Slurry pump motion monitoring management system based on data analysis

Abstract

The invention relates to the technical field of vibration monitoring, in particular to a slurry pump motion monitoring management system based on data analysis, which comprises a multi-source acquisition module, a period judgment module, a trend focusing module and a dynamic management and control module. According to the method, vibration data are collected through a three-axis acceleration sensor, a fundamental frequency amplitude is extracted through FFT, time-frequency characteristics and periodic distribution are fused in combination with an extreme value interval sequence, the data representation dimension is enhanced, the rising slope is processed through moving average, the periodic variation is calculated, random noise is restrained, and trend continuity is enhanced. Monotonicity test is combined with dynamic threshold judgment to improve anomaly recognition sensitivity; a sliding window is used for segmenting multi-cycle data; a sudden change interval is positioned based on a pressure amplification rate second derivative; the limitation of a frequency domain on transient feature recognition is broken through, a pressure amplification rate three-dimensional vector is constructed, cosine similarity is calculated, and a working condition difference degree is quantified to dynamically bind a maintenance instruction; and closed-loop feedback is formed to improve the early warning and decision matching degree.

Description

Technical Field

[0001] The present invention relates to the technical field of vibration monitoring, and in particular to a mud pump motion monitoring and management system based on data analysis. Background Art

[0002] The field of vibration monitoring technology involves the real-time collection, analysis, and identification of mechanical vibration signals generated during equipment operation in order to determine the equipment status, identify abnormal operation, and perform maintenance management. The core content of this technical field is to collect the vibration signals of the equipment through devices such as acceleration sensors, velocity sensors, or displacement sensors, and input the collected signals into the data analysis system for time domain analysis, frequency domain analysis, and feature extraction. This field covers the monitoring of key components in industrial production such as rotating machinery and reciprocating machinery, and is mainly used for fault diagnosis, life prediction, and preventive maintenance. In the vibration monitoring process, the key lies in the identification and analysis of vibration frequency, amplitude, and vibration mode. The equipment status is usually judged by combining technical means such as sensor measurement, electrical signal conversion, data transmission, and computational analysis.

[0003] Among them, the mud pump motion monitoring and management system refers to a system that monitors and manages the operating status of mud pumps used in oil drilling and other operations in real time. Its patent subject mainly covers the working status monitoring of key moving components such as the mud pump piston rod, connecting rod, and crankshaft. Vibration data of the components during operation is obtained by deploying vibration acceleration sensors, and a frequency domain analysis method based on Fourier transform is used to identify abnormal vibration patterns. At the same time, a threshold judgment mechanism is combined to determine the operating status. The system also uses a digital signal acquisition card to continuously collect vibration data and input the data into an embedded processing unit via a wired method for analysis and processing. The entire process relies on mechanical vibration spectrum recognition methods and motion trend modeling and analysis methods to comprehensively detect the operating status of the mud pump.

[0004] Existing technologies use Fourier transforms for frequency domain analysis. Their spectral resolution is limited by the fixed sampling window length, and they are not sensitive enough to short-term shocks or slowly changing anomalies in non-stationary signals. Fixed threshold judgment mechanisms have poor adaptability in load fluctuation scenarios, and the false alarm rate increases significantly with the complexity of the operating conditions. Single-dimensional vibration signal analysis lacks collaborative modeling with multiple physical quantities such as pressure and torque, making it difficult to effectively distinguish between normal operating fluctuations and early anomalies. Maintenance strategies rely on periodic detection data, and there is a time lag from data collection to decision execution, making it impossible to respond to sudden failures in real time. For example, intermittent impact vibrations caused by microcracks in the crankshaft bearings of mud pumps are difficult to identify at an early stage with existing technologies because their energy does not reach a fixed threshold and their spectral characteristics are masked by background noise. By the time the vibration energy is significantly enhanced, chain damage has already occurred. Summary of the Invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a mud pump motion monitoring and management system based on data analysis.

[0006] In order to achieve the above objectives, the present invention adopts the following technical solutions: A mud pump motion monitoring and management system based on data analysis includes: A multi-source acquisition module is used to acquire vibration data through a three-axis acceleration sensor, call FFT analysis to extract the amplitude of the fundamental frequency component as the acceleration rising slope, identify extreme points to generate an extreme value interval sequence, and transmit the acceleration rising slope and the extreme value interval sequence to a period discrimination module; a period discrimination module, configured to perform sliding average processing on the acceleration rising slope, calculate the variation between adjacent periods, verify the monotonicity of the extreme value interval sequence, generate a precursor node identifier when a threshold condition is met, and transmit five consecutive periods of raw vibration data containing the precursor node identifier and the corresponding pressure increase rate to a trend focusing module; a trend focusing module, configured to segment the five-cycle raw vibration data using a sliding window segmentation algorithm, locate the central period based on the precursor node identifier, calculate the second-order derivative of the pressure increase rate, and output the pressure mutation interval to the dynamic control module; The dynamic control module is used to construct a three-dimensional vector containing the pressure increase rate, calculate the cosine similarity of adjacent cycles, and when the similarity is lower than a critical value, bind the pressure mutation interval with a maintenance instruction and output it to the control terminal.

[0007] As a further solution of the present invention, the acceleration rising slope is specifically the amplitude of the fundamental frequency component, the pressure mutation interval includes the second-order derivative value of the pressure increase rate and the sliding window segment, and the maintenance instruction is specifically a three-dimensional vector model, a cosine similarity index, and an instruction binding parameter.

[0008] As a further solution of the present invention, the dynamically adjusted periodic change threshold is obtained by optimizing historical fault data using a gradient descent method. The specific correlation relationship is that when the pressure increase rate increases by 5%, the periodic change threshold is correspondingly increased by 10%; The sliding window length is set to 1.5 times the standard vibration period of the device, and the step size parameter is set to 0.5 times the standard vibration period; The critical value is quantified as 0.85 through regression analysis of historical failure data.

[0009] As a further solution of the present invention, the multi-source acquisition module includes: The vibration acquisition submodule uses a three-axis acceleration sensor to collect equipment vibration signals at a fixed sampling frequency, separates and stores the three-axis acceleration data according to the X / Y / Z axes, performs mean filtering on each axis data to eliminate high-frequency noise interference and generate the original vibration signal; The slope calculation submodule performs Hanning window function windowing on the original vibration signal, calculates the multi-axial spectrum energy distribution using fast Fourier transform, locates the frequency point amplitude corresponding to the fundamental frequency component, establishes a discrete sequence of fundamental frequency amplitude changes over time, and fits the sequence slope parameters using the least squares method to generate the acceleration rise slope; The extreme value sequence generation submodule sets a dynamic threshold in the acceleration rising slope sequence, marks three consecutive data points as maximum values ​​when the middle point is greater than the two adjacent points, calculates the difference between the timestamps corresponding to the adjacent maximum values, constructs a difference set in chronological order, and generates an extreme value interval sequence.

[0010] As a further solution of the present invention, the amplitude of the fundamental frequency component is positively correlated with the piston stroke frequency of the mud pump, and the correlation coefficient is greater than 0.92.

[0011] As a further solution of the present invention, the period determination module includes: The trend smoothing submodule arranges the acceleration rising slopes in chronological order, sets a sliding window to cover the current point and every two data points before and after it, performs a weighted average operation on the data in the window, and the weight coefficient decreases linearly according to the distance from the center point to generate a smoothed acceleration value; The variation calculation submodule calculates the absolute difference between the acceleration values ​​in two adjacent periods in the smoothed acceleration value sequence, accumulates and sums three consecutive differences, and when the accumulated value exceeds the dynamically adjusted period change threshold, records the time stamp of the position and generates a period change mark; The node identification submodule performs a monotonicity test on the extreme value interval sequence, counts the number of continuously increasing or decreasing items in the sequence, and when the number of consecutive changes in the same direction reaches a set threshold, extracts the periodic change mark of the corresponding time period, combines the adjacent period pressure increase rate collected by the pressure sensor, and generates five consecutive cycles of raw vibration data of the precursor node identification.

[0012] As a further solution of the present invention, the set threshold is that the same direction changes occur more than three times within five consecutive cycles.

[0013] As a further solution of the present invention, the trend focus module includes: The data segmentation submodule obtains the five-cycle original vibration data, uses a sliding window segmentation algorithm to set window length and step size parameters, divides the data into segments according to time series, extracts the vibration amplitude mean, peak-to-peak value, and energy spectral density characteristics within multiple windows, and generates a vibration data segment set; The center positioning submodule, based on the precursor node identifier, calls the pressure increase rate of the target window and its two adjacent windows in the vibration data segment concentration, takes the natural logarithm of the increase rate of the three windows for arithmetic average, calculates the geometric mean of the antilogarithm value, selects the window index corresponding to the maximum geometric mean, and locates the center period coordinate; The mutation determination submodule calls the pressure increase rate of the five windows before and after the central period coordinate using the formula: ; Calculate the second-order derivative of the pressure increase rate, detect the location of the extreme point, and take the ±2σ time span centered on the extreme point, marking it as the pressure mutation interval; in, Represents the pressure mutation interval, Represents the absolute value of the pressure change in the kth window, in MPa. is the window time span coefficient, ranging from 0.8 to 1.2. is the ratio of window vibration energy to total energy, is the time coordinate of the window center, is the time value corresponding to the center period coordinate, in seconds, is the center window index, and σ is the standard deviation of historical data.

[0014] As a further solution of the present invention, the combined calculation of the natural logarithm and the geometric mean can improve the recognition accuracy of the pressure increase abnormal window by 17.3%.

[0015] As a further solution of the present invention, the dynamic control module includes: The pressure amplification vector construction submodule collects the real-time data stream of the pressure sensor, intercepts the pressure amplification rate parameter at a fixed period, linearly arranges the amplification rate values ​​of three consecutive periods according to the time dimension, and forms a pressure amplification vector through three-dimensional coordinate mapping; The period similarity calculation submodule uses a vector window to intercept adjacent period data segments based on the pressure increase vector, calculates the angle in the three-dimensional vector space using the cosine theorem, converts the ratio of the vector dot product to the module length product into a similarity quantization value, and generates a period similarity value. The mutation interval binding submodule compares the cycle similarity value with the system preset threshold. When the detection value is lower than the critical standard, it extracts the start and end timestamps of the corresponding cycle, encapsulates the time interval data and the maintenance instruction protocol into a data packet, and outputs the pressure mutation interval to the control terminal. When the similarity quantification value is lower than 0.85, the corresponding equipment sealing failure probability exceeds 83%.

[0016] Compared with the prior art, the advantages and positive effects of the present invention are: In the present invention, the vibration data collected by the three-axis acceleration sensor is subjected to FFT analysis to extract the amplitude of the fundamental frequency component, which is combined with the generation of the extreme value interval sequence to fuse the time-frequency characteristics and the periodic distribution characteristics to enhance the data characterization dimension. The sliding average processes the acceleration rising slope and calculates the change in adjacent cycles to suppress random noise interference and enhance the trend continuity. The extreme value interval sequence monotonicity test is combined with the dynamic threshold judgment to improve the sensitivity of periodic abnormal feature recognition. The sliding window segmentation algorithm performs segmented processing on multi-cycle data and locates the mutation interval based on the second-order derivative of the pressure increase rate, breaking through the limitations of traditional frequency domain analysis on the recognition of transient features. The three-dimensional vector of the pressure increase rate is constructed and the cosine similarity of adjacent cycles is calculated to quantify the difference in working conditions. The dynamic binding of the mutation interval and the maintenance instruction is achieved through critical value judgment to form a closed-loop feedback mechanism to improve the timeliness of warning and the matching degree of decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is a system flow chart of the present invention; Figure 2 This is a flow chart of the multi-source acquisition module of the present invention; Figure 3 This is a flow chart of the cycle discrimination module of the present invention; Figure 4 This is a flow chart of the trend focusing module of the present invention; Figure 5 This is a flow chart of the dynamic control module of the present invention. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0019] In the description of the present invention, it should be understood that the terms "length," "width," "up," "down," "front," "back," "left," "right," "vertical," "horizontal," "top," "bottom," "inside," "outside," and the like, indicating positions or relationships, are based on the positions or relationships shown in the accompanying drawings and are intended only to facilitate the description of the present invention and simplify the description. They do not indicate or imply that the devices or elements referred to must have a specific orientation, be constructed, or operate in a specific orientation. Therefore, they should not be construed as limiting the present invention. Furthermore, in the description of the present invention, "plurality" means two or more, unless otherwise expressly and specifically defined.

[0020] Example 1 See also Figure 1 , a mud pump motion monitoring and management system based on data analysis includes: The multi-source acquisition module is used to acquire vibration data through a three-axis acceleration sensor, call FFT analysis to extract the amplitude of the fundamental frequency component as the acceleration rise slope, identify the extreme points to generate the extreme interval sequence, and transmit the acceleration rise slope and extreme interval sequence to the period discrimination module; The cycle identification module is used to perform sliding average processing on the acceleration rising slope, calculate the variation of adjacent cycles, and test the monotonicity of the extreme value interval sequence. When the threshold condition is met, a precursor node identifier is generated. The raw vibration data of five consecutive cycles containing the precursor node identifier and the corresponding pressure increase rate are transmitted to the trend focus module. The trend focusing module is used to segment the five-cycle raw vibration data using a sliding window segmentation algorithm, locate the central cycle based on the precursor node identifier, calculate the second-order derivative of the pressure increase rate, and output the pressure mutation interval to the dynamic control module; The dynamic control module is used to construct a three-dimensional vector containing the pressure increase rate and calculate the cosine similarity of adjacent cycles. When the similarity is lower than the critical value, the pressure mutation interval is bound to the maintenance instruction and output to the control terminal.

[0021] The acceleration rising slope is specifically the amplitude of the fundamental frequency component, the pressure mutation interval includes the second-order derivative value of the pressure increase rate and the sliding window segment, and the maintenance instruction is specifically the three-dimensional vector model, cosine similarity index, and instruction binding parameters.

[0022] The dynamically adjusted periodic change threshold is obtained by optimizing historical fault data using the gradient descent method. The specific correlation relationship is that when the pressure increase rate increases by 5%, the periodic change threshold is correspondingly increased by 10%; The sliding window length is set to 1.5 times the standard vibration period of the device, and the step size parameter is set to 0.5 times the standard vibration period; The critical value is quantified as 0.85 through regression analysis of historical failure data.

[0023] See also Figure 2 , the multi-source acquisition module includes: The vibration acquisition submodule uses a three-axis acceleration sensor to collect equipment vibration signals at a fixed sampling frequency, separates and stores the three-axis acceleration data according to the X / Y / Z axes, performs mean filtering on each axis data to eliminate high-frequency noise interference and generate the original vibration signal; The vibration collection submodule is installed at a specific key position of the mud pump body, specifically a piezoelectric triaxial acceleration sensor model CA-YD-186 just below the center of the crosshead slideway, and its fixed sampling frequency is set to The selection of this sampling frequency is based on the analysis of the highest expected fault characteristic frequency of the mud pump, considering that the basic frequency of the mud pump piston stroke is about to The upper limit of the potential fault characteristic frequency of its high harmonics and rolling bearings, gears and other components can reach , according to the Nyquist sampling theorem, the sampling frequency is set to the target highest frequency More than twice, that is Above, choose To ensure that there is no signal aliasing and provide sufficient spectrum analysis resolution, the sensor continuously and synchronously collects vibration acceleration signals in three mutually perpendicular directions, namely, X-axis, Y-axis, and Z-axis, during the normal operation of the mud pump, forming three independent digital time series. The collected three-axis acceleration data are then separated according to the X-axis data sequence, Y-axis data sequence, and Z-axis data sequence, and are independently stored in the non-volatile solid-state storage drive configured by the system. At a specific sampling moment, the acceleration value recorded in the X-axis is , the Y axis is , the Z axis is These values ​​are recorded as a data point in the corresponding axial time series. For each axial raw acceleration data time series, such as the acceleration data sequence fragment of the X-axis is [0.527, 0.481, 2.603, 0.552, 0.629, 0.515, 0.498] m / s², mean filtering is performed, and the window length of the mean filter is Set to 5 sampling points, the window length The selection is based on filtering out frequencies above the sampling frequency (Right now ) while retaining the lower frequency features associated with equipment failures. data points , its filtered value By calculating Central The arithmetic mean of the data points is obtained, which is calculated as follows: , taking the third data point in the sequence For example, the filtered value is ,Through this calculation process, a single high-amplitude noise point in the sequence caused by electrical interference or transient impact (such as , if it is confirmed to be noise) is significantly weakened, the data sequences of the Y-axis and Z-axis also perform exactly the same mean filtering process to generate the filtered Y-axis original vibration signal and the Z-axis original vibration signal respectively. These three signals processed by mean filtering together constitute the original vibration signal used for subsequent advanced analysis.

[0024] The slope calculation submodule performs Hanning windowing on the original vibration signal, calculates the multi-axial spectrum energy distribution using fast Fourier transform, locates the frequency amplitude corresponding to the fundamental frequency component, establishes a discrete sequence of fundamental frequency amplitude changes over time, and fits the sequence slope parameters using the least squares method to generate the acceleration rise slope. The slope calculation submodule receives the three original vibration signals of the X-axis, Y-axis and Z-axis, and calculates the original vibration signal time series after mean filtering along the Z-axis. As the processing object, the data segment length 12800 data points are selected, which corresponds to the sampling frequency. The 5-second data below ensures that it contains at least 4 to 5 complete mud pump stroke cycles (assuming the stroke frequency is to Fluctuates between ), for which the length is of The Hanning window function is applied to the signal, and the Hanning window function is defined as ,in The value range is from 0 to ,Will Each data point in and the window function value at the corresponding position Multiply to get the windowed signal , specifically, for the first data points , and its windowed value is The purpose of this operation is to reduce the spectrum leakage effect caused by signal truncation during the subsequent fast Fourier transform (FFT). The fast Fourier transform algorithm in the standard library is used to transform the windowed Z-axis signal. Perform the transformation, calculate its energy or amplitude distribution in the frequency domain, obtain the complex spectrum sequence of the Z axis, and take its modulus to obtain the single-amplitude spectrum , the spectrum clearly shows the different frequency points The same Hanning windowing and FFT processing process is performed on the original vibration signals of the X-axis and Y-axis to obtain the corresponding vibration amplitudes. and According to the design parameters of the mud pump and its normal operating speed range, the vibration fundamental frequency generated by the reciprocating motion of the piston is predetermined to be to In the narrow band range, the calculated Z-axis spectrum In this to Accurately search for the frequency point with the maximum amplitude within the frequency range, and the amplitude corresponding to this frequency point is identified as the amplitude of the Z-axis fundamental frequency component of the current data segment , and record the exact frequency value of the fundamental frequency , in one calculation, if to Search within the range The spectrum amplitude at is the largest, and its value is , then the current data segment , set a fixed time step of 1 second, and repeat the above calculation process from selecting new data segments, windowing, FFT to extracting the fundamental frequency amplitude every 1 second, thereby obtaining a discrete sequence of Z-axis fundamental frequency amplitudes that changes over time, at five consecutive 1-second time points The Z-axis fundamental frequency amplitude sequence obtained is (unit ), the corresponding time series is (in seconds), for these two sequences and Apply the least squares method to perform linear fitting, with the goal of determining the slope parameter and the intercept parameter , so that the fitting straight line The sum of squares of the residuals between the actual data points is minimized, and the slope parameter The calculation formula is ; in is the number of data points in the sequence, where , substitute the data for calculation: ; ; ; ; Substitute these summary values ​​into the slope formula: (unit ); This calculated slope parameter The value of , which is the quantitative representation of the acceleration rising slope in the current 5-second time window.

[0025] The extreme value sequence generation submodule sets a dynamic threshold in the acceleration slope sequence. When three consecutive data points satisfy the requirement that the middle point is greater than the adjacent two points, they are marked as maximum values. The difference between the timestamps corresponding to the adjacent maximum values ​​is calculated, and a difference set is constructed in chronological order to generate an extreme value interval sequence. The extreme value sequence generation submodule receives the continuously generated acceleration rising slope sequence, which is composed of the slope calculation submodule outputting one value per second. During a period of continuous monitoring time, the slope value sequence obtained is (unit ), each slope value is associated with a unique timestamp, which is recorded as , in which the slope sequence Set a dynamic threshold in To assist in identifying statistically significant local maxima, the specific method for setting the dynamic threshold is: first, select the most recent The calculated slope values ​​constitute a reference data set, and the arithmetic mean of the reference data set is calculated. and standard deviation , in an actual calculation, we get , , dynamic threshold Defined as ,in is a preset sensitivity adjustment coefficient, which is set to 1.5. This coefficient value is determined by performing parameter optimization experiments on historical fault data. The goal is to maximize the fault precursor detection rate while controlling the false alarm rate within 5%. , traverse the slope sequence currently obtained , for each data point in the sequence except the first and last two data points (in ), to determine whether it satisfies the conditions at the same time and , if this condition is met, then It is initially marked as a local maximum point. Further, the local maximum point The value must also be greater than or equal to the dynamic threshold calculated above , is finally confirmed as a valid maximum value. middle: ,satisfy and But due to , so Not considered a valid maximum value, continue checking, ,satisfy and ,and ,therefore (corresponding timestamp ) is confirmed as a valid maximum, in this way, from the sequence All valid maximum points and their corresponding timestamps are screened out. Assume that after this screening process, the timestamp sequence of the valid maximum values ​​is , whose specific value is (The time unit is seconds, relative to the monitoring starting point), calculate the timestamp difference between these adjacent valid maximum points, and obtain the difference sequence , substitute the value into , these calculated differences are arranged in the order of their occurrence in time to form a difference set, which is the extreme value interval sequence. (Unit: seconds).

[0026] The amplitude of the fundamental frequency component is positively correlated with the piston stroke frequency of the mud pump, and the correlation coefficient is above 0.92.

[0027] See also Figure 3 , the period discrimination module includes: The trend smoothing submodule arranges the acceleration rising slope in chronological order, sets a sliding window to cover the current point and every two data points before and after it, performs a weighted average operation on the data in the window, and the weight coefficient decreases linearly according to the distance from the center point to generate a smoothed acceleration value; The trend smoothing submodule receives the acceleration rising slope sequence generated by the slope calculation submodule and arranged in time order, that is, (unit ), set a sliding window with a fixed length of 5 data points, which will cover the current data point to be processed and its two immediately preceding and following data points. The index is Data points (which requires To ensure the window is complete), the data contained in the sliding window is , perform weighted average operation on the five data points in the window to obtain the smoothed value, and the setting of the weight coefficient follows the distance from the center point The closer the point, the greater the weight, and the weight coefficient decreases linearly and symmetrically from the center point to both sides. The specific weight coefficient vector is set as ,in Corresponding center point , and Corresponding to the point 1 position away from the center point, and For points 2 positions away from the center point, in order to achieve linear decrease and make the sum of weights 1, the basic weight unit can be set to , let the weight be , then the total weight , in order to make ,but , so the weight coefficient for practical application is , its approximate decimal representation is , in sequence (assuming index starts at 0) Take the example of calculation, the corresponding data in the sliding window is ,but Smoothed acceleration value of the point The calculation is as follows: , for the sequence All data points that can form a complete 5-point window in the 5-point window are subjected to this weighted average operation, thereby obtaining a smooth acceleration value sequence with smaller fluctuations and more obvious trends. ,like .

[0028] The variation calculation submodule calculates the absolute difference between the acceleration values ​​in two adjacent periods in the smoothed acceleration value sequence, accumulates and sums three consecutive differences, and when the accumulated value exceeds the dynamically adjusted period change threshold, it records the timestamp of the position and generates a period change mark; The variation calculation submodule acts on the smoothed acceleration value sequence generated by the trend smoothing submodule , the sequence is as First, calculate the absolute difference of the smoothed acceleration values ​​of two adjacent time points (or periods) in the sequence to form an absolute difference sequence ,in , for the above fragment, which The sequence is: Then, this absolute difference sequence The three consecutive differences are accumulated and summed to form a cumulative sum sequence , which is calculated as , the first cumulative sum (This cumulative sum reflects arrive The overall change activity between (reflect arrive and so on. , when a certain cumulative sum value is calculated Exceeding a dynamically adjusted period change threshold When Of the three corresponding differences, the middle one The timestamp of this dynamic threshold The setting is based on the cumulative sum sequence calculated under the historical normal operation status of the equipment Statistical analysis, specifically, collecting at least 200 normal working conditions value, calculate its mean and standard deviation , an actual statistical analysis shows , , then the periodic change threshold Defined as ,in is an adjustment coefficient, which is set to 2.5 based on the retrospective analysis of historical fault data to balance the sensitivity and false alarm rate of detection. , in the above example, Does not exceed 0.002050, and exceeds 0.002050, so The corresponding middle difference (The value is 0.000766, corresponding to the smoothed value and timestamp of the change between The timestamp recorded is a periodic change mark.

[0029] The node identification submodule performs a monotonicity test on the extreme value interval sequence, counting the number of continuously increasing or decreasing items in the sequence. When the number of consecutive changes in the same direction reaches a set threshold, it extracts the periodic change mark of the corresponding time period and combines it with the adjacent period pressure increase rate collected by the pressure sensor to generate five consecutive cycles of raw vibration data for the precursor node identification. The node identification submodule generates the extreme value interval sequence generated by the extreme value sequence generation submodule To process, the sequence , perform a monotonicity test, the specific operation is to traverse the sequence, count the number of consecutive increasing or consecutive decreasing items in the sequence, and The comparison starts with the second item: (Incremental, continuous increment count is 1), (turn to decrement, continuous decrement count is 1), (Decrementing, continuous decrement count is 2), (Decrementing, continuous decrement count is 3), (Decrease, the continuous decrease count is 4). At this time, the continuous decrease times reaches 4 times. When there are more than 3 (that is, at least 3) same-direction changes in 5 consecutive extreme value interval periods, or more strictly, when the number of continuous same-direction changes (whether increasing or decreasing) reaches a set threshold, the threshold is set to 4 times, then the monotonic trend is judged to be significant. This threshold of 4 times is based on the correlation analysis of the fault precursors and the monotonic length of the extreme value interval sequence in a large amount of historical data. It is found that in most early fault indications, the extreme value interval has at least 4 continuous same-direction changes. Therefore, 4 is selected as the judgment threshold to improve the reliability of recognition. In the above example, when the extreme value interval sequence appears When starting from the second item to the fifth item End, there have been 4 consecutive decreases ( arrive , arrive , arrive , arrive ), the condition of "the number of consecutive same-direction changes reaches the set threshold of 4 times" is met, and the end timestamp of the time period corresponding to the end of the fourth decrease (i.e., the interval value of 3.8s) is recorded. Similarly, in the subsequent part of the sequence In, from arrive If there are five consecutive increases, the condition is also met. The periodic variation mark generated by the variation calculation submodule during the time period of this significant monotonic trend is extracted, and combined with the adjacent period pressure increase rate collected and calculated in real time by the pressure sensor model MPM489 installed on the mud pump discharge pipeline during the same time period, the pressure increase rate is defined as ,in is the average pressure in a stroke cycle. If the pressure increase rate is lower than -0.1 or higher than 0.1, the pressure change is considered significant. During the period when the extreme interval sequence is detected to decrease for 4 consecutive times, if there are one or more cycle change marks, and the pressure increase rate sequence in this period is , where 0.12 indicates that the pressure has a significant increase of more than 0.1. Combining this information (strong monotonicity of the extreme interval sequence, the presence of periodic variation markers, and significant changes in the pressure increase rate), the time or time period that meets the conditions is identified as a precursor node. Once the precursor node is determined, its corresponding time point is set to , then from that point in time Start from (or take it as the center and trace back some data), extract the original vibration data of five consecutive stroke cycles that have been stored previously (that is, the X, Y, Z three-axis vibration signal data collected by the three-axis acceleration sensor and completed the mean filter processing). If the original vibration data of a stroke cycle contains data points (corresponding to a 1-second period and a 2560Hz sampling rate), the total length of the extraction is The triaxial vibration data series with data points is used as the input for subsequent analysis.

[0030] Table 1 Vibration and pressure monitoring characteristic parameters (precursor identification stage) As shown in Table 1, the table lists some key characteristic parameter values ​​of the mud pump at different monitoring time points, including the extreme value interval duration, the pressure increase rate of the corresponding period, and the cumulative sum of the three-point changes in the smoothed acceleration slope, as well as the periodic change mark determined by whether the cumulative sum exceeds the threshold (in this case, the threshold is 0.002050). These data are used together to identify precursor nodes.

[0031] The threshold is set to more than three changes in the same direction within five consecutive cycles.

[0032] See also Figure 4 , the trend focus module includes: The data segmentation submodule obtains five cycles of raw vibration data, uses a sliding window segmentation algorithm to set the window length and step size parameters, divides the data into segments according to the time series, extracts the vibration amplitude mean, peak-to-peak value, and energy spectrum density characteristics within multiple windows, and generates a vibration data segment set; The data segmentation submodule obtains the original vibration data of five consecutive stroke cycles corresponding to the precursor node determined by the node identification submodule. Taking the original vibration data in the Z axis as an example, the duration of one stroke cycle is 1 second and the sampling frequency is , so one cycle contains 2560 vibration data points, and the total length of the data for five cycles is data points, and this data sequence is recorded as , using sliding window segmentation algorithm to The data is divided into small parts and the length of the sliding window is set. The data points are 640, which is 1 / 4 of a stroke cycle (2560 points). It is designed to capture the local vibration characteristics within the cycle and set the step size of the sliding window. The window length is 320 data points, which is half of the window length, to achieve a 50% window overlap rate. This overlapping setting ensures the continuity of information between adjacent windows and avoids missing key change features. The parameter combination of window length 640 points and step length 320 points is determined after a feature extraction comparison experiment on historical fault data. The experiment compared window lengths of 320 points, 640 points, and 1280 points, with step lengths of 1 / 4, 1 / 2, and 3 / 4 of the corresponding lengths, and found that the configuration of 640 points and 320 points achieved the best balance between the sensitivity of the feature to early fault signals and computational efficiency. Starting from the 0th data point of the sequence, the first length is intercepted Data segment , and then move the window's starting position forward (ie 320 data points), intercept the second data segment , and so on, until the tail of the sliding window reaches or exceeds The end of the sequence, for each successfully intercepted data segment , calculate the following three characteristic parameters of its internal vibration signal: Mean Amplitude, calculated as the arithmetic mean of the absolute values ​​of all data points in the data segment, that is, ,in From 0 to ; Peak-to-Peak Value, calculated as the difference between the maximum and minimum values ​​of the data points in the data segment, that is, Energy is calculated as the sum of the squares of the instantaneous values ​​of all data points in the data segment divided by the number of data points, that is, , if Calculation shows that the sum of the absolute values ​​of the internal data points is 320m / s, so the mean amplitude is , if its maximum value is , the minimum value is , then its peak-to-peak value is , if the sum of squares of its internal data points is 288 , then its energy is ,The three types of characteristic parameters (amplitude mean, peak-to-peak value, and energy) calculated from all windows, together with the start and end time information of the window, are organized together to form a vibration data segment set containing multiple feature vectors for subsequent modules to call.

[0033] The center positioning submodule, based on the precursor node identification, calls the pressure increase rate of the target window and its two adjacent windows in the vibration data segmentation concentration, takes the natural logarithm of the increase rate of the three windows for arithmetic average, calculates the geometric mean of the antilogarithm value, selects the window index corresponding to the maximum geometric mean, and locates the center period coordinates; The center positioning submodule uses the approximate time range indicated by the precursor node identifier and calls the vibration data segment set generated by the data segmentation submodule. Each window in the segment set They are all associated with a set of vibration characteristic parameters, and correspond to a clear time period and the pressure increase rate measured by the pressure sensor during that time period. (The pressure increase rate here is calculated for each small window period, not the entire stroke cycle). An initial reference window is selected near the time center point indicated by the precursor node, and its index is set to , call the reference window and its previous window in the time series and the next window on the time series The corresponding pressure increase rate, specifically, if the window The pressure increase rate is ,window The pressure increase rate is ,window The pressure increase rate is , take the natural logarithm of the pressure increase rate of these three windows (make sure it is positive, if it is negative or zero, add a very small positive bias such as 0.0001) and get , , , calculate the arithmetic mean of these three logarithms , and then calculate the arithmetic mean of Calculate its antilogarithm (i.e. natural exponent function), and obtain the geometric mean of the three original growth rates , for all windows obtained by segmenting the entire five-cycle data (or within an analysis segment pre-delineated near the precursor node, for example, containing 20 consecutive windows), the same geometric mean calculation process is performed on each window with itself as the temporary center, together with one adjacent window before and after it, for a total of three windows, to obtain a geometric mean sequence ,in For the total number of windows that can be calculated in this analysis segment, the one with the largest value in the geometric mean sequence is selected. , the index of the corresponding center window The index is identified as the core area where the pressure changes most dramatically. This is the central period coordinate for positioning in this step.

[0034] The mutation determination submodule calls the pressure increase rate of the five windows before and after the central period coordinate using the formula: ; Calculate the second-order derivative of the pressure increase rate, detect the location of the extreme point, and take the ±2σ time span centered on the extreme point, marking it as the pressure mutation interval; in, Represents the pressure mutation interval, Represents the absolute value of the pressure change in the kth window, in MPa. is the window time span coefficient, ranging from 0.8 to 1.2. is the ratio of window vibration energy to total energy, is the time coordinate of the window center, is the time value corresponding to the center period coordinate, in seconds, is the center window index, σ is the standard deviation of historical data; The mutation determination submodule receives the center period coordinates determined by the center positioning submodule, that is, the index of the center window (hereinafter referred to as Indicates the center window index), and calls the center window , its first two windows (index , ) and the next two windows (index , ), the absolute value of the pressure change in the total five consecutive windows and other related parameters, using the formula ; To ensure the consistency of the physical meaning of the calculation, all time-related measurements, such as the center time of each window and the time step between windows , all use seconds (s) as the basic time unit for recording and calculation, and the pressure change MPa is uniformly used as the basic pressure unit, and vibration energy by The unit is , and subsequent calculations are based on this unit system. Is a quantitative indicator that characterizes the comprehensive parameter in the brackets (denoted as ) with time (or equivalent window index), used to identify the sudden change characteristics of the pressure accumulation effect. In discrete data points, approximate calculation is performed through second-order central difference. If With center window The changes form a sequence , and each center window The corresponding time is , the average time span (i.e. time step) of each window is ,but ; Summation symbol Indicates the center window of the current analysis As the benchmark, the relevant items of five consecutive windows before and after it, including itself, are accumulated. Representative The absolute value of the pressure change in a window (MPa) is obtained by taking the continuous readings of the pressure sensor within the window time period (such as 0.25 seconds) and calculating the absolute value of the difference between the maximum and minimum pressure values ​​recorded in the window. For example, for the window , the monitored pressure changes from 10.12MPa to 10.45MPa, then MPa. For the The time span correction factor (unitless) of a window is the actual duration of the window. Relative to a standard window time length (Here A normalized adjustment of 0.25 seconds, corresponding to the number of window data points and the sampling rate, is calculated as , if The actual duration of the window is measured by the high-precision timing module to be 0.245 seconds, so ; If it is 0.25 seconds, it is 1.0; if it is 0.255 seconds, it is 1.02. The value range is limited to between 0.8 and 1.2. Window data outside this range is considered invalid or requires special processing. This coefficient is introduced to adapt to the tiny stroke cycle fluctuations that may occur during the actual operation of the mud pump. For the The Z-axis vibration energy of each window With these five windows ( arrive ) total vibration energy The ratio of (unitless), that is ,in Calculated by the data segmentation submodule, , this weight reflects the The relative importance of the intensity of each window vibration in the current analysis group. For the The exact time coordinate (s) of the center point of each window relative to the starting point of the entire five-cycle data segment. If the starting time of the five-cycle data segment is , each window width is 0.25 seconds, window This is the first of the five-period data Sliding window (step length 0.125s), then its starting time is , center time . The window that is the center when the summation is currently being calculated The central time coordinate (s), that is The denominator Representative The center time of each window is the same as the center window of the current sequence (Right now ) is the absolute value of the difference between the central time (s), when This item is 0 when , in order to avoid division by zero, when When , the term is replaced by a preset positive constant representing a very small time interval replace, The value of is set to 1% of the window time span, that is, , this item makes the distance from the center window The farther the window is, the less its contribution to the total is attenuated as the time distance increases. Parameter assignment and calculation process: Assume that the central window of the current analysis Central time Seconds (relative to the starting point of the five-period data), window time span Seconds, time step Seconds (because we calculate When the value is , the center is moved window by window. Here is Table 2 Formula Parameter assignment table As shown in Table 2, this table is used to calculate the window Central When the value is The specific values ​​of each parameter are required. First, calculate the total vibration energy of the five windows . Then calculate the energy weight of each window : ; ; ; ; ; Calculate the sum of each term within the brackets : ; ; ; ; ; For the above The sum of the values ​​is This is when the analysis center is the window Calculated when To calculate the second-order derivative, it is also necessary to analyze the window and Calculated when Value, that is and , these values ​​are obtained by moving the center of the entire calculation process to and And resubstitute the data of each window (including recalculating the total energy and weights of the 5 windows) to obtain the result. Here, the calculation results are given directly: (The corresponding analysis center is the window , the time is ) (The corresponding analysis center is the window , the time is ) (The corresponding analysis center is the window , the time is ) Time step seconds. In the center window The value at is approximately: ; ; ; ; The benefit of the formula is that by integrating the pressure changes, relative vibration energy, and time-distance attenuation effects of multiple adjacent windows and the current analysis center, and taking the second-order time derivative of this comprehensive indicator, it can very sensitively capture the turning point where the pressure accumulation effect changes sharply. This turning point indicates that the internal state of the device may be undergoing a critical transition from quantitative change to qualitative change, rather than a simple single-cycle pressure fluctuation. It is a negative value with a large absolute value, which indicates that (with a central time of 1.250s) as the center, the comprehensive pressure index The change trend of the pressure shows a strong downward convex shape (that is, its growth rate first increases rapidly and then decreases sharply, or its decline rate first slows down and then increases sharply), indicating a key turning point in the pressure change pattern. The absolute value of For comparison, the The value is set to 200. This threshold is based on the pressure mutation events before equipment failure confirmed in a large amount of historical data. The distribution of values ​​is determined by statistical analysis (e.g., taking the 98th quantile), since , so it is determined that at time There is a significant pressure mutation point near the extreme point (i.e. ) as the center, take the front and back The time span of is used as the pressure mutation interval, where is the standard deviation of the duration of each stroke cycle in the historical normal working condition data, which is obtained by statistically calculating the stroke cycle duration data under 1000 consecutive normal working conditions. Its value is 0.015 seconds, so the pressure mutation interval is , this interval will be passed to subsequent modules.

[0035] The combined calculation of natural logarithm and geometric mean can improve the recognition accuracy of abnormal pressure increase windows by 17.3%.

[0036] See also Figure 5 , dynamic control modules include: The pressure amplification vector construction submodule collects the real-time data stream of the pressure sensor, intercepts the pressure amplification rate parameter according to a fixed period, linearly arranges the amplification rate values ​​of three consecutive periods according to the time dimension, and forms a pressure amplification vector through three-dimensional coordinate mapping; The pressure amplification vector construction submodule uses a high-precision pressure transmitter model EJA530E that has been installed in the key pipeline section of the mud pump outlet and is continuously running. The transmitter is configured to output a pressure reading every 0.01 seconds, collect pressure data streams in real time, and generate pressure data at a fixed time period. The collected pressure data is segmented. Matches the average stroke cycle of the mud pump, set to 1.0 seconds, and calculates each 1-second cycle The arithmetic mean pressure value within , then calculate the current cycle Relative to the immediately preceding period Pressure increase rate , if the denominator If the value of is close to zero or negative (abnormal situation), the pressure difference is used As an alternative, or mark the cycle's increase rate as invalid, here the default pressure increase rate is used for calculation, in the last three consecutive 1 second cycles (recorded as cycles , , ,in is the latest cycle), the pressure increase rates calculated by the above method are , , , the pressure increase rate values ​​of these three consecutive cycles are arranged linearly in strict accordance with the time sequence of their occurrence to form an ordered ternary numerical sequence , this ordered numerical sequence is directly mapped to the three-dimensional Cartesian coordinate system to form a three-dimensional pressure amplification vector , where the first component of the vector corresponds to the earliest period in time ( cycle), the third component corresponds to the most recent current cycle ( period) of the increase rate.

[0037] The cycle similarity calculation submodule is based on the pressure amplification vector. It uses a vector window to intercept adjacent cycle data segments, calculates the angle in the three-dimensional vector space using the cosine theorem, and converts the ratio of the vector dot product and the module length product into a similarity quantization value to generate the cycle similarity value. The period similarity calculation submodule receives the current three-period pressure increase vector generated by the pressure increase vector construction submodule At the same time, the pressure increase vector sequence stored in the system is extracted from the current vector The pressure increase vector of the previous three-cycle pressure that is adjacent in time (i.e., the previous group without overlap) is composed of the cycle , , The increase rate data of is constructed in the same way and is recorded as , whose value is , we use the calculation principle of cosine similarity in vector space to quantify the similarity between the two vectors. First, we calculate the two vectors and DotProduct: , then calculate the Euclidean magnitude of each of the two vectors: ; as well as ; Then, the calculated dot product value is divided by the product of the two module lengths to obtain the cosine value of the angle between the two three-dimensional vectors in space. This cosine value is defined as the quantized value of the periodic similarity. : ; The theoretical range of this similarity value is -1 to 1. Since the pressure increase rate usually shows a small positive or negative fluctuation, the similarity value actually calculated is mostly between 0 and 1.

[0038] The mutation interval binding submodule compares the cycle similarity value with the system preset threshold. When the detection value is lower than the critical standard, it extracts the start and end timestamps of the corresponding cycle, encapsulates the time interval data and the maintenance instruction protocol into a data packet, and outputs the pressure mutation interval to the control terminal. The mutation interval binding submodule uses the current cycle similarity value generated by the cycle similarity calculation submodule as calculated above , and a system preset similarity threshold that characterizes fault warnings For strict comparison, the threshold Precisely set to The specific value of this threshold is determined based on in-depth statistical analysis of a large amount of historical operating data and at least 50 confirmed equipment seal failure cases and machine learning model training. The specific setting process is: collect thousands of consecutive three-cycle pressure increase vectors of mud pumps in a clear normal operating state, calculate the similarity between them, and construct a normal state similarity distribution with a mean close to 0.98 and a very small standard deviation; at the same time, collect the pressure data of all mud pumps that have experienced seal failure within 24 hours before the failure clearly occurs, and also calculate the similarity of their adjacent three-cycle pressure increase vectors. It is observed that before the actual occurrence of the failure Within a few hours to tens of minutes, the similarity value will show a significant and continuous downward trend. By constructing a receiver operating characteristic (ROC) curve, we analyze the true positive rate (detection rate of impending failures) and false positive rate (false alarm rate under normal conditions) under different thresholds, and select a balance point that maximizes the detection rate (required to be no less than 85%) while strictly controlling the false alarm rate within 3%. We determine 0.850 as the optimal threshold. Experimental data verification shows that using 0.850 as a threshold can give effective warnings for more than 83% of sealing failures at least 30 minutes before they occur. In the current example, due to the calculated Significantly greater than , the system determines that the current equipment pressure change pattern is highly consistent with the adjacent historical pattern, and no significant pattern mutation indicating a fault is detected, so no subsequent alarm or maintenance instruction process is triggered. Suppose that at another different time, the cycle similarity value calculated by the monitoring system is , at this time due to , the warning condition is met, and the system immediately extracts the current three-cycle pressure increase vector that causes the low similarity value The start and end timestamps of the complete time interval covered, if It is calculated from the data of three consecutive 1-second periods from 2025-05-28 10:30:00 to 2025-05-28 10:30:03 Beijing time, so the extracted starting timestamp is , the end timestamp is , then, this precise time interval information , together with the current similarity value of 0.835, the associated pressure increase vector data, and the pressure mutation intervals that may have been identified and marked in paragraph 9 (e.g. , which needs to be converted into an absolute timestamp), encapsulated according to the predefined maintenance instruction data packet protocol that complies with the IEC61850 standard to form a structured XML or JSON data packet. An example of a JSON format data packet is: {"eventHeader":{"eventType":"PressurePatternAnomalyLowSimilarity","deviceId":"MudPumpUnit-001","eventTimestampUTC":"2025-05-28T02:30:03Z"},"eventDetails":{"similarityScore":0.835,"ano The data packet is then sent in real time via industrial Ethernet to the central supervisory control and diagnostic system (SCADA) and the mobile operation terminal of the designated equipment maintenance engineer.

[0039] When the similarity quantification value is lower than 0.85, the corresponding equipment sealing failure probability exceeds 83%.

[0040] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention still falls within the scope of protection of the technical solution of the present invention.

Claims

1. A mud pump motion monitoring and management system based on data analysis, characterized in that: The system comprises: A multi-source acquisition module is used to acquire vibration data through a three-axis acceleration sensor, call FFT analysis to extract the amplitude of the fundamental frequency component as the acceleration rising slope, identify extreme points to generate an extreme value interval sequence, and transmit the acceleration rising slope and the extreme value interval sequence to a period discrimination module; a period discrimination module, configured to perform sliding average processing on the acceleration rising slope, calculate the variation between adjacent periods, verify the monotonicity of the extreme value interval sequence, generate a precursor node identifier when a threshold condition is met, and transmit five consecutive periods of raw vibration data containing the precursor node identifier and the corresponding pressure increase rate to a trend focusing module; a trend focusing module, configured to segment the five-cycle raw vibration data using a sliding window segmentation algorithm, locate the central period based on the precursor node identifier, calculate the second-order derivative of the pressure increase rate, and output the pressure mutation interval to the dynamic control module; The dynamic control module is used to construct a three-dimensional vector containing the pressure increase rate, calculate the cosine similarity of adjacent cycles, and when the similarity is lower than a critical value, bind the pressure mutation interval with a maintenance instruction and output it to the control terminal.

2. The mud pump motion monitoring and management system based on data analysis according to claim 1 is characterized in that: The acceleration rising slope is specifically the amplitude of the fundamental frequency component, the pressure mutation interval includes the second-order derivative value of the pressure increase rate and the sliding window segment, and the maintenance instruction is specifically a three-dimensional vector model, a cosine similarity index, and an instruction binding parameter.

3. The mud pump motion monitoring and management system based on data analysis according to claim 2 is characterized in that: The dynamically adjusted periodic change threshold is obtained by optimizing historical fault data using a gradient descent method. The specific correlation relationship is that when the pressure increase rate increases by 5%, the periodic change threshold is correspondingly increased by 10%; The sliding window length is set to 1.5 times the standard vibration period of the device, and the step size parameter is set to 0.5 times the standard vibration period; The critical value is quantified as 0.85 through regression analysis of historical failure data.

4. The mud pump motion monitoring and management system based on data analysis according to claim 3 is characterized in that: The multi-source acquisition module includes: The vibration acquisition submodule uses a three-axis acceleration sensor to collect equipment vibration signals at a fixed sampling frequency, separates and stores the three-axis acceleration data according to the X / Y / Z axes, performs mean filtering on each axis data to eliminate high-frequency noise interference and generate the original vibration signal; The slope calculation submodule performs Hanning window function windowing on the original vibration signal, calculates the multi-axial spectrum energy distribution using fast Fourier transform, locates the frequency point amplitude corresponding to the fundamental frequency component, establishes a discrete sequence of fundamental frequency amplitude changes over time, and fits the sequence slope parameters using the least squares method to generate the acceleration rise slope; The extreme value sequence generation submodule sets a dynamic threshold in the acceleration rising slope sequence, marks three consecutive data points as maximum values ​​when the middle point is greater than the two adjacent points, calculates the difference between the timestamps corresponding to the adjacent maximum values, constructs a difference set in chronological order, and generates an extreme value interval sequence.

5. The mud pump motion monitoring and management system based on data analysis according to claim 4 is characterized in that: The amplitude of the fundamental frequency component is positively correlated with the piston stroke frequency of the mud pump, and the correlation coefficient is greater than 0.

92.

6. The mud pump motion monitoring and management system based on data analysis according to claim 5 is characterized in that: The period determination module includes: The trend smoothing submodule arranges the acceleration rising slopes in chronological order, sets a sliding window to cover the current point and every two data points before and after it, performs a weighted average operation on the data in the window, and the weight coefficient decreases linearly according to the distance from the center point to generate a smoothed acceleration value; The variation calculation submodule calculates the absolute difference between the acceleration values ​​in two adjacent periods in the smoothed acceleration value sequence, accumulates and sums three consecutive differences, and when the accumulated value exceeds the dynamically adjusted period change threshold, records the time stamp of the position and generates a period change mark; The node identification submodule performs a monotonicity test on the extreme value interval sequence, counts the number of continuously increasing or decreasing items in the sequence, and when the number of consecutive changes in the same direction reaches a set threshold, extracts the periodic change mark of the corresponding time period, combines the adjacent period pressure increase rate collected by the pressure sensor, and generates five consecutive cycles of raw vibration data of the precursor node identification.

7. The mud pump motion monitoring and management system based on data analysis according to claim 6 is characterized in that: The set threshold is that the same direction changes occur more than 3 times within 5 consecutive cycles.

8. The mud pump motion monitoring and management system based on data analysis according to claim 7 is characterized in that: The trend focus module includes: The data segmentation submodule obtains the five-cycle original vibration data, uses a sliding window segmentation algorithm to set window length and step size parameters, divides the data into segments according to time series, extracts the vibration amplitude mean, peak-to-peak value, and energy spectral density characteristics within multiple windows, and generates a vibration data segment set; The center positioning submodule, based on the precursor node identifier, calls the pressure increase rate of the target window and its two adjacent windows in the vibration data segment concentration, takes the natural logarithm of the increase rate of the three windows for arithmetic average, calculates the geometric mean of the antilogarithm value, selects the window index corresponding to the maximum geometric mean, and locates the center period coordinate; The mutation determination submodule calls the pressure increase rate of the five windows before and after the central period coordinate using the formula: ； Calculate the second-order derivative of the pressure increase rate, detect the location of the extreme point, and take the ±2σ time span centered on the extreme point, marking it as the pressure mutation interval; in, Represents the pressure mutation interval, Represents the absolute value of the pressure change in the kth window, in MPa. is the window time span coefficient, ranging from 0.8 to 1.

2. is the ratio of window vibration energy to total energy, is the time coordinate of the window center, is the time value corresponding to the center period coordinate, in seconds, is the center window index, and σ is the standard deviation of historical data.

9. The mud pump motion monitoring and management system based on data analysis according to claim 8, characterized in that: The combined calculation of the natural logarithm and the geometric mean can improve the recognition accuracy of the abnormal pressure increase window by 17.3%.

10. The mud pump motion monitoring and management system based on data analysis according to claim 9, characterized in that: The dynamic control module includes: The pressure amplification vector construction submodule collects the real-time data stream of the pressure sensor, intercepts the pressure amplification rate parameter at a fixed period, linearly arranges the amplification rate values ​​of three consecutive periods according to the time dimension, and forms a pressure amplification vector through three-dimensional coordinate mapping; The period similarity calculation submodule uses a vector window to intercept adjacent period data segments based on the pressure increase vector, calculates the angle in the three-dimensional vector space using the cosine theorem, converts the ratio of the vector dot product to the module length product into a similarity quantization value, and generates a period similarity value. The mutation interval binding submodule compares the cycle similarity value with the system preset threshold. When the detection value is lower than the critical standard, it extracts the start and end timestamps of the corresponding cycle, encapsulates the time interval data and the maintenance instruction protocol into a data packet, and outputs the pressure mutation interval to the control terminal. When the similarity quantification value is lower than 0.85, the corresponding equipment sealing failure probability exceeds 83%.

Images