Visualization method for monitoring operation state of labyrinth compressor

Abstract

The invention belongs to the technical field of state monitoring of labyrinth compressors, and particularly relates to a visualization method for monitoring the running state of a labyrinth compressor. The method comprises the steps that vibration and key phase signals of the labyrinth compressor are collected, the Hurst index of instantaneous frequency and the kurtosis of the vibration signals are calculated based on the key phase signals so as to obtain the bandwidth of a phase-locked loop, then the instantaneous phase of a rotating shaft is extracted, and equal-angle resampling is achieved; performing angle domain resampling on the vibration signal by adopting a mixed interpolation core, extracting order energy and dual-coherence spectrum characteristics through synchronous compression transformation, and constructing a characteristic tensor taking the rotating speed and the order as indexes; through polar coordinate distribution diagram visualization, different state clusters are identified by means of density peak clustering, states are distinguished by colors, and the degree of closeness to a cluster center is represented by brightness. The technical problem that it is difficult to accurately and comprehensively monitor the operation state of the labyrinth compressor in the prior art is solved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of state monitoring of labyrinth compressors, and particularly relates to a visual method for monitoring the operating state of a labyrinth compressor. BACKGROUND

[0002] A labyrinth compressor is a reciprocating piston compressor using non-contact labyrinth sealing technology, mainly used for gas compression and delivery of flammable, explosive and toxic gases under high pressure conditions, and widely used in key industrial fields such as petrochemical industry, natural gas delivery and energy processing. As a core fluid delivery device, the stability of the operating state of the labyrinth compressor directly determines the continuity and safety of the production process. Due to the complex working environment, the labyrinth compressor is prone to faults such as shaft imbalance, abnormal sealing gap and component rubbing during long-term operation. If not identified in time, it may cause equipment downtime, production interruption, and even safety accidents. Therefore, real-time and accurate operating state monitoring and fault warning of the labyrinth compressor have become the core demand in the field of industrial equipment operation and maintenance.

[0003] Vibration signal analysis is the main means to monitor the operating state of the labyrinth compressor and diagnose potential faults. The core principle is that changes in the operating state of the labyrinth compressor will directly reflect in the frequency, amplitude, phase and other characteristics of the vibration signal, and the health condition of the equipment can be deduced in reverse through analysis of the vibration signal. However, the actual operation process of the labyrinth compressor is not stable. During the start-stop stage, the speed will linearly rise from zero to the rated value. Under variable load conditions, the speed needs to be dynamically adjusted according to production requirements, resulting in continuous changes in the frequency components of the vibration signal over time. This characteristic makes the traditional frequency spectrum analysis method based on fast Fourier transform (FFT) have problems such as frequency spectrum blurring and energy leakage, which cannot accurately extract the fault features related to the speed, resulting in a significant decrease in fault diagnosis accuracy. To solve the above problems, the industry gradually adopts order tracking analysis technology. Order tracking analysis technology converts the vibration signal from the time domain to the angle domain, eliminating the influence of speed variation on frequency spectrum analysis, so that the order related to the rotation of the shaft becomes stable in the angle domain, which is convenient for fault diagnosis.

[0004] However, the accuracy of order tracking analysis depends largely on two key links: instantaneous phase estimation and equi-angle resampling. The current acquisition of instantaneous phase mainly relies on the key phase signal synchronous with the rotating shaft, combined with the phase-locked loop technology for processing. However, the loop filter bandwidth of the traditional phase-locked loop is a fixed value, which cannot adapt to the complex operating conditions of the compressor. If a wideband filter is used, although it can quickly track the speed mutation, it has weak anti-noise ability and is easy to misjudge the environmental interference signal as a speed change, leading to phase estimation deviation; if a narrowband filter is used, although it can suppress noise, the response speed is slow and cannot follow the rapid fluctuations of the rotating speed in time, also causing phase calculation error, ultimately affecting the accuracy of order tracking. In the equi-angle resampling link, commonly used methods such as linear interpolation and cubic spline interpolation are simple to implement, but when the original vibration signal contains impact components caused by faults, the fixed interpolation kernel function may smooth out the key transient characteristics, introducing amplitude and phase errors, and affecting the accuracy of subsequent feature extraction. The traditional visualization methods (such as time domain waveform diagram and frequency domain spectrum diagram) can only display features in a single dimension and cannot effectively fuse multi-dimensional features, making it difficult for operation and maintenance personnel to intuitively and comprehensively judge the overall state of the equipment, increasing the complexity and time cost of fault positioning. SUMMARY

[0005] In view of this, the purpose of the present application is to provide a visualization method for monitoring the operating state of a labyrinth compressor, to solve the technical problem that the prior art cannot accurately and comprehensively monitor the operating state of a labyrinth compressor.

[0006] The present application provides a visualization method for monitoring the operating state of a labyrinth compressor, comprising the following steps: obtaining the vibration signal of the labyrinth compressor during operation and the key phase signal synchronous with the rotating shaft; based on the key phase signal, calculating the Hurst index of the instantaneous frequency in the time window and the kurtosis of the vibration signal, and obtaining the loop filter bandwidth of the phase-locked loop according to the Hurst index and the kurtosis, wherein the loop filter bandwidth is negatively correlated with the Hurst index and has a pre-set piecewise function relationship with the kurtosis; processing the key phase signal using the phase-locked loop configured with the loop filter bandwidth to obtain the instantaneous phase information of the rotating shaft; determining the equi-angle sampling time according to the instantaneous phase information; calculating the peak factor in the neighborhood of each sampling time for the vibration signal, and obtaining the mixed weight of the Gaussian kernel function and the Lagrange basis function according to the peak factor, wherein the weight of the peak factor and the Gaussian kernel function is negatively correlated, and the weight of the Lagrange basis function is positively correlated; resampling the vibration signal using the mixed interpolation kernel composed of the linear superposition of the weighted kernel functions to obtain the angle domain signal; The angle domain signal is synchronously compressed and transformed to extract the energy of the preset fundamental frequency order and harmonic order, as well as the biphase spectrum values ​​between each order. The features extracted at different speeds are constructed into feature tensors indexed by speed and order. A polar coordinate state distribution map is generated for visualization, with the polar radial axis and polar angle axis representing the rotational speed and order, respectively; density peak clustering is performed on the feature vectors in the feature tensor to divide the state clusters; the color of the coordinate point in the polar coordinate state distribution map is determined by the index of the state cluster to which the corresponding feature vector belongs, and the brightness is determined by the reciprocal of the distance from the feature vector to the center of the state cluster to which the vector belongs.

[0007] Further, the step of calculating the Hearst exponent of the instantaneous frequency within the time window and the kurtosis of the vibration signal based on the key phase signal, and obtaining the loop filter bandwidth of the phase-locked loop based on the Hearst exponent and the kurtosis, includes: Obtain a first kurtosis threshold and a second kurtosis threshold, wherein the second kurtosis threshold is greater than the first kurtosis threshold; Calculate the Hearst exponent H of the instantaneous frequency within the time window and the kurtosis K of the vibration signal; Based on the range of kurtosis K, the corresponding bandwidth coefficient C is selected: when K is less than or equal to the first kurtosis threshold, the first bandwidth coefficient C1 is selected; when K is greater than the first kurtosis threshold and less than or equal to the second kurtosis threshold, the second bandwidth coefficient C2 is selected; when K is greater than the second kurtosis threshold, the third bandwidth coefficient C3 is selected; where C3... <C2<C1; The bandwidth B of the loop filter is calculated according to the formula. Calculations show that This is the baseline bandwidth.

[0008] Furthermore, the hybrid interpolation kernel determines whether there is an impact in the vibration signal by calculating the peak factor; when the vibration signal is stable and the peak factor is low, the weight of the Gaussian kernel function is increased to smooth the noise; when the vibration signal contains an impact and the peak factor is high, the weight of the Lagrange basis function is increased to preserve the transient waveform characteristics.

[0009] Further, for the vibration signal, calculating the peak factor in the neighborhood at each sampling time, and obtaining the mixed weight of the Gaussian kernel function and the Lagrange basis function based on the peak factor, includes: For each equiangular sampling moment, a preset number of time sampling points before and after that moment are taken to form the neighborhood of that moment; Calculate the peak factor CF of the vibration signal within the neighborhood; Using the preset slope parameter k and center threshold Through function The weights of the Lagrange basis function are calculated. ; Weights of the Gaussian kernel functions According to the formula The calculation result is obtained.

[0010] Further, the synchronous compression transformation of the angle domain signal, extraction of energy of preset fundamental frequency order and harmonic order, and bicoherence spectrum value between orders, comprises: From the synchronous compression transformation result of the angle domain signal, signal components of fundamental frequency 1 times frequency order, 2 times frequency order and 3 times frequency order are extracted; The amplitude square of the signal components of 1 times frequency, 2 times frequency and 3 times frequency order is integrated respectively to obtain the energy of each order 、 、 ; The bicoherence spectrum value between the signal components of 1 times frequency order and 2 times frequency order is calculated ; The energy 、 、 and the bicoherence spectrum value are combined into a feature vector for representing the running state under the current rotating speed.

[0011] Further, the density peak clustering of the feature vectors in the feature tensor is performed to divide state clusters, comprising: A cut-off distance is set as an analysis radius to calculate the Euclidean distance between any two feature vectors in the feature tensor; The local density of each feature vector is calculated, and the local density represents the number of other feature vectors with an Euclidean distance less than the cut-off distance from the current feature vector; The minimum distance δ of each feature vector is calculated, and the minimum distance δ represents the minimum Euclidean distance from the current feature vector among other feature vectors with a larger local density than the current feature vector; The feature vectors with a large local density and a large minimum distance δ are selected as the center points of the state clusters, and the remaining feature vectors are assigned to the state clusters to which the feature vectors with the nearest distance from the center points and a larger local density than the center points belong, to complete the division of the state clusters.

[0012] Further, the state clusters comprise a normal state cluster, an unbalance state cluster and a rub-impact state cluster.

[0013] Further, the vibration signal during the operation of the labyrinth compressor is acquired, and the key phase signal synchronized with the rotating shaft is acquired, the vibration signal is collected by an acceleration sensor, and the key phase signal is collected by an eddy current sensor aiming at the key groove or protrusion on the rotating shaft of the labyrinth compressor.

[0014] Further, the Hurst index is calculated by using a re-scaled range analysis method.

[0015] Further, the determination mode of the equi-angle sampling time is that the time corresponding to the equi-interval angle is calculated by linear interpolation inverse solution according to the instantaneous phase information, and the time is the equi-angle sampling time.

[0016] The beneficial effects of the present application are: The present application adopts an adaptive bandwidth adjustment strategy, judges the trend of the rotating speed change through the Hurst index, judges the strength of the impact interference in the signal through the kurtosis, and the two jointly determine the loop filter bandwidth of the phase-locked loop. The dynamic adjustment makes the phase-locked loop narrow the bandwidth to enhance the anti-interference when the rotating speed is stable and the noise is large, and widens the bandwidth to realize fast tracking when the rotating speed changes rapidly, and then more accurate instantaneous phase information of the rotating shaft can be acquired. In the equi-angle resampling link, an adaptive mixed interpolation kernel is adopted, the peak factor in the neighborhood of each sampling point is calculated to judge whether there is an impact, when the peak factor is high, the weight of the Lagrange basis function is increased to retain the transient waveform, and when the peak factor is low, the weight of the Gaussian kernel function is increased to smooth the noise, so that the impact characteristics caused by the fault cannot be incorrectly filtered out or passivated in the key process of converting the signal from the time domain to the angle domain. Furthermore, on the basis of extracting the energy of each order, the present application introduces a double-coherence spectrum value as a feature, which can capture deep fault information that cannot be found by energy analysis alone, realizes multi-dimensional and high-sensitivity feature extraction and representation of the operating state of the labyrinth compressor, and improves the accuracy of fault identification. Finally, the present application can automatically identify and divide the operating state of the equipment through the density peak clustering algorithm and the polar coordinate state distribution diagram, and can also fuse the multi-dimensional feature information into a single view through a novel visualization method, thereby improving the comprehensiveness and intuitiveness of the labyrinth compressor state monitoring. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 A step flow chart of a visualization method for labyrinth compressor operating state monitoring according to the present application; Figure 2 A signal acquisition schematic diagram; Figure 3 An adaptive setting schematic diagram of the phase-locked loop bandwidth; Figure 4 A mixed interpolation kernel weight setting schematic diagram; Figure 5constructing a schematic diagram for the feature tensor; Figure 6 polar coordinate state distribution map. DETAILED DESCRIPTION

[0018] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application.

[0019] A specific embodiment of a visualization method for labyrinth compressor operating state monitoring of the present application: As Figure 1 shown, the visualization method for labyrinth compressor operating state monitoring of the present application comprises the following steps: S1: acquiring a vibration signal during labyrinth compressor operation and a key phase signal synchronous with a rotating shaft; based on the key phase signal, calculating a Hurst index of an instantaneous frequency in a time window and a kurtosis of the vibration signal, and acquiring a loop filter bandwidth of a phase-locked loop according to the Hurst index and the kurtosis, the loop filter bandwidth being negatively correlated with the Hurst index and being in a preset piecewise function relationship with the kurtosis; processing the key phase signal by using the phase-locked loop configured with the loop filter bandwidth to obtain instantaneous phase information of the rotating shaft; Specifically, in this step, an acceleration sensor installed at a specified position of the labyrinth compressor is used to collect radial or axial vibration signals, and a keyway or a protrusion is arranged on the rotating shaft, so that a pulse signal is generated every time the rotating shaft rotates one revolution by aligning the eddy current sensor with the mark on the rotating shaft. The two signals are synchronously collected to obtain discrete time series signals, as shown in Figure 2 .

[0020] The instantaneous frequency sequence of the rotating shaft is obtained by calculating the time interval between adjacent pulses of the key phase signal. The Hurst index H of the instantaneous frequency sequence in a time window is calculated by using the re-indexed range analysis method, and the Hurst index reflects the trend of the rotating speed change. At the same time, the kurtosis value of the original vibration signal in the same time window is calculated, and the kurtosis value reflects the strength of the impact component in the signal. The bandwidth of the phase-locked loop can be determined according to the Hurst index, the kurtosis and the reference bandwidth. The calculated bandwidth is configured into the digital phase-locked loop, and the phase-locked loop includes a phase detector, a loop filter and a voltage-controlled oscillator. The phase detector compares the phase difference between the input key phase pulse signal and the internal oscillation signal generated by the voltage-controlled oscillator, generates an error signal, and sends the error signal to the loop filter for processing. The bandwidth of the loop filter is configured according to the Hurst index and the kurtosis, and can determine whether to smooth the noise or quickly respond to the change according to the rotating speed stability and the vibration impact. The filtered error signal adjusts the frequency and phase of the voltage-controlled oscillator, and the output of the voltage-controlled oscillator is the high-resolution instantaneous phase information of the rotating shaft, as shown in Figure 3 .

[0021] S2: determining equal-angle sampling time according to the instantaneous phase information; calculating a peak factor in a neighborhood of each sampling time for the vibration signal, and obtaining a mixed weight of a Gaussian kernel function and a Lagrange basis function according to the peak factor, wherein the weight of the peak factor and the Gaussian kernel function is negatively correlated, and the weight of the peak factor and the Lagrange basis function is positively correlated; re-sampling the vibration signal by using a mixed interpolation kernel composed of linear superposition of the weighted kernel functions to obtain an angle domain signal. According to the instantaneous phase information obtained in step S1, the phase reaches the equal-interval angle by linear interpolation inverse solution, such as the time corresponding to 1024 points per revolution. These time points are the equal-angle sampling time. For each equal-angle sampling time, a neighborhood window containing dozens of original sampling points is taken before and after it, and the peak factor of the vibration signal in the neighborhood window is calculated, that is, the ratio of the maximum value of the signal absolute value to the signal effective value. The weight of the Gaussian kernel function and the weight of the Lagrange basis function are obtained, and the sum of the two is 1. Preferably, the weight of the Lagrange basis function is determined by the peak factor through the S-type function. When the vibration signal is smooth and the peak factor is small, the weight of the Lagrange basis function is close to 0, and the weight of the Gaussian kernel function is close to 1; when the vibration signal contains an impact and the peak factor is large, the weight of the Lagrange basis function is close to 1, and the weight of the Gaussian kernel function is close to 0. The amplitude of the original time domain vibration signal at each equal-angle sampling time is calculated by the mixed interpolation kernel composed of the weight of the Gaussian kernel function multiplied by the Gaussian kernel function plus the weight of the Lagrange basis function multiplied by the Lagrange basis function. Specifically, the vibration amplitude at the target time is calculated by the mixed interpolation kernel, and the amplitudes calculated at all equal-angle time points are arranged in order to obtain an angle domain signal, as shown in FIG. 2. Figure 4 Exemplarily, for the equal-angle sampling points that need to be calculated, such as 10°, 20°, etc., their positions on the time axis are found, the mixed interpolation kernel is called, the original time sampling values around the target time point are taken as input, and the vibration amplitude at the target time point is calculated by the mixed interpolation kernel; the amplitudes calculated at all angle points are arranged in order according to the angle to obtain a vibration signal sequence with angle as the horizontal coordinate, that is, the angle domain signal.

[0022] S3: performing synchronous compression transformation on the angle domain signal, extracting the energy of the preset fundamental frequency order and harmonic order, and the bi-coherence spectrum value between each order, and constructing the extracted features at different speeds into a feature tensor indexed by speed and order.

[0023] In this step, the angle domain signal after resampling is subjected to synchronous compression transformation to obtain an angle-order spectrum. From the angle-order spectrum, the energy amplitude is calculated by integrating along a specific order line, such as 1 times frequency, 2 times frequency, 3 times frequency, etc., to obtain the energy features of each order. At the same time, the angle domain signal is segmented for Fourier transformation, and then the bi-coherence spectrum value between the preset order pairs is calculated, which reflects the nonlinear correlation between the orders. All the order energies and bi-coherence spectrum values extracted at a specific speed are combined into a feature vector, and the above process is repeated at multiple different speed points during the speed-up and speed-down process of the labyrinth compressor to form a three-dimensional feature tensor, three dimensions respectively representing speed, order and features composed of energy and bi-coherence spectrum value, as shown in Figure 5 .

[0024] S4: Generate a polar coordinate state distribution map, with the polar radius axis and the polar angle axis representing the speed and the order respectively; perform density peak clustering on the feature vectors in the feature tensor to divide the state clusters; the color of the coordinate points in the polar coordinate state distribution map is determined by the state cluster index to which the corresponding feature vector belongs, and the brightness is determined by the reciprocal of the distance from the feature vector to the center of the state cluster to which the vector belongs.

[0025] In this step, all feature vectors in the feature tensor are taken as input, and a density peak clustering algorithm is applied to identify the clustering centers by calculating the local density of each feature vector and the distance to higher local density points, and then all feature vectors are divided into different state clusters. For example, normal state cluster, unbalance state cluster, rubbing state cluster. A polar coordinate state distribution map is created, preferably, the polar radius from the center to the outer edge represents the speed from low to high, and the polar angle is evenly divided into 360 degrees for each order analyzed. Each coordinate point in the polar coordinate state distribution map corresponds to a specific speed and order in the polar radius and polar angle respectively. The color of the coordinate point is obtained according to the state cluster index to which the corresponding feature vector is divided, such as green for normal, yellow for unbalance, and red for rubbing. The Euclidean distance from the feature vector to the center of the state cluster to which it belongs is calculated, and the reciprocal of the Euclidean distance is mapped to the brightness of the point. The closer the distance, the more typical the coordinate point is as a representative of the state of the labyrinth compressor, and the higher the brightness; the farther the distance, the darker the brightness, indicating that it may be in a state boundary or abnormal working condition, as shown in Figure 6 .

[0026] In order to cope with the change of speed stability of the equipment in different states, in an optional embodiment, based on the keyphasor signal, the HURST index of the instantaneous frequency in the time window and the kurtosis of the vibration signal are calculated, and the loop filter bandwidth of the phase-locked loop is obtained according to the HURST index and the kurtosis, including: obtaining a first kurtosis threshold and a second kurtosis threshold, wherein the second kurtosis threshold is greater than the first kurtosis threshold; calculating a HURST index H of the instantaneous frequency in the time window and a kurtosis K of the vibration signal; According to the value range of the kurtosis K, a corresponding bandwidth coefficient C is selected: when K is less than or equal to a first kurtosis threshold, a first bandwidth coefficient C1 is selected; when K is greater than the first kurtosis threshold and less than or equal to a second kurtosis threshold, a second bandwidth coefficient C2 is selected; when K is greater than the second kurtosis threshold, a third bandwidth coefficient C3 is selected; wherein C3 < C2 < C1; The bandwidth B of the loop filter is calculated according to the formula The HURST index H is calculated according to the formula The reference bandwidth is 0.5.

[0027] Exemplarily, the first kurtosis threshold is 3.5 and the second kurtosis threshold is 6. When the device runs smoothly and there is no obvious impact in the vibration signal, the calculated kurtosis K may be 2.8, which is less than 3.5. At this time, it indicates that the speed fluctuation is small, and a larger first bandwidth coefficient C1, such as 0.8, can be selected to enhance the anti-noise ability of the phase-locked loop and avoid misjudging random noise as speed change.

[0028] If the device has early failure and produces a slight impact, the kurtosis K is 5.0, which is between 3.5 and 6.0. At this time, a moderate second bandwidth coefficient C2, such as 0.5, is selected to enable the phase-locked loop to respond to the real speed fluctuation caused by the fault impact in time. When the device has serious failure and produces a severe impact, the kurtosis K is greater than 6, for example, K is 8.1. At this time, a smaller third bandwidth coefficient C3, such as 0.2, is selected to ensure that the phase-locked loop can quickly track the speed change. At the same time, the HURST index H reflects the smoothness of the instantaneous frequency. If the H value is 0.9, it indicates that the frequency sequence is very smooth, and the bandwidth B will be reduced accordingly. If the H value is 0.6, it indicates that the frequency sequence fluctuates violently, and the bandwidth B will be increased, realizing fine double regulation of the bandwidth.

[0029] In an optional embodiment, for the vibration signal, a peak factor in a neighborhood of each sampling time is calculated, and a mixed weight of a Gaussian kernel function and a Lagrange basis function is obtained according to the peak factor, including: For each equiangular sampling time, a preset number of time sampling points before and after the time are taken to form a neighborhood of the time; The peak factor CF of the vibration signal in the neighborhood is calculated; A preset slope parameter k and a center threshold value The weight of the Lagrange basis function is calculated by the function ; The weight of the Gaussian kernel function is calculated according to the formula .

[0030] ​​For example, suppose interpolation is performed on a sampling point at a constant angle, defining the neighborhood of the sampling point as 16 time sampling points before and after it. If the vibration signal waveform within the neighborhood of the sampling point is smooth and there is no fault impact, the calculated peak factor CF may be 2.5. A preset center threshold for the Logistic function is used. The slope parameter k is 5, and the peak factor of 2.5 is less than the center threshold of 4.0. Since the peak factor of 2.5 is less than the center threshold of 4.0, the calculated result of the Logistic function is... It is very close to 0. Therefore, the weights of the Gaussian kernel function are... The value will be very close to 1, in which case the Gaussian kernel function will be mainly used for interpolation, and its smoothing properties can filter out high-frequency noise in the signal. Conversely, if the neighborhood contains a strong impact caused by bearing pitting, the peak factor CF is 7.0, which is greater than the center threshold of 4.0. In this case, the calculated value will be... It will be very close to 1, and If the value is close to 0, then the Lagrange basis function will be mainly used for interpolation, which can reconstruct the transient waveform of the impact signal and avoid the fault feature passivation that may be caused by Gaussian smoothing.

[0031] In an optional embodiment, the step of synchronously compressing and transforming the angle domain signal to extract the energy of the preset fundamental frequency order and harmonic orders, as well as the biphase spectral values ​​between each order, includes: From the synchronous compression transformation result of the angle domain signal, extract the signal components of the first, second, and third harmonics of the fundamental frequency; Integrating the squared amplitudes of the signal components at the first, second, and third harmonics orders respectively yields the energy of each order. , , ; Calculate the biphase spectrum values ​​between the first and second harmonic order signal components. ; Energy , , and bicoherence spectral values These are combined into a feature vector to represent the operating state at the current speed.

[0032] Assuming the fundamental frequency of the rotating shaft is a first harmonic, and the gear meshing frequency exhibits higher harmonics, the analysis involves performing a synchronous compression transform on a segment of the angle domain signal to obtain its order spectrum. The first harmonic signal is then extracted from this spectrum, and its energy is calculated to obtain... The value is 12.5 units. Similarly, the second and third harmonic orders of the signal are extracted, and the energy is calculated. It is 4.8 units. is 2.1 units. The three energy values reflect the strength of the fundamental and harmonic components. The bi-coherence value between the 1x and 2x components is calculated as 0.78, which indicates the strength of the non-linear correlation between the two components, and such correlation is often enhanced by faults such as gear tooth wear or crack. The four values are combined into a feature vector, i.e., [12.5, 4.8, 2.1, 0.78], which serves as a fingerprint of the gearbox operation condition at the current rotational speed.

[0033] In an alternative embodiment, the performing density peak clustering on the feature vectors in the feature tensor to partition the condition clusters comprises: setting a cutoff distance calculating the Euclidean distance between any two feature vectors in the feature tensor as the analysis radius; calculating the local density of each feature vector, where the local density represents the number of other feature vectors whose Euclidean distance to the current feature vector is less than the cutoff distance ; calculating the minimum distance δ of each feature vector, where the minimum distance δ represents the minimum Euclidean distance to the current feature vector among other feature vectors whose local density is greater than that of the current feature vector; selecting the feature vectors that have both large local density and large minimum distance δ as the center points of the condition clusters, and assigning the remaining feature vectors to the condition cluster to which the center point belongs and whose local density is greater than that of the center point, to complete the partition of the condition clusters.

[0034] Illustratively, assume that 1000 feature vectors are collected, and the cutoff distance is set to 0.8. For one of the feature vectors, point A, it is found that there are 60 other vectors whose distance to it is less than 0.8, so the local density of point A is 60. All the vector points whose local density is greater than 60 are found, which are the high local density points, and the distance of point A to these high local density points is calculated, where the minimum distance is 2.5, which is the minimum distance δ of point A. This calculation is performed for all the 1000 points, and it is found that some points, such as point X, have a very high value, e.g., 150, and a large δ value, e.g., 3.0, which indicates that point X itself is in a dense area and is far away from the center of other denser areas, so it is likely to be the center of an independent condition cluster, representing a typical operation condition, such as normal operation of the device. While another The point Y with both value and delta value being large can represent another state, such as an early fault state. After all the center points are determined, the remaining points are classified into the state clusters according to rules, and the automatic classification of the equipment running state is completed.

[0035] Compared with the prior art, the application solves the problem of phase estimation deviation under response lag or noise interference when the speed fluctuates by dynamically adjusting the phase-locked loop bandwidth with the Hurst index of the instantaneous frequency of the key phase signal and the kurtosis of the vibration signal, and improves the precision of instantaneous phase estimation; in view of the defect that the traditional fixed interpolation kernel resampling is easy to lose the fault impact characteristics, the application dynamically allocates the mixed weight of the Gaussian kernel and the Lagrange basis function based on the signal peak value factor in the neighborhood of the sampling time, effectively avoids the blunting of the fault impact component, and ensures that the early slight fault characteristics are not lost; in view of the limitation that the prior art is difficult to fuse multi-dimensional features, the application extracts the order energy and the two-phase coherence spectrum value between orders by synchronous compression transformation, constructs a feature tensor indexed by speed and order, completely retains the state information under different working conditions, and reduces the fault misjudgment and omission caused by single-dimensional features; in view of the problem that the traditional visualization method is not intuitive enough, the application uses the polar coordinate state distribution map combined with the density peak value clustering, uses color to distinguish state types and brightness to reflect the distance from the cluster center, fuses multi-dimensional features in a single intuitive view, greatly reduces the judgment difficulty of operation and maintenance personnel, shortens the fault positioning time, and the overall technical scheme is more suitable for the complex operation requirements of the labyrinth compressor in the industrial scene, and realizes a significant breakthrough in the accuracy of the labyrinth compressor running state monitoring, the feature integrity, the state representation comprehensiveness and the operation and maintenance ease of use.

[0036] Although the present specification has shown and described multiple embodiments of the present application, it will be apparent to those of ordinary skill in the art that such embodiments are provided by way of example only. Those of ordinary skill in the art will think of many changes, modifications and alternatives without departing from the spirit and principles of the present application.

Claims

1. A visualization method for monitoring the operating status of a labyrinth compressor, characterized in that, Includes the following steps: The vibration signal and the key phase signal synchronized with the shaft during the operation of the labyrinth compressor are acquired. Based on the key phase signal, the Hearst exponent of the instantaneous frequency within the time window and the kurtosis of the vibration signal are calculated. The loop filter bandwidth of the phase-locked loop is obtained by combining the Hearst exponent and the kurtosis. The loop filter bandwidth is negatively correlated with the Hearst exponent and has a preset piecewise function relationship with the kurtosis. The key phase signal is processed using the phase-locked loop configured with the loop filter bandwidth to obtain the instantaneous phase information of the shaft. The equiangular sampling time is determined based on the instantaneous phase information; for the vibration signal, the peak factor in the neighborhood of each sampling time is calculated, and the mixed weight of the Gaussian kernel function and the Lagrange basis function is obtained based on the peak factor, wherein the peak factor is negatively correlated with the weight of the Gaussian kernel function and positively correlated with the weight of the Lagrange basis function; the vibration signal is resampled using a hybrid interpolation kernel composed of the linear superposition of the weighted kernel functions to obtain the angular domain signal; The angle domain signal is synchronously compressed and transformed to extract the energy of the preset fundamental frequency order and harmonic order, as well as the biphase spectrum values ​​between each order. The features extracted at different speeds are constructed into feature tensors indexed by speed and order. A polar coordinate state distribution map is generated for visualization, with the polar radial axis and polar angle axis representing the rotational speed and order, respectively; density peak clustering is performed on the feature vectors in the feature tensor to divide the state clusters; the color of the coordinate point in the polar coordinate state distribution map is determined by the index of the state cluster to which the corresponding feature vector belongs, and the brightness is determined by the reciprocal of the distance from the feature vector to the center of the state cluster to which the vector belongs.

2. The visualization method for monitoring the operating status of a labyrinth compressor according to claim 1, characterized in that, The step of calculating the Hearst exponent of the instantaneous frequency within the time window and the kurtosis of the vibration signal based on the key phase signal, and obtaining the loop filter bandwidth of the phase-locked loop based on the Hearst exponent and the kurtosis, includes: Obtain a first kurtosis threshold and a second kurtosis threshold, wherein the second kurtosis threshold is greater than the first kurtosis threshold; Calculate the Hearst exponent H of the instantaneous frequency within the time window and the kurtosis K of the vibration signal; Based on the range of kurtosis K, the corresponding bandwidth coefficient C is selected: when K is less than or equal to the first kurtosis threshold, the first bandwidth coefficient C1 is selected; when K is greater than the first kurtosis threshold and less than or equal to the second kurtosis threshold, the second bandwidth coefficient C2 is selected; when K is greater than the second kurtosis threshold, the third bandwidth coefficient C3 is selected; where C3... <C2<C1; The bandwidth B of the loop filter is calculated according to the formula. Calculations show that This is the baseline bandwidth.

3. The visualization method for monitoring the operating status of a labyrinth compressor according to claim 1, characterized in that, The hybrid interpolation kernel determines whether there is an impact in the vibration signal by calculating the peak factor; when the vibration signal is stable and the peak factor is low, the weight of the Gaussian kernel function is increased to smooth the noise; when the vibration signal contains an impact and the peak factor is high, the weight of the Lagrange basis function is increased to preserve the transient waveform characteristics.

4. The visualization method for monitoring the operating status of a labyrinth compressor according to claim 3, characterized in that, For the vibration signal, the peak factor in the neighborhood at each sampling time is calculated, and the mixed weight of the Gaussian kernel function and the Lagrange basis function is obtained based on the peak factor, including: For each equiangular sampling moment, a preset number of time sampling points before and after that moment are taken to form the neighborhood of that moment; Calculate the peak factor CF of the vibration signal within the neighborhood; Using the preset slope parameter k and center threshold Through function The weights of the Lagrange basis functions are calculated. ; The weights of the Gaussian kernel function According to the formula Calculated.

5. The visualization method for monitoring the operating status of a labyrinth compressor according to claim 1, characterized in that, The synchronous compression transformation of the angle domain signal to extract the energy of the preset fundamental frequency order and harmonic orders, as well as the biphase coherence spectrum values ​​between each order, includes: From the synchronous compression transformation result of the angle domain signal, extract the signal components of the first, second, and third harmonics of the fundamental frequency; Integrating the squared amplitudes of the signal components at the first, second, and third harmonics orders respectively yields the energy of each order. , , ; Calculate the biphase spectrum values ​​between the first and second harmonic order signal components. ; Energy , , and bicoherence spectral values These are combined into a feature vector to represent the operating state at the current speed.

6. The visualization method for monitoring the operating status of a labyrinth compressor according to claim 5, characterized in that, Performing density peak clustering on the feature vectors in the feature tensor to divide state clusters includes: Set a cutoff distance As the analysis radius, the Euclidean distance between any two eigenvectors in the feature tensor is calculated. Calculate the local density of each eigenvector Local density This means the Euclidean distance to the current feature vector is less than the cutoff distance. The number of other feature vectors; Calculate the minimum distance δ for each feature vector. The minimum distance δ represents the minimum Euclidean distance to the current feature vector among other feature vectors with a local density greater than the current feature vector. Selecting those with large local density The feature vector with the largest minimum distance δ is used as the center point of the state cluster. The remaining feature vectors are assigned to the state clusters to which the feature vectors closest to the center point and whose local density is greater than that of the center point belong, thus completing the state cluster partitioning.

7. The visualization method for monitoring the operating status of a labyrinth compressor according to claim 6, characterized in that, The state clusters include normal state clusters, unbalanced state clusters, and rubbing state clusters.

8. The visualization method for monitoring the operating status of a labyrinth compressor according to claim 1, characterized in that, The vibration signal and the key phase signal synchronized with the shaft of the labyrinth compressor are obtained during operation. The vibration signal is acquired by an accelerometer, and the key phase signal is acquired by an eddy current sensor aligned with the keyway or protrusion on the shaft of the labyrinth compressor.

9. The visualization method for monitoring the operating status of a labyrinth compressor according to claim 1, characterized in that, The Hearst exponent was calculated using the rescaled range analysis method.

10. The visualization method for monitoring the operating status of a labyrinth compressor according to claim 1, characterized in that, The method for determining the equal-angle sampling time is as follows: based on the instantaneous phase information, the time corresponding to the phase reaching the equal-interval angle is calculated by linear interpolation inverse solution. This time is the equal-angle sampling time.

Images