A method and system for monitoring the operating status of a gas compressor

Abstract

The present invention relates to the technical field of equipment operation status monitoring, and in particular to a method and system for monitoring the operation status of a gas compressor. The method respectively obtains a real-time monitoring data sequence of the gas compressor under each preset dimension in the current compression cycle, and for any sampling moment of the current compression cycle, obtains the basic weight of each preset dimension at any sampling moment based on a preset number of groups of historical data; divides the current compression cycle into stages to obtain a stage weight set of each stage, and adjusts the basic weight of each preset dimension at any sampling moment based on the stage weight set to obtain a dynamic weight; uses the dynamic weight to perform weighted processing on the real-time monitoring data of each preset dimension to obtain an equipment health index at any sampling moment; and issues an abnormal warning to the gas compressor based on the change rate of the equipment health index at adjacent sampling moments in the current compression cycle, so as to make the operation status monitoring of the gas compressor more accurate.

Description

Technical Field

[0001] The present invention relates to the technical field of equipment operating status monitoring, and in particular to a method and system for monitoring the operating status of a gas compressor. Background Art

[0002] In industrial production, gas compressors are key power equipment in numerous process flows. The stability and reliability of their operating status are directly linked to the safe and efficient operation of the entire production system and the consistent quality of products. Gas compressors play an indispensable role in gas transportation and compression in chemical production, pressurization in oil and gas extraction, and refrigerant circulation in refrigeration and air-conditioning systems. However, traditional methods for monitoring the operating status of gas compressors struggle to comprehensively and accurately assess their operating conditions and potential failure risks.

[0003] In the existing technology, some methods rely on only a single type of sensor for monitoring, for example: monitoring the exhaust pressure of the compressor only through a pressure sensor, or using only a vibration sensor to detect the vibration of the equipment. However, the data dimensions obtained by such methods are too single and cannot fully reflect the complex operating status of the gas compressor. Moreover, these single data have not established an effective correlation model with the key performance indicators of the gas compressor and the operating parameters of the overall production system, making it difficult to achieve an in-depth assessment of the health status of the gas compressor. In addition, some methods use an alarm mechanism based on a fixed threshold, which triggers an alarm when the monitoring data exceeds the preset threshold. However, this mechanism lacks flexibility and intelligence, and cannot be dynamically adjusted according to the actual operating conditions and historical data of the gas compressor. It is easy to cause false alarms or missed alarms, and it is difficult to meet the requirements of modern industrial production for precise and intelligent monitoring of the operating status of gas compressors. Summary of the Invention

[0004] In view of this, an embodiment of the present invention provides a method and system for monitoring the operating status of a gas compressor to solve the problem of how to dynamically optimize the operating status monitoring of the gas compressor.

[0005] In a first aspect, an embodiment of the present invention provides a method for monitoring the operating status of a gas compressor, the method comprising the following steps:

[0006] Acquire a real-time monitoring data sequence of the gas compressor in each preset dimension during the current compression cycle, wherein the preset dimensions include pressure data, triaxial vibration data, and sound data;

[0007] For any sampling moment of the current compression cycle, a preset number of groups of historical data matching any sampling moment are obtained from the historical monitoring data sequence under each preset dimension within the historical compression cycle, and a basic weight of each preset dimension at any sampling moment is obtained based on the preset number of groups of historical data;

[0008] Based on the similarity between the historical monitoring data sequence and the real-time monitoring data sequence of the historical compression cycle, the current compression cycle is divided into stages, and a stage weight set consisting of the stage weight of each preset dimension in each stage is obtained. According to the stage weight set corresponding to the stage to which any sampling moment belongs, the basic weight of each preset dimension at any sampling moment is adjusted to obtain the dynamic weight of each preset dimension at any sampling moment;

[0009] Based on the dynamic weight of each preset dimension at any sampling moment, the real-time monitoring data of each preset dimension at any sampling moment is weightedly processed to obtain the equipment health index at any sampling moment; according to the change rate of the equipment health index at adjacent sampling moments in the current compression cycle, an abnormal warning is issued for the gas compressor.

[0010] In a second aspect, an embodiment of the present invention provides a gas compressor operating status monitoring system, comprising a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements a gas compressor operating status monitoring method as described in the first aspect.

[0011] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0012] The present invention obtains a real-time monitoring data sequence of a gas compressor under each preset dimension in a current compression cycle, wherein the preset dimensions include pressure data, triaxial vibration data, and sound data. For any sampling moment in the current compression cycle, a preset number of groups of historical data matching the any sampling moment are obtained from the historical monitoring data sequence under each preset dimension in the historical compression cycle, and a basic weight for each preset dimension at the any sampling moment is obtained based on the preset number of groups of historical data. The current compression cycle is divided into stages based on the similarity between the historical monitoring data sequence and the real-time monitoring data sequence of the historical compression cycle, and a stage weight set consisting of the stage weights of each preset dimension in each stage is obtained. The basic weight of each preset dimension at any sampling moment is adjusted based on the stage weight set corresponding to the stage to which the any sampling moment belongs, to obtain a dynamic weight for each preset dimension at the any sampling moment. Based on the dynamic weight for each preset dimension at the any sampling moment, the real-time monitoring data of each preset dimension at the any sampling moment is weighted to obtain an equipment health index at the any sampling moment. An abnormality warning is issued for the gas compressor based on the rate of change of the equipment health index at adjacent sampling moments in the current compression cycle. Among them, the compression process of the current compression cycle is divided into real-time stages through historical monitoring data in historical compression cycles to ensure that the weight of the key preset dimension in the corresponding stage is increased, and then the stage weight set composed of the stage weight of each preset dimension in each stage is used to perform real-time dynamic adjustment on the basic weight of each preset dimension at any sampling moment in the current compression cycle, so as to obtain a dynamic weight that can more accurately reflect the importance of the preset dimension in equipment abnormality monitoring. Therefore, by performing weighted fusion analysis on real-time monitoring data under multiple dimensions, the judgment deviation of monitoring the operating status of the gas compressor based on monitoring data under a single dimension is reduced, making the operating status monitoring of the gas compressor more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0014] Figure 1 This is a flow chart of a method for monitoring the operating status of a gas compressor provided in Example 1 of the present invention. DETAILED DESCRIPTION

[0015] The embodiments of the present disclosure are described in detail below, and examples of the embodiments are shown in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and intended to be used to explain the present disclosure, but should not be understood as limiting the present disclosure.

[0016] It should be noted that the terms "first," "second," and the like in the specification of the present disclosure and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Instead, they are merely examples of devices and methods consistent with some aspects of the present disclosure.

[0017] In order to illustrate the technical solution of the present invention, specific embodiments are provided below.

[0018] See also Figure 1 , is a flow chart of a method for monitoring the operating status of a gas compressor provided by the first embodiment of the present invention, such as Figure 1 As shown, a method for monitoring the operating status of a gas compressor may include:

[0019] Step S101 : obtaining a real-time monitoring data sequence of the gas compressor in each preset dimension during the current compression cycle.

[0020] In an embodiment of the present invention, different types of sensors are installed on a gas compressor to collect monitoring data of multiple preset dimensions, where the preset dimensions include but are not limited to pressure data, triaxial vibration data, and sound data. The specific monitoring data collection process for each preset dimension is as follows: a pressure sensor is installed on the exhaust pipe, with a range of 0-10MPa and a frequency of 200Hz; a triaxial vibration sensor is installed on the cylinder housing of the gas compressor, collecting three acceleration signals with a range of ±50g, where g represents the acceleration of gravity, and a frequency of 2kHz; and an acoustic sensor is installed on the outside of the crankcase (non-contact) to collect audio waveforms.

[0021] Since the gas compressor compresses gas in a fixed compression cycle, the embodiment of the present invention needs to collect monitoring data of each preset dimension of the gas compressor in a compression cycle when performing real-time abnormal monitoring of the operating status of the gas compressor in a compression cycle. Since the amount of monitoring data in each preset dimension is inconsistent, in order to achieve time synchronization, a time window is used to obtain feature data in each preset dimension to characterize the operating status data of the gas compressor in a compression cycle. Therefore, in the embodiment of the present invention, the length of the time window is set to 0.1 seconds, that is, one feature data is output every 0.1 seconds, and the time corresponding to the output feature data is recorded as the sampling time, that is, one sampling time corresponds to one feature data.

[0022] For pressure data, the pressure value is collected by the pressure sensor and recorded as T. For example, if the pressure sensor collects 10 data points every 0.1 second (100 Hz × 0.1s = 10), the 10th data point is taken as the characteristic data T(t) at the corresponding sampling time, and the characteristic data sequence within a compression cycle is obtained, which is recorded as the pressure value sequence within a compression cycle T.

[0023] For triaxial vibration data, the original collection is three acceleration signals. The energy entropy in the time window is calculated by wavelet packet decomposition. For example, 200 data points are collected every 0.1 second (2kHz×0.1s=200), and the wavelet packet energy entropy of a time window is calculated. , recorded as the characteristic data at the corresponding sampling time , corresponding to the characteristic data sequence within a compression cycle, recorded as the energy entropy sequence E within a compression cycle, where the general process of wavelet packet decomposition is: Since mechanical vibration signals usually have non-stationary characteristics, the embodiment of the present invention decomposes the three-way acceleration signal collected by the three-axis vibration sensor layer by layer, finely divides the frequency band, captures the time domain and frequency domain characteristics of the signal, and analyzes transient faults. Since the acquisition frequency is 2kHz, according to the Nyquist theorem, the effective analysis frequency band is 0-1kHz. In actual industrial applications, the characteristic frequency of piston wear is usually in the low-frequency region, and the characteristic frequency of valve leakage is usually in the high-frequency region. The fifth layer decomposition divides the signal into frequency bands, each with a width of Through this decomposition method, the fault characteristic frequencies can be allocated to independent frequency bands as much as possible to avoid multiple fault frequencies from being mixed in the same frequency band. Taking the x-axis as an example, the original signal of the x-axis is [0.12, 0.15, -0.08, 0.23, ... 0.17]. 200 data points are collected every 0.1 seconds. The collected data point sequence is low-pass filtered to remove high-frequency noise. The above signal is framed into 50ms windows, 100 data points as one frame, and the step size is 25ms, that is, 50 data points. The specific framing is as follows: , perform five-layer wavelet packet decomposition on each frame to generate 32 frequency band nodes, and calculate the energy of each node in the fifth layer of each frame signal, that is:

[0024]

[0025] Where, represents the energy of the kth node in the jth layer, represents the coefficient value of the kth node in the jth layer at the i-th time point, which is obtained by wavelet decomposition. N represents the total number of wavelet packet coefficients contained in the node, so the energy entropy of the x-axis acceleration signal is for:

[0026]

[0027] in, Indicates the total energy of the current frame.

[0028] Similarly, the energy entropy of the Y-axis and Z-axis acceleration signals are obtained and recorded as , where x and y axes are more important, so the weight distribution is 0.4:0.4:0.2. The energy entropy of the three axes is weighted summed and normalized to obtain the wavelet packet energy entropy corresponding to a time window, which is recorded as .

[0029] For sound data, the original collection is the audio waveform. The MFCC distance within the time window is calculated by MFCC+DTW. Specifically, assuming that 1600 points (16kHz×0.1s=1600) are collected every 0.1 second, they are segmented and processed with a frame length of 25ms, that is, 400 data points. After segmentation, for any frame, the frame (400 data points) is fast Fourier transformed, and the amplitude spectrum of the frequency point is output. The obtained data is input into the Mel filter bank, and M-dimensional energy values ​​are obtained through frequency conversion. M depends on the number of filters. If the number of filters is 26, 26 Mel energy values ​​are output. The output Mel energy values ​​are logarithmically transformed to compress the dynamic range and then discrete cosine transformed to extract the MFCC coefficients of the frame, and then an MFCC coefficient sequence corresponding to a time window is obtained. Similarly, an MFCC coefficient sequence corresponding to each time window of a compression period is obtained, and a set of MFCC coefficient sequences within a compression period is formed in time sequence.

[0030] Since the gas compressor is in a normal operating state at the beginning when it is used for the first time, as the gas compressor is used, its operating state will gradually become abnormal as the equipment is used for a longer time. Therefore, the embodiment of the present invention defaults to starting to monitor the operating state of the gas compressor after it has been used m times. Considering that the set m value is too large, it is easy to cause abnormal data in the standard data obtained subsequently. Therefore, m is preferably set to 5 times, and there is no restriction here. Correspondingly, for the operating state monitoring of the gas compressor under any compression cycle, the MFCC coefficient sequence set obtained within the m compression cycles is used as the standard data. In the MFCC coefficient sequence set within the m compression cycles, the MFCC coefficient sequence under each same time sequence number is averaged respectively, and each corresponding mean value sequence constitutes a standard MFCC coefficient sequence set, which is used to characterize the sound data template of a compression cycle under normal operation. Then, obtain the MFCC coefficient sequence set within any compression period that needs to be monitored for operating status, and perform DTW distance calculation on each MFCC coefficient sequence in the set according to the time sequence number in the MFCC coefficient sequence set and the corresponding mean sequence in the standard MFCC coefficient sequence set, and obtain the DTW distance corresponding to each MFCC coefficient sequence in the MFCC coefficient sequence set, which is recorded as the feature data at the corresponding sampling time. , the corresponding feature data sequence within any compression period is obtained, which is recorded as the MFCC distance sequence D of any compression period.

[0031] It's worth noting that the purpose of MFCC is to extract acoustic features relevant to human perception and suppress irrelevant noise. The purpose of DTW calculation is to eliminate timing alignment discrepancies and improve fault detection robustness. The wavelet packet decomposition and MFCC coefficient extraction described above are both existing technologies and will not be detailed here.

[0032] In an embodiment of the present invention, any compression cycle is taken as an example and recorded as the current compression cycle. According to the data collection and processing method of each preset dimension mentioned above, a real-time monitoring data sequence of the gas compressor under each preset dimension in the current compression cycle is obtained respectively, wherein the real-time monitoring data sequence includes: a real-time pressure value sequence corresponding to the pressure data, a real-time energy entropy sequence corresponding to the three-axis vibration data, and a real-time MFCC distance sequence corresponding to the sound data, which is used to monitor the operating status of the gas compressor in the current compression cycle in real time according to the real-time monitoring data sequence.

[0033] It should be noted that since the operating status of the gas compressor in the current compression cycle is monitored in real time, and the current compression cycle is not necessarily completely completed, the length of the acquired real-time monitoring data sequence is not necessarily the complete sequence length.

[0034] Step S102: For any sampling moment of the current compression cycle, obtain a preset number of groups of historical data matching any sampling moment in the historical monitoring data sequence under each preset dimension within the historical compression cycle, and obtain the basic weight of each preset dimension at any sampling moment based on the preset number of groups of historical data.

[0035] Since the existing technology only monitors the operating status of the gas compressor based on monitoring data in a single dimension, which is prone to judgment deviation, the embodiment of the present invention improves the accuracy of monitoring the operating status of the gas compressor by fusing and analyzing the monitoring data in multiple dimensions. However, the compression process includes three operating stages: startup, stable operation, and unloading. The sensor representations in different operating stages are different, and the importance of the corresponding monitoring data is different. For example, in the startup stage of the compression process, the pressure continues to rise; while in the stable operation stage of the compression process, the pressure fluctuates slightly, and in the unloading stage, the pressure drops sharply, which makes the monitoring data in different dimensions have different weights when finally judging the operating status of the gas compressor. In the embodiment of the present invention, first, based on the historical monitoring data sequence in the historical compression cycle, the basic weight of each preset dimension in the current compression cycle is analyzed. The basic weight is used to characterize the baseline importance of each preset dimension of the gas compressor in the historical compression cycle.

[0036] Taking any sampling moment in the current compression period as an example, any sampling moment is recorded as the i-th sampling moment. According to the historical monitoring data sequence under each preset dimension in the historical compression period, the basic weight of each preset dimension at the i-th sampling moment is obtained. The specific acquisition method is:

[0037] Obtain the time series number of any sampling moment in the current compression cycle, and for any historical compression cycle, obtain the elements corresponding to the element number identical to the time series number in the historical monitoring data sequence under each preset dimension within any historical compression cycle to form a group of historical data, and obtain a preset number of groups of historical data. For example: Assume that the i-th sampling moment corresponds to the 10th element in any real-time monitoring data sequence of the current compression cycle. Similarly, in each historical monitoring sequence of any historical compression cycle, obtain the 10th element to form a group of historical data, and the historical data includes pressure value, energy entropy and MFCC distance. Therefore, multiple groups of historical data can be obtained in all historical compression cycles before the current compression cycle or a preset number (preferably 100 historical compression cycles) of historical compression cycles. The larger the number of historical data, the more dispersed the effective information. Therefore, the 100 historical compression cycles closest to the current compression cycle are selected.

[0038] After obtaining the preset number of groups of historical data, the covariance matrix of the preset number of groups of historical data is obtained. The number of rows of the covariance matrix is ​​the number of groups of historical data, and the number of columns of the covariance matrix is ​​the number of preset dimensions. The covariance matrix is ​​subjected to eigenvalue decomposition to obtain the eigenvalues ​​corresponding to each preset dimension, which are respectively recorded as ,in, They correspond to pressure value, energy entropy and MFCC distance respectively. It should be noted that covariance matrix operation and eigenvalue decomposition belong to the existing technology and will not be described in detail here.

[0039] The eigenvalues ​​corresponding to all preset dimensions are accumulated to obtain the eigenvalue accumulation value, and the ratio of the eigenvalue corresponding to each preset dimension to the eigenvalue accumulation value is calculated and recorded as the basic weight of the corresponding preset dimension. Among them, the calculation formulas for the basic weights of pressure value, energy entropy and MFCC distance are:

[0040]

[0041]

[0042]

[0043] in, is the basic weight of the pressure value, is the basic weight of energy entropy, is the basic weight of MFCC distance.

[0044] At this point, the basic weight of each preset dimension at the i-th sampling moment is obtained.

[0045] Step S103: Divide the current compression cycle into stages according to the similarity between the historical monitoring data sequence and the real-time monitoring data sequence of the historical compression cycle, and obtain a stage weight set composed of the stage weight of each preset dimension in each stage. According to the stage weight set corresponding to the stage to which any sampling moment belongs, adjust the basic weight of each preset dimension at any sampling moment to obtain the dynamic weight of each preset dimension at any sampling moment.

[0046] The covariance matrix reflects the correlation between monitoring data under different preset dimensions. Through eigenvalue decomposition, larger eigenvalues ​​indicate greater independence of the monitoring data under the corresponding dimension, and thus a higher ability to explain the operating status of the gas compressor. Sensor requirements vary at different process stages. Therefore, the current compression cycle is divided into stages based on process requirements. The basic weight ratio of each preset dimension at the i-th sampling moment is adjusted to ensure that the key preset dimensions have an increased weight in the corresponding stage.

[0047] In an embodiment of the present invention, the pressure value sequence collected by the pressure sensor is used as the main data for analysis. Since the inflection point of stage division requires the entire compression process to be completed before analysis can be performed based on the complete monitoring data, and the operating status of the gas compressor is monitored in real time, it cannot be guaranteed that the monitoring data within the current compression cycle can be completely acquired. Therefore, it is difficult to accurately divide the stages using the real-time pressure value sequence of the current compression cycle as the analysis object. Therefore, in an embodiment of the present invention, considering that the closer the compression interval, the more similar the operating status, the most recent historical compression cycle is used as a reference, that is, the previous compression cycle of the current compression cycle is obtained as the target cycle, and the inflection point of stage division is obtained based on the historical pressure value sequence of the target cycle, thereby completing the real-time stage division of the current compression cycle.

[0048] A historical pressure value sequence of the target period is obtained, and a low-pass filter is performed on the historical pressure value sequence with a cutoff frequency of 5 Hz to eliminate high-frequency noise. The historical pressure value sequence of the target period is evenly divided into two subsequences. Second-order difference processing is then performed on each subsequence to obtain a corresponding second-order difference sequence. For any second-order difference sequence, based on the absolute value of each difference value in the any second-order difference sequence, the difference value corresponding to the maximum absolute value is obtained as the inflection point, thereby obtaining two inflection points in the historical pressure sequence of the target period.

[0049] It is known that each differential value in the second-order difference sequence is calculated from three consecutive historical pressure values ​​in the historical pressure value sequence of the target period. Therefore, for any inflection point, the three historical pressure values ​​corresponding to the any inflection point are obtained in the historical pressure value sequence of the target period. According to the sampling time corresponding to each of the three historical pressure values ​​in the target period, the middle sampling time is selected as the dividing time.

[0050] Since the pressure continues to rise during the startup phase, the energy distribution of the vibration signal is dispersed (high energy entropy and small fluctuations), while the pressure fluctuates slightly during the stable operation phase, the mechanical load is stable, and the energy of the vibration signal is concentrated (low energy entropy but the fluctuations may be small). If the inflection point division is correct, the energy entropy during the startup phase should have a low variance (because the compression mode of the startup process is relatively consistent). If the inflection point division is incorrect, there will be abnormal interference. Therefore, the rationality of the inflection point division can be verified by the energy entropy sequence before the inflection point. Specifically: in the historical energy entropy sequence of the target period, obtain all the historical energy entropies before the demarcation moment, and obtain the historical energy entropy variance ; Set the variance threshold , if the historical energy entropy variance Less than the preset variance threshold , it means that the inflection point division is correct, and any of the above demarcation moments is taken as the target demarcation moment. If the historical energy entropy variance Greater than or equal to the preset variance threshold , it means that the inflection point division is wrong, and the reference period needs to be replaced. Then, the compression period before the previous compression period of the current compression period is used as the target period, and the method of obtaining the historical energy entropy variance is repeated until the historical energy entropy variance is less than the preset variance threshold. The historical pressure value sequence of the target period with correct inflection point division and the demarcation moment corresponding to the inflection point of the historical pressure value sequence are obtained, which are recorded as the target demarcation moment.

[0051] Furthermore, after obtaining the historical pressure value sequence of the target period with the correct inflection point division, the similarity between the historical pressure value sequence of the target period and the real-time pressure value sequence of the current compression period is calculated. If the data points of the pressure value sequence of a compression period are 1600, but the data points of the real-time pressure value sequence of the current compression period are 1200, then only the 1200 data points of the real-time pressure value sequence of the current compression period need to be calculated for similarity. The similarity calculation method is: subtract the two elements at each identical position between the real-time pressure value sequence and the historical pressure value sequence to obtain the absolute value of the difference, perform the average operation on all the absolute values ​​of the difference to obtain the mean of the absolute values ​​of the difference, and obtain the similarity based on the mean of the absolute values ​​of the difference. The larger the mean of the absolute values ​​of the difference, the smaller the similarity. The similarity calculation can also use DTW distance or Pearson correlation coefficient, which are both existing technologies and will not be described in detail here.

[0052] After obtaining the similarity between the historical pressure value sequence of the target cycle and the real-time pressure value sequence of the current compression cycle, the similarity threshold is set to 0.9. If the similarity is greater than or equal to the preset similarity threshold, it means that the similarity between the two pressure value sequences is high, and the availability of the target demarcation moment is strong, thereby obtaining the time sequence number of the target demarcation moment in the target cycle, and obtaining the sampling moment corresponding to the same time sequence number in the current compression cycle according to the time sequence number of the sub-target demarcation moment in the target cycle, which is recorded as the current demarcation moment; if the similarity is less than the preset similarity threshold, it means that the availability of the target demarcation moment is poor, and the reference cycle needs to be replaced, and then the previous compression cycle of the previous compression cycle of the current compression cycle is used as the target cycle, and the method of obtaining the similarity between the historical pressure value sequence of the target cycle and the real-time pressure value sequence of the current compression cycle is repeated until the similarity is greater than or equal to the preset similarity threshold, and the current demarcation moment is obtained.

[0053] It should be noted that the rule for replacing the reference period is to gradually replace the historical compression period that is farther and farther away from the current compression period. Assuming that the current compression period is j, the target periods are j-1, j-2, j-3, etc.

[0054] Similarly, the current demarcation moment corresponding to each inflection point is obtained. After determining the two current demarcation moments, the current compression cycle is divided into three stages according to each of the current demarcation moments, namely the startup stage, the stable operation stage and the unloading stage. Then, according to the divided stages, the priority ratio of the preset dimensions in each stage is determined, that is, the importance ratio of each preset dimension in different stages. The startup stage focuses on pressure surge prevention and requires priority attention, while the stable operation stage focuses on vibration wear monitoring and the unloading stage focuses on the monitoring of valve leakage sound. Therefore, in an embodiment of the present invention, the sensor with a high priority is set to twice that of the normal situation. Therefore, the priority ratio of pressure data, vibration data, and acoustic data in the startup stage is 2:1:1, and the priority ratio of pressure data, vibration data, and acoustic data in the stable operation stage is 1:2:1. That is, the priority ratio of the preset dimensions in the startup stage is 2:1:1, the priority ratio of the preset dimensions in the stable operation stage is 1:2:1, and the priority ratio of pressure data, vibration data, and acoustic data in the unloading stage is .

[0055] Then, based on the sensor history corresponding to each preset dimension, the performance parameters of the measurement equipment for each preset dimension were obtained. Specifically, through simulation experiments, the sensitivity and specificity indicators of the measurement equipment for each preset dimension were obtained. For sensitivity, during normal operation of the equipment, an abnormal situation was artificially introduced, such as a temporary increase in load. The monitoring data collected by the sensor was recorded to see whether it exceeded the data threshold and triggered an alarm. The number of missed detections and the number of detections were counted. The sensitivity of each sensor was obtained based on the number of missed detections and the number of detections. Sensitivity was calculated as the number of detections / (number of missed detections + number of detections). For specificity, the specificity of the sensor during the experiment was obtained by counting the number of correct judgments and the number of false alarms. The number of false alarms was determined through a voting mechanism. If only one sensor was abnormal while the other sensors were normal, it was considered a false alarm. The specificity indicator was: number of correct judgments / (number of normal judgments + number of false alarms). In the simulation experiment, the performance parameters of the measurement equipment for each preset dimension were updated after each mechanical overhaul to avoid inaccurate performance judgments due to performance degradation after long-term use. Thus, the sensitivity and specificity indicators of the measurement equipment (i.e., sensor) for each preset dimension were obtained. Then, according to the priority ratio of the preset dimensions in each stage and the performance parameters of the measurement equipment in each preset dimension, a stage weight set consisting of the stage weight of each preset dimension in each stage is obtained, which is used to characterize the stage coefficient of each preset dimension in different stages to adjust the basic weight of each preset dimension.

[0056] According to the priority ratio of the preset dimensions in each stage and the performance parameters of the measurement equipment in each preset dimension, a stage weight set consisting of the stage weight of each preset dimension in each stage is obtained, including:

[0057] For any stage, according to the priority ratio of the preset dimensions at any stage, obtain the priority ratio value corresponding to each preset dimension, and multiply the priority ratio value corresponding to each preset dimension and the sensitivity and specificity index of the corresponding measuring device as the measuring device score of each preset dimension at any stage;

[0058] The measurement equipment scores of all preset dimensions in any stage are accumulated to obtain the score accumulation value, and the ratio between the measurement equipment score of each preset dimension in any stage and the score accumulation value is recorded as the stage weight of each preset dimension in any stage, and the stage weights of all preset dimensions in any stage are combined into a stage weight set under any stage.

[0059] In one embodiment, the measurement device score of each preset dimension at the yth stage = sensitivity × specificity index × priority, then the sensor scores corresponding to the pressure data, vibration data, and sound data at the yth stage are respectively recorded as Then, the score of each sensor in the yth stage is normalized to obtain the corresponding stage weight. The calculation formulas for the stage weights of pressure value, energy entropy and MFCC distance are:

[0060]

[0061]

[0062]

[0063] in, Indicates the stage weight of the pressure value in the yth stage, is the stage weight of energy entropy in the yth stage, is the stage weight of the MFCC distance in the yth stage.

[0064] The stage weights of the pressure value, energy entropy and MFCC distance in the yth stage are combined into a stage weight set. Similarly, the stage weight set corresponding to the pressure value, energy entropy and MFCC distance in each stage is obtained. Since the basic weight is to ensure that the dynamic weight does not deviate from the law of historical data, and the specific requirements of different process stages will have certain deviations, in order to respond to real-time requirements, the basic weight of each preset dimension at the i-th sampling moment is fine-tuned through the stage weight set corresponding to each stage. The fine-tuning method is as follows: for any preset dimension at the i-th sampling moment, in the stage weight set corresponding to the stage to which the i-th sampling moment belongs, the stage weight corresponding to any preset dimension is obtained, the stage weight corresponding to any preset dimension is used as the numerator, the basic weight of any preset dimension is used as the denominator, and the obtained ratio is recorded as the adjustment coefficient; the product of the basic weight of any preset dimension and the adjustment coefficient is used as the dynamic weight of any preset dimension.

[0065] The calculation formulas for the dynamic weights of pressure value, energy entropy and MFCC distance are:

[0066]

[0067]

[0068]

[0069] in, Represents the dynamic weight of the pressure value at the i-th sampling moment, represents the dynamic weight of energy entropy at the i-th sampling moment, Represents the dynamic weight of the MFCC distance at the i-th sampling moment, The stage weight of the pressure value at the stage to which the i-th sampling moment belongs, is the stage weight of the energy entropy of the stage to which the i-th sampling moment belongs, is the stage weight of the MFCC distance under the stage to which the i-th sampling moment belongs.

[0070] At this point, the dynamic weight of each preset dimension at the i-th sampling moment in the current compression period is obtained. Similarly, the dynamic weight of each preset dimension at each sampling moment in the current compression period is obtained.

[0071] In step S104, based on the dynamic weight of each preset dimension at any sampling moment, the real-time monitoring data of each preset dimension at any sampling moment is weighted to obtain the equipment health index at any sampling moment; and according to the change rate of the equipment health index at adjacent sampling moments in the current compression cycle, an abnormal warning is issued to the gas compressor.

[0072] After obtaining the dynamic weight of each preset dimension at each sampling moment in the current compression cycle, the real-time comprehensive health status of the equipment is quantified by integrating the real-time monitoring data of three types of sensors: pressure data, vibration data, and acoustic data. The specific process is as follows:

[0073] Before quantifying the real-time comprehensive health status of the equipment, in order to eliminate the dimension and make the real-time monitoring data of different sensors weighted superposition possible, it is necessary to obtain the standard monitoring data corresponding to each preset dimension under the normal operation of the gas compressor, that is, the standard pressure value , standard energy entropy and standard MFCC distance ; Among them, the standard pressure value The method for obtaining is: obtain the mode of the pressure value sequence in the historical compression period as the standard pressure value ; Standard energy entropy The method for obtaining is: by obtaining the energy entropy sequence E within a compression cycle under the normal state of the device, calculating the mean of the energy entropy sequence E, and recording the mean as the standard energy entropy ; Standard MFCC distance The method for obtaining the distance is as follows: obtain the MFCC distance sequence within a compression cycle when the device is in normal state, and take the maximum value in the MFCC distance sequence as the standard MFCC distance .

[0074] Taking the i-th sampling moment of the current compression cycle as an example, the ratio between the real-time monitoring data of each preset dimension and the corresponding standard monitoring data at the i-th sampling moment is calculated to obtain a ratio sequence. Based on the dynamic weight of each preset dimension at the i-th sampling moment, the ratios in the ratio sequence are weighted and summed to obtain the device health index at the i-th sampling moment. The calculation expression of the device health index at the i-th sampling moment is:

[0075]

[0076] in, Represents the device health index at the i-th sampling moment.

[0077] Similarly, the equipment health index at each sampling moment in the current compression cycle is obtained, and an abnormal warning is issued to the gas compressor based on the change rate of the equipment health index at adjacent sampling moments in the current compression cycle. The specific abnormal warning method is as follows:

[0078] Based on the device health index at each sampling moment in the current compression cycle, the cumulative value of the abnormal trend of the entire cycle in the current compression cycle is calculated. The calculation formula for the cumulative value of the abnormal trend of the entire cycle is:

[0079]

[0080] in, Indicates the accumulated value of abnormal trend of the entire cycle within the current compression cycle. Represents the device health index at the t-th sampling moment, represents the device health index at the t-1th sampling moment, || represents the absolute value symbol, M represents the number of sampling moments, Indicates the rate of change of the device health index at adjacent sampling times.

[0081] It should be noted that The larger the value is, the greater the possibility of abnormality in the gas compressor, and the larger the corresponding abnormal trend cumulative value.

[0082] At this time, the equipment health index at each sampling moment in the current compression cycle and the cumulative value of the abnormal trend of the entire cycle in the current compression cycle are obtained. Since the rate of change of the single equipment health index is usually small, long-term accumulation can indicate potential faults. For example, the HI increases by 0.5% every day in the initial stage of mechanical wear, but it is still within the normal threshold range. In the embodiment of the present invention, if the equipment health index HI at any sampling moment in the current compression cycle is greater than or equal to the HI threshold of 1.5, an alarm is directly issued to perform equipment maintenance. This situation is often aimed at emergencies; if the TAI of the current compression cycle is greater than the TAI threshold of 2.0, a first-level warning is issued to indicate potential risks, and it is recommended to check the equipment log and recent data; when the TAI of the current compression cycle is greater than the TAI threshold of 3.0, a second-level warning is issued, and forced shutdown maintenance is required to avoid the expansion of the fault.

[0083] It should be noted that the focus of the embodiments of the present invention is on how to set dynamic weights when fusing the real-time monitoring data of each preset dimension, rather than on the setting of the HI threshold and the TAI threshold. In this embodiment, they are only example values, and the thresholds can be dynamically adjusted according to the equipment life cycle or the needs of the implementer, such as reducing the threshold of old equipment by 10%.

[0084] Based on the same inventive concept as the above method, an embodiment of the present invention also provides a gas compressor operating status monitoring system, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of any one of the above-mentioned gas compressor operating status monitoring methods.

[0085] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention, and should all be included in the scope of protection of the present invention.

Claims

1. A method for monitoring the operating status of a gas compressor, characterized in that: The method comprises: Acquire a real-time monitoring data sequence of the gas compressor in each preset dimension during the current compression cycle, wherein the preset dimensions include pressure data, triaxial vibration data, and sound data; For any sampling moment of the current compression cycle, a preset number of groups of historical data matching any sampling moment are obtained from the historical monitoring data sequence under each preset dimension within the historical compression cycle, and a basic weight of each preset dimension at any sampling moment is obtained based on the preset number of groups of historical data; Based on the similarity between the historical monitoring data sequence and the real-time monitoring data sequence of the historical compression cycle, the current compression cycle is divided into stages, and a stage weight set consisting of the stage weight of each preset dimension in each stage is obtained. According to the stage weight set corresponding to the stage to which any sampling moment belongs, the basic weight of each preset dimension at any sampling moment is adjusted to obtain the dynamic weight of each preset dimension at any sampling moment; Based on the dynamic weight of each preset dimension at any sampling moment, the real-time monitoring data of each preset dimension at any sampling moment is weighted to obtain the equipment health index at any sampling moment; based on the change rate of the equipment health index at adjacent sampling moments in the current compression cycle, an abnormality warning is issued for the gas compressor; The step of obtaining a preset number of groups of historical data matching any one of the sampling moments from the historical monitoring data sequence under each preset dimension within the historical compression period includes: Obtaining a time sequence number of any sampling moment in the current compression cycle, and for any historical compression cycle, obtaining, in the historical monitoring data sequence of each preset dimension within the any historical compression cycle, elements corresponding to element numbers identical to the time sequence number to form a group of historical data, and obtaining a preset number of groups of historical data based on the group of historical data for each historical compression cycle; Obtaining the basic weight of each preset dimension at any sampling moment according to the preset number of groups of historical data includes: Obtaining a covariance matrix of the preset number of groups of historical data, where the number of rows of the covariance matrix is ​​the number of groups of historical data, and the number of columns of the covariance matrix is ​​the number of preset dimensions, performing eigenvalue decomposition on the covariance matrix to obtain an eigenvalue corresponding to each preset dimension; Accumulate the eigenvalues ​​corresponding to all preset dimensions to obtain an eigenvalue cumulative value, calculate the ratio of the eigenvalue corresponding to each preset dimension to the eigenvalue cumulative value, and record it as the basic weight of the corresponding preset dimension; The real-time monitoring data sequence includes: a real-time pressure value sequence corresponding to the historical data, a real-time energy entropy sequence corresponding to the triaxial vibration data, and a real-time MFCC distance sequence corresponding to the sound data. Then, according to the similarity between the historical monitoring data sequence and the real-time monitoring data sequence of the historical compression period, the current compression period is divided into stages, and a stage weight set consisting of the stage weights of each preset dimension in each stage is obtained, including: The compression cycle before the current compression cycle is recorded as the target cycle. The historical pressure value sequence of the target cycle is evenly divided into two subsequences. Second-order difference processing is performed on each subsequence to obtain the corresponding second-order difference sequence. For any second-order difference sequence, based on the absolute value of each difference value in the second-order difference sequence, the difference value corresponding to the maximum absolute value is obtained as the inflection point, thereby obtaining two inflection points in the historical pressure sequence of the target cycle. For any inflection point, obtain three historical pressure values ​​corresponding to the inflection point in the historical pressure value sequence of the target period, and select the middle sampling time as the demarcation time based on the sampling times corresponding to the three historical pressure values ​​in the target period; In the historical energy entropy sequence of the target period, all historical energy entropies before the demarcation moment are obtained to obtain the historical energy entropy variance; if the historical energy entropy variance is less than a preset variance threshold, the demarcation moment is used as the target demarcation moment, and the similarity between the historical pressure value sequence of the target period and the real-time pressure value sequence of the current compression period is calculated; if the similarity is greater than or equal to the preset similarity threshold, the time sequence number of the target demarcation moment in the target period is obtained, and according to the time sequence number of the target demarcation moment in the target period, the sampling moment corresponding to the same time sequence number is obtained in the current compression period and recorded as the current demarcation moment; Obtaining the current demarcation moment corresponding to each inflection point, dividing the current compression cycle into three stages according to each current demarcation moment, obtaining the preset dimension priority ratio in each stage, and obtaining a stage weight set consisting of the stage weight of each preset dimension in each stage based on the preset dimension priority ratio in each stage and the measurement device performance parameter in each preset dimension; The step of obtaining a stage weight set consisting of the stage weights of each preset dimension in each stage according to the preset dimension priority ratio in each stage and the measurement device performance parameter in each preset dimension includes: Through simulation experiments, the sensitivity and specificity indicators of the measurement device corresponding to each preset dimension are obtained. For any stage, the priority ratio value corresponding to each preset dimension is obtained according to the priority ratio of the preset dimensions at any stage. The product of the priority ratio value corresponding to each preset dimension and the sensitivity and specificity indicators of the corresponding measurement device is recorded as the measurement device score of each preset dimension at any stage; Accumulate the measurement device scores of all preset dimensions at any stage to obtain a score accumulation value, record the ratio between the measurement device score of each preset dimension at any stage and the score accumulation value as the stage weight of each preset dimension at any stage, and form the stage weight set of all preset dimensions at any stage; The step of adjusting the basic weight of each preset dimension at any sampling moment according to the stage weight set corresponding to the stage to which any sampling moment belongs, to obtain the dynamic weight of each preset dimension at any sampling moment, includes: For any preset dimension at any sampling moment, in the stage weight set corresponding to the stage to which any sampling moment belongs, obtain the stage weight corresponding to any preset dimension, use the stage weight corresponding to any preset dimension as the numerator, and use the basic weight of any preset dimension as the denominator, and the obtained ratio is recorded as the adjustment coefficient; the product of the basic weight of any preset dimension and the adjustment coefficient is used as the dynamic weight of any preset dimension.

2. A method for monitoring the operating status of a gas compressor according to claim 1, characterized in that: After obtaining the historical capability entropy variance, it also includes: If the historical energy entropy variance is greater than or equal to the preset variance threshold, the previous compression cycle of the previous compression cycle of the current compression cycle is taken as the target cycle, and the method of obtaining the historical energy entropy variance is repeated until the historical energy entropy variance is less than the preset variance threshold, and the target demarcation moment is obtained.

3. A method for monitoring the operating status of a gas compressor according to claim 1, characterized in that: After calculating the similarity between the historical pressure value sequence of the target cycle and the real-time pressure value sequence of the current compression cycle, the following is also included: If the similarity is less than a preset similarity threshold, the previous compression cycle of the previous compression cycle of the current compression cycle is used as the target cycle, and the method of obtaining the similarity between the historical pressure value sequence of the target cycle and the real-time pressure value sequence of the current compression cycle is repeated until the similarity is greater than or equal to the preset similarity threshold, and the current demarcation moment is obtained.

4. A method for monitoring the operating status of a gas compressor according to claim 1, characterized in that: The weighted processing of the real-time monitoring data of each preset dimension at any sampling moment based on the dynamic weight of each preset dimension at any sampling moment to obtain the device health index at any sampling moment includes: Obtain the standard monitoring data corresponding to each preset dimension under the normal operating state of the gas compressor, calculate the ratio between the real-time monitoring data of each preset dimension at any sampling moment and the corresponding standard monitoring data, and obtain a ratio sequence. According to the dynamic weight of each preset dimension at any sampling moment, perform weighted summation on the ratios in the ratio sequence to obtain the equipment health index at any sampling moment.

5. A gas compressor operating status monitoring system, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the steps of the gas compressor operating status monitoring method according to any one of claims 1 to 4 are implemented.

Images