Chemical pump set fault early warning method, system and device based on big data

Abstract

The invention belongs to the technical field of chemical pump set fault early warning, and particularly provides a chemical pump set fault early warning method, system and device based on big data, and the method mainly comprises the steps: obtaining the axial vibration waveform and inlet and outlet pressure difference of a chemical pump set; calculating a peak ratio and a zero offset index of each section of signal; calculating a waveform slope accumulated value of the axial vibration waveform by adopting a specified window width; generating a two-dimensional state vector containing the shaft system unbalance degree and the sealing failure probability; generating joint features of the main frequency amplitude ratio and the phase difference distribution; and processing the inlet and outlet pressure difference and the two-dimensional state vector by adopting a double-flow random forest model, and outputting an early warning result. According to the method, the mechanical vibration and the fluid pressure signals are fused, the dynamic features are extracted, the cross-modal association rule is quantified, hidden fault recognition is enhanced through shunt processing and feature crossing, the misjudgment risk can be reduced, accurate early warning is achieved in combination with dynamic modeling, sudden shutdown and leakage accidents are avoided, and equipment safety and production continuity are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of chemical pump group fault warning, and particularly relates to a chemical pump group fault warning method, system and device based on big data. Background Art

[0002] Chemical pump groups operate in high-temperature, high-pressure and corrosive medium environments for a long time. Mechanical components are prone to hidden damage due to alternating loads and fluid impacts. Their early fault signals are weak and highly mixed with noise, making it difficult to effectively identify them through conventional means.

[0003] The traditional maintenance of chemical pump groups relies on regular inspections and threshold alarms, making it difficult to capture the co-evolution of shaft system offset and main frequency phase offset of pressure, resulting in the lack of cross-modal feature intersections. The key signals of progressive faults are ignored, leading to delayed fault warnings. Fault-induced chain reactions are likely to cause sudden drops in efficiency, sudden shutdowns and medium leakage accidents, resulting in significant economic losses and safety risks. Summary of the Invention

[0004] This application effectively solves the problem in the prior art that the traditional maintenance of chemical pump groups relies on regular inspections and threshold alarms, making it difficult to capture the co-evolution of shaft system offset and main frequency phase offset of pressure, resulting in the lack of cross-modal feature intersections, and the key signals of progressive faults are ignored, thus leading to delayed fault warnings. It can reduce the risk of misjudgment, achieve accurate warnings through dynamic modeling, avoid sudden shutdowns and leakage accidents, and improve equipment safety and production continuity.

[0005] To achieve the above object, the present invention adopts the following technical solutions: In a first aspect, this application provides a chemical pump group fault warning method based on big data, including: obtaining the axial vibration waveform and the inlet and outlet pressure difference of the chemical pump group; dividing the axial vibration waveform according to the rotation speed period, and calculating the peak ratio and zero bias index of each segment of the signal; calculating the cumulative value of the waveform slope of the axial vibration waveform using a specified window width; generating a two-dimensional state vector including the shaft system imbalance degree and the seal failure probability according to the peak ratio, zero bias index and the cumulative value of the waveform slope; aligning the two-dimensional state vector with the inlet and outlet pressure difference to generate a joint feature of the main frequency amplitude ratio and the phase difference distribution; processing the inlet and outlet pressure difference and the two-dimensional state vector using a dual-stream random forest model. The dual-stream random forest model extracts the hourly pressure volatility of the inlet and outlet pressure difference and performs feature crossing on the two-dimensional state vector and the joint feature, and then imports the obtained relevant features into a random forest classifier for processing to output a warning result.

[0006] Further, obtain the axial vibration waveform and the inlet / outlet pressure difference of the chemical pump group, including: obtaining the original vibration waveform collected by the axially installed vibration sensor and the rotation speed pulse signal output by the rotation speed sensor; according to the standard deviation of the interval time of the rotation speed pulse signal, screening out the axial vibration waveform segments with a rotation speed fluctuation less than a specified percentage within a continuous first specified time period, and marking them as the first waveform segments; obtaining the monitored inlet / outlet pressure difference, and when the instantaneous value of the inlet / outlet pressure difference exceeds a specified multiple of the average pressure difference in the first specified time period, retaining the axial vibration waveform segments in the front and back second specified time periods, and marking them as the second waveform segments; taking the intersection of the first waveform segments and the second waveform segments to generate the axial vibration waveform and the inlet / outlet pressure difference with the same time series as the axial vibration waveform.

[0007] Further, divide the axial vibration waveform by the rotation speed period, including: calculating the start time point and the end time point of each complete rotation period according to the rotation speed pulse signal; aligning the start time point of the rotation period with the start time point of the axial vibration waveform to generate a vibration waveform segment with phase marks; checking whether the axial vibration waveform segment contains an incomplete rotation period according to the phase marks, and if so, extending or truncating the incomplete rotation period to obtain the axial vibration waveform segment of the complete rotation period and the inlet / outlet pressure difference with the same time series as the axial vibration waveform.

[0008] Further, calculate the peak ratio and zero offset index of each signal segment, including: determining the maximum amplitude value and the minimum amplitude value of each segment of the axial vibration waveform segment; extracting the axial vibration waveform segment collected in the stationary state of the chemical pump group and calculating its signal mean value to obtain the reference zero point; dividing the maximum amplitude value by the minimum amplitude value to generate the peak ratio of each segment; calculating the absolute difference between the signal mean value of each segment of the axial vibration waveform segment and the reference zero point, and multiplying the absolute difference by one hundred to obtain the zero offset index.

[0009] Further, calculate the cumulative value of the waveform slope of the axial vibration waveform by using a specified window width, including: determining the window width according to the zero offset index value of the axial vibration waveform segment; calculating the absolute value of the first derivative of each sampling point of the axial vibration waveform segment within the window width to generate a local slope sequence; accumulating and summing the values of the local slope sequence to generate the cumulative value of the waveform slope.

[0010] Further, determining the window width according to the zero offset index value of the axial vibration waveform segment includes: setting a first value and a second value, and both the first value and the second value are positive numbers; for each increase of the first value in the zero offset index, the window width is shortened by the second value.

[0011] Further, a two-dimensional state vector including the shafting unbalance degree and the seal failure probability is generated according to the peak ratio, the zero-bias index, and the cumulative value of the waveform slope, including: aligning the peak ratio, the zero-bias index, and the cumulative value of the waveform slope according to the time stamp to form a three-channel input data group; processing the three-channel input data group by using a pressure-coupled convolutional network, the pressure-coupled convolutional network being based on a neural network model, determining the convolutional kernel interval unit number according to the difference between the current inlet and outlet pressure difference and the average pressure difference in the third specified time period before, determining the pooling granularity according to the peak ratio, and mapping the pooled feature to the shafting unbalance degree and the seal failure probability through a Sigmoid activation function to generate a two-dimensional state vector including the shafting unbalance degree and the seal failure probability.

[0012] Further, the two-dimensional state vector is aligned with the inlet and outlet pressure difference to generate a joint feature of the main frequency amplitude ratio and the phase difference distribution, including: aligning the two-dimensional state vector with the inlet and outlet pressure difference according to a unified time stamp; performing a short-time Fourier transform on the inlet and outlet pressure difference to extract the amplitude of the main frequency component of the pressure waveform; comparing the main frequency pressure amplitude with the historical average amplitude under the same working conditions in the fourth specified time period before to calculate the main frequency amplitude ratio; calculating the phase difference distribution between the shafting unbalance degree change curve of the two-dimensional state vector and the main frequency component waveform of the pressure; and combining the main frequency amplitude ratio and the phase difference distribution into a joint feature.

[0013] In a second aspect, the present application provides a chemical pump unit fault warning system based on big data, which includes: A data acquisition module: acquiring the axial vibration waveform and the inlet and outlet pressure difference of the chemical pump unit.

[0014] A vibration signal analysis module: dividing the axial vibration waveform according to the rotation speed period, calculating the peak ratio and the zero-bias index of each segment of the signal; calculating the cumulative value of the waveform slope of the axial vibration waveform by using a specified window width.

[0015] A state vector generation module: generating a two-dimensional state vector including the shafting unbalance degree and the seal failure probability according to the peak ratio, the zero-bias index, and the cumulative value of the waveform slope.

[0016] A joint feature generation module: aligning the two-dimensional state vector with the inlet and outlet pressure difference to generate a joint feature of the main frequency amplitude ratio and the phase difference distribution.

[0017] A warning analysis module: processing the inlet and outlet pressure difference and the two-dimensional state vector by using a two-stream random forest model, the two-stream random forest model obtaining relevant features by extracting the hourly pressure volatility of the inlet and outlet pressure difference and performing feature crossing between the two-dimensional state vector and the joint feature, and then importing the relevant features into a random forest classifier for processing to output a warning result.

[0018] In a third aspect, the present application provides a chemical pump unit fault warning device based on big data, which includes a memory and a processor; the memory is used to store computer programs; the processor is used to execute the steps of the chemical pump unit fault warning method based on big data described in the first aspect.

[0019] In a fourth aspect, the present application provides a readable storage medium, in which computer program instructions are stored. When the computer program instructions are read and run by a processor, the steps of the chemical pump unit fault warning method based on big data described in the first aspect are executed.

[0020] Advantages of the present invention: The present application integrates mechanical vibration and fluid pressure signals, extracts dynamic features and quantifies cross-modal correlation laws, enhances the identification of hidden faults through shunt processing and feature crossing, and effectively solves the problem that in the prior art, the traditional maintenance of chemical pump units relies on regular inspections and threshold alarms, making it difficult to capture the co-evolution of shaft system offset and pressure main frequency phase offset, resulting in the lack of cross-modal feature crossing, the key signals of progressive faults being ignored, and thus the fault warning being lagged. It can reduce the risk of misjudgment, achieve accurate warning in combination with dynamic modeling, avoid sudden shutdowns and leakage accidents, and improve equipment safety and production continuity.

[0021] Other features and advantages of the present invention will be described in the following specification, and, in part, will be obvious from the specification, or will be understood by implementing the present invention. The objectives and other advantages of the present invention can be realized and obtained by the structures pointed out in the specification and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0023] Figure 1 Shows a schematic flowchart of the chemical pump unit fault warning method based on big data of the present invention; Figure 2 Shows a schematic block diagram of the chemical pump unit fault warning system based on big data of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0024] To solve the problems raised in the background art, this application extracts dynamic features and quantifies cross-modal correlation laws by integrating mechanical vibration and fluid pressure signals, enhances implicit fault recognition through shunt processing and feature crossing, can reduce the risk of misjudgment, realizes precise early warning in combination with dynamic modeling, avoids sudden shutdowns and leakage accidents, and improves equipment safety and production continuity.

[0025] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0026] In some embodiments, as Figure 1 shown, this application provides a fault early warning method for chemical pump sets based on big data, including: S100. Obtain the axial vibration waveform and the inlet and outlet pressure difference of the chemical pump set.

[0027] S200. Divide the axial vibration waveform according to the rotation speed period, calculate the peak ratio and zero bias index of each segment of the signal; calculate the cumulative value of the waveform slope of the axial vibration waveform using a specified window width.

[0028] S300. Generate a two-dimensional state vector including the shafting imbalance degree and the seal failure probability according to the peak ratio, zero bias index, and cumulative value of the waveform slope.

[0029] S400. Align the two-dimensional state vector with the inlet and outlet pressure difference to generate a joint feature of the main frequency amplitude ratio and phase difference distribution.

[0030] S500. Process the inlet and outlet pressure difference and the two-dimensional state vector using a dual-stream random forest model. The dual-stream random forest model extracts the hourly pressure volatility of the inlet and outlet pressure difference and performs feature crossing between the two-dimensional state vector and the joint feature, and after obtaining the relevant features, imports them into the random forest classifier for processing to output the early warning result.

[0031] In some embodiments, S100. Obtain the axial vibration waveform and the inlet and outlet pressure difference of the chemical pump set, including: S110. Obtain the original vibration waveform collected by the axially installed vibration sensor and the rotation speed pulse signal output by the rotation speed sensor.

[0032] The detection of the shafting imbalance degree and the seal failure probability is more significant axially. The original vibration waveform collected by the axially installed vibration sensor and the rotation speed pulse signal output by the rotation speed sensor can be obtained. Each pulse corresponds to one revolution of the pump shaft.

[0033] S120. Screen out the axial vibration waveform segments with rotational speed fluctuations less than a specified percentage within a continuous first specified time period according to the standard deviation of the interval time of the rotational speed pulse signals, and mark them as the first waveform segments.

[0034] First, calculate the average time according to the interval time sequence of adjacent rotational speed pulses within a continuous first specified time period (such as 10 minutes). , and then calculate the standard deviation of the interval time of the rotational speed pulse signals according to the standard deviation formula. , if the standard deviation of the interval time is less than the average time of a specified percentage (such as 5%). , it is considered that the rotational speed fluctuation in this period meets the requirements and is marked as the first waveform segment.

[0035] S130. Obtain the monitored inlet and outlet pressure difference. When the instantaneous value of the inlet and outlet pressure difference exceeds a specified multiple of the average pressure difference in the first specified time period, retain the axial vibration waveform segments in the second specified time period before and after, and mark them as the second waveform segments.

[0036] When the instantaneous pressure difference exceeds a specified multiple (such as 1.5 times) of the average pressure difference in the first specified time period (such as 30 minutes). , retain the axial vibration waveform segments in the second specified time period (such as 30 seconds) before and after this moment, and mark them as the second waveform segments.

[0037] S140. Take the intersection of the first waveform segments and the second waveform segments to generate the axial vibration waveform and the inlet and outlet pressure difference with the same time series as the axial vibration waveform.

[0038] For example, if the first waveform segment is from 10:00 to 10:10 and the second waveform segment is from 10:05 to 10:15, then the intersection is the vibration waveform segment from 10:05 to 10:10, and the pressure difference data for this time period is intercepted synchronously.

[0039] By screening the intersection of the rotation speed stable segments (the first waveform segments) and the pressure mutation associated segments (the second waveform segments), the data quality is significantly improved, providing input data with a high signal-to-noise ratio for subsequent fault warnings.

[0040] In some embodiments, S200. Segment the axial vibration waveform according to the rotation speed period, including: Sa210. Calculate the start time point and end time point of each complete rotation period according to the rotational speed pulse signals.

[0041] Exemplarily, the rotational speed of the chemical pump group is 600 RPM, and the ideal period is 0.1 second. The actual pulse timestamp sequence is =0.0s, = 0.101 s, = 0.199 s, then the first cycle = 0.101 s, the second cycle = 0.098 s, indicating the existence of slight rotational speed fluctuations.

[0042] Sa220. Align the starting time point of the rotation period with the starting time point of the axial vibration waveform to generate a vibration waveform segment with phase markings.

[0043] Taking the starting time point of the rotational speed pulse as a reference, intercept the data segment corresponding to the time interval in the axial vibration waveform.

[0044] Add labels to each waveform segment , indicating the phase position of the segment in the rotation period. For example = 0°, corresponding to the starting point.

[0045] By aligning with the rotational speed pulse, the phase interference in the calculation of shafting imbalance characteristics can be eliminated.

[0046] Sa230. Check whether the axial vibration waveform segment contains an incomplete rotation period according to the phase markings. If so, extend or truncate the incomplete rotation period to obtain an axial vibration waveform segment of a complete rotation period and the inlet - outlet pressure difference with the same time series as the axial vibration waveform.

[0047] If the time length of the waveform segment deviates from the theoretically calculated rotation period by more than the tolerance range. For example, | | > 0.05 , where 0.05 represents the tolerance range, then it is determined as an incomplete cycle.

[0048] If the end time of the waveform segment is close to the next pulse, such as the remaining time is less than 5% of the cycle length, supplement the subsequent sampling points to a complete cycle; if the starting time of the waveform segment lags, such as more than 5% of the cycle length, discard the first half of the segment and re - intercept from the starting point of the nearest complete cycle.

[0049] Exemplarily, if the theoretical period is 0.1 s and the actual duration of a waveform segment is 0.095 s with a deviation of 5%, it is determined as a complete cycle; if the duration is 0.08 s, then 0.02 s of vibration data needs to be extended forward or truncated backward to 0.08 s.

[0050] Adjusting the waveform segment according to the actual cycle length can avoid the truncation error of the fixed window.

[0051] According to the start time point of the vibration waveform segment and the end time point , intercept the segment of the same time interval from the pressure difference data.

[0052] In some embodiments, calculating the peak ratio and zero offset index of each signal segment includes: Sb210. Determine the maximum amplitude value and minimum amplitude value of each axial vibration waveform segment; extract the axial vibration waveform segment collected under the stationary state of the chemical pump unit, and calculate the signal mean value thereof to obtain the reference zero point.

[0053] The maximum amplitude value is the maximum value of the absolute values of all sampling points in the waveform segment, and the minimum amplitude value is the minimum value of the absolute values of all sampling points in the waveform segment.

[0054] Under the stationary state of the chemical pump unit (without rotation and without fluid transportation), collect the axial vibration waveform segment, and take the mean value of all sampling points as the reference zero point.

[0055] Eliminate the inherent deviation of the sensor through the stationary state data.

[0056] Sb220. Divide the maximum amplitude value by the minimum amplitude value to generate the peak ratio of each segment; calculate the absolute difference between the signal mean value of each axial vibration waveform segment and the reference zero point, and multiply the absolute difference by one hundred to obtain the zero offset index.

[0057] When the shafting is unbalanced, the rotational centrifugal force causes the vibration amplitude to fluctuate violently periodically, and the peak ratio increases significantly. For example, the peak ratio is 2.0 - 3.0 under normal conditions, and can reach above 5.0 under unbalanced conditions.

[0058] Dividing the maximum amplitude value by the minimum amplitude value can generate the peak ratio characterizing the degree of shafting imbalance.

[0059] Take the mean value of all sampling points of the axial vibration waveform segment, and calculate the zero offset index according to the zero offset index formula.

[0060] When the seal fails, the small displacement of the shafting or fluid leakage causes the overall shift of the vibration signal. For example, the mean value shifts from 0.02 mV to 0.12 mV, and the zero offset index directly reflects such shift amount.

[0061] The peak ratio and zero offset index can effectively avoid dimension interference and adapt to multi-source data fusion.

[0062] In some embodiments, calculating the cumulative value of the waveform slope of the axial vibration waveform using a specified window width includes: Sc210. Determine the window width according to the zero offset index value of the axial vibration waveform segment.

[0063] Sc220. Calculate the absolute value of the first derivative of each sampling point in the axial vibration waveform segment within the window width to generate a local slope sequence.

[0064] For adjacent sampling points and , calculate the absolute value of the derivative . The absolute values of the derivatives of all sampling points within the window form a local slope sequence S = { , , }, where represents the nth absolute value of the derivative, characterizing the nth local slope.

[0065] The local slope reflects the change rate of the vibration waveform. When the seal wears, a small leakage causes high-frequency jitter of the waveform, and the slope value increases significantly.

[0066] Sc230. Accumulate and sum the values of the local slope sequence to generate a cumulative waveform slope value.

[0067] The cumulative waveform slope value quantifies the overall severity of the change in the vibration waveform within the window. When the seal fails, this value continues to rise. For example, in the normal state, it is 1.5 - 2.0, and in the failure state, it can reach above 3.0.

[0068] In some embodiments, determining the window width according to the zero - bias index value of the axial vibration waveform segment includes: Set a first value and a second value, and both the first value and the second value are positive numbers; for each increase of the first value in the zero - bias index, the window width is shortened by the second value.

[0069] Set the first value and the second value . For each increase of α in the zero - bias index, the window width is shortened by β. The window width is: , where represents the initial window width, represents the zero - bias index, represents rounding down.

[0070] Exemplarily, when the first value is 5 and the second value is 10 ms, if the zero - bias index is 8, then the window width is 90 ms.

[0071] When the seal fails, the increase in the zero - bias index reflects an increase in the shafting offset. Shortening the window width can focus on high - frequency micro - vibrations, avoid the loss of features caused by the long - window smoothing effect, and at the same time improve the discrimination of the cumulative slope value when the seal fails.

[0072] In some embodiments, S300. Generate a two-dimensional state vector including the shafting imbalance degree and the seal failure probability according to the peak ratio, the zero-bias index, and the cumulative value of the waveform slope, including: S310. Align the peak ratio, the zero-bias index, and the cumulative value of the waveform slope according to the time stamp to form a three-channel input data group.

[0073] The three-channel input data group is a time series matrix, and the three-channel input data group at each time point includes the peak ratio, the zero-bias index, and the cumulative value of the waveform slope at the same moment.

[0074] S320. Process the three-channel input data group using a pressure-coupled convolutional network. The pressure-coupled convolutional network is based on a neural network model. Determine the convolutional kernel interval unit number according to the difference between the current inlet and outlet pressure difference and the average pressure difference in the third specified time period before, determine the pooling granularity according to the peak ratio, and map the pooled feature map to the shafting imbalance degree and the seal failure probability through the Sigmoid activation function to generate a two-dimensional state vector including the shafting imbalance degree and the seal failure probability.

[0075] For example, the third specified time period can be 10 minutes. If the inlet and outlet pressure difference and the average pressure difference in the previous ten minutes The difference ΔP is:

[0076] For example, the interval unit number can be: max(1, ) where represents rounding down.

[0077] Exemplarily, if is 0.5Mpa, the interval unit number is 2, that is, the convolutional kernel slides once every 2 sampling points.

[0078] The higher the peak ratio, the smaller the pooling window, and the shafting imbalance details are retained. The pooling granularity formula can be expressed as: where represents the pooling granularity, represents rounding down, represents the peak ratio.

[0079] Exemplarily, if the peak ratio is 4, the pooling granularity is 6, that is, max pooling is performed every 6 sampling points.

[0080] The convolutional layer extracts local correlation features, the pooling layer compresses the data dimension, flattens the pooled features and inputs them into the fully connected layer, processes the output of the fully connected layer through the Sigmoid function, and finally two neurons respectively output the shaft imbalance degree and the seal failure probability , and the range of both is [0,1].

[0081] The three-channel input simultaneously covers the vibration amplitude, offset, and slope trend. The pressure difference drives the convolutional kernel interval, and the peak ratio drives the pooling granularity, which can improve the pertinence of feature extraction. The Sigmoid output directly quantifies the fault probability without the need for manual threshold setting.

[0082] In some embodiments, in S400, the two-dimensional state vector is aligned with the inlet and outlet pressure difference to generate joint features of the main frequency amplitude ratio and phase difference distribution, including:[[]] S410. Align the two-dimensional state vector with the inlet and outlet pressure difference according to the unified timestamp.

[0083] For example, taking the phase mark after the vibration waveform segmentation as the reference, intercept the pressure difference data in the same time interval.

[0084] If the timestamp of the two-dimensional state vector is from 10:00:00 to 10:00:10 (one data point per second), then synchronously intercept the inlet and outlet pressure difference data in this time period to form 10 groups of aligned data.

[0085] S420. Perform short-time Fourier transform on the inlet and outlet pressure difference to extract the amplitude of the main frequency component of the pressure waveform.

[0086] The Hanning window can be used for the short-time Fourier transform. For example, the window length is set to 1 second and the overlap rate is 50%, that is, the spectrum is calculated every 0.5 seconds.

[0087] After performing the short-time Fourier transform to obtain the spectrum, take the frequency component with the largest amplitude in the spectrum as the main frequency component and record the amplitude of the main frequency component .

[0088] S430. Compare the main frequency amplitude of the pressure with the historical average amplitude under the same working conditions in the fourth specified time period in the past, calculate the main frequency amplitude ratio; calculate the phase difference distribution between the shaft imbalance degree change curve of the two-dimensional state vector and the waveform of the main frequency component of the pressure.

[0089] The fourth specified time period can be 24 hours, and the main frequency amplitude ratio can be:[[]] , where,[[]] represents the current main frequency amplitude,[[]] represents the historical average amplitude.

[0090] For the shaft imbalance degree change curve (t) and the waveform of the main frequency component of pressure Perform normalization processing, then calculate the cross-correlation function, and take the time offset corresponding to the maximum correlation coefficient , and convert it to a phase difference , is: , where represents the main frequency identified after the short-time Fourier transform of the pressure waveform, that is, the frequency component with the largest amplitude in the spectrum, represents the time offset corresponding to the maximum correlation coefficient of the two cross-correlation functions.

[0091] The phase difference distribution can quantify the real-time coupling relationship between shafting imbalance and pressure fluctuation. The main frequency amplitude ratio is dynamically calculated based on historical working conditions, which can reduce the false alarm rate.

[0092] S440. Combine the main frequency amplitude ratio and the phase difference distribution into a joint feature.

[0093] The joint feature is a two-dimensional vector ( , ), and each time point corresponds to a set of values. If the window contains multiple time points, then take the mean value and the standard deviation as the joint feature.

[0094] In some embodiments, the two-stream random forest model in S500 consists of two independent data processing streams and a random forest classifier. Among them, the pressure difference stream is used to process the inlet and outlet pressure difference data, extract the hourly pressure volatility, and the two-dimensional state vector stream processes the two-dimensional state vector and the joint feature for feature crossing. The random forest classifier is used to integrate the two-stream features and output the fault warning result.

[0095] The fault warning results include "normal", "shafting imbalance warning", and "seal failure warning".

[0096] Statistically analyze the fluctuation characteristics of the inlet and outlet pressure difference data according to an hourly window, , where represents the hourly pressure volatility, represents the maximum value of the hourly pressure difference, represents the minimum value of the hourly pressure difference, represents the average value of the hourly pressure difference.

[0097] Feature crossing includes multiplication crossing and difference crossing. Multiplication crossing: , ; Difference crossing: | |, | |.

[0098] The dual-stream features (pressure volatility rate, cross features) are input into a random forest classifier. The input features include the hourly pressure volatility rate and 4 cross features, forming a 5-dimensional feature vector. For example, a forest can be composed of 100 decision trees. Each tree selects split nodes through the Gini index, and finally votes to output the warning result.

[0099] For example, if the input feature vector is [40%, 1.28, 0.75, 0.2, 0.5], and 60 trees in the random forest vote for "seal failure warning" while 40 trees vote for "normal", then the final output is "seal failure warning".

[0100] The pressure difference flow focuses on time-domain statistics, and the two-dimensional state vector flow captures the state-pressure correlation. By means of multiplication and difference operations to quantify the fault coupling effect, it can improve the clarity of the classification boundary.

[0101] In some embodiments, as Figure 2 shown, the present application provides a chemical pump unit fault warning system based on big data, which includes: Data acquisition module: Obtain the axial vibration waveform and the inlet and outlet pressure difference of the chemical pump unit.

[0102] Vibration signal analysis module: Segment the axial vibration waveform according to the rotation speed period, calculate the peak ratio and zero offset index of each segment of the signal; calculate the cumulative value of the waveform slope of the axial vibration waveform using a specified window width.

[0103] State vector generation module: Generate a two-dimensional state vector including the shafting imbalance degree and the seal failure probability according to the peak ratio, zero offset index and the cumulative value of the waveform slope.

[0104] Joint feature generation module: Align the two-dimensional state vector with the inlet and outlet pressure difference to generate joint features of the main frequency amplitude ratio and the phase difference distribution.

[0105] Warning analysis module: Use a dual-stream random forest model to process the inlet and outlet pressure difference and the two-dimensional state vector. The dual-stream random forest model extracts the hourly pressure volatility rate of the inlet and outlet pressure difference and performs feature crossing between the two-dimensional state vector and the joint features, and then imports the relevant features into the random forest classifier for processing to output the warning result.

[0106] This embodiment has all the advantages of a chemical pump unit fault warning method based on big data and can automatically implement the steps of a chemical pump unit fault warning method based on big data.

[0107] In some embodiments, the present application provides a chemical pump unit fault warning device based on big data, which includes a memory and a processor; the memory is used to store a computer program; the processor is used to implement the steps of a chemical pump unit fault warning method based on big data when executing the computer program.

[0108] In some embodiments, the present application provides a storage medium storing computer program instructions, which, when read and executed by a processor, perform the steps of the intelligent monitoring and anomaly recognition method for strength training.

[0109] Among them, any reference to a memory, storage, database, or other medium used in the embodiments provided by the present invention may include non-volatile and / or volatile memories. The non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. The volatile memory may include random access memory (RAM) or an external cache memory.

[0110] It should be noted that in this text, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising one..." does not exclude the existence of additional identical elements in the process, method, article or device comprising the element.

[0111] Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A chemical pump group fault warning method based on big data, characterized in that, include: Obtain the axial vibration waveform and inlet and outlet pressure difference of the chemical pump group; The axial vibration waveform is divided into sections according to the rotational speed period, and the peak ratio and zero bias index of each signal segment are calculated; the waveform slope cumulative value of the axial vibration waveform is calculated using a specified window width; A two-dimensional state vector including the shafting imbalance and seal failure probability is generated based on the peak ratio, zero bias index and waveform slope accumulation value. The two-dimensional state vector is aligned with the inlet and outlet pressure difference to generate the joint features of the main frequency amplitude ratio and phase difference distribution; A two-stream random forest model is used to process the inlet and outlet pressure difference and the two-dimensional state vector. The two-stream random forest model extracts the hourly pressure fluctuation rate of the inlet and outlet pressure difference and performs feature cross-feature analysis on the two-dimensional state vector and the joint feature. The obtained relevant features are then imported into the random forest classifier for processing and output of the warning results. The two-dimensional state vector is aligned with the inlet and outlet pressure difference to generate the joint features of the main frequency amplitude ratio and phase difference distribution, including: Align the two-dimensional state vector and the inlet and outlet pressure difference according to a unified timestamp; Perform short-time Fourier transform on the inlet and outlet pressure difference to extract the amplitude of the main frequency component of the pressure waveform; Compare the pressure main frequency amplitude with the historical average amplitude under the same working condition in the past fourth specified time period to calculate the main frequency amplitude ratio; calculate the phase difference distribution between the shaft system imbalance change curve of the two-dimensional state vector and the pressure main frequency component waveform; The dominant frequency amplitude ratio and phase difference distribution are combined into a joint feature.

2. The method for fault early warning of a chemical pump unit based on big data according to claim 1, wherein Obtain the axial vibration waveform and inlet and outlet pressure difference of the chemical pump group, including: Obtain the original vibration waveform collected by the axially mounted vibration sensor and the speed pulse signal output by the speed sensor; According to the standard deviation of the interval time of the speed pulse signal, the axial vibration waveform segment in which the speed fluctuation is less than a specified percentage within a first specified continuous time period is selected and marked as the first waveform segment; Obtain the monitored inlet and outlet pressure difference. When the instantaneous value of the inlet and outlet pressure difference exceeds a specified multiple of the average pressure difference in the first specified time period, retain the axial vibration waveform segment in the second specified time period before and after, and mark it as the second waveform segment. The intersection of the first waveform segment and the second waveform segment is taken to generate an axial vibration waveform and an inlet and outlet pressure difference in the same time sequence as the axial vibration waveform.

3. The method for fault early warning of chemical pump sets based on big data according to claim 2, wherein, The axial vibration waveform is divided according to the speed cycle, including: Calculate the start and end time of each complete rotation cycle based on the speed pulse signal; Align the starting time point of the rotation period with the starting time point of the axial vibration waveform to generate a vibration waveform segment with a phase mark; Check whether the axial vibration waveform segment contains an incomplete rotation cycle according to the phase mark. If so, extend or truncate the incomplete rotation cycle to obtain an axial vibration waveform segment of a complete rotation cycle and the inlet and outlet pressure difference in the same time series as the axial vibration waveform.

4. The method for fault early warning of chemical pump sets based on big data according to claim 1, characterized in that, Calculate the peak ratio and zero bias index of each signal segment, including: Determine the maximum and minimum amplitude values of each axial vibration waveform segment; extract the axial vibration waveform segment collected when the chemical pump group is in a stationary state, calculate its signal average value to obtain the reference zero point; The maximum amplitude value is divided by the minimum amplitude value to generate the peak ratio of each segment; the absolute difference between the signal mean value and the reference zero point of each axial vibration waveform segment is calculated, and the absolute difference is multiplied by 100 to obtain the zero bias index.

5. The method for fault early warning of chemical pump sets based on big data according to claim 1, wherein, Calculates the cumulative slope of the axial vibration waveform using a specified window width, including: Determine the window width based on the zero bias index value of the axial vibration waveform segment; Calculate the absolute value of the first-order derivative of each sampling point of the axial vibration waveform segment within the window width to generate a local slope sequence; The values of the local slope sequence are accumulated and summed to generate the waveform slope cumulative value.

6. The method for fault early warning of chemical pump sets based on big data according to claim 5, characterized in that, According to the zero bias index value of the axial vibration waveform segment, the window width is determined including: Set a first value and a second value, and both the first value and the second value are positive numbers; Every time the zero bias index increases by a first value, the window width is shortened by a second value.

7. The method for fault early warning of chemical pump sets based on big data according to claim 1, characterized in that, Based on the peak ratio, zero bias index and waveform slope accumulation value, a two-dimensional state vector containing the shafting imbalance and seal failure probability is generated, including: Align the peak ratio, zero bias index and waveform slope accumulation values according to timestamps to form a three-channel input data group; A pressure-coupled convolutional network is used to process the three-channel input data group. The pressure-coupled convolutional network is based on a neural network model. The number of convolution kernel interval units is determined according to the difference between the current inlet and outlet pressure difference and the average pressure difference in the third specified time period. The pooling granularity is determined according to the peak ratio. The pooled features are mapped into shaft imbalance and seal failure probability through the Sigmoid activation function, generating a two-dimensional state vector containing the shaft imbalance and seal failure probability.

8. A chemical pump group fault warning system based on big data, characterized in that, It includes: Data acquisition module: obtains the axial vibration waveform and inlet and outlet pressure difference of the chemical pump group; Vibration signal analysis module: divides the axial vibration waveform into segments according to the speed cycle, calculates the peak ratio and zero deviation index of each segment signal; calculates the waveform slope cumulative value of the axial vibration waveform using a specified window width; State vector generation module: Generates a two-dimensional state vector including shafting imbalance and seal failure probability based on the peak ratio, zero bias index, and waveform slope accumulation value; Joint feature generation module: aligns the two-dimensional state vector with the inlet and outlet pressure difference to generate joint features of the main frequency amplitude ratio and phase difference distribution; Early warning analysis module: A dual-stream random forest model is used to process the inlet and outlet pressure difference and the two-dimensional state vector. The dual-stream random forest model extracts the hourly pressure fluctuation rate of the inlet and outlet pressure difference and performs feature cross-feature analysis on the two-dimensional state vector and the joint feature. The obtained relevant features are then imported into the random forest classifier for processing and output of the early warning result. The two-dimensional state vector is aligned with the inlet and outlet pressure difference to generate the joint features of the main frequency amplitude ratio and phase difference distribution, including: Align the two-dimensional state vector and the inlet and outlet pressure difference according to a unified timestamp; Perform short-time Fourier transform on the inlet and outlet pressure difference to extract the amplitude of the main frequency component of the pressure waveform; Compare the pressure main frequency amplitude with the historical average amplitude under the same working condition in the past fourth specified time period to calculate the main frequency amplitude ratio; calculate the phase difference distribution between the shaft system imbalance change curve of the two-dimensional state vector and the pressure main frequency component waveform; The dominant frequency amplitude ratio and phase difference distribution are combined into a joint feature.

9. A chemical pump unit fault warning device based on big data, characterized in that, It includes a memory and a processor; the memory is used for storing a computer program; the processor is used for implementing the steps of the big data-based chemical pump group fault warning method according to any one of claims 1-7 when executing the computer program.

Images