A fault diagnosis method and device for a floating platform reciprocating compressor

Abstract

The application discloses a kind of floating platform reciprocating compressor fault diagnosis method and device, including for the vibration signal to be treated decomposition obtains IMF component;By index, some orders of IMF containing obvious impact characteristics are screened out;The screened IMF component is filtered respectively, and non-impact component is filtered out;The IMF component after adaptive variable scale morphological filtering is reconstructed to obtain noise reduction signal;Amplify the mutation component in noise reduction signal;Solve the envelope of mutation component, and utilize filter to carry out smooth filtering to envelope line;Peak value detection is carried out to extract the phase feature of impact component in signal;The phase feature of impact is acquired.Diagnosis device includes: acquisition module, processing module, extraction module and diagnosis module, establishes reciprocating compressor fault feature knowledge base, and carries out fault diagnosis according to the impact feature of vibration signal captured by extraction module.And support and guarantee are provided for the predictive maintenance of future offshore platform reciprocating compressor.

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis of floating platform reciprocating compressors, and in particular to a method and apparatus for fault diagnosis by extracting vibration signal impact features of floating platform reciprocating compressors. Background Technology

[0002] Large reciprocating piston compressors, as power sources for compressing and transporting media, are widely used in petroleum, chemical, and refrigeration industries, and are also key equipment on offshore platforms. Their safe, reliable, stable, and long-term operation is crucial. When certain faults occur in a reciprocating compressor, they often lead to an increase in the number of impacts detected by vibration sensors in structures such as the compressor body, valve cover, and crankcase, or a shift in the impact phase. By monitoring the number and phase of impacts within a single cycle in real time, a preliminary fault diagnosis can be made. For example, when the small end bearing experiences wear, additional impacts will occur near the piston reversing dead center; when the intake or exhaust valves experience spring failure or valve plate breakage, the opening or closing of the intake and exhaust valves will be premature or delayed, resulting in a shift in the acceleration impact phase at the valve cover and cylinder block; when the large end bearing experiences wear, the crankcase speed signal will generate two impacts per cycle. By monitoring the number, peak value, and phase of impacts in the compressor body, valves, or crankcase within a single cycle, a preliminary fault diagnosis can be made.

[0003] The volume acceleration signal in a reciprocating compressor contains harmonic signals related to the crank rotation cycle, impact signals related to the equipment's operating state, and environmental noise signals. To accurately extract the impact component from the signal, in addition to noise reduction, it is also necessary to suppress the harmonic components. Using traditional linear filters to filter out harmonic components requires sufficient prior knowledge to accurately predict the frequencies of all harmonics in order to apply an effective notch filter for suppression. Furthermore, if the frequencies of the harmonics overlap with the frequency components contained in the impact signal, it will cause a loss of impact characteristics, contradicting the goal of accurately extracting these characteristics. Summary of the Invention

[0004] The technical problem to be solved by the present invention is a method and device for fault diagnosis of a floating platform reciprocating compressor, which extracts the impact characteristics of the vibration signal of the floating platform reciprocating compressor and realizes fault diagnosis.

[0005] To address the aforementioned technical problems, this invention provides a fault diagnosis device for a floating platform reciprocating compressor, comprising:

[0006] The data acquisition module is used to acquire vibration data of the reciprocating compressor on the floating platform and transmit it to the host computer for processing.

[0007] The processing module, connected to the acquisition module, processes the acquired vibration signal through EEMD-adaptive variable-scale morphological filtering.

[0008] The extraction module, connected to the processing module, is used to perform peak detection to extract the phase characteristics of the impact component in the signal and capture the parametric characteristics of the impact.

[0009] The diagnostic evaluation module, connected to the extraction module, establishes a fault feature knowledge base for reciprocating compressors, and performs fault diagnosis by comparing the vibration signal impact features captured by the extraction module with the fault feature knowledge base and outputting the diagnostic results.

[0010] The visualization module, connected to the diagnostic evaluation module and the data acquisition module, visualizes the diagnostic results and acquired data. It also enables remote data processing and retrieval through interactive web access via a remote terminal.

[0011] Another technical solution is: a fault diagnosis method for a floating platform reciprocating compressor, characterized by comprising the following steps:

[0012] S01: Decompose the vibration signal to be processed to obtain several IMF components;

[0013] S02: Select several IMF components by using indicators to identify IMFs of certain orders that contain obvious shock characteristics;

[0014] S03: Perform adaptive variable-scale morphological filtering on the selected IMF components to filter out non-impact components;

[0015] S04: Reconstruct the denoised signal from the IMF components after adaptive variable-scale morphological filtering;

[0016] S05: Amplify the amplitude-frequency abrupt change components in the noise-reduced signal;

[0017] S06: Obtain the envelope of the mutation component and smooth the envelope using a smoothing filter;

[0018] S07: Perform peak phase detection to extract the phase characteristics of the impact component in the signal;

[0019] S08: After obtaining the phase characteristics of the impact, capture the parameter characteristics to further diagnose the reciprocating compressor fault.

[0020] S09: Visualize diagnostic results and collected data, and remotely process and retrieve data.

[0021] Furthermore, the kurtosis-correlation index is used as the indicator in step S02, including:

[0022] Step 1, kurtosis index:

[0023] For discrete signal x(i)={x i |i=1,2,…,n}, where n is the number of signal points, and the kurtosis K is defined as:

[0024]

[0025] In the formula Let x(i) be the mean, defined as:

[0026]

[0027] The kurtosis values ​​of each component are calculated separately, and components with kurtosis values ​​less than 5.0 are removed, thus completing the first step of screening IMF components.

[0028] The second step is to analyze correlation indicators.

[0029] Calculate the correlation coefficients between each order of IMF components and the original function. Assuming there are two continuous time-domain signals x(t) and y(t), their correlation coefficient R(xy) is defined as:

[0030]

[0031] Select IMFs with a correlation coefficient greater than 10% to complete the second screening step.

[0032] Further: In step S03, the adaptive variable-scale morphological filtering includes:

[0033] For a discrete signal f(n) with a total of N points, where the independent variable n = 0, 1, ..., N-1, and a structuring element sequence g(m) with a total of M points, where the independent variable m = 0, 1, ..., M-1, and N < M, define an expansion operator. and corrosion operator for:

[0034]

[0035]

[0036] in, This represents the expansion of f(n) with respect to g(m). This represents the erosion of f(n) with respect to g(m).

[0037] The morphological opening operator (°) and the morphological closing operator (·) are defined by the sequential combination of the dilation operator and the erosion operator:

[0038]

[0039]

[0040] Where (f°g)(n) represents the morphological opening operator of f(n) with respect to g(m), and (f·g)(n) represents the morphological closing operator of f(n) with respect to g(m).

[0041] Furthermore, the morphological opening-closing (OC) and morphological closing-opening (CO) operators are defined by sequential combinations of the morphological opening and closing operators:

[0042] OC[f(n)]=(f°g·g)(n)

[0043] CO[f(n)]=(f·g°g)(n)

[0044] Where OC[f(n)] represents the morphological opening and closing operator of f(n), and CO[f(n)] represents the morphological closing and opening operator of f(n).

[0045] A cascaded morphological filter Γ[f(n)] consisting of morphological opening-closing operators and morphological closing-opening operators:

[0046]

[0047] Furthermore, the filter structuring element uses a flat structure with a height of 0. For a discrete signal f(n) (n = 0, 1, ..., N-1), a structuring element sequence g(n, m, k) is defined (n = 0, 1, ..., N-1; m = 0, 1, ..., k-1), where N is the number of signal points and k is the width of the structuring element at the nth sampling point. The width k is determined as follows:

[0048] (1) Let the phase and amplitude of the signal at each point be respectively... x(i)={x i |i=1,2,…,N}, where N is the number of signal points;

[0049] (2) Calculate all local extrema of the signal, and denote their phase and amplitude as φ(j) = {φj}. j |j=1,2,…,M}、y(j)={y j |j=1,2,…,M}, where M is the number of extreme points;

[0050] (3) Linearly normalize the amplitude of the signal extreme points y norm (j):

[0051]

[0052] Among them, y j Let represent the magnitude of the j-th extreme point, min{y(j)} be the minimum value of the extreme point sequence y(j), and max{y(j)} be the maximum value of the extreme point sequence y(j).

[0053] (4) Calculate the signal waveform scale s:

[0054]

[0055] (5) Define nonlinear mapping By blurring the boundary between impact and non-impact components and altering the normalized local extremum amplitude distribution, the width of the structuring element corresponding to each point in the signal, k(i), is calculated as follows: i |i=1,2,…,N} is determined by the following formula:

[0056]

[0057] Among the symbols Indicates rounding up, mapping Defined as:

[0058]

[0059] In the formula, 'a' is a variable parameter that controls the "curvature" of the mapping curve, used to adjust the amplitude distribution at extreme points. The recommended value range is a∈[2,8]. Using k(i) as the element width, cascaded morphological filtering is applied to the signal to effectively suppress non-impact components.

[0060] Further, in step S05, the abrupt change components in the denoised signal are amplified using the Teager energy operator:

[0061] Specifically, for discrete signals, for a discrete signal x(i) (i = 1, 2, ..., n) with a total number of points n, a three-point symmetric differential energy operator is constructed to smooth it. The energy operator Ψ[x(i)] is defined as:

[0062]

[0063] Further, in step 06: the Hilbert envelope of the mutation component is obtained, and the envelope is smoothed using a Savitzky-Golay filter.

[0064] Specifically, for a continuous-time signal x(t), its Hilbert transform Defined as:

[0065]

[0066] Constructing analytic signals

[0067]

[0068] Then the Hilbert envelope of signal x(t) is:

[0069] Divide the envelope of the Savitzky-Golay smoothed signal into n segments in the angular domain graph, calculate the average amplitude of the signal in each segment, and arrange them into a sequence l[i], i = 0, 1, ..., n-1. Calculate the mean of this sequence. Using the mean square error σ, the 3-σ rule is applied to remove segments with obvious impact characteristics, leaving a sequence of m segments. Characterizes the relatively stationary components in a signal. Let the average value of the sequence r[j] be... The difference between the maximum and minimum values ​​is Δr. The weighting coefficient α is used to adjust the threshold, thus defining the impact detection threshold as follows:

[0070] Furthermore, the number of segments is n = 72, meaning each segment is 5°.

[0071] The technical advantages of this invention are as follows: This application proposes a method to obtain several intrinsic mode functions (IMFs) by performing overall average empirical mode decomposition on vibration signals. An adaptive variable-scale morphological filter is used to configure an appropriate structuring element width for each signal point. Conversely, the excellent performance of cascaded morphological filters in pulse suppression is utilized to retain the impact component in the IMF while suppressing the non-impact component. Finally, an adaptive impact peak detection method is used to automatically determine the phase characteristics of the impact component in the signal. This device and method can effectively extract the impact characteristics from the signal, providing a basis for fault diagnosis of reciprocating compressors and supporting and guaranteeing predictive maintenance of reciprocating compressors on offshore platforms in the future.

[0072] A method and apparatus for extracting impact features from vibration signals of a floating platform reciprocating compressor based on an EEMD-adaptive variable-scale morphological filter is proposed. This method uses mathematical morphological filtering to suppress harmonic components in the IMF of each order to be reconstructed. Compared with morphological filtering of the denoised signal after EEMD decomposition and reconstruction, it has a better harmonic suppression effect, enabling accurate extraction of impact features from the vibration signals of the floating platform reciprocating compressor and achieving fault diagnosis. Attached Figure Description

[0073] Figure 1 A flowchart for a fault diagnosis method that extracts impact features from vibration signals of a reciprocating compressor on a floating platform;

[0074] Figure 2 A schematic diagram of a fault diagnosis device for extracting vibration signal impact features of a reciprocating compressor on a floating platform.

[0075] Figures 3a-3d To simulate the impact, harmonics, noise, and their mixed vibration signals of a reciprocating compressor;

[0076] Figure 4 To reconstruct the noise-reduced signal after adaptive variable-scale morphological filtering for simulating a reciprocating compressor;

[0077] Figure 5 To simulate the impact threshold calculation and peak detection results of a reciprocating compressor;

[0078] Figure 6 The actual reciprocating compressor contains body acceleration signals and dynamic pressures at the crankshaft end (CE) and cylinder head end (HE);

[0079] Figure 7 This is the actual body acceleration signal in a reciprocating compressor;

[0080] Figure 8 The noise-reduced signal is reconstructed after adaptive variable-scale morphological filtering for a real reciprocating compressor.

[0081] Figure 9 This is the filtering result of the Teager energy operator for a real reciprocating compressor;

[0082] Figure 10 This is the result of the actual impact threshold calculation and peak detection for reciprocating compressors. Detailed Implementation

[0083] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely represents selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0084] Based on this, this application proposes a method to obtain several intrinsic mode functions (IMFs) by performing ensemble empirical mode decomposition (EEMD) on vibration signals. An adaptive variable scale morphological filter (AVSMF) is used to configure an appropriate structuring element width for each signal point. The excellent pulse suppression performance of the cascaded morphological filter is then utilized to retain the impact component in the IMF while suppressing the non-impact component. Finally, an adaptive impact peak detection method is used to automatically determine the phase characteristics of the impact component in the signal. This device and method can effectively extract the impact characteristics from the signal, providing a basis for fault diagnosis of reciprocating compressors and supporting and guaranteeing the predictive maintenance of reciprocating compressors on offshore platforms in the future.

[0085] A method and apparatus for extracting impact features from vibration signals of a floating platform reciprocating compressor based on an EEMD-adaptive variable-scale morphological filter is proposed. This method uses mathematical morphological filtering to suppress harmonic components in the IMF of each order to be reconstructed. Compared with morphological filtering of the denoised signal after EEMD decomposition and reconstruction, it has a better harmonic suppression effect, enabling accurate extraction of impact features from the vibration signals of the floating platform reciprocating compressor and achieving fault diagnosis.

[0086] Figure 1 A flowchart illustrating a fault diagnosis method for extracting vibration signal impact features of a reciprocating compressor on a floating platform, provided in an embodiment of this application. See also... Figure 1 As shown in the embodiments of this application, the method for extracting vibration signal impact features of a floating platform reciprocating compressor for fault diagnosis specifically includes:

[0087] Step 1: Perform EEMD decomposition on the vibration signal to be processed to obtain several IMFs.

[0088] Specifically, EEMD decomposition involves introducing Gaussian white noise into the original signal, performing multiple EMD decompositions, and averaging the IMF components from the multiple decompositions to obtain the final IMF.

[0089] Each IMF must meet the following two conditions:

[0090] (1) Throughout the entire signal, the difference between the number of extreme points and the number of zero-crossing points of the signal is less than or equal to 1;

[0091] (2) At any given time, the upper envelope formed by the local maxima and the lower envelope formed by the local minima are locally symmetrical with respect to the time axis.

[0092] Main steps of EMD decomposition:

[0093] (1) Locate the upper and lower extreme points of the vibration signal and draw the upper and lower envelope lines;

[0094] (2) Calculate the mean of the upper and lower envelopes to obtain the mean envelope;

[0095] (3) Subtract the mean envelope from the original vibration signal to obtain the intermediate signal;

[0096] (4) Determine whether the intermediate signal meets the two conditions of IMF. If it does, the intermediate signal is recorded as an IMF component. If it does not meet the conditions, repeat steps (1) to (4) for the intermediate signal.

[0097] Step 2: Select several IMF components by using the kurtosis-correlation index to identify IMF components of several orders that contain obvious shock characteristics.

[0098] Specifically, the first step is the principle of kurtosis index:

[0099] Kurtosis is a time-domain statistical dimensionless index, often used to detect impulse characteristics in signals. For a discrete signal x(i) = {x i |i=1,2,…,n}, where n is the number of signal points, and the kurtosis K is defined as:

[0100]

[0101] In the formula Let x(i) be the mean, defined as:

[0102]

[0103] The kurtosis values ​​of each component are calculated separately, and components with kurtosis values ​​less than 5.0 are removed, thus completing the first step of screening IMF components.

[0104] The second step is to analyze correlation indicators.

[0105] During EEMD decomposition, due to factors such as computational errors and edge effects, the number of IMF components obtained is greater than the actual components of the original signal. These extra IMF components are called "pseudo-components". If these pseudo-components are included in the signal reconstruction, it will introduce new noise, so these pseudo-components must be removed.

[0106] Calculate the correlation coefficients between each order of IMF components and the original function. Assuming there are two continuous time-domain signals x(t) and y(t), their correlation coefficient R(xy) is defined as:

[0107]

[0108] Select IMFs with a correlation coefficient greater than 10% to complete the second screening step.

[0109] Step 3: Perform adaptive variable-scale morphological filtering on the selected IMF components to filter out non-impact components and improve the signal-to-noise ratio of the impact signal.

[0110] For a discrete signal f(n) with a total of N points, where the independent variable n = 0, 1, ..., N-1, and a structuring element sequence g(m) with a total of M points, where the independent variable m = 0, 1, ..., M-1, N < M, define an expansion operator. and corrosion operator

[0111]

[0112]

[0113] in, This represents the expansion of f(n) with respect to g(m). This represents the erosion of f(n) with respect to g(m);

[0114] The morphological opening operator (°) and the morphological closing operator (·) are defined by the sequential combination of the dilation operator and the erosion operator:

[0115]

[0116]

[0117] Where (f°g)(n) represents the morphological opening operator of f(n) with respect to g(m), and represents the morphological closing operator of f(n) with respect to g(m);

[0118] Furthermore, the morphological opening-closing (OC) and morphological closing-opening (CO) operators are defined by sequential combinations of morphological opening and morphological closing operators:

[0119] OC[f(n)]=(f°g·g)(n)

[0120] CO[f(n)]=(f·g°g)(n)

[0121] Where OC[f(n)] represents the morphological opening and closing operator of f(n), and CO[f(n)] represents the morphological closing and opening operator of f(n).

[0122] The morphological opening-closing operator can suppress positive pulses in the signal and eliminate sharp "peaks" in the signal, while the morphological closing-opening operator can suppress negative pulses in the signal and fill in the low "valleys" in the signal. In order to remove both positive and negative bidirectional pulses in the signal at the same time, a cascaded morphological filter Γ[f(n)] composed of morphological opening-closing and morphological closing-opening operators can be defined.

[0123]

[0124] Traditional morphological filters use structuring elements with fixed widths, while the morphological filter proposed in this method uses structuring element widths that adaptively vary with the amplitude of local extrema at each peak (trough) in the signal. The structuring element uses a flat structure with a height of 0. For a discrete signal f(n) (n = 0, 1, ..., N-1), the structuring element sequence g(n, m, k) (n = 0, 1, ..., N-1; m = 0, 1, ..., k-1) is defined, where N is the number of signal points, and k is the width of the structuring element at the nth sampling point. The width k is determined as follows:

[0125] (1) Calculate the phase and amplitude of all local extrema in the signal, denoted as φ(j)={φ j |j=1,2,…,M}、y(j)={y j |j=1,2,…,M}, where M is the number of local extrema;

[0126] (2) Calculate all local extrema of the signal, and denote their phase and amplitude as φ(j) = {φj}. j |j=1,2,…,M}、y(j)={y j |j=1,2,…,M}, where M is the number of extreme points;

[0127] (3) Linearly normalize the amplitude of the signal at local extreme points. norm (j):

[0128]

[0129] Among them, y j Let represent the magnitude of the j-th extreme point, min{y(j)} be the minimum value of the extreme point sequence y(j), and max{y(j)} be the maximum value of the extreme point sequence y(j).

[0130] (4) Calculate the signal waveform scale s:

[0131]

[0132] (5) Define nonlinear mapping By blurring the boundary between impact and non-impact components and altering the normalized local extremum amplitude distribution, the width of the structuring element corresponding to each point in the signal, k(i), is calculated as follows: i |i=1,2,…,N} is determined by the following formula:

[0133]

[0134] Among the symbols Indicates rounding up, mapping Defined as:

[0135]

[0136] In the formula, α is a variable parameter that controls the "curvature" of the mapping curve and is used to adjust the amplitude distribution of extreme points. The recommended value range is a∈[2,8]. Cascaded morphological filtering is performed on the signal with k(i) as the element width to effectively suppress non-impact components.

[0137] Step 4: Reconstruct the denoised signal from the IMF components after adaptive variable-scale morphological filtering.

[0138] Step 5: Use the Teager energy operator to amplify the amplitude-frequency abrupt change components in the denoised signal.

[0139] Specifically, for discrete signals, for a discrete signal x(i) (i = 1, 2, ..., n) with a total number of points n, a three-point symmetric differential energy operator is constructed to smooth it. The energy operator Ψ[x(i)] is defined as:

[0140]

[0141] Step 6: Calculate the Hilbert envelope of the mutation component and smooth the envelope using the Savitzky-Golay filter.

[0142] Specifically, for a continuous-time signal x(t), its Hilbert transform Defined as:

[0143]

[0144] Constructing analytic signals

[0145]

[0146] Then the Hilbert envelope of signal x(t) is:

[0147] The Savitzky-Golay filter is a widely used data stream smoothing and noise reduction method. It is a method that uses a polynomial in the time domain and employs the least squares method to achieve the best fit by moving the window.

[0148] Divide the envelope of the Savitzky-Golay smoothed signal into n segments in the angular domain graph, calculate the average amplitude of the signal in each segment, and arrange them into a sequence l[i], i = 0, 1, ..., n-1. Calculate the mean of this sequence. Using the mean square error σ, the 3-σ rule is applied to remove segments with obvious impact characteristics, leaving a sequence of m segments. Characterizes the relatively stationary components in a signal. Let the average value of the sequence r[j] be... The difference between the maximum and minimum values ​​is Δr, and the weighting coefficient 'a' is used to adjust the threshold. Therefore, the impact detection threshold can be defined as follows: In this embodiment, the number of segments is n = 72, that is, each 5° is a small segment.

[0149] Step 7: Perform peak phase detection to extract the phase characteristics of the impact component in the signal.

[0150] Specifically, by performing zero-crossing detection on the smoothed envelope with a threshold value as the zero point, all "peaks" above the threshold line can be extracted, which are the impact peaks. False impacts are then eliminated by determining the angle of duration of each impact peak within the angular domain map, with a duration angle of 5°. Finally, the angle in the angular domain map where the peak value of each impact peak is located is determined as the phase feature of the impact.

[0151] Step 8: After obtaining the phase characteristics of the impact, capture other characteristics such as the impact amplitude and energy to further diagnose the reciprocating compressor fault.

[0152] Figure 2 This is a schematic diagram of a fault diagnosis device for extracting vibration signal impact features of a reciprocating compressor on a floating platform. (Reference) Figure 2 As shown in the embodiment of this application, the fault diagnosis device for extracting vibration signal impact features of a floating platform reciprocating compressor specifically includes:

[0153] The data acquisition module is used to acquire vibration data of the reciprocating compressor on the floating platform and transmit it to the host computer for processing.

[0154] The acquisition module specifically includes: a key phase sensor, a multi-point vibration sensor, and a signal conditioning and acquisition device. The key phase sensor is used to acquire key phase pulse signals from impact locations such as the cylinder valve cover, body, and crankshaft of the reciprocating compressor. By setting the trigger value, the time interval of one working cycle of the compressor can be obtained. The multi-point vibration sensor can capture vibration signals from multiple key locations on the compressor. The signal conditioning and acquisition device includes: data acquisition hardware using the NI 9263 sound and vibration voltage input module, and an NI Compact RIO 9047 chassis paired with the NI 9263 module for low-level data acquisition and TCP / IP network data transmission. First, data is read from the FIFO, and then the data is transmitted to the host computer via the TCP protocol. The TCP protocol supports data transmission to multiple clients, i.e., transmission to multiple host computers.

[0155] The acquisition module can provide the raw vibration signal data required by the subsequent processing module and provide the basis for multi-source fault diagnosis for the subsequent diagnostic module.

[0156] The processing module, connected to the acquisition module, includes the storage and processing of acquired vibration signal data, and processes the acquired vibration signals through EEMD-adaptive variable-scale morphological filtering.

[0157] The extraction module, connected to the processing module, is used to perform peak detection to extract the phase characteristics of the impact component in the signal, and to capture other characteristics such as the amplitude and energy of the impact.

[0158] The diagnostic evaluation module, connected to the extraction module, establishes a knowledge base of reciprocating compressor fault characteristics and performs fault diagnosis based on the vibration signal impact characteristics captured by the extraction module.

[0159] Specifically, the extraction module extracts impact features based on continuously collected data from the field in real time, compares the multi-source feature values ​​with the reciprocating compressor fault feature knowledge base, and outputs diagnostic results. The time interval for diagnostic output results can be set, i.e., the diagnostic interval for the compressor's working cycle can be customized.

[0160] The visualization module, connected to the diagnostic evaluation module and the data acquisition module, visualizes the diagnostic results and acquired data. It also enables remote data processing and retrieval through interactive web access via a remote terminal.

[0161] Specifically, the module is embedded in the actual 3D model of the compressor, visualizing the collected measurement results in real time and highlighting the location of diagnosed faults. Furthermore, considering the characteristics of reciprocating compressors on offshore platforms that require remote monitoring and storage of large amounts of data, the web-based diagnostic module is set up locally, allowing interactive access via a remote terminal, avoiding bandwidth issues introduced by large data backhaul. A remote data download port (on-demand download) is provided, enabling experts who have installed the client to perform in-depth analysis, achieving remote monitoring and diagnosis of the compressor system.

[0162] The floating platform reciprocating compressor vibration signal impact feature extraction fault diagnosis device can collect vibration signals from the acquisition module within each compressor working cycle. After processing and extraction, the device obtains the impact phase feature values ​​of the vibration signals. These feature values ​​are then matched against a knowledge base in the diagnostic evaluation module to provide a compressor status diagnosis result. A visualization module displays the results along with the monitored and acquired values. The entire device has the capability to remotely acquire data and monitor the compressor status, ensuring the safe and stable operation of the reciprocating compressor.

[0163] Example 1:

[0164] A continuous time-domain model of the volume acceleration signal in the reciprocating compressor is established as follows:

[0165]

[0166] In the formula, τ i A i f represents the time and magnitude of the i-th impact. ni Let f be a certain natural frequency of the body-sensor system in the compressor, ζ be the corresponding system damping ratio, and f be the frequency of the body-sensor system. j For the harmonic frequency components in the signal, 1(t-τ) i ) is the unit step function, and p(t) is the noise signal.

[0167] Figures 3a-3d To simulate the impact, harmonics, noise, and mixed vibration signals of a reciprocating compressor, reference Figures 3a-3d The reciprocating compressor speed is set to 900 RPM, and the accelerometer sampling rate f is... s =51200Hz, assuming the signal contains three different types of impulse signals, representing high-frequency high-amplitude impulses, low-frequency low-amplitude impulses, and high-frequency low-amplitude impulses, respectively. Their amplitudes A i The system's natural frequency f is 160, 90, and 90 respectively. ni The damping ratios are 3000, 1500, and 3000 respectively. i The values ​​are 0.1, 0.1, and 0.1 respectively, representing the phase τ of the impact. i The simulated signal references are 60°, 120°, and 270° respectively, for one cycle. Figure 3a Meanwhile, to simulate the actual operating conditions of a compressor, harmonic components with amplitudes of 20 and 10, and frequencies of 90Hz and 180Hz, are introduced, assuming they are in phase. Figure 3b Finally, a Gaussian white noise reference with a signal-to-noise ratio of -3dB is introduced. Figure 3c The final synthesized simulation signal reference Figure 3d .

[0168] Since it takes some time for an impact to reach its peak value, the actual peak phase of the impact lags slightly behind the phase in which the impact occurs.

[0169] The phases of the three impact peaks in this example are shown in Table 1.

[0170] Table 1 Simulated Impact Phase

[0171]

[0172] Impact feature extraction was performed on the simulated mixed vibration signal of the reciprocating compressor, following these steps:

[0173] The mixed signal is decomposed using EEMD to obtain several orders of IMF;

[0174] The kurtosis of each IMF and its correlation coefficient with the original signal are calculated, and finally the 2nd, 3rd and 4th order IMFs are selected.

[0175] Adaptive variable-scale morphological filtering is applied to the selected IMFs, and the impulsive components in the filtered signal are preserved while the non-impulsive components are suppressed.

[0176] The filtered signal is reconstructed to obtain the denoised signal. At this point, both the harmonic and noise signals of the original signal are well suppressed. See [link to documentation]. Figure 4 The abrupt changes in the signal are amplified using the Teager energy operator;

[0177] Envelope smoothing yields the final signal to be detected;

[0178] The impact phase is obtained by adaptively calculating the impact threshold and performing peak detection. See [link to relevant documentation]. Figure 5 The test results are shown in Table 2.

[0179] Table 2 Impact test results

[0180]

[0181] The impact phase detection results show that both high-amplitude and low-amplitude impacts, as well as high-frequency and low-frequency impacts, can be accurately detected, and the phase error of the impact peak is less than 0.5 degrees.

[0182] Example 2:

[0183] For example, a newly commissioned flash vapor three-stage double-acting reciprocating compressor has two primary low-pressure cylinders and two secondary high-pressure cylinders, arranged symmetrically at both ends. Each cylinder has a PCB-EX603C01 accelerometer installed in its central section to monitor the vibration signal. The acquisition device used is the NI Compact RIO 9047 reconfigurable embedded measurement and control system, paired with an NI 9263 sound and vibration input module, enabling high-speed acquisition of acceleration signals. In this example, the compressor operates at 1200 RPM, and the sampling rate is set to 10240 Hz. Acceleration signals from two cycles at the central section of the #1 primary low-pressure cylinder, along with the dynamic gas pressures at the cylinder head (HE) and crankshaft (CE) sides, are transformed into a angular domain diagram with a phase range of 0° to 720°, referencing... Figure 6 As shown, the acceleration signal is extracted separately as a reference. Figure 7 As shown. The denoised signal reference is obtained after EEMD-adaptive variable-scale morphological denoising processing. Figure 8 As shown, it is evident that the non-impact components in the signal are significantly suppressed. This is achieved after Teager energy operator filtering (reference...). Figure 9 (As shown), after envelope smoothing filtering, the impact threshold is adaptively identified, and finally the impact phase is determined by peak phase detection, with reference to... Figure 10 As shown in the figure, the horizontal line represents the impact detection threshold line, and the vertical line represents the impact peak detection result. The impact peak phase detection results are shown in Table 3.

[0184] Table 3 Impact test results

[0185]

[0186] Based on the impact peak phase detection results and the dynamic pressure inside the HE and CE cylinders, it can be clearly seen that the stable phases of impact 1 and impact 2 indicate that the exhaust valve at the CE end and the intake valve at the HE end are seated, respectively; the phases of impact 3 and 5 are relatively stable, indicating that the exhaust valve at the HE end and the intake valve at the CE end are seated; while the phase of impact 4 fluctuates greatly. Combined with the fact that the dynamic pressure inside the HE end cylinders shows a brief abnormal rise in air pressure around 280° in a single cycle, it is determined that the exhaust valve at the HE end vibrates when it is seated, causing the valve plate to impact the valve seat multiple times, thus introducing additional impact.

[0187] The above-described embodiments are merely preferred embodiments provided to fully illustrate the present invention, and the scope of protection of the present invention is not limited thereto. Equivalent substitutions or modifications made by those skilled in the art based on the present invention are all within the scope of protection of the present invention. The scope of protection of the present invention is defined by the claims.

Claims

1. A fault diagnosis method for a floating platform reciprocating compressor, characterized in that, It includes the following steps: S01: Decompose the vibration signal to be processed to obtain several IMF components; S02: Select several IMF components by using indicators to identify IMFs of certain orders that contain obvious shock characteristics; S03: Perform adaptive variable-scale morphological filtering on the selected IMF components to filter out non-impact components; S04: Reconstruct the denoised signal from the IMF components after adaptive variable-scale morphological filtering; S05: Amplify the amplitude-frequency abrupt change components in the noise-reduced signal; S06: Obtain the envelope of the mutation component and smooth the envelope using a smoothing filter; S07: Perform peak phase detection to extract the phase characteristics of the impact component in the signal; S08: After obtaining the phase characteristics of the impact, capture the parameter characteristics to further diagnose the reciprocating compressor fault. S09: Visualize diagnostic results and collected data, and remotely process and retrieve data; Step 06: Obtain the Hilbert envelope of the mutation component and smooth the envelope using the Savitzky-Golay smoothing filter; Specifically, for a continuous-time signal x(t), its Hilbert transform Defined as: Constructing analytic signals Then the Hilbert envelope of signal x(t) is: Divide the envelope of the Savitzky-Golay smoothed signal into n segments in the angular domain graph, calculate the average amplitude of the signal in each segment, and arrange them into a sequence l[i], i = 0, 1, ..., n-1. Calculate the mean of this sequence. Using the mean square error σ, the 3-σ rule is applied to remove segments with obvious impact characteristics, leaving a sequence of m segments. To characterize the relatively stationary components in a signal, let the average value of the sequence r[j] be... The difference between the maximum and minimum values ​​is Δr, and the weighting coefficient 'a' is used to adjust the threshold. Therefore, the impact detection threshold can be defined as follows:

2. The diagnostic method as described in claim 1, characterized in that: The kurtosis-correlation index used in step S02 includes: Step 1, kurtosis index: For discrete signal x(i)={x i |i=1,2,…,n}, where n is the number of signal points, and the kurtosis K is defined as: In the formula Let x(i) be the mean, defined as: The kurtosis values ​​of each order of IMF components are calculated separately, and components with kurtosis values ​​less than 5.0 are removed, thus completing the first step of IMF component screening. The second step is to analyze correlation indicators. Calculate the correlation coefficients between each order of IMF components and the original function. Assuming there are two continuous time-domain signals x(t) and y(t), their correlation coefficient R(xy) is defined as: Select IMFs with a correlation coefficient greater than 10% to complete the second screening step.

3. The diagnostic method as described in claim 1, characterized in that: In step S03, the adaptive variable-scale morphological filtering includes: For a discrete signal f(n) with a total of N points, where the independent variable n = 0, 1, ..., N-1, and a structuring element sequence g(m) with a total of M points, where the independent variable m = 0, 1, ..., M-1, and N < M, define an expansion operator. and corrosion operator for: in, This represents the expansion of f(n) with respect to g(m). This represents the erosion of f(n) with respect to g(m); The morphological opening operator (°) and the morphological closing operator (·) are defined by the sequential combination of the dilation operator and the erosion operator: Where (f°g)(n) represents the morphological opening operator of f(n) with respect to g(m), and (f·g)(n) represents the morphological closing operator of f(n) with respect to g(m); Furthermore, the morphological opening-closing (OC) operator and the morphological closing-opening (CO) operator are defined by sequential combination of the morphological opening operator and the morphological closing operator: OC[f(n)]=(f°g·g)(n) CO[f(n)]=(f·g°g)(n) Where OC[f(n)] represents the morphological opening-closing operator of f(n), and CO[f(n)] represents the morphological closing-opening operator of f(n); A cascaded morphological filter Γ[f(n)] consisting of morphological opening-closing operators and morphological closing-opening operators:

4. The diagnostic method as described in claim 3, characterized in that: The filter's structuring elements use a flat structure with a height of 0. For a discrete signal f(n) (n = 0, 1, ..., N-1), a structuring element sequence g(n, m, k) (n = 0, 1, ..., N-1; m = 0, 1, ..., k-1) is defined, where N is the number of signal points, and k is the width of the structuring element at the nth sampling point. The width k is determined as follows: (1) Let the phase and amplitude of the signal at each point be respectively... x(i)={x i |i=1,2,…,N}, where N is the number of signal points; (2) Calculate the phase and amplitude of all local extrema in the signal, denoted as φ(j)={φ j |j=1,2,…,M}、y(j)={y j |j=1,2,…,M}, where M is the number of local extrema; (3) Linearly normalize the amplitude of the signal at local extreme points. norm (j): Among them, y j Let represent the magnitude of the j-th extreme point, min{y(j)} be the minimum value of the extreme point sequence y(j), and max{y(j)} be the maximum value of the extreme point sequence y(j). (4) Calculate the signal waveform scale s: (5) Define nonlinear mapping By blurring the boundary between impact and non-impact components and altering the normalized local extremum amplitude distribution, the width of the structuring element corresponding to each point in the signal, k(i), is calculated as follows: i |i=1,2,…,N} is determined by the following formula: Among the symbols Indicates rounding up, mapping Defined as: In the formula, a is a variable parameter used to adjust the amplitude distribution of extreme points. The recommended value range is a∈[2,8]. With k(i) as the element width, cascaded morphological filtering is performed on the signal to effectively suppress non-impact components.

5. The diagnostic method as described in claim 1, characterized in that: Step S05: Amplify the amplitude-frequency abrupt change component in the denoised signal using the Teager energy operator: Specifically, for discrete signals, for a discrete signal x(i) (i = 1, 2, ..., n) with a total number of points n, a three-point symmetric differential energy operator is constructed to smooth it. The energy operator Ψ[x(i)] is defined as:

6. The diagnostic method as described in claim 5, characterized in that: The number of segments n = 72, meaning each segment is 5°.

7. A fault diagnosis device for a floating platform reciprocating compressor, employing the method described in claim 1, characterized in that, It includes: The acquisition module is used to acquire vibration signal data of the reciprocating compressor of the floating platform and transmit it to the host computer for processing. The processing module, connected to the acquisition module, processes the acquired vibration signal through EEMD-adaptive variable-scale morphological filtering. The extraction module, connected to the processing module, is used to extract the phase characteristics of the impact component in the signal by peak detection and to capture the parametric characteristics of the impact. The diagnostic evaluation module, connected to the extraction module, establishes a fault feature knowledge base for reciprocating compressors, and performs fault diagnosis by comparing the vibration signal impact features captured by the extraction module with the fault feature knowledge base and outputting the diagnostic results. The visualization module, connected to the diagnostic evaluation module and the data acquisition module, visualizes the diagnostic results and acquired data, and enables remote data processing and retrieval through interactive web access via a remote terminal.

Images