-domain and frequency-domain analysis, it eliminates interference from non-rotating stall clusters such as sensor malfunctions and airflow turbulence, providing a prerequisite for subsequent targeted separation of oscillations and decay signals. Its core value is ensuring that subsequent processing is only performed on oscillations caused by rotating stall clusters, avoiding distortion in efficiency assessments due to ineffective operations, and guaranteeing the accuracy and engineering applicability of the overall solution.

[0079] Step S20: If it is determined that the stall cluster oscillation characteristic frequency band is calculated by the real-time rotor speed of the compressor and the propagation coefficient of the rotating stall cluster, and the low-frequency characteristic frequency band of efficiency decay is calibrated to achieve dual-frequency band division;

[0080] In step S20, the dual-band division process is as follows:

[0081] Obtain the real-time rotor speed N (r / min) of the compressor, and calculate the theoretical oscillation frequency by combining it with the propagation coefficient k of the rotating stall cluster (value range 0.2-0.8, calibrated according to the compressor model and measured data under near-stall conditions, default value is 0.5). Based on the theoretical oscillation frequency, the characteristic frequency band of the stall cluster oscillation is set as the center frequency. It covers the frequency fluctuation range that may occur in the rotating stall cluster, ensuring complete coverage of the oscillation signal, and calibrates the low-frequency characteristic band of efficiency degradation as [0Hz, 0.001Hz].

[0082] It should be noted that the calibration basis for the low-frequency characteristic band of efficiency degradation is as follows: combining the characteristics of efficiency degradation as a slow trend of low frequency small amplitude, and referring to engineering measured data and signal analysis rules, the low-frequency characteristic band of efficiency degradation is calibrated as [0Hz, 0.001Hz]. This range can accurately capture the slow degradation signal of 0.1%-0.3% per thousand hours, while avoiding low-frequency noise interference.

[0083] Understandably, the significance of step S20 lies in: completing the precise dual-band division of the compressor efficiency signal under near-stall conditions; calculating the stall cluster oscillation characteristic frequency band that can completely encapsulate the oscillation signal through real-time rotor speed and stall cluster propagation coefficient; and simultaneously calibrating the efficiency decay low-frequency characteristic frequency band that can accurately capture small, slowly decaying signals and avoid low-frequency noise by combining the slow-varying characteristics of efficiency decay with engineering measurement laws. This lays the foundation for frequency domain division for subsequent wavelet packet decomposition to achieve precise separation of oscillation and decay signals, ensuring that subsequent decomposition operations have clear frequency domain targeting and avoiding mutual confusion of different characteristic signals in the frequency domain.

[0084] Step S30: Collect the compressor efficiency oscillation + decay composite signal, and perform wavelet packet decomposition according to the divided dual frequency bands, including narrow window width decomposition for the oscillation frequency band and wide window width decomposition for the decay frequency band.

[0085] In step S30, the process of acquiring the combined signal of compressor efficiency oscillation and decay is as follows:

[0086] Collect (using the same collection frequency as in step S10) the real-time efficiency signal of the compressor under near-stall conditions. The real-time efficiency signal is a composite signal formed by superimposing the oscillation signal (corresponding to the [F-5Hz, F+5Hz] frequency band defined in S20) and the decay signal (corresponding to the [0Hz, 0.001Hz] frequency band).

[0087] In step S30, the process of performing wavelet packet decomposition according to the divided dual frequency bands is as follows:

[0088] Determination of wavelet packet fundamental and decomposition level: The db4 wavelet packet is selected as the fundamental (balancing signal decomposition accuracy and computational efficiency). The number of decomposition levels is calculated based on the dual-band characteristics to ensure that the two frequency band signals can be accurately distinguished after decomposition. Based on a sampling frequency of 1kHz, the number of decomposition levels is set to 8. At this time, the frequency domain resolution can reach 0.390625Hz, which meets the requirements for accurate decomposition of [0Hz, 0.001Hz] and [F-5Hz, F+5Hz].

[0089] Dual-band targeted window width decomposition: Perform 8-level wavelet packet decomposition on the composite signal, and use differentiated window width settings for the two frequency bands:

[0090] Stall coil oscillation characteristic frequency band [F-5Hz, F+5Hz]: A narrow window width (window width value set to 8) is used to improve frequency resolution, accurately lock the frequency domain range of the oscillation signal, and avoid confusion with signals in other frequency bands;

[0091] Low-frequency characteristic band of efficiency degradation [0Hz, 0.001Hz]: A wide window width (window width value set to 32) is adopted to improve the time resolution, fully preserve the slow trend details of the degradation signal, and avoid loss of trend information;

[0092] After decomposition, wavelet packet coefficients of the corresponding stall clique oscillation characteristic frequency band and efficiency decay low-frequency characteristic frequency band are located and extracted to form two independent frequency band signals.

[0093] Understandably, the significance of step S30 lies in achieving targeted wavelet packet decomposition of the compressor efficiency oscillation-decay composite signal. First, the efficiency composite signal under near-stall conditions is acquired synchronously. Then, an appropriate db4 wavelet packet is selected, and the number of decomposition layers that meets the dual-band splitting accuracy is set. Differentiated window width decomposition strategies are adopted for the different signal characteristics of the oscillation band and the decay band. Narrow window width improves the frequency resolution of the oscillation band to accurately lock the oscillation signal, while wide window width improves the time resolution of the decay band to fully preserve the details of the slow-changing trend. Finally, the composite signal is accurately split in the frequency domain and two independent frequency band signals are extracted, completing the physical separation of oscillation interference and decay signal, and providing interference-free basic signal data for subsequent reconstruction of the pure efficiency decay trend curve.

[0094] Step S40: Remove the oscillation characteristic frequency band signal after decomposition, retain only the efficiency degradation characteristic frequency band signal, and reconstruct the pure efficiency degradation trend curve without oscillation interference through spline interpolation + moving average.

[0095] In step S40, the process of reconstructing the pure efficiency degradation trend curve without oscillation interference is as follows:

[0096] Based on the two independent frequency band signals extracted by S30, the oscillation characteristic frequency band is located. The wavelet packet coefficients corresponding to the 0Hz and 0.001Hz are removed to completely eliminate oscillation interference, and only the wavelet packet coefficients of the low-frequency characteristic band of efficiency degradation ([0Hz, 0.001Hz]) are retained.

[0097] Perform inverse wavelet packet transform on the wavelet packet coefficients of the retained low-frequency characteristic band of efficiency degradation to restore them to the time domain signal and obtain the degradation signal sequence;

[0098] A cubic spline interpolation algorithm is used to process signal gaps, with the interpolation node spacing set to 1ms (consistent with the sampling interval). A smooth interpolation function is constructed using three adjacent valid data points as support to fill discrete gaps.

[0099] The interpolated decay signal sequence was processed by a moving average, with the moving window size set to 50 data points (corresponding to a 50ms time window, balancing smoothing and trend preservation), according to the formula. (m is a sliding window, For the value of the j-th original data point, The value of the i-th data point is calculated point by point to reduce residual minor noise interference.

[0100] The smoothed decay signal sequence is sorted by timestamp to construct a pure efficiency decay trend curve with the horizontal axis representing running time (in seconds) and the vertical axis representing efficiency value (in %).

[0101] Understandably, the significance of step S40 lies in: reconstructing a pure efficiency degradation trend curve free from oscillation interference and noise; completely eliminating oscillation interference by removing wavelet packet coefficients corresponding to the oscillation characteristic frequency band; restoring the degradation time-domain signal through inverse wavelet packet transform; and finally constructing a realistic efficiency degradation time-efficiency value curve by combining cubic spline interpolation to fill signal gaps and moving average to weaken residual minor noise. This completely eliminates the masking and distortion of the efficiency degradation trend by interference factors such as rotating stall clique oscillation and random noise, providing a real, continuous, and smooth standardized data carrier for subsequent accurate fitting and calculation of degradation rate and analysis of degradation trend pattern, ensuring the accuracy of the data source for trend determination.

[0102] Step S50: Based on the pure efficiency decline trend curve, perform fitting calculation and analysis of linear / nonlinear decline rate, and combine with the shape analysis of the pure efficiency decline trend curve to determine the compressor efficiency decline trend.

[0103] In step S50, the process of determining the compressor efficiency decline trend is as follows:

[0104] Based on the pure efficiency decline trend curve, a standardized dataset is constructed with compressor operating time as the independent variable and the efficiency value η (unit: %) at the corresponding moment as the dependent variable. Where n is the number of runtime points;

[0105] Linear fitting is performed on the standardized dataset using the least squares method to obtain the linear fitting function model, i.e.: ,in, The linear decay rate (unit: % / h, negative values ​​indicate efficiency decay). The initial efficiency intercept (unit: %) is used to simultaneously calculate the goodness of fit R1;

[0106] By fitting an exponential function to the standardized dataset, we obtain an exponential fitting model, i.e. ( For coefficients, (The coefficient is negative; the larger the absolute value of the coefficient, the more significant the acceleration trend).

[0107] By fitting a power function to the standardized dataset, a power function fitting model is obtained, i.e. ( For coefficients, (The coefficient is negative; the larger the absolute value of the coefficient, the more significant the acceleration trend).

[0108] The coefficients of the exponential fitting model and the power function fitting model were solved using the least squares method. After solving, the goodness of fit R2 of the exponential fitting model and the goodness of fit R3 of the power function fitting model were calculated respectively.

[0109] Extracting the linear decay rate from the linear fitting function model Converted to decay rate per thousand hours (Unit: % / kh), simultaneously calculate The fluctuation range;

[0110] Taking the first derivative of the exponential fitting model yields the instantaneous decay rate. The change in rate over the entire time period is then calculated. If the change is greater than 0 and continuously increasing, it indicates nonlinear accelerated decay. The average instantaneous rate over the entire time period is then converted to a thousand-hour average rate. ;

[0111] Taking the first derivative of the power function fitting model yields the instantaneous decay rate. When the terminal rate approaches 0 (≤0.001% / kh), it indicates nonlinear stationary decay. The average instantaneous rate over the entire period is then converted to a thousand-hour average rate. ;

[0112] The criteria for determining the compressor efficiency decline trend are set, specifically including:

[0113] Linear slow decay: Among the goodness-of-fit R1, R2, and R3, R1 is the highest and ≥0.85, and the decay rate per thousand hours is... Furthermore, the fluctuation range is ≤ ±0.02% / 1000h, and the pure efficiency decline trend curve is approximately a straight line (the slope deviation of each period is ≤5%, and the residual between the actual efficiency value and the linear fitting value is ≤ ±0.1%).

[0114] Nonlinear accelerated decay: Among the goodness-of-fit R1, R2, and R3, R2 has the highest value and is ≥0.85, representing the decay rate over thousands of hours. Moreover, it continues to increase, and the pure efficiency decline trend curve shows a concave shape;

[0115] Nonlinear stationary decay: Among the goodness-of-fit R1, R2, and R3, R3 has the highest value and is ≥0.85, representing the decay rate over thousands of hours. It gradually decreases to below 0.001% / kh, and the curve shows an upward convex shape and tends to flatten out (the terminal rate approaches 0 (≤0.001% / kh)).

[0116] Understandably, the significance of step S50 lies in achieving a core step for the scientific and quantitative prediction of compressor efficiency degradation trends. It involves constructing a standardized dataset based on an interference-free pure efficiency degradation trend curve, using the least squares method to complete the fitting and goodness-of-fit evaluation of multiple models (linear, exponential, and power functions), accurately calculating the degradation rate and rate change characteristics per thousand hours under different fitting models, and setting multi-dimensional, quantifiable degradation trend judgment conditions based on curve morphology characteristics. This enables accurate classification and judgment of three typical trends: linear slow degradation, nonlinear accelerated degradation, and nonlinear steady degradation. It not only quantifies the rate indicators of efficiency degradation but also predicts the development and evolution of degradation patterns, providing quantitative and trend-based core decision-making basis for subsequent compressor operation optimization, maintenance, and life assessment.

[0117] Example 2, please refer to Figure 3 As shown in the embodiment of the present invention, a compressor efficiency evaluation system based on decline trend prediction includes the following modules:

[0118] High-frequency oscillation determination module: Collects real-time monitoring data of intake and exhaust pressure, flow rate and efficiency under near-stall state of compressor, and makes a comprehensive judgment on whether the high-frequency oscillation of efficiency signal is caused by rotating stall cluster through time domain analysis of high-frequency periodic oscillation and frequency domain analysis of characteristic peaks.

[0119] Dual-band segmentation module: If the determination is yes, the stall cluster oscillation characteristic frequency band is calculated by the real-time rotor speed of the compressor and the propagation coefficient of the rotating stall cluster, and the low-frequency characteristic frequency band of efficiency decay is calibrated to achieve dual-band segmentation;

[0120] Dual-band decomposition module: Acquires compressor efficiency oscillation + decay composite signal, performs wavelet packet decomposition according to the divided dual-band, including narrow window decomposition for the oscillation band and wide window decomposition for the decay band;

[0121] Efficiency decay reconstruction module: Removes the oscillation characteristic frequency band signal after decomposition, retains only the efficiency decay characteristic frequency band signal, and reconstructs the pure efficiency decay trend curve without oscillation interference through spline interpolation + moving average;

[0122] Efficiency degradation analysis module: Based on the pure efficiency degradation trend curve, it performs fitting calculation and analysis of linear / nonlinear degradation rates, and combines the shape analysis of the pure efficiency degradation trend curve to determine the compressor efficiency degradation trend.

[0123] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A compressor efficiency evaluation method based on decline trend prediction, characterized in that: Includes the following steps: Step S10: Collect real-time monitoring data of intake and exhaust pressure, flow rate and efficiency of the compressor under near stall conditions, and judge whether the high-frequency oscillation of efficiency signal is caused by the rotating stall cluster through time domain analysis of high-frequency periodic oscillation and frequency domain analysis of characteristic peaks. Step S20: If the determination is yes, the stall cluster oscillation characteristic frequency band is calculated by the real-time rotor speed of the compressor and the propagation coefficient of the rotating stall cluster, and the low-frequency characteristic frequency band of efficiency decay is calibrated to achieve dual-frequency band division; Step S30: Collect the compressor efficiency oscillation + decay composite signal, and perform wavelet packet decomposition according to the divided dual frequency bands, including narrow window width decomposition for the oscillation frequency band and wide window width decomposition for the decay frequency band. Step S40: Remove the oscillation characteristic frequency band signal after decomposition, retain only the efficiency degradation characteristic frequency band signal, and reconstruct the pure efficiency degradation trend curve without oscillation interference through spline interpolation + moving average. The process of reconstructing the pure efficiency decline trend curve without oscillation interference is as follows: Perform inverse wavelet packet transform on the wavelet packet coefficients of the retained low-frequency characteristic band of efficiency degradation to restore them to the time domain signal and obtain the degradation signal sequence; A cubic spline interpolation algorithm is used to process the signal gap. The moving average processing is performed on the interpolated and completed decay signal sequence. The smoothed decay signal sequence is sorted by timestamp to construct a pure efficiency decay trend curve with the horizontal axis being the running time and the vertical axis being the efficiency value. Step S50: Based on the pure efficiency decline trend curve, perform fitting calculation and analysis of linear / nonlinear decline rate, and combine with the shape analysis of the pure efficiency decline trend curve to determine the compressor efficiency decline trend. The process for determining the compressor efficiency decline trend is as follows: Based on the pure efficiency decline trend curve, a standardized dataset was constructed with running time and efficiency value. The least squares method was used to fit linear, exponential and power function models respectively, and the goodness of fit of each model was calculated. Based on the fitting results, the linear decay rate is extracted and converted into a decay rate per thousand hours, or the instantaneous rate change is analyzed by taking the derivative of the nonlinear model. The overall shape characteristics of the pure efficiency decline trend curve are used as an auxiliary factor for judgment. By comparing the goodness of fit, quantifying the decline rate, and analyzing the curve shape, the specific category of efficiency decline can be determined as linear slow decline, nonlinear accelerated decline, or nonlinear steady decline.

2. The compressor efficiency evaluation method based on decline trend prediction according to claim 1, characterized in that: The process of collecting real-time monitoring data on compressor inlet and outlet pressure, flow rate, and efficiency under near-stall conditions is as follows: Under near-stall conditions and when the anti-surge system is not triggered, data acquisition is triggered. The data format includes timestamp, intake pressure, exhaust pressure, intake flow rate, intake temperature, and exhaust temperature. Outliers are removed using the 3σ criterion, environmental random noise is filtered out using a 5th-order Butterworth low-pass filter, and the real-time efficiency value is obtained by isentropic efficiency calculation.

3. The compressor efficiency evaluation method based on decline trend prediction according to claim 2, characterized in that: The time-domain analysis process for high-frequency periodic oscillations is as follows: The collected pressure, flow, and efficiency data are divided into data segments, and the fluctuation amplitude, fluctuation coefficient, and oscillation period consistency of each data segment are calculated. When the fluctuation amplitude and fluctuation coefficient of the pressure, flow, and efficiency signals all exceed the preset threshold and the oscillation period deviation rate is not greater than 10%, the time domain analysis is deemed to have passed. Otherwise, if any indicator does not meet the requirements, the analysis is deemed to have failed.

4. The compressor efficiency evaluation method based on decline trend prediction according to claim 3, characterized in that: The frequency domain analysis process of the characteristic spikes is as follows: Fast Fourier Transform is performed on the segmented data to extract the characteristic frequency range related to the rotor speed and the propagation coefficient of the stall cluster. If there is a significant peak in the characteristic frequency range and the peak amplitude ratio is greater than or equal to 5, and the peak frequency deviation of multiple consecutive data segments does not exceed 1Hz, then the frequency domain analysis is considered to be passed. Otherwise, if any indicator is not met, it is considered to be failed. Only when both the time domain and frequency domain judgments are passed is it determined that the high-frequency oscillation of the efficiency signal is caused by the rotating stall cluster.

5. The compressor efficiency evaluation method based on decline trend prediction according to claim 1, characterized in that: The process of dividing the dual-band is as follows: The real-time rotor speed of the compressor is obtained, and the theoretical oscillation frequency is calculated by combining the propagation coefficient of the rotating stall cluster. The oscillation characteristic frequency band is set with the theoretical oscillation frequency as the center, and its width covers the frequency fluctuation range. At the same time, the low-frequency characteristic frequency band is calibrated according to the slow variation characteristics of efficiency decay.

6. The compressor efficiency evaluation method based on decline trend prediction according to claim 5, characterized in that: The process of performing wavelet packet decomposition is as follows: Determination of wavelet packet fundamental and decomposition level: The db4 wavelet packet is selected as the fundamental. The number of decomposition levels is calculated based on the dual-band characteristics. Wavelet packet decomposition is performed on the composite signal, and differentiated window widths are set for the two frequency bands. The stall clique oscillation characteristic frequency band uses a narrow window width, while the efficiency degradation low-frequency characteristic frequency band uses a wide window width. After decomposition, the wavelet packet coefficients of the corresponding stall clique oscillation characteristic frequency band and efficiency degradation low-frequency characteristic frequency band are extracted to form two independent frequency band signals.

7. The compressor efficiency evaluation method based on decline trend prediction according to claim 1, characterized in that: The process of comprehensively determining whether efficiency decline belongs to a specific category of linear slow decline, nonlinear accelerated decline, or nonlinear steady decline is as follows: Linear slow decay determination: The linear fitting model has the highest goodness of fit and is not lower than 0.

85. The calculated decay rate per thousand hours is less than 0.1% and the fluctuation range is limited. At the same time, the pure efficiency decay trend curve is linear, and the slope deviation and fitting residual of each period do not exceed the preset threshold. Nonlinear accelerated decay determination: The goodness of fit of the exponential fitting model is the highest and not less than 0.

85. The calculated average decay rate per thousand hours is not less than 0.1% and shows a continuous increasing characteristic. At the same time, the pure efficiency decay trend curve shows a concave shape. Nonlinear steady-state decay determination: The power function fitting model has the highest goodness of fit and is not lower than 0.

85. The calculated average decay rate per thousand hours gradually decreases and approaches zero. At the same time, the pure efficiency decay trend curve shows an upward convex shape and the terminal segment tends to be flat.

8. A compressor efficiency evaluation system based on decline trend prediction, characterized in that: Includes the following modules: High-frequency oscillation determination module: Collects real-time monitoring data of intake and exhaust pressure, flow rate and efficiency under near-stall state of compressor, and makes a comprehensive judgment on whether the high-frequency oscillation of efficiency signal is caused by rotating stall cluster through time domain analysis of high-frequency periodic oscillation and frequency domain analysis of characteristic peaks. Dual-band segmentation module: If the determination is yes, the stall cluster oscillation characteristic frequency band is calculated by the real-time rotor speed of the compressor and the propagation coefficient of the rotating stall cluster, and the low-frequency characteristic frequency band of efficiency decay is calibrated to achieve dual-band segmentation; Dual-band decomposition module: Acquires compressor efficiency oscillation + decay composite signal, performs wavelet packet decomposition according to the divided dual-band, including narrow window decomposition for the oscillation band and wide window decomposition for the decay band; Efficiency decay reconstruction module: Removes the oscillation characteristic frequency band signal after decomposition, retains only the efficiency decay characteristic frequency band signal, and reconstructs the pure efficiency decay trend curve without oscillation interference through spline interpolation + moving average; The process of reconstructing the pure efficiency decline trend curve without oscillation interference is as follows: Perform inverse wavelet packet transform on the wavelet packet coefficients of the retained low-frequency characteristic band of efficiency degradation to restore them to the time domain signal and obtain the degradation signal sequence; A cubic spline interpolation algorithm is used to process the signal gap. The moving average processing is performed on the interpolated and completed decay signal sequence. The smoothed decay signal sequence is sorted by timestamp to construct a pure efficiency decay trend curve with the horizontal axis being the running time and the vertical axis being the efficiency value. Efficiency degradation analysis module: Based on the pure efficiency degradation trend curve, it performs fitting calculation and analysis of linear / nonlinear degradation rate, and combines the shape analysis of the pure efficiency degradation trend curve to determine the compressor efficiency degradation trend. The process for determining the compressor efficiency decline trend is as follows: Based on the pure efficiency decline trend curve, a standardized dataset was constructed with running time and efficiency value. The least squares method was used to fit linear, exponential and power function models respectively, and the goodness of fit of each model was calculated. Based on the fitting results, the linear decay rate is extracted and converted into a decay rate per thousand hours, or the instantaneous rate change is analyzed by taking the derivative of the nonlinear model. The overall shape characteristics of the pure efficiency decline trend curve are used as an auxiliary factor for judgment. By comparing the goodness of fit, quantifying the decline rate, and analyzing the curve shape, the specific category of efficiency decline can be determined as linear slow decline, nonlinear accelerated decline, or nonlinear steady decline.

Images