Compressor cylinder vibration monitoring method based on angle period and polar coordinates

Abstract

The application discloses a compressor cylinder vibration monitoring method based on angle period and polar coordinates, relates to the field of intelligent monitoring of reciprocating mechanical operation states, and comprises the following steps: collecting cylinder vibration signals, resampling the signals according to the crankshaft angle to obtain angle domain signals, extracting impact representation quantities, normalizing the impact representation quantities by using historical references, constructing polar coordinate impact points and working condition boundary sets, locking the impact points to adhere to the boundary according to the angle around difference, combining the boundary influence half-width to generate dimensionless boundary development coordinates, screening and constructing an effective boundary development impact point set, taking the boundary development longitudinal coordinates as weights, calculating spatial weighted centroids to generate tongue ridge offsets, calculating third-order spatial skewness to generate tongue bending amounts, judging event types according to the tongue ridge offsets and the adhered boundary, combining average impact intensity and historical quantile probability to generate alarm levels, and outputting abnormal distribution zone monitoring results. The application overcomes the attribution jump caused by the phase migration of gas valves under variable working conditions, and improves the discrimination accuracy.

Description

Technical Field

[0001] This invention relates to the field of intelligent monitoring of the operating status of reciprocating machinery, specifically to a method for monitoring the vibration of compressor cylinders based on angle period and polar coordinates. Background Technology

[0002] Reciprocating compressors operate under continuous high loads in industrial plants such as petrochemical, coal chemical, and natural gas booster systems. The cylinder surface vibration signals generated during operation exhibit significant periodicity and impact, and are highly correlated with the current operating conditions. The high-frequency impacts on the cylinder surface are primarily caused by the thermal processes of gas compression and expansion within the cylinder, piston reversal impacts, and the opening and closing actions of the intake and exhaust valves. These impacts are mapped onto the time axis to the mechanical angle of the crankshaft and the piston stroke. To intuitively monitor the compressor's operating status, existing polar coordinate monitoring methods acquire crankshaft speed signals, using a crankshaft mechanical angle period from 0 to 360 degrees, converting the time-domain cylinder vibration waveform into an angle-domain waveform, and presenting the impact intensity values ​​on a polar coordinate graph. The conventional approach is to pre-define four static stroke sectors—expansion, intake, compression, and exhaust—within the polar coordinate graph, and simultaneously define the theoretical opening and closing boundary lines of the intake and exhaust valves, thereby assigning various vibration impact points accordingly.

[0003] The existing fixed-angle zone monitoring system is based on the assumption that the actual operating angle of the valve perfectly matches the preset theoretical boundary. In the online operation scenario of the compressor, the equipment is often in a state of variable operating condition adjustment. In addition, compressors mostly use self-acting valves that rely on the pressure difference between inside and outside the cylinder to automatically open and close. The process of building up and releasing the cylinder pressure is easily affected by the combined interference of multiple dynamic factors such as load rate adjustment, pipeline pressure fluctuation, changes in fluid medium composition, valve wear, and spring stiffness decay. The above dynamic factors will cause the actual opening and closing action of the intake and exhaust valves to deviate from the design theoretical boundary, resulting in premature or delayed dynamic phase migration. When the high-energy impact point triggered by the valve action drifts on the polar coordinate diagram, the generated vibration energy band is very likely to adhere to or even cross the preset static stroke boundary line. If the static sector standard is still used to forcibly divide the impact point at this time, it will directly sever the continuous boundary impact energy, causing serious attribution deviation. The static zoning judgment mechanism makes it easy for the phase drift caused by normal variable operating conditions to be misjudged as mechanical failures such as valve leakage, and the actual early abnormal valve impact will also be missed because it is incorrectly classified into the non-corresponding operating condition area. Phase migration across the boundary of the same cylinder block under different loads will trigger disordered cross alarms in adjacent operating zones, ultimately causing frequent alarm level jumps and distorted diagnostic results. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention proposes a compressor cylinder vibration monitoring method based on angle period and polar coordinates. This method solves the problems caused by the dynamic migration of valve action phase under varying operating conditions, which leads to the fragmentation of boundary impact energy by static sectors, resulting in incorrect polar coordinate impact point attribution, frequent alarm level jumps, and missed or misjudged early mechanical faults.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] The original cylinder block vibration signal is resampled based on the original crankshaft angle reference signal to obtain the angular domain vibration signal;

[0007] The impact characterization quantity represented by the analysis window within the angular domain vibration signal is extracted. The impact characterization quantity represented by the analysis window is normalized by combining the historical health benchmark impact characterization quantity to generate the normalized impact radius. The polar coordinate impact point set and the working condition boundary set are then constructed.

[0008] Based on the minimum angular circumference difference, the set of polar coordinate impact points is determined to belong to the attached boundary in the set of working condition boundaries. The directional angular offset relative to the attached boundary is calculated. The circumferential angular distance between the adjacent boundary corresponding to the attached boundary is extracted to generate the boundary influence half-width. The dimensionless boundary expansion abscissa is generated by combining the directional angular offset and the boundary influence half-width. The normalized impact radius is assigned as the boundary expansion ordinate to construct the expansion coordinate points. The expansion coordinate points that meet the impact enhancement conditions and reach the preset number threshold are selected to construct the effective boundary expansion impact point set.

[0009] Based on the effective boundary, the set of impact points is expanded. Using the boundary expansion ordinate as the weight, the spatial weighted centroid is calculated to generate the tongue-shaped offset, the third-order spatial skewness is calculated to generate the tongue-shaped bending amount, and the mean of the boundary expansion ordinate is calculated to generate the average normalized impact intensity.

[0010] Based on the tongue ridge offset and attachment boundary, the boundary event type is generated. The boundary event alarm quantity is calculated by combining the tongue bending amount and the average normalized impact intensity. The historical health alarm quantity sample set is compared to generate the historical ranked normalized quantile probability, and the final monitoring result is output.

[0011] Furthermore, a discrete angle sampling point sequence is constructed based on a preset angle resolution, and an interpolation algorithm is used to convert the original time-domain cylinder vibration signal into an angle-domain vibration signal with an aligned angle sampling point sequence, thus completing the resampling.

[0012] An angle calculation window is established with each angle sampling point in the angle sampling point sequence as the center. The root mean square value of the vibration signal under the angle calculation window within a single cycle is calculated to generate a single-cycle impact characterization quantity. The median value of the single-cycle impact characterization quantity at the same angle position within multiple consecutive mechanical cycles is extracted to generate an analysis window representing the impact characterization quantity.

[0013] Furthermore, a two-dimensional operating condition matrix is ​​constructed by dividing the full range of load rate and pressure ratio into continuous operating interval matrices. The operating condition category is defined by discretizing the corresponding grids of the two-dimensional operating condition matrix according to the actual load rate and pressure ratio falling into the matrix.

[0014] When the actual load rate change rate or pressure ratio change rate is detected to exceed the preset threshold and enter a transient switch, the normalization process is paused.

[0015] When there is no corresponding historical health benchmark impact characterization quantity for a working condition category, the historical health benchmark impact characterization quantity corresponding to the most matching adjacent working condition category in the coordinate neighborhood of the two-dimensional working condition matrix is ​​called and the substitution operation is performed.

[0016] Furthermore, the working condition boundary of the stroke and the theoretical operating line of the valve are integrated and compared.

[0017] When the distance between the theoretical action line of the valve and the circumferential angle of the boundary of the adjacent stroke is less than or equal to twice the preset angle resolution, the midpoint of the circumferential angle of the theoretical action line of the valve and the boundary of the adjacent stroke is extracted to form a composite boundary.

[0018] By using composite boundaries to replace the valve's theoretical action line and adjacent stroke working condition boundaries corresponding to the conflict, the composite boundaries and independent boundaries are incorporated into the working condition boundary set.

[0019] Furthermore, the polar angles of each polar impact point within the set of polar impact points are extracted, and the original angle difference between the polar angle of the polar impact point and each working condition boundary within the set of working condition boundaries is calculated.

[0020] If the original angle difference is greater than 180°, perform a conversion operation by subtracting 360° from the original angle difference.

[0021] If the original angle difference is less than or equal to -180°, perform a 360° conversion operation on the original angle difference;

[0022] Based on the conversion operation results, the absolute value of the angle surrounding difference is obtained, a global traversal comparison is performed, and the working condition boundary corresponding to the minimum absolute value of the angle surrounding difference is selected as the attachment boundary to which the polar coordinate impact point uniquely belongs.

[0023] When the absolute value of the circumferential difference between the polar coordinate impact point and the two adjacent working condition boundaries is equal, the front boundary in the circumferential direction is determined as the unique attachment boundary to which the polar coordinate impact point belongs.

[0024] Furthermore, a numerical traversal is performed on all unfolded coordinate points belonging to the same attachment boundary, removing background baseline points whose unfolded ordinate is less than 1, and retaining unfolded coordinate points whose unfolded ordinate is greater than or equal to 1.

[0025] When the number of retained unfolded coordinate points is greater than or equal to the preset number threshold, the retained unfolded coordinate points will be aggregated to construct a set of effective boundary unfolded impact points corresponding to the attachment boundary;

[0026] When the number of retained unfolded coordinate points is less than the preset threshold, the set corresponding to the attached boundary is marked as an empty set, thus preventing the empty set from participating in the calculation of tongue-shaped morphological parameters.

[0027] Furthermore, extract the dimensionless boundary expansion abscissa and boundary expansion ordinate within the effective boundary expansion impact point set corresponding to the attached boundary;

[0028] Calculate the sum of the products of the dimensionless boundary expansion x-coordinate and the boundary expansion y-coordinate within the effective boundary expansion impact point set, and calculate the sum of the y-coordinates of all boundary expansions within the effective boundary expansion impact point set;

[0029] The tongue ridge offset is generated by dividing the sum of the products of the x-coordinates and y-coordinates of each dimensionless boundary expansion by the sum of the y-coordinates of all boundary expansions.

[0030] Furthermore, the difference between the dimensionless boundary expansion abscissa and the tongue ridge offset of each dimensionless boundary expansion impact point set within the effective boundary expansion is calculated, and the cube of each difference is used to generate the corresponding deviation term.

[0031] Multiply the deviation term by the corresponding boundary's ordinate and perform a global summation to generate the summation value;

[0032] Divide the summed value by the sum of the ordinates of all boundaries within the effective boundary expansion impact point set to generate the tongue-shaped bending amount.

[0033] Furthermore, the positive or negative value of the tongue ridge offset is determined. When the tongue ridge offset is greater than zero, it is determined to extend outward to the working area on the positive side of the attachment boundary. When the tongue ridge offset is less than zero, it is determined to extend outward to the working area on the negative side of the attachment boundary.

[0034] When the attachment boundary is the dividing line of the stroke working condition, the extension to the positive side of the attachment boundary working condition area is defined as the impact event of the beginning side boundary of the adjacent subsequent working condition, and a boundary event type is generated. The extension to the negative side of the attachment boundary working condition area is defined as the impact event of the end side boundary of the adjacent preceding working condition, and a boundary event type is generated.

[0035] When the attachment boundary is the theoretical action line of the valve, the outward extension of the working condition area on the positive side of the attachment boundary is defined as the valve action rear boundary impact event, and a boundary event type is generated. The outward extension of the working condition area on the negative side of the attachment boundary is defined as the valve action front boundary impact event, and a boundary event type is generated.

[0036] When the lingual ridge bias is zero, no unilateral abduction boundary events are generated.

[0037] Furthermore, the absolute value of the tongue bending amount is extracted, and the absolute value of the tongue bending amount is multiplied by the average normalized impact intensity to generate the boundary event alarm amount;

[0038] Retrieve the historical health alarm sample set corresponding to the boundary event type, count the number of samples in the historical health alarm sample set that are less than or equal to the boundary event alarm, divide the number of samples that are less than or equal to the boundary event alarm by the total number of historical health alarm samples, and generate the historical sorted normalized quantile probability.

[0039] The historical sorted normalized quantile probability is divided into a preset continuous warning interval to generate alarm levels. The boundary event type, boundary event alarm quantity, historical sorted normalized quantile probability, alarm level, attachment boundary and tongue ridge offset are summarized to generate the final monitoring results.

[0040] On the polar coordinate graph, based on the geometric position of the attached boundary, the abnormal distribution zone corresponding to the set of impact points of the effective boundary is displayed using highlighted graphics, and the boundary event type and alarm level are marked simultaneously.

[0041] Compared with existing technologies, it has the following advantages:

[0042] This proposed method for monitoring compressor cylinder vibration based on angle period and polar coordinates changes the underlying judgment logic of traditional fixed sector division using polar coordinates, establishing a morphological topology reconstruction mechanism that dynamically anchors the impact point in polar coordinates to the operating condition boundary. When the compressor is under complex and variable operating conditions and the self-acting valve's opening and closing action causes dynamic phase shifts, conventional static boundary lines easily sever continuous vibration impact energy bands, leading to incorrect attribution judgments. This application uses the absolute value of the angle surrounding difference to lock the unique attribution boundary of the target, and combines the boundary influence half-width to perform dimensionless measurement transformation, reconstructing the entire impact point set into a local rectangular coordinate space unfolding around the boundary. This transformation mechanism overcomes the interference caused by the mechanical angle differences of multiple cylinders and equipment load fluctuations, eliminates the attribution jumps and disordered alarms caused by vibration energy crossing the theoretical boundary line under extreme operating conditions, and establishes a solid judgment foundation adapted to dynamic operating conditions.

[0043] To address the challenge of distinguishing between early mechanical anomalies and normal operating condition fluctuations, this solution introduces a high-order spatial morphology operator for in-depth analysis of boundary anomaly distribution zones. Conventional second-order variance can only characterize the symmetry and dispersion of data distribution, failing to identify valve leaks or abnormal impacts with a clear unidirectional extension tendency. This application uses the boundary expansion ordinate as a weighting parameter to calculate the spatial weighted centroid, generating a tongue-shaped offset, and extracts the third-order spatial skewness to generate the tongue-shaped bending amount. The aforementioned operator accurately quantifies the unilateral outward tailing evolution trend of impact energy, multiplying it positively with the average normalized impact intensity to construct a comprehensive alarm value, which is then used for adaptive rating based on historical quantile probabilities using nonparametric statistics. The entire calculation logic deeply integrates the geometric asymmetric extension properties of the impact zone with the amplitude energy properties, effectively eliminating fluid pulsation disturbances and achieving highly sensitive identification of hidden anomalies such as valve plate damage or valve malfunction. Attached Figure Description

[0044] Figure 1 This is a schematic diagram of the method flow of the present invention. Detailed Implementation

[0045] 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 embodiments of the present invention, and not all embodiments. 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.

[0046] Please see Figure 1 This application provides a method for monitoring compressor cylinder vibration based on angle period and polar coordinates;

[0047] The method specifically includes the following steps:

[0048] Step 1: The data acquisition module synchronously acquires the original cylinder block vibration signal and the original crankshaft angle reference signal. The data processing unit performs angle resampling according to the preset angle resolution to obtain the angle domain vibration signal.

[0049] Specifically, the data processing unit sets the starting position of the original crankshaft angle reference signal to the top dead center of the piston in the specified cylinder block or the key phase reference point. Based on a preset angle resolution, for example, between 0.5° and 2°, it calculates the total number of angle sampling points within the mechanical cycle, i.e., dividing a complete 360° mechanical circumference angle by this angle resolution, thereby constructing a discrete sequence of angle sampling points. When the original sampling time is not aligned with a given angle sampling point, a linear interpolation algorithm is used to convert the original cylinder block vibration signal in the time domain into an angle domain vibration signal for the target mechanical cycle. By introducing a mechanical phase reference to perform equal-angle resampling, the distortion of the time domain waveform period length caused by dynamic fluctuations in rotational speed is eliminated, ensuring that vibration signals at different speeds are uniformly aligned to the same mechanical coordinate axis.

[0050] It should be noted that the linear interpolation algorithm is used only when the original sampling frequency of the cylinder vibration sensor is much higher than the compressor's mechanical operating fundamental frequency, and when at least one valid original sampling point is included within a time interval corresponding to one angular resolution, to prevent high-frequency impact signal aliasing or loss. When pulse loss or sensor disconnection is detected, causing the reference point interval to deviate from the theoretical value of the current speed by more than a preset tolerance, such as exceeding 10% of the theoretical interval period, the unit will discard the resampling operation for that abnormal period. In addition, in other embodiments, cubic spline interpolation or Sinc interpolation algorithms can also be used for resampling to further preserve the peak characteristics of high-frequency impacts.

[0051] The data processing unit establishes an angle calculation window centered on each angle sampling point, extracts the single-cycle impact characterization quantity reflecting the local energy intensity, and combines the statistical median value of multiple cycles to generate an analysis window representative impact characterization quantity.

[0052] Specifically, an angle calculation window spanning a preset number of sampling points is established centered on the current angle sampling point. The root mean square (RMS) value of the vibration signal within this angle calculation window is calculated within a single cycle and used as the single-cycle impact characterization quantity. Subsequently, for a preset analysis window length, such as 3 to 10 consecutive mechanical cycles, the statistical median value of multiple consecutive single-cycle impact characterization quantities at the same angle position is extracted to generate the representative impact characterization quantity of the analysis window. Extracting the RMS value of energy in local angle intervals and combining it with median filtering effectively filters out non-mechanically regular random high-frequency noise while retaining the continuous impact characteristics generated by valve seating and piston impact.

[0053] The data processing unit determines the operating condition category of the current analysis window in real time, retrieves the matching historical health benchmark impact characterization quantity, normalizes the impact characterization quantity representing the analysis window, and obtains the normalized impact radius.

[0054] Specifically, the operating condition category is discretized based on the current actual load rate and pressure ratio of the compressor. The historical health benchmark impact characteristic quantity matching this operating condition category is then used to calculate the normalized impact radius using the following formula:

[0055]

[0056] In the formula, Indicates the corresponding angle sampling point Normalized impact radius at the location; The analysis window represents the impact characteristic quantity; u represents the working condition category defined by the discretization of the actual load rate and pressure ratio; This represents the historical health baseline impact characterization extracted under operating condition category u. The calculation logic refines the baseline to specific phase points and operating condition categories, separating baseline fluctuations caused by normal fluid pulsation and valve opening / closing under different production loads. The output normalized impact radius is a dimensionless parameter used to map the relative enhancement factor of abnormal impacts.

[0057] It should be noted that, to prevent the program from diverging due to a zero denominator, when the historical health baseline impact characterization at a certain angle sampling point is identified as equal to zero or lower than the preset system background noise threshold, the data processing unit replaces the historical health baseline impact characterization at that angle sampling point with a preset minimal positive bias term before performing the above division operation. The dimension of this minimal positive bias term is consistent with the impact characterization, and its value is set to a minimal multiple of the median value of the impact characterization of the entire healthy sample at all angles, such as ten to the power of negative six. This conditional substitution logic effectively suppresses abnormal calculation amplification in mechanical dead zones or areas of weak background vibration.

[0058] It should be noted that the discretization method for defining operating condition categories involves pre-obtaining the full range of load rate and pressure ratio during normal operation and dividing them into a continuous operating interval matrix. For example, the load rate can be divided in 10% increments, and the pressure ratio in 0.5 increments. The grid where the actual operating parameters fall corresponds to a uniquely determined operating condition category. When the load rate or pressure ratio fluctuation rate exceeds a preset threshold and enters a transient switch, the normalization calculation is paused or the previous steady-state benchmark is used. During the suspension period, the system can output a transient operation indicator and disable alarm logic to prevent false alarms. When it is identified that there is no corresponding historical health benchmark impact characteristic quantity for the current operating condition category, the data processing unit will pause the normalization calculation of the current analysis window, trigger the benchmark self-learning program for the new operating condition category, or temporarily call the health benchmark of the adjacent operating condition category with the best matching coordinate neighborhood in the two-dimensional operating condition matrix for substitution calculation until the high-quality health data segment for the new operating condition is collected.

[0059] It should also be noted that the historical health baseline impact characterization values ​​called upon need to be extracted and statistically generated from high-quality data segments that have been confirmed as fault-free and have been running continuously and stably under the corresponding operating condition category for more than a preset time, in order to prevent the risk of early minor faults being absorbed and masked as normal baselines.

[0060] The data processing unit constructs a set of polar coordinate impact points by combining the angle sampling points and the normalized impact radius, and integrates the four preset stroke working condition boundaries and the four theoretical action lines of the air valves to construct a set of working condition boundaries.

[0061] Specifically, the angle sampling point is set as the polar angle, and the dimensionless normalized impact radius is set as the polar diameter, mapped to a two-dimensional polar coordinate impact point, thus constructing a set of polar coordinate impact points within the current analysis window. Simultaneously, the four pre-set stroke condition boundaries in the equipment design parameters—expansion and intake boundary, intake and compression boundary, compression and exhaust boundary, and exhaust and expansion boundary—as well as the four theoretical valve action lines—intake valve opening and closing, and exhaust valve opening and closing—are integrated and compared. If the circumferential angular distance between any theoretical valve action line and the adjacent stroke condition boundary is determined to be less than or equal to twice the angular resolution, the midpoint of their circumferential angles is taken to synthesize a unique composite boundary, which replaces the two conflicting boundaries. All independent and composite boundaries are ultimately incorporated into the condition boundary set. The midpoint fusion mechanism of adjacent boundaries eliminates the risk of repeated jumps and incorrect attribution of the same abnormal incremental impact point between two extremely close theoretical boundaries under extreme conditions, accurately segmenting the polar coordinate view.

[0062] In practical engineering applications, let's take a specific data extrapolation process as an example. Assuming a set angular resolution of 1°, one mechanical cycle is divided into 360 angular sampling points. The data acquisition module continuously collects vibration data for five mechanical cycles to construct an analysis window. At the 180° angular sampling point, the data processing unit extracts an angular calculation window from 178° to 182°, calculating the root mean square values ​​of the vibration signals within this window for the five cycles as 2.1, 2.3, 10.5, 2.2, and 2.4, respectively, where 10.5 represents random noise caused by occasional fluid pulses. After median smoothing, the impact characterization value at this 180° position is effectively filtered out of random noise and determined to be 2.3. If the current equipment is operating under full load, the historical health baseline impact characterization value at 180° under this condition is retrieved as 1.0, and the normalized impact radius is calculated as 2.3 through ratio calculation. The dimensionless numerical value is ultimately transformed into two-dimensional polar coordinate feature points with spatial geometric properties, which are then input into the subsequent boundary expansion calculation.

[0063] Step 2: The data processing unit traverses the set of polar coordinate impact points and the set of working condition boundaries, calculates the angular distance between each polar coordinate impact point and the attachment boundary, and then determines the attachment boundary to which each polar coordinate impact point belongs.

[0064] Specifically, the data processing unit introduces angle surround difference conversion logic to calculate the minimum directed difference between any two angles. It extracts the polar angles of each polar impact point within the set of polar impact points and calculates the original angle difference between the polar angle of the polar impact point and each working boundary within the set of working boundary conditions. If the original angle difference is greater than 180°, it subtracts 360°; if the original angle difference is less than or equal to -180°, it adds 360°, thus limiting the angle difference to between -180° and +180°. Based on the conversion result, it obtains the absolute value of the angle surround difference, performs a global traversal comparison, and selects the working boundary corresponding to the smallest absolute value of the angle surround difference, determining it as the unique attachment boundary to which the polar impact point belongs.

[0065] It should be noted that the introduced angular circumference difference conversion logic bridges the numerical discontinuity at the boundary between 360° and 0° of the mechanical circumference, preventing adjacent impact points with the same mechanical phase angle from being incorrectly classified due to periodic abrupt changes in their mathematical expression. Furthermore, when the absolute value of the angular circumference difference between a polar coordinate impact point and two adjacent working condition boundaries is equal, it is classified according to preset rules to the boundary located ahead of it in the circumferential direction, thus avoiding program anomalies caused by classification jumps.

[0066] The data processing unit calculates the directional angular offset of each polar coordinate impact point relative to its corresponding attached boundary, and performs dimensionless processing in conjunction with the boundary influence half-width to generate dimensionless boundary unfolded abscissa and boundary unfolded ordinate.

[0067] Specifically, the direction along which the crankshaft angle increases is defined as the positive side of the boundary, and the direction along which the crankshaft angle decreases is defined as the negative side of the boundary. For any polar coordinate impact point, the angle difference between its polar angle and its attachment boundary is calculated using the aforementioned angle surround difference conversion logic, and this value is taken as the directional angular offset. A directional angular offset value greater than zero indicates that it is located on the positive side of the attachment boundary, less than zero indicates that it is located on the negative side of the attachment boundary, and equal to zero indicates that it falls exactly on the attachment boundary line.

[0068] Subsequently, all boundaries in the set of working condition boundaries are pre-sorted into a circumferential sequence according to their angle values ​​from smallest to largest. The two adjacent boundaries in the circumferential direction of the currently attached boundary are extracted, and the circumferential angular distances from the current attached boundary to the adjacent boundaries are calculated. Half of the smaller angle value is selected as the boundary influence half-width. The directional angular offset is divided by this boundary influence half-width to obtain the dimensionless boundary unfolded abscissa. Since the dimension of the boundary influence half-width is consistent with the directional angular offset, this division operation effectively eliminates the angular dimension. The normalized impact radius corresponding to the polar coordinate impact point is directly assigned as the boundary unfolded ordinate, which inherits the dimensionless property of the normalized impact radius.

[0069] It should be noted that by constraining the upper limit of the available expansion space of each boundary to half the shortest distance between adjacent boundaries, the logical loophole of repeatedly reading and triggering a second alarm for the same impact anomaly point on two adjacent boundaries is avoided. Dimensionless processing eliminates the interference of differences in the mechanical angles between stroke intervals of a multi-cylinder compressor on subsequent geometric reconstruction, ensuring that all valve boundary events are in a unified mathematical measurement space.

[0070] The data processing unit removes background noise points and discrete invalid points, and collects the expanded coordinate points that meet the impact enhancement conditions and reach the calculation quantity threshold into a set of effective boundary expanded impact points.

[0071] Specifically, a numerical traversal is performed on all unfolded coordinate points belonging to the same attachment boundary. Background baseline points with an unfolded boundary ordinate less than 1 are removed, and only coordinate points with an unfolded boundary ordinate greater than or equal to 1 are retained. When the number of coordinate points satisfying this ordinate value condition is greater than or equal to a preset threshold, such as 3, preferably 5 to 15, these coordinate points are included in the set of effective boundary unfolding impact points corresponding to the attachment boundary.

[0072] It should be noted that the ordinate value of 1 for the boundary expansion represents the natural contour line after normalization of the prior health baseline. Coordinate points greater than or equal to 1 are extracted to characterize abnormal energy accumulation areas that show no energy decay relative to the health baseline but exhibit impact enhancement. The algorithm's requirement of at least 3 coordinate points is based on the fact that subsequent tongue-shaped feature recognition involves third-order spatial skewness calculations. In actual industrial implementation, 5 to 15 consecutive effective points are preferred to enhance the algorithm's robustness against single-point high-frequency noise. Discrete point sets below this threshold cannot mathematically and topologically support the reconstruction of a tongue-shaped outward morphology with a definite direction. When the number of qualifying coordinate points is less than 3, it is determined that there are no obvious abnormal outward events near the attachment boundary, and the corresponding set is marked as empty or dormant, no longer participating in subsequent tongue-shaped morphology parameter calculations.

[0073] Based on the coordinate transformation logic described above, the following is an example of boundary expansion for a specific feature point. Suppose there exists a polar impact point in the current analysis window's set of polar impact points, with a polar angle of 185° and a normalized impact radius of 2.0. After the data processing unit circularly sorts the set of working condition boundaries, it determines that the angle between the nearest intake and compression zone boundary is 180°, thus identifying its corresponding attachment boundary as 180°. The directional angular offset of this impact point relative to the 180° boundary is calculated as +5° using the angular difference logic. If the adjacent preceding boundary of this 180° boundary is 120° and the following boundary is 240°, the minimum angular distance to the adjacent boundary is 60°. Taking half of this value yields a corresponding boundary influence half-width of 30°. Through ratio calculation, the point's relative to the attachment boundary is converted to a dimensionless boundary expansion abscissa of approximately 0.167. Since its boundary expansion ordinate inherits the normalized impact radius of 2.0, the condition that this value is greater than 1 is met. The feature point is identified as a valid abnormal protrusion on the positive side of the boundary, i.e., the compression zone side, and is included in the set of valid boundary unfolding impact points corresponding to the 180° boundary, and input into the subsequent parameter extraction logic.

[0074] Step 3: The data processing unit extracts tongue-shaped morphological features from the impact point set for each effective boundary, and calculates the tongue ridge offset, tongue bending amount, and average normalized impact intensity in sequence.

[0075] Specifically, the dimensionless boundary expansion x-coordinates and y-coordinates are extracted from the set of impact points corresponding to the attached boundary expansion; the sum of the products of each dimensionless boundary expansion x-coordinate and y-coordinate in the set of impact points is calculated, and the sum of all boundary expansion y-coordinates in the set of impact points is calculated; the sum of the products of each dimensionless boundary expansion x-coordinate and y-coordinate is divided by the sum of all boundary expansion y-coordinates to generate the tongue spine offset.

[0076]

[0077] In the formula, This indicates the tongue ridge offset corresponding to the attachment boundary with boundary number j. The dimensionless x-coordinate of the effective boundary expansion impact point; The ordinate of the boundary expansion at the impact point represents the effective boundary expansion. This represents the set of effective boundary expansion impact points attached to the boundary. The calculation operation involves determining the weighted spatial centroid of this set of effective boundary expansion impact points, using the boundary expansion ordinate as a weighting factor. This ensures that feature points with higher impact intensity contribute more to the overall outward expansion direction of the anomaly distribution zone. A tongue-shaped offset value greater than zero indicates outward expansion towards the positive side of the boundary's working condition zone, while a value less than zero indicates outward expansion towards the negative side of the boundary's working condition zone.

[0078] Subsequently, the data processing unit calculates the tongue-shaped bending amount of the effective boundary unfolding impact point set based on the aforementioned tongue ridge offset.

[0079] Specifically, the difference between the dimensionless boundary expansion abscissa and the tongue-shaped offset of each dimensionless boundary expansion impact point set within the effective boundary expansion impact point set is calculated, and the cube of the difference is used to generate an deviation term; the deviation term is multiplied by the corresponding boundary expansion ordinate and a global summation is performed to generate an accumulated sum; the accumulated sum is divided by the sum of the ordinates of all boundary expansions within the effective boundary expansion impact point set to generate the tongue-shaped bending amount.

[0080]

[0081] In the formula, This represents the amount of tongue-shaped bending corresponding to the attachment boundary. The calculation logic is to obtain the third-order spatial skewness of the data distribution. The reason for setting the deviation term as the cube of the difference is that the second-order central moment or variance can only characterize the symmetrical discrete width of the impact distribution and cannot identify the unilateral tailing phenomenon with a clear direction; while the third-order spatial skewness can accurately quantify the asymmetry of the data distribution, thus matching the geometric outward extension characteristic of the impact distribution extending from the working condition boundary to a single adjacent working condition area.

[0082] Simultaneously, the data processing unit calculates the average normalized impact intensity of the set of impact points expanded from the effective boundary to quantify the overall energy level of the anomalous distribution zone.

[0083] Specifically, the sum of the ordinates of the expanded boundary coordinates of all valid boundary points in the set is calculated, and then divided by the total number of coordinate points in the set to generate the average normalized impact intensity.

[0084]

[0085] In the formula, This represents the average normalized impact intensity corresponding to the attachment boundary; This represents the total number of coordinate points included in the set of impact points expanded from the effective boundary. This calculation characterizes the overall energy level of the impact zone and is used in subsequent diagnostic logic to help identify pseudo-anomaly distributions that only have geometric expansion but extremely weak impact energy.

[0086] It should be noted that the algorithm using the boundary expansion ordinate as the core weighting coefficient is based on the fact that the dynamic drift of the valve action phase under varying operating conditions is often accompanied by the nonlinear release of fluid pressure, resulting in the core abnormal impact point having a significantly high energy characteristic. Energy weighting can effectively suppress the interference of residual low-amplitude background disturbances near the boundary on morphological reconstruction.

[0087] To verify the quantification effect of the above morphological features, the following discrete data points were used for calculation. Suppose the data processing unit obtains a set of effective boundary expansion impact points corresponding to a certain attachment boundary. This set contains three effective feature points, whose dimensionless boundary expansion x-coordinates and y-coordinates correspond to the following values: first point x-coordinate 0.1, y-coordinate 1.5; second point x-coordinate 0.2, y-coordinate 2.0; third point x-coordinate 0.3, y-coordinate 2.5. First, the sum of the y-coordinates of each point is calculated to be 6.0. The sum of the products of the x-coordinates and y-coordinates is 0.15 + 0.40 + 0.75, which equals 1.30. By performing a ratio calculation, dividing 1.30 by 6.0 yields a ridge offset of approximately 0.217. This positive value indicates that the impact zone extends outwards towards the positive side of the boundary. Subsequently, the difference between the x-coordinates of each point and the 0.217 tongue-shaped ridge offset is calculated. The cube of each difference is then multiplied by the corresponding y-coordinate, summed, and divided by the sum of the y-coordinates (6.0) to obtain the tongue-shaped bending amount reflecting the unilateral outward extension intensity. Finally, the sum of the y-coordinates of this set (6.0) is divided by the total number of coordinate points (3) to obtain the average normalized impact intensity of 2.0. These three quantitative parameters together constitute a complete feature vector describing the morphology and energy of this anomalous distribution zone, which is then input into the next level of boundary event determination logic.

[0088] Step 4: The data processing unit combines the direction symbols of the attachment boundary and the tongue offset to generate the boundary event type.

[0089] Specifically, the sign of the tongue offset value is determined. When the tongue offset is greater than zero, it is determined to extend outward into the working area on the positive side of the attachment boundary; when the tongue offset is less than zero, it is determined to extend outward into the working area on the negative side of the attachment boundary. The boundary event type is determined according to a preset boundary event mapping rule. The specific logic of this mapping rule is as follows: when the attachment boundary is the stroke working condition boundary line, extending outward into the working area on the positive side of the attachment boundary is defined as the boundary event type generated by the impact event on the beginning side boundary of the adjacent subsequent working condition, and extending outward into the working area on the negative side of the attachment boundary is defined as the boundary event type generated by the impact event on the end side boundary of the adjacent preceding working condition; when the attachment boundary is the theoretical valve action line, extending outward into the working area on the positive side of the attachment boundary is defined as the boundary event type generated by the impact event on the side boundary after valve action, and extending outward into the working area on the negative side of the attachment boundary is defined as the boundary event type generated by the impact event on the side boundary before valve action; when the tongue offset is equal to zero, no single-sided extension type boundary event is generated.

[0090] It should be noted that when the lingual ridge offset is equal to zero or its absolute value is less than the range of numerical calculation accuracy, it is determined that the impact band is attached to the boundary but has no clear unilateral outward direction. In this case, a unilateral outward boundary event is not generated to avoid misdiagnosis triggered by ordinary vibration waves without directionality.

[0091] The data processing unit calculates the boundary event alarm quantity based on the tongue bending amount and the average normalized impact intensity.

[0092] Specifically, the absolute value of the tongue-shaped bending amount is extracted and multiplied by the average normalized impact intensity to directly generate the boundary event alarm value:

[0093]

[0094] In the formula, Indicates the number of boundary event alarms; Indicates the absolute value of the tongue-shaped bending amount; This represents the average normalized impact intensity. The algorithm using this product form is based on the fact that the mechanical risk of boundary-attached anomaly distribution zones stems not only from the increase in local vibration intensity but also from their abnormal geometric outward projection morphology that crosses or adheres to the working condition boundary. By positively multiplying the parameter characterizing the spatial outward projection morphology with the parameter characterizing the energy intensity, a comprehensive dimensionless alarm quantity is constructed, overcoming the technical deficiency of traditional single-amplitude thresholds that cannot distinguish between normal variable working condition phase drift and actual mechanical faults.

[0095] The data processing unit compares the number of boundary event alarms with the historical health alarm sample set to generate the historical ranked normalized quantile probability.

[0096] Specifically, the system retrieves a sample set of historical health alarms corresponding to the boundary event type. It then counts the number of samples in the historical health alarm sample set corresponding to alarms less than or equal to the boundary event alarm count. This number is divided by the total number of historical health alarm samples to generate a historically ranked normalized quantile probability. This calculation logic determines the relative quantile ranking of the current boundary event alarm count within the historical health distribution.

[0097] It should be noted that the historical health alarm sample set is a baseline data set calculated using the same morphological feature extraction algorithm as described above, during a period when the compressor is confirmed to be in a fault-free state and operating continuously and stably. The number of health alarm samples for each boundary event type is set to no less than 100 analysis windows to ensure the confidence level of the nonparametric statistical distribution. If the number of historical samples is insufficient, the corresponding alarm level will be marked as insufficient baseline samples and the formal alarm output will be suspended.

[0098] The alarm output unit generates alarm levels based on the normalized quantile probability of historical sorting, and the visualization display unit outputs the final monitoring results.

[0099] Specifically, the alarm output unit divides the historical ranking normalized quantile probability into multiple consecutive warning intervals. In a preferred embodiment, a normal level is generated when the historical ranking normalized quantile probability is less than 0.80, a concern level is generated when it is greater than or equal to 0.80 and less than 0.90, a warning level is generated when it is greater than or equal to 0.90 and less than 0.97, and an alarm level is generated when it is greater than or equal to 0.97. The quantile threshold is set based on the abnormal distribution probability of nonparametric statistics, with 0.97 corresponding to the outlier extreme value in the top 3% of the historical healthy distribution. Boundary event type, boundary event alarm quantity, historical ranking normalized quantile probability, alarm level, attachment boundary and tongue ridge offset are summarized to generate the final monitoring result. The visualization display unit receives the final monitoring result and displays the abnormal distribution band corresponding to the set of impact points of the effective boundary on the polar coordinate graph according to the geometric position of the attachment boundary through a highlighted graphic mark that is different from ordinary impact points, and simultaneously marks the boundary event type and alarm level.

[0100] Furthermore, a dynamic quantile classification mechanism based on relative sorting replaces the fixed rigid amplitude threshold, enabling the alarm logic to adapt to the fundamental differences in normal impact responses between different cylinder blocks, different operating conditions, and different boundary event types. The output of the visualization unit presents the boundary-attached anomaly distribution band as an independent boundary event, preventing maintenance personnel from misinterpreting it as an impact within the normal operating condition zone.

[0101] Based on the above monitoring links, let's take the alarm triggering process of a certain boundary anomaly as an example. Assume that a set of effective boundary impact points attached to the 180° boundary between the intake and compression zones is obtained, and the calculated tongue ridge offset is 0.217, tongue bending amount is 1.5, and average normalized impact intensity is 2.0. Based on the tongue ridge offset being greater than zero, the impact band is determined to extend outwards towards the positive side of the compression zone. Then, according to the mapping rule, the boundary event type is defined as a compression start-side boundary impact event. The absolute value of the tongue bending amount, 1.5, is extracted and multiplied by the average normalized impact intensity, 2.0, to obtain a boundary event alarm quantity of 3.0. A historical health alarm quantity sample set containing 200 homogeneous calculation samples is retrieved. After traversal and comparison, 188 historical samples are found to be less than or equal to 3.0. The historical ranking normalized quantile probability is calculated to be 0.94 through ratio calculation. The alarm output unit generates a warning level based on the condition that this 0.94 quantile value falls within the 0.90 to 0.97 interval. The visualization unit ultimately renders a highlighted abnormal distribution band on the positive side of the 180° boundary of the polar coordinate graph and outputs a comprehensive status indicator with the compression start side boundary impact event and warning level.

[0102] The above embodiments are only used to illustrate the technical methods of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical methods of the present invention without departing from the spirit and scope of the technical methods of the present invention.

Claims

1. A method for monitoring compressor cylinder vibration based on angular period and polar coordinates, characterized in that, include: The original cylinder block vibration signal is resampled based on the original crankshaft angle reference signal to obtain the angular domain vibration signal; The impact characterization quantity represented by the analysis window within the angular domain vibration signal is extracted. The impact characterization quantity represented by the analysis window is normalized by combining the historical health benchmark impact characterization quantity to generate the normalized impact radius. The polar coordinate impact point set and the working condition boundary set are then constructed. Based on the minimum angular circumference difference, the set of polar coordinate impact points is determined to belong to the attached boundary in the set of working condition boundaries. The directional angular offset relative to the attached boundary is calculated. The circumferential angular distance between the adjacent boundary corresponding to the attached boundary is extracted to generate the boundary influence half-width. The dimensionless boundary expansion abscissa is generated by combining the directional angular offset and the boundary influence half-width. The normalized impact radius is assigned as the boundary expansion ordinate to construct the expansion coordinate points. The expansion coordinate points that meet the impact enhancement conditions and reach the preset number threshold are selected to construct the effective boundary expansion impact point set. Based on the effective boundary, the set of impact points is expanded. Using the boundary expansion ordinate as the weight, the spatial weighted centroid is calculated to generate the tongue-shaped offset, the third-order spatial skewness is calculated to generate the tongue-shaped bending amount, and the mean of the boundary expansion ordinate is calculated to generate the average normalized impact intensity. Among them, the dimensionless boundary expansion abscissa and boundary expansion ordinate are extracted from the effective boundary expansion impact point set corresponding to the attached boundary. Calculate the sum of the products of the dimensionless boundary expansion x-coordinate and the boundary expansion y-coordinate within the effective boundary expansion impact point set, and calculate the sum of the y-coordinates of all boundary expansions within the effective boundary expansion impact point set; The tongue ridge offset is generated by dividing the sum of the products of the x-coordinates of each dimensionless boundary expansion and the y-coordinates of the boundary expansion by the sum of the y-coordinates of all boundary expansions. Calculate the difference between the dimensionless boundary expansion x-coordinate and the tongue ridge offset of each dimensionless boundary expansion impact point set within the effective boundary expansion impact point set, and calculate the cube of each difference to generate the corresponding deviation term; Multiply the deviation term by the corresponding boundary's ordinate and perform a global summation to generate the summation value; Divide the summed value by the sum of the ordinates of all boundary expansions within the effective boundary expansion impact point set to generate the tongue-shaped bending amount; Based on the tongue ridge offset and attachment boundary, the boundary event type is generated. The boundary event alarm quantity is calculated by combining the tongue bending amount and the average normalized impact intensity. The historical health alarm quantity sample set is compared to generate the historical ranked normalized quantile probability, and the final monitoring result is output.

2. The compressor cylinder vibration monitoring method based on angular period and polar coordinates according to claim 1, characterized in that, include: A discrete angle sampling point sequence is constructed based on a preset angle resolution. An interpolation algorithm is used to convert the original cylinder vibration signal in the time domain into an angle domain vibration signal with an aligned angle sampling point sequence, thus completing the resampling. An angle calculation window is established with each angle sampling point in the angle sampling point sequence as the center. The root mean square value of the vibration signal under the angle calculation window within a single cycle is calculated to generate a single-cycle impact characterization quantity. The median value of the single-cycle impact characterization quantity at the same angle position within multiple consecutive mechanical cycles is extracted to generate an analysis window representing the impact characterization quantity.

3. The compressor cylinder vibration monitoring method based on angular period and polar coordinates according to claim 1, characterized in that, include: A two-dimensional operating condition matrix is ​​constructed by dividing the full range of load rate and pressure ratio into continuous operating interval matrices. The operating condition category is defined by discretizing the corresponding grids of the two-dimensional operating condition matrix according to the actual load rate and pressure ratio falling into the matrix. When the actual load rate change rate or pressure ratio change rate is detected to exceed the preset threshold and enter a transient switch, the normalization process is paused. When there is no corresponding historical health benchmark impact characterization quantity for a working condition category, the historical health benchmark impact characterization quantity corresponding to the most matching adjacent working condition category in the coordinate neighborhood of the two-dimensional working condition matrix is ​​called and the substitution operation is performed.

4. The compressor cylinder vibration monitoring method based on angular period and polar coordinates according to claim 1, characterized in that, include: Integrate and compare the working condition boundary of the stroke and the theoretical action line of the valve; When the distance between the theoretical action line of the valve and the circumferential angle of the boundary of the adjacent stroke is less than or equal to twice the preset angle resolution, the midpoint of the circumferential angle of the theoretical action line of the valve and the boundary of the adjacent stroke is extracted to form a composite boundary. By using composite boundaries to replace the valve's theoretical action line and adjacent stroke working condition boundaries corresponding to the conflict, the composite boundaries and independent boundaries are incorporated into the working condition boundary set.

5. The compressor cylinder vibration monitoring method based on angular period and polar coordinates according to claim 1, characterized in that, include: Extract the polar angle of each polar impact point in the set of polar impact points, and calculate the original angle difference between the polar angle of the polar impact point and the original angle difference between each working condition boundary in the set of working condition boundaries. If the original angle difference is greater than 180°, perform a conversion operation by subtracting 360° from the original angle difference. If the original angle difference is less than or equal to -180°, perform a 360° conversion operation on the original angle difference; Based on the conversion operation results, the absolute value of the angle surrounding difference is obtained, a global traversal comparison is performed, and the working condition boundary corresponding to the minimum absolute value of the angle surrounding difference is selected as the attachment boundary to which the polar coordinate impact point uniquely belongs. When the absolute value of the circumferential difference between the polar coordinate impact point and the two adjacent working condition boundaries is equal, the front boundary in the circumferential direction is determined as the unique attachment boundary to which the polar coordinate impact point belongs.

6. The compressor cylinder vibration monitoring method based on angular period and polar coordinates according to claim 5, characterized in that, include: For all unfolded coordinate points belonging to the same attachment boundary, perform numerical traversal, remove background baseline points with unfolded boundary ordinates less than 1, and retain unfolded coordinate points with unfolded boundary ordinates greater than or equal to 1. When the number of retained unfolded coordinate points is greater than or equal to the preset number threshold, the retained unfolded coordinate points will be aggregated to construct a set of effective boundary unfolded impact points corresponding to the attachment boundary; When the number of retained unfolded coordinate points is less than the preset threshold, the set corresponding to the attached boundary is marked as an empty set, thus preventing the empty set from participating in the calculation of tongue-shaped morphological parameters.

7. The compressor cylinder vibration monitoring method based on angular period and polar coordinates according to claim 1, characterized in that, include: Determine the sign of the tongue ridge offset value. When the tongue ridge offset is greater than zero, it is determined to extend outward into the working area on the positive side of the attachment boundary. When the tongue ridge offset is less than zero, it is determined to extend outward into the working area on the negative side of the attachment boundary. When the attachment boundary is the dividing line of the stroke working condition, the extension to the positive side of the attachment boundary working condition area is defined as the impact event of the beginning side boundary of the adjacent subsequent working condition, and a boundary event type is generated. The extension to the negative side of the attachment boundary working condition area is defined as the impact event of the end side boundary of the adjacent preceding working condition, and a boundary event type is generated. When the attachment boundary is the theoretical action line of the valve, the outward extension of the working condition area on the positive side of the attachment boundary is defined as the valve action rear boundary impact event, and a boundary event type is generated. The outward extension of the working condition area on the negative side of the attachment boundary is defined as the valve action front boundary impact event, and a boundary event type is generated. When the lingual ridge bias is zero, no unilateral abduction boundary events are generated.

8. The compressor cylinder vibration monitoring method based on angular period and polar coordinates according to claim 7, characterized in that, include: Extract the absolute value of the tongue bending amount, and multiply the absolute value of the tongue bending amount by the average normalized impact intensity to generate the boundary event alarm value; Retrieve the historical health alarm sample set corresponding to the boundary event type, count the number of samples in the historical health alarm sample set that are less than or equal to the boundary event alarm, divide the number of samples that are less than or equal to the boundary event alarm by the total number of historical health alarm samples, and generate the historical sorted normalized quantile probability. The historical sorted normalized quantile probability is divided into a preset continuous warning interval to generate alarm levels. The boundary event type, boundary event alarm quantity, historical sorted normalized quantile probability, alarm level, attachment boundary and tongue ridge offset are summarized to generate the final monitoring results. On the polar coordinate graph, based on the geometric position of the attached boundary, the abnormal distribution zone corresponding to the set of impact points of the effective boundary is displayed using highlighted graphics, and the boundary event type and alarm level are marked simultaneously.

Images