A fracturing equipment state monitoring and fault diagnosis system and method

Abstract

The application discloses a fracturing equipment state monitoring and fault diagnosis system and method, and belongs to the intelligent technology field of oil and gas field fracturing equipment. The application aims to solve the problems of single signal monitoring, lagging fault early warning and high misjudgment rate in the prior art. The method comprises the following steps: collecting multi-source operation data of a fracturing pump in real time; establishing a theoretical pressure indicator diagram, and comparing the theoretical pressure indicator diagram with an actual indicator diagram generated by real-time data to realize first-stage fault judgment; processing signals by using an improved wavelet threshold denoising method, and extracting time domain and frequency domain features; and inputting the processed feature data into a joint diagnosis model to realize second-stage fault judgment. The joint diagnosis model combines a principal component analysis (PCA) model for uncalibrated data anomaly detection and a BP neural network model for calibrated data fault classification. Through deep fusion of a mechanism model and a data-driven model, the application effectively resists environmental interference, and significantly improves diagnosis accuracy and operation and maintenance efficiency.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of intelligentization of oil and gas field fracturing equipment, and particularly relates to a fracturing equipment state monitoring and fault diagnosis system and method. BACKGROUND

[0002] As the core equipment for the development of unconventional resources such as shale gas and tight oil and gas, the fracturing pump bears the key task of injecting high-pressure fracturing fluid (sand-carrying fluid, acid fluid, etc.) into the formation to form a complex fracture network, and its technical performance directly determines the construction efficiency, safety and economy of hydraulic fracturing operation; with the growth of global energy demand and the development of unconventional oil and gas resources towards deep layer (buried depth greater than 3500m), super high pressure (pressure greater than 100MPa) and large displacement (displacement greater than 3m³ / min), the fracturing pump faces multiple technical challenges under extreme working conditions.

[0003] Typical fracturing pumps are mostly three-cylinder or five-cylinder plunger type structures, composed of a power end (including a crankshaft, a gear box, bearings, etc.) and a hydraulic end (including a pump head, a plunger, a suction valve, a discharge valve, etc.); during operation, the power end bears huge alternating loads, while the hydraulic end is long-term exposed to a combined stress environment of high pressure, high impact, high abrasion and chemical corrosion; such extreme working conditions easily lead to premature failure of key components (such as plunger seals, pump valves, crankshaft bearings), and the failure mode often exhibits multi-dimensional coupling characteristics of mechanical and hydraulic systems. At present, the industry generally uses an alarm system based on a single signal and fixed threshold for the state monitoring of fracturing pumps, mainly relying on the monitoring of discrete data of vibration, pressure or temperature; this method has significant technical limitations, and existing systems mostly analyze the power end bearing vibration or the hydraulic end pressure fluctuation independently, lacking systematic evaluation of mechanical-hydraulic coupling characteristics; research shows that when the vibration signal exceeds the threshold (such as acceleration >10g), the bearing wear has entered the middle and late stages (clearance >0.5mm), and the risk of sudden shutdown increases by more than 3 times; in addition, a single signal is easily disturbed by environmental interference such as pipe manifold vibration conduction and fracturing truck group coordinated operation load fluctuation, with a misjudgment rate of more than 25%; the fault data of North American shale gas fields in 2021 showed that the cases of pressure pulse misjudgment as valve failure accounted for 32% of the invalid maintenance amount; the conservative threshold strategy also forces the operation party to adopt a "preventive shutdown" mode, with an average annual unplanned downtime of a single fracturing pump of 120 hours, and maintenance costs accounting for 18%-25% of the total equipment operation cost.

[0004] Although there have been attempts to introduce supplementary monitoring means such as infrared thermal imaging, acoustic emission, etc., there are still difficulties in multi-source heterogeneous data fusion, low precision in fault feature extraction, and lack of effective residual life prediction models, etc. Therefore, the industry urgently needs a new technology system that can deeply integrate high-precision sensing data, multi-physical field coupling models and intelligent algorithms to realize the health management of the fracturing pump throughout its life cycle, improve the fault recognition rate and reduce the operation and maintenance cost. SUMMARY

[0005] The present application aims to solve the problems of relying on single-dimensional monitoring signals, fault early warning lag and high operation and maintenance cost in the existing fracturing pump monitoring technology, and provides a fracturing equipment state monitoring and fault diagnosis system and method based on multi-source information fusion and dynamic coupling analysis.

[0006] The purpose of the present application is achieved by the following technical solutions:

[0007] The fracturing equipment state monitoring and fault diagnosis system comprises a multi-source information acquisition module, a mechanism model analysis module, a signal processing module and a joint fault diagnosis module. The multi-source information acquisition module is composed of a sensor group, a PLC, a voltage / current multi-channel acquisition card, a switch and a gateway, and is used for acquiring real-time data of each measuring point of the fracturing equipment and integrating and forwarding; the mechanism model analysis module generates a theoretical pressure indicator diagram and an actual pressure indicator diagram through the rated parameters and actual parameters of the measured equipment respectively, and makes a first-level fault state judgment by comparing the differences; the signal processing module is used for processing high-frequency vibration and high-frequency pressure signals, using an improved wavelet algorithm for signal denoising, and then extracting fault feature data in the time-frequency domain and reducing the dimension; the joint fault diagnosis module is used for receiving the fault feature data and making a second-level fault state judgment, and at least two diagnostic models are integrated inside, and according to whether the data has a pre-labeled fault type, a corresponding diagnostic model is selected for analysis.

[0008] The present application also provides a corresponding fracturing equipment state monitoring and fault diagnosis method, characterized by comprising the following steps:

[0009] Step one: real-time acquisition of the operating data of the fracturing equipment, including vibration, pressure, axial force and key phase displacement;

[0010] Step two: establishment of a theoretical pressure indicator diagram and a diagnostic model of the fracturing equipment, generation of an actual pressure indicator diagram according to the acquired real-time operating data, first-level fault state judgment by comparing the characteristic differences between the theoretical pressure indicator diagram and the actual pressure indicator diagram;

[0011] The theoretical pressure calculation formula of the liquid in the working chamber during suction and discharge is:

[0012]

[0013]

[0014] wherein Pf and Pe are the liquid pressure in the working chamber during the suction and discharge processes, respectively; Pb and Pc are the reference pressures on the suction side and the discharge side, respectively; g is the specific weight of the fluid; hL and hR are the frictional head losses of the fluid in the suction and discharge lines, respectively; hL1 and hR1 are the local head losses of the fluid in the suction and discharge lines, respectively; hL2 and hR2 are the internal losses of the fluid in the pump during the suction and discharge processes, respectively; V is the average flow velocity in the pipe; V1 is the flow velocity in the suction and discharge pipes; g is the acceleration due to gravity;

[0015] Step three: signal processing on the collected high-frequency data, including noise reduction processing, feature extraction, and feature dimension reduction;

[0016] (1) Wavelet noise reduction processing

[0017] Since the high-frequency signal in the working process of the fracturing equipment is sometimes a non-stationary signal, and the wavelet transform belongs to time-frequency analysis and has self-adaptability, it is very suitable for processing non-stationary signals. In order to solve the problem that the fixed threshold noise reduction method is easy to cause loss of effective information when processing multi-layer wavelet coefficients, a local adaptive threshold is used in the present application, which can be dynamically adjusted according to the number of wavelet decomposition layers, and its calculation formula is:

[0018]

[0019]

[0020] wherein is the original vibration signal; is the total number of wavelet decomposition layers; is the current wavelet decomposition layer number; is the nth layer low-frequency coefficient; is the ith layer high-frequency coefficient; is the adaptive threshold of the ith layer; is the noise standard deviation; is the length of the processed signal;

[0021] In order to solve the problems of signal discontinuity caused by hard threshold function, constant deviation produced by soft threshold function, and complex calculation of traditional semi-soft threshold function, an improved semi-soft threshold function is used to quantize the wavelet coefficients, and its expression is:

[0022]

[0023] wherein is the current wavelet decomposition layer number; is a sign function; is the threshold filtered i-th layer high frequency coefficient; is the i-th layer high frequency coefficient; is the threshold value; is a natural constant.

[0024] (2) Time-frequency domain feature extraction

[0025] After denoising the high-frequency signal, in order to comprehensively represent the equipment state from the running data, multi-dimensional feature extraction is performed on the signal. Specifically, time domain features that can reflect the statistical distribution and energy characteristics of the signal are extracted from the signal through the calculation formula of the time domain features, including maximum value, minimum value, peak-to-peak value, mean value, variance, standard deviation, root mean square, kurtosis, skewness; and the frequency domain features that can reflect the frequency structure and energy distribution of the signal are further extracted by fast Fourier transform FFT conversion to the frequency domain, including mean square frequency, center of gravity frequency, frequency variance, thereby forming a multi-dimensional feature vector.

[0026] (3) Feature dimension reduction

[0027] Preferably, for feature dimension reduction, the PCA dimension reduction method is adopted to map the original high-dimensional feature space to a low-dimensional principal component space, while retaining the variance information of the original data to the greatest extent, eliminating the correlation between features, and providing high-quality input for the subsequent diagnostic model.

[0028] Step four: input the fault feature data into the joint fault diagnosis module for second-level fault state judgment, the joint fault diagnosis module calls the corresponding diagnostic model integrated internally according to whether the fault feature data has a pre-labeled fault type, analyzes and outputs the diagnostic result. The joint fault diagnosis module integrates two models with different mechanisms:

[0029] The diagnostic model based on labeled fault data preferably adopts a BP neural network model. The BP neural network includes an input layer, multiple hidden layers and an output layer, and is trained through a back propagation algorithm, which can effectively learn and identify the fault patterns of complex systems, and perform accurate fault classification and prediction. The construction process is as follows:

[0030] (1) Forward propagation calculation: input the training set, and calculate the output value through the weight and threshold value, the calculation formula is:

[0031]

[0032] wherein Index of the network layer; Index of the node of the current layer; Output of the i-th node of the k-th layer; Activation function; Input of the i-th node of the k-th layer; Index of the node of the previous layer; Number of nodes of the previous layer; Connection weight of the j-th node of the previous layer to the i-th node of the k-th layer; Output of the j-th node of the k-1-th layer; Bias of the i-th node of the k-th layer.

[0033] (2) Error backpropagation: calculate the error between the actual output value and the predicted output value, adjust the weight and threshold value by gradient descent method to reduce the error, and the error backpropagation formula is:

[0034]

[0035] wherein is the update amount of the connection weight of the i-th neuron to the j-th neuron; is the learning rate; is the loss function the partial derivative of the weight , that is, the gradient;

[0036] The weight adjustment formula is:

[0037]

[0038] wherein is the weight connecting the i-th neuron and the j-th neuron; is the number of iterations; is the weight value at the t-th iteration; is the update amount of the weight; is the learning rate; is the loss function the partial derivative of the weight , that is, the gradient.

[0039] (3) BP neural network predicts the fault type, and the calculation formula is:

[0040]

[0041] wherein is the predicted output of the BP neural network; is the trained BP neural network model; is the normalized test data.

[0042] Based on the diagnosis model of uncalibrated fault data, preferably using principal component analysis PCA model, the ratio between control limit and statistics is used for judgment, and the process is as follows:

[0043] (1) Calculate Q control limit, the calculation formula is:

[0044]

[0045] Wherein Q control limit; First order moment; Second order moment; Shape parameter; Critical value of standard normal distribution.

[0046] (2) Calculate SPE statistics, the calculation formula is:

[0047]

[0048] Wherein SPE statistics of the ith sample; The ith test set sample after standardization; Projection matrix; Transpose of P matrix.

[0049] (3) Calculate T² control limit and statistics, the calculation formula is:

[0050]

[0051]

[0052] Wherein T² control limit; Selected principal component number; Sample number of training data; 95% quantile of F distribution, based on k and N-k degrees of freedom; T² statistics of the ith sample; The ith test sample after standardization; Projection matrix; Transpose of P matrix; Inverse matrix of diagonal matrix, containing the first k eigenvalues; Transposition of vector form.

[0053] (4) Compare control limit and statistics: if the data sample is detected for three times in succession > Or > ​, it is illustrated that the sample data is an abnormal data sample, the storage of all abnormal data samples is calculated, and the formula is:

[0054]

[0055] Wherein The residual error of the i-th sample is residual error of the i-th sample; The normalized i-th test sample is normalized i-th test sample; The projection of the i-th test sample in the principal component space is the projection of the i-th test sample in the principal component space; The transpose of the projection matrix is the transpose of the projection matrix.

[0056] (5) Contribution analysis: according to the obtained fault data sample, the contribution of each variable in the sample is calculated, and the variable with the largest contribution is extracted for analysis, so as to realize accurate fault positioning, and the calculation formula is:

[0057]

[0058] Wherein The contribution of the j-th variable of the i-th sample is the contribution of the j-th variable of the i-th sample; The residual error of the j-th variable in the i-th sample is the residual error of the j-th variable in the i-th sample.

[0059] Further technical solutions also include a fault strategy scheduling module and a remote control module, which further improve the fault handling efficiency and monitoring efficiency; after the fault pre-alarm occurs, the fault strategy scheduling module can automatically generate the vibration and pressure image corresponding to the current fault point, detailed text description and maintenance strategy; the remote control module uploads data to the cloud at regular intervals through the upper computer deployed in the field instrument room, so that remote query and control instructions can be realized.

[0060] The fracturing equipment state monitoring and fault diagnosis system and method provided by the application has the following beneficial effects:

[0061] (1) Improve fault diagnosis accuracy and early identification ability: by integrating vibration, pressure, temperature and flow sensor network, a multi-level dynamic fault feature extraction model is constructed to realize accurate identification of small faults;

[0062] (2) Significantly enhance system anti-interference ability and reliability: the false alarm rate is reduced from the industry average of 25% to less than 8%, and the system remains stable under extreme working conditions such as ultrahigh pressure and high sand ratio;

[0063] (3) Improve intelligent level and self-adaptive ability: the dynamic PCA threshold correction mechanism can adapt to displacement fluctuation, pressure change and other complex working conditions for self-adaptive adjustment;

[0064] (4) The verification effect is remarkable and has popularization and application value: when the training sample amount is more than 4000, the false positive rate is stable within 8%. BRIEF DESCRIPTION OF DRAWINGS

[0065] Figure 1 The composition schematic diagram of the fracturing equipment monitoring system proposed by the application is shown in the figure;

[0066] Figure 2 The structure schematic diagram of the fracturing equipment monitoring system proposed by the application is shown in the figure;

[0067] Figure 3 The fault diagnosis flowchart of the fracturing equipment proposed by the application is shown in the figure;

[0068] Figure 4 The theoretical and actual pressure force comparison diagram of the fracturing equipment is shown in the figure;

[0069] Figure 5 The actual vibration data comparison diagram of each pump cavity of the liquid end of the fracturing pump is shown in the figure;

[0070] Figure 6 The theoretical and actual discharge pressure comparison diagram of each pump cavity of the liquid end of the fracturing pump is shown in the figure;

[0071] Figure 7 The suction pressure actual data scatter diagram of the liquid end of the fracturing pump is shown in the figure;

[0072] Figure 8 The false positive rate change curve diagram of the fracturing equipment fault diagnosis is shown in the figure. DETAILED DESCRIPTION

[0073] In order to deepen the understanding of the application, the application will be further described in combination with examples below, and the examples are only used to explain the application and do not constitute the limitation to the protection scope of the application. The running data of a GD500 type fracturing pump is taken as an example to verify the fault diagnosis method described in the application, and the main parameters of the equipment during running are shown in Table 1:

[0074] Table 1: GD5000 type fracturing pump running parameters

[0075]

[0076] As Figure 3As shown, the pressure equipment fault diagnosis process proposed by the application adopts a staged progressive structure as a whole: on the left, historical data and real-time data are collected and classified, and PCA model and BP neural network model are respectively used for training to form a multi-dimensional fault diagnosis basic model library; on the right, the multi-source fusion fault diagnosis process forms a closed-loop intelligent diagnosis system, which generates a preliminary diagnosis result after inputting the to-be-tested data, and triggers a branch process through abnormal state judgment - if the detection is abnormal, the fault type is accurately output and the result verification link is entered; if the detection is not abnormal, it is determined as normal state, and the feedback mechanism of the verification link can return new data to the PCA model to realize dynamic optimization, ensuring that the system continuously adapts to the working condition changes.

[0077] As shown in Figure 4 , in the case of actual discharge pressure of 80 MPa, the actual measured value rises slowly in the initial stage, does not reach the theoretical pressure in the stable stage, and the pressure decreases synchronously in the final stage, but due to the influence of leakage, the rate and end pressure of the descending process also differ from the theoretical value; Figure 5 is a comparison chart of actual vibration data of each pump cavity of the fracturing pump liquid end, which shows the actual vibration data of each pump cavity of the fracturing pump liquid end; Figure 6 is a comparison chart of theoretical discharge pressure and actual discharge pressure of each pump cavity of the fracturing pump liquid end, which shows the theoretical discharge pressure of each pump cavity at different times, and the actual discharge pressure is drawn at the top, and the comparison between the two can determine whether there is abnormal pressure; Figure 7 is a scatter plot of actual suction pressure data of the fracturing pump liquid end; each actual data is taken from the same time axis and corresponds to each other. Figure 6 The measured discharge pressure shows periodic abnormal pressure fluctuations, which do not match the theoretical pressure fluctuations of each cylinder, combined with Figure 5 the vibration curve, the pump cavities also show periodic and varying degrees of spikes, and Figure 7 the measured suction pressure does not show large fluctuations, indicating that the pump cavity has the risk of early failure and is most likely a discharge valve failure; the final model diagnosis shows that the fault is the discharge valve leakage of No. 3 pump cavity, and timely generates the vibration and pressure image corresponding to the current fault point, detailed text description and maintenance strategy.

[0078] It can be seen from Figure 8 that when the training sample size is 1000-10000, the false positive rate can be stabilized at 7%-10%.

Claims

1. A fracturing equipment condition monitoring and fault diagnosis system, characterized in that, include: The multi-source information acquisition module consists of a sensor group, a PLC, a voltage / current type multi-channel acquisition card, a switch, and a gateway. It is used to collect real-time data from various measuring points of the fracturing equipment and integrate and forward it. The mechanism model analysis module is used to establish a theoretical pressure indicator diagram based on the design parameters and working conditions of the fracturing pump, and to generate an actual pressure indicator diagram based on the real-time operating data collected by the multi-source information acquisition module. By comparing the characteristic differences between the theoretical pressure indicator diagram and the actual pressure indicator diagram, the first-level fault state is judged. The signal processing module is used to process high-frequency vibration and high-frequency pressure signals. It uses an improved wavelet algorithm to reduce signal noise, thereby extracting fault feature data in the time and frequency domain and reducing its dimensionality. The joint fault diagnosis module is used to receive the fault feature data and perform a second-level fault status judgment. The joint fault diagnosis module integrates at least two diagnostic models. Depending on whether the data has a pre-defined fault type, the corresponding diagnostic model is selected for analysis. Diagnostic models based on calibrated fault data: BP neural network model, used to classify and identify data with predefined fault type labels; Diagnostic models based on uncalibrated fault data: Principal Component Analysis (PCA) model, used for anomaly detection of data not identified as known fault types.

2. The fracturing equipment condition monitoring and fault diagnosis system according to claim 1, characterized in that: It also includes a fault strategy scheduling module and a remote control module; after a fault pre-alarm occurs, the fault strategy scheduling module can automatically generate vibration and pressure images, detailed text descriptions and maintenance strategies corresponding to the current fault point; the remote control module uploads data to the cloud periodically through the host computer deployed in the field instrument room, enabling remote issuance of query and control commands.

3. A method for condition monitoring and fault diagnosis of fracturing equipment, characterized in that: Includes the following steps: Step 1: Real-time acquisition of operational data from fracturing equipment, including vibration, pressure, axial force, and key phase displacement; Step 2: Establish the theoretical pressure indicator diagram and diagnostic model of the fracturing equipment, and generate the actual pressure indicator diagram based on the collected real-time operating data. By comparing the characteristic differences between the theoretical pressure indicator diagram and the actual pressure indicator diagram, the first-level fault state judgment is performed. Step 3: Perform signal processing on the collected high-frequency data, including noise reduction, feature extraction, and feature dimensionality reduction; 1) Wavelet noise reduction processing: To address the issue of effective information loss in fixed-threshold denoising methods when processing multi-level wavelet coefficients, a locally adaptive threshold is employed. This threshold can be dynamically adjusted according to the different wavelet decomposition levels. in This is the original vibration signal; This represents the total number of wavelet decomposition levels. This represents the current wavelet decomposition level. These are the low-frequency coefficients of the nth layer; For the i-th layer, the high-frequency coefficients are: The adaptive threshold for the i-th layer; The standard deviation of noise; The length of the processed signal; An improved semi-soft thresholding function is also used to quantize the wavelet coefficients, and its expression is as follows: in This represents the current wavelet decomposition level. It is a symbolic function; These are the high-frequency coefficients of the i-th layer after threshold filtering; For the i-th layer, the high-frequency coefficients are: The threshold value is used; It is a natural constant; 2) Time-frequency domain feature extraction The time-domain features that reflect the statistical distribution and energy characteristics of the denoised signal are extracted from the time-domain feature calculation formula; the signal is then transformed to the frequency domain by Fast Fourier Transform (FFT) to further extract the frequency domain features that reflect the frequency structure and energy distribution, thus forming a multi-dimensional feature vector. 3) Feature dimensionality reduction Principal component analysis (PCA) maps the original high-dimensional feature space to a low-dimensional principal component space, preserving the variance information of the original data to the greatest extent while eliminating the correlation between features, thus providing high-quality input for subsequent diagnostic models. Step 4: Input the dimensionality-reduced feature data into the joint fault diagnosis module for the second-level fault status judgment. The joint fault diagnosis module calls the corresponding internally integrated diagnostic model to analyze and output the diagnostic results based on whether the dimensionality-reduced feature data has a pre-defined fault type. Diagnostic models based on calibrated fault data: BP neural network model, used to classify and identify data with predefined fault type labels; Diagnostic models based on uncalibrated fault data: Principal Component Analysis (PCA) model, used for anomaly detection of data not identified as known fault types.

4. The method for monitoring the condition and diagnosing faults of fracturing equipment according to claim 3, characterized in that, The BP neural network model includes an input layer, multiple hidden layers, and an output layer. It is trained using the backpropagation algorithm and can effectively learn and identify fault modes in complex systems, enabling accurate fault classification and prediction. in This is the predicted output of the BP neural network; The trained BP neural network model; This is the standardized test data.

5. The method for monitoring the condition and diagnosing faults of fracturing equipment according to claim 3, characterized in that, The PCA principal component analysis model calculates the SPE statistic and T² statistic of the test sample: in Let SPE statistic be the value of the i-th sample. This is the standardized i-th test set sample; The projection matrix; for Transpose of a matrix; Let T² be the T-statistic for the i-th sample; for Transpose of a vector; It is the inverse of the diagonal matrix; The statistic is compared with the control limits obtained from training to determine whether the sample data is abnormal. If abnormal, the sample residuals are calculated. in Let be the residual of the i-th sample; This is the standardized i-th test sample; Let be the projection of the i-th test sample into the principal component space; This is the transpose of the projection matrix; Further contribution analysis was used to pinpoint the key variables causing the anomalies: in The contribution of the j-th variable to the i-th sample; Let be the residual of the j-th variable in the i-th sample.

Images