Cementing pump fault diagnosis method and device

Abstract

The application discloses a cementing pump fault diagnosis method and device, and the method comprises the following steps: acquiring historical data of a cementing pump rotating speed sensor, historical data of a vibration sensor and historical operation conditions in cementing operation; extracting multi-dimensional historical characteristics of the vibration sensor based on time domain and frequency domain; combining the multi-dimensional historical characteristics of the vibration sensor with the historical data of the rotating speed sensor to obtain historical input characteristics of the cementing pump; calculating the relative inter-class dispersion distance of the historical input characteristics of the cementing pump, constructing a feature importance factor and performing weighted processing; inputting the weighted processed historical input characteristics of the cementing pump and the historical operation conditions into a preset weighted width learning model, training the preset weighted width learning model, obtaining the trained preset weighted width learning model as a fault classifier, and predicting the fault diagnosis result according to real-time data of the cementing pump rotating speed sensor and the vibration sensor in the cementing operation process.

Description

Technical Field

[0001] This invention relates to the field of safety monitoring technology for offshore oil cementing equipment, specifically to a method and device for diagnosing cementing pump faults. Background Technology

[0002] In offshore oil production, cementing operations are a crucial link between drilling and oil production. They play a vital role in ensuring the safe, efficient, and environmentally friendly production of oil and gas wells, and have a direct impact on the stable operation and economic benefits of the entire oil and gas industry chain.

[0003] The core of a cementing pump injection system is the cementing pump, whose main function is to inject cement slurry into the wellbore to fix the wellbore and prevent oil and gas leakage. Continuous and stable operation of the cementing pump helps shorten operation time, improve efficiency, and reduce costs. Offshore platforms operate in harsh environments, and the cementing pump has many wear parts (such as valves, plungers, and packing). Equipment failures (such as valve leakage or severe packing wear) can lead to uneven cement slurry injection and unstable pressure, affecting the overall wellbore fixation effect and potentially causing significant marine ecological disasters. Therefore, real-time and accurate monitoring of cementing pump malfunctions is crucial to avoid sudden shutdowns and severe losses during actual use.

[0004] In recent years, with the development of information technology for cementing equipment, a large amount of cementing pump operation data can be collected in real time at the operation site and stored in a database for data-driven research on cementing pump fault diagnosis. However, existing neural network-based fault diagnosis methods for offshore cementing pumps have shortcomings: on the one hand, traditional neural networks use iterative optimization methods, resulting in low training efficiency; on the other hand, existing methods ignore the differences in data features, treating all features the same, making it difficult to leverage the outstanding role of some effective features. Summary of the Invention

[0005] In view of the above problems, embodiments of the present invention are proposed to provide a cementing pump fault diagnosis method and apparatus that overcomes or at least partially solves the above problems.

[0006] According to one aspect of the present invention, a method for diagnosing faults in a cementing pump is provided, the method comprising:

[0007] Acquire historical data from cementing pump speed sensor, vibration sensor, and historical operating conditions during cementing operations; historical operating conditions include normal operating conditions and fault operating conditions.

[0008] For any vibration sensor's historical data, multi-dimensional historical features of the vibration sensor are extracted based on the time and frequency domains; the multi-dimensional historical features of the vibration sensor are combined with the historical data of the rotation speed sensor to obtain the historical input features of the cementing pump.

[0009] Calculate the relative inter-class dispersion distance of each historical input feature in the historical input features of the cementing pump, and construct a feature importance factor based on the relative inter-class dispersion distance, so as to weight the historical input features of the cementing pump according to the feature importance factor;

[0010] The weighted historical input features and historical operating conditions of the cementing pump are input into a preset weighted width learning model. The preset weighted width learning model is trained to obtain a fault classifier, which is used to predict fault diagnosis results based on real-time data from the cementing pump speed sensor and vibration sensor during cementing operations.

[0011] According to another aspect of the present invention, a cementing pump fault diagnosis device is provided, comprising:

[0012] The historical data acquisition module is suitable for acquiring historical data from cementing pump speed sensor, vibration sensor, and historical operating conditions during cementing operations; historical operating conditions include normal operating conditions and fault operating conditions.

[0013] The multi-dimensional extraction module is suitable for extracting multi-dimensional historical features of any vibration sensor based on time and frequency domains from historical data of the vibration sensor; and combining the multi-dimensional historical features of the vibration sensor with historical data of the speed sensor to obtain the historical input features of the cementing pump.

[0014] The weighted processing module is suitable for calculating the relative inter-class dispersion distance of each historical input feature in the historical input features of the cementing pump, and constructing a feature importance factor based on the relative inter-class dispersion distance, so as to perform weighted processing on the historical input features of the cementing pump according to the feature importance factor;

[0015] The model training and prediction module is suitable for inputting the weighted historical input features and historical operating conditions of the cementing pump into a preset weighted width learning model, training the preset weighted width learning model, and obtaining the trained preset weighted width learning model as a fault classifier, so as to predict the fault diagnosis results based on the real-time data of the cementing pump speed sensor and vibration sensor during the cementing operation.

[0016] According to another aspect of the present invention, a computing device is provided, comprising: a processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other through the communication bus;

[0017] The memory is used to store at least one executable instruction, which causes the processor to perform the operation corresponding to the above-described cementing pump fault diagnosis method.

[0018] According to another aspect of the present invention, a computer storage medium is provided, the storage medium storing at least one executable instruction, the executable instruction causing a processor to perform an operation corresponding to the above-described cementing pump fault diagnosis method.

[0019] According to another aspect of the present invention, a computer program product is provided, including at least one executable instruction that causes a processor to perform an operation corresponding to the above-described cementing pump fault diagnosis method.

[0020] The cementing pump fault diagnosis method and apparatus provided in this embodiment of the invention fully extracts multi-dimensional data features of the cementing pump's vibration sensor from both time-frequency and domain-frequency perspectives. When weighting the feature importance factors based on the multi-dimensional features, important features closely related to the fault condition can be assigned larger weights, while secondary features with low correlation can be given smaller weights. This fully utilizes the differences in information contained in different features to improve the modeling effect of the preset weighted width learning model, thereby improving the accuracy of predicting the fault classification of the cementing pump.

[0021] The above description is merely an overview of the technical solutions of the embodiments of the present invention. In order to better understand the technical means of the embodiments of the present invention and to implement them in accordance with the contents of the specification, and to make the above and other objects, features and advantages of the embodiments of the present invention more apparent and understandable, specific implementation methods of the embodiments of the present invention are described below. Attached Figure Description

[0022] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the embodiments of the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0023] Figure 1 A flowchart of a cementing pump fault diagnosis method according to an embodiment of the present invention is shown;

[0024] Figure 2 A schematic diagram of the pre-weighted width learning model structure is shown;

[0025] Figure 3 The diagram shows the confusion of classification results from a traditional learning model.

[0026] Figure 4 The classification confusion graph of the feature-weighted width learning model is shown;

[0027] Figure 5 A schematic diagram of a cementing pump fault diagnosis device according to an embodiment of the present invention is shown;

[0028] Figure 6 A schematic diagram of the structure of a computing device according to an embodiment of the present invention is shown. Detailed Implementation

[0029] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this invention will be thorough and complete, and will fully convey the scope of the invention to those skilled in the art.

[0030] Figure 1 A flowchart of a cementing pump fault diagnosis method according to an embodiment of the present invention is shown, such as... Figure 1 As shown, the method includes the following steps:

[0031] Step S101: Obtain historical data from the cementing pump speed sensor, historical data from the vibration sensor, and historical operating conditions during cementing operations.

[0032] During historical cementing operations, the operating data of the cementing pump can be recorded and stored in a database. This database contains historical data on the cementing pump's operation during past cementing operations. From this database, historical data from the cementing pump's speed sensor, vibration sensor, and historical operating conditions can be retrieved. Since the cementing pump has multiple vibration sensors, historical data for each sensor needs to be acquired. Historical operating conditions include normal and fault conditions, and different numerical classifications can be set for each type of fault condition.

[0033] Specifically, retrieve historical operating data of cementing pumps containing n samples from the database: {s(t), v1(t), v2(t), ... v i (t),…v m y(t), y(t), 1≤t≤n, where t represents the sample number, s(t) represents the historical data of the speed sensor, v1(t), v2(t), ..., v my(t) represents the historical data of m vibration sensors installed at different locations, and y(t) represents the operating condition type of the cementing pump's historical operating data. Its value can be an integer such as {0, 1, ..., C}, corresponding to normal operating conditions and various fault conditions, respectively. The historical operating condition data values ​​are given as examples; specific settings should be made according to the implementation situation and are not limited here. The historical operating conditions correspond to the historical data of the cementing pump's speed sensor and vibration sensor in the sample. These historical operating conditions can be labeled with the historical data of the speed sensor and vibration sensor for use in the subsequent training of the pre-set weighted width learning model.

[0034] Step S102: For the historical data of any vibration sensor, extract the multi-dimensional historical features of the vibration sensor based on the time domain and frequency domain, and combine the multi-dimensional historical features of the vibration sensor with the historical data of the rotation speed sensor to obtain the historical input features of the cementing pump.

[0035] Existing technologies treat all features equally and indiscriminately when processing historical data from vibration sensors. This approach fails to effectively utilize the differences among various features of historical vibration sensor data and cannot reflect the degree to which different data features influence operating conditions. Considering these issues, this embodiment analyzes the historical data of any vibration sensor from a cementing pump from both time and frequency domain perspectives, extracting multi-dimensional historical features of the vibration sensor. These multi-dimensional historical features include the maximum, minimum, mean, standard deviation, peak-to-peak value, RMS value, skewness, kurtosis, centroid frequency, root mean square frequency, mean frequency, and frequency standard deviation of the vibration signal. These multi-dimensional historical features can be obtained by analyzing historical data from multiple vibration sensors. i The result is obtained by calculating (t). Where 1≤i≤m, m is the number of vibration sensors, and t is the t-th sample.

[0036] For the maximum value The following can be calculated using formula (1):

[0037]

[0038] in, Let R be the data points in the t-th sample from the i-th vibration sensor.

[0039] For the minimum value The following can be calculated using formula (2):

[0040]

[0041] For the mean The following can be calculated using formula (3):

[0042]

[0043] For the standard deviation It is calculated using formula (4):

[0044]

[0045] For the peak-to-peak value It is calculated using formula (5):

[0046]

[0047] For the effective value It is calculated using formula (6):

[0048]

[0049] For the skewness It is calculated using formula (7):

[0050]

[0051] For the kurtosis It is calculated using formula (8):

[0052]

[0053] For the center frequency It is calculated using formula (9):

[0054]

[0055] where d i,j (1 < j < J) are J frequency values in the spectrum of v i (t), and P(d i,j ) is the power spectrum of the vibration signal.

[0056] For the root mean square frequency It is calculated using formula (10):

[0057]

[0058] For the mean frequency It is calculated using formula (11):

[0059]

[0060] For the frequency standard deviation It is calculated using formula (12):

[0061]

[0062] The multi-dimensional historical characteristics of each cementing pump vibration sensor are obtained by calculating the historical data of each vibration sensor using the above formulas (1)-(12). The historical input characteristics of the cementing pump are obtained by combining the multi-dimensional historical features of the vibration sensor with the historical data of the speed sensor. Among them, the historical input features are obtained by combining the multi-dimensional historical features of the vibration sensor with the historical data of the speed sensor. The total amount is related to the number of dimensions 'a' of the multi-dimensional historical features and the number of vibration sensors 'm', with the total amount M = am + 1, f k (t) represents the k-th historical input feature, where 1 ≤ k ≤ M.

[0063] Step S103: Calculate the relative inter-class dispersion distance of each historical input feature in the historical input features of the cementing pump, and construct a feature importance factor based on the relative inter-class dispersion distance, so as to perform weighted processing on the historical input features of the cementing pump according to the feature importance factor.

[0064] The historical input features of the cementing pump include multi-dimensional historical features from vibration sensors and historical features from speed sensors. However, the importance of each historical feature varies under different operating conditions. If the historical input features of the cementing pump are directly used for model training, the differences between different data features are ignored, resulting in some data features failing to play an effective role during model training. Therefore, it is necessary to further process the differences in important features in the historical output features to improve the accuracy of the model.

[0065] Specifically, the relative dispersion analysis method is used to calculate the first and second class dispersions of the cementing pump under each historical operating condition based on the mean of the historical input characteristics of the cementing pump under each historical operating condition and the covariance matrix under all historical operating conditions. The first class dispersion is the overall inter-class dispersion of the cementing pump, and the second class dispersion is the inter-class dispersion of the cementing pump after removing any feature.

[0066] Specifically, the first type of inter-class dispersion (disp) all The overall inter-class dispersion, which includes all historical features of various working conditions, is calculated according to formula (13):

[0067]

[0068] Among them, mf x ,mf y ...

[0069] The second inter-class dispersion (disp) k The inter-class scatter after removing the k-th feature, where k is any feature, is calculated according to formula (14):

[0070]

[0071] in, Let represent the mean values ​​of the historical feature data for the x-th and y-th historical conditions after removing the k-th feature, respectively. Both are row vectors. covf k This is the covariance matrix of all operating condition data after removing the k-th feature.

[0072] The relative inter-class dispersion distance (dist) of each feature in the historical input features of the cementing pump can be calculated based on the first and second class dispersion. k According to formula (15), the result is:

[0073]

[0074] Relative inter-class dispersion distance k The larger the value, the more important the corresponding feature; the feature importance factor imp. k According to formula (16):

[0075]

[0076] Specifically, the feature importance factor is determined based on the comparison between the first-class and second-class dispersion. When the first-class dispersion is greater than the second-class dispersion, the feature importance factor is determined by the relative inter-class dispersion distance; otherwise, the feature importance factor is determined by a preset factor. The preset factor γ can be set according to the implementation situation and is not limited here.

[0077] After obtaining the feature importance factors, construct a diagonal matrix Imp based on the feature importance factors, as shown in formula (17):

[0078] Imp = diag{[imp1 … imp k … imp M ]} (17)

[0079] Where, diag represents the diagonal matrix operator, and M is the total amount of historical input features of the cementing pump, which is determined by referring to the description in step S102.

[0080] The historical input features f(t) of the cementing pump are weighted using the diagonal matrix Imp to obtain the weighted historical input features f. w (t), as shown in formula (18):

[0081] f w (t)=f(t)×Imp (18)

[0082] Step S104: Input the weighted historical input features and historical operating conditions of the cementing pump into the preset weighted width learning model, train the preset weighted width learning model, and use the trained preset weighted width learning model as a fault classifier to predict the fault diagnosis result based on the real-time data of the cementing pump speed sensor and vibration sensor during the cementing operation.

[0083] An input matrix is ​​constructed based on the weighted historical input features of the cementing pump, wherein the input matrix F is obtained by weighting the historical input features of its n samples. w =[f w (1); f w (2); ...; f w [(n)]; n is the number of samples. The historical operating conditions y(t) are processed by one-hot encoding to obtain the output matrix Y.

[0084] For the input matrix F w Perform a primary mapping to obtain primary mapping nodes. Specifically, this is done based on a preset nonlinear function φ for the primary mapping and a preset first set of weight parameters. It can be generated randomly. For the input matrix F w Perform a primary mapping to obtain the primary mapping node Z. p There are multiple primary mapping nodes, such as P, where 1≤p≤P, as shown in formula (19):

[0085]

[0086] Merge multiple primary mapping nodes, Z all =[Z1,Z2,…Z p …,Z P Based on the preset nonlinear function ξ of the enhanced mapping and the preset second set of weight parameters... It can be generated randomly. For Z all Perform augmented mapping to obtain augmented mapping node H. q Among them, there are multiple enhanced mapping nodes, such as Q nodes, where 1 ≤ q ≤ Q, as shown in formula (20):

[0087]

[0088] The primary mapping node and the enhanced mapping node are merged to obtain the first matrix [Z]. all H1H2…H Q] Calculate the generalized inverse of the first matrix, and calculate the output weight parameter B based on the generalized inverse and the output matrix Y. As shown in formula (21):

[0089] B = [Z] all H1 H2 … H Q ] * Y (21)

[0090] Here, * indicates finding the generalized inverse of the matrix.

[0091] The preset weighted width WBL (Web-Based Learning) model is constructed using the above formulas (19)-(21). Its model expression is as follows:

[0092]

[0093] Here, g() represents the global model function.

[0094] The structure diagram of the pre-defined weighted width learning model is as follows: Figure 2 As shown, based on historical data from n samples of speed and vibration sensors, such as speed data and vibration data, feature vectors of historical input features are obtained through steps S101-S102. A feature importance factor is constructed based on the relative inter-class dispersion distance, and the feature vectors are weighted to obtain a weighted feature vector. This weighted feature vector is then obtained based on a nonlinear function of a preset primary mapping and random weight parameters. That is, a first set of weight parameters is preset to obtain primary mapping nodes. The merged primary mapping nodes are then processed according to the nonlinear function of the enhancement mapping and random weight parameters. The process involves using a pre-defined second set of weight parameters for enhanced mapping, resulting in enhanced mapping nodes. Based on the primary mapping nodes, enhanced mapping nodes, and the output matrix Y (the result of the operating condition classification), optimized output weight parameters B are obtained. This completes the training of the pre-defined weighted width learning model, which serves as the fault classifier. The fault classifier can be trained offline using historical data. Furthermore, it can be further optimized by collecting historical data, etc., but this is not limited here.

[0095] After obtaining the fault classifier, it can be conveniently used to perform real-time condition prediction for cementing operations, diagnose faults in the operating conditions, and determine the fault results. Corresponding to real-time prediction, real-time data from the cementing pump speed sensor and the vibration sensor can be obtained during the cementing operation. new ),v1(t new ),v2(t new ),…,v m (t new )},t newThis indicates the sequence number of the real-time data. For any vibration sensor's real-time data, referring to step S102, multi-dimensional real-time features of the vibration sensor are extracted based on the time and frequency domains. The multi-dimensional real-time features of the vibration sensor are combined with the real-time features of the speed sensor's real-time data to obtain the real-time input feature f(t) of the cementing pump. new A diagonal matrix Imp is constructed based on feature importance factors. The real-time input features of the cementing pump are weighted and then input into the fault classifier to predict the fault diagnosis results.

[0096] This embodiment achieves fault prediction and diagnosis through two phases: offline and online. In the offline training phase, historical data of cementing pumps under various operating conditions is collected. For the historical data from vibration sensors, time-domain and frequency-domain features are extracted. The importance of each feature is analyzed using relative dispersion analysis, and each feature is weighted. Offline training yields a feature-weighted width learning model to establish a fault classifier. In the online testing phase, real-time data of the cementing pump is collected, and corresponding time-domain and frequency-domain features are extracted. Features are weighted using feature importance factors and then input into the fault classifier. The classification result for the operating conditions is obtained based on the output.

[0097] Using a cementing pump unit from a certain company as the test object, four different operating conditions were tested during the unit's operation: normal operating condition, wear on the upper valve of cylinder #1, wear on the upper valve of cylinder #3, and wear on the lower valve of cylinder #3, labeled as operating condition 1, operating condition 2, operating condition 3, and operating condition 4, respectively. A total of 3582 samples were collected from all operating conditions and divided into two groups: one group containing 60% of the samples served as the model training dataset, and the other group containing 40% of the samples served as the model testing dataset. Fault classifiers were established using both a traditional learning model (such as a width learning model) and the feature-weighted width learning model of this embodiment, respectively. Figure 3 The diagram shown is a confusion graph of the classification results from a traditional learning model. Figure 4 The diagram shows the classification confusion of the feature-weighted width learning model. Figure 3 , Figure 4 The horizontal axis represents the true class of the working condition, and the vertical axis represents the output class of the model as a classifier. Figure 3 In the example, the traditional learning model correctly identified 192 samples for scenario 2. Figure 4 The feature-weighted width learning model can correctly identify 216 samples for working condition 1, which is a significant improvement. Figure 4 The identification results for working condition 4 in the middle are also better than Figure 3 The accuracy of the traditional model and the model in this embodiment under different working conditions is significantly improved, resulting in a better improvement effect. Table 1 below shows a comparison of the recognition accuracy of the traditional model and the model in this embodiment under different working conditions:

[0098] Table 1

[0099]

[0100] In Table 1, the overall recognition accuracy of the traditional model is 89.92%, while the recognition accuracy of the feature-weighted width learning model in this embodiment can reach 92.02%, an improvement of 2.1%, indicating that this embodiment can effectively improve the fault classification accuracy of cementing pumps.

[0101] According to the cementing pump fault diagnosis method provided in the embodiments of the present invention, starting from time frequency and domain frequency, the multi-dimensional data features of the vibration sensor of the cementing pump are fully extracted. When the feature importance factor is determined and weighted according to the multi-dimensional features, important features closely related to the fault condition can be assigned a larger weight, while secondary features with low correlation can be given a smaller weight. The differences in information contained in different features are fully utilized to improve the modeling effect of the preset weighted width learning model, thereby improving the accuracy of predicting the fault classification of the cementing pump.

[0102] Figure 5 A schematic diagram of the structure of the cementing pump fault diagnosis device provided in an embodiment of the present invention is shown. Figure 5 As shown, the device includes:

[0103] The historical data acquisition module 510 is suitable for acquiring historical data from the cementing pump speed sensor, historical data from the vibration sensor, and historical operating conditions during cementing operations; historical operating conditions include normal operating conditions and fault operating conditions.

[0104] The multi-dimensional extraction module 520 is suitable for extracting multi-dimensional historical features of any vibration sensor based on time and frequency domains from historical data of the vibration sensor; and combining the multi-dimensional historical features of the vibration sensor with historical data of the speed sensor to obtain the historical input features of the cementing pump.

[0105] The weighted processing module 530 is suitable for calculating the relative inter-class dispersion distance of each historical input feature in the historical input features of the cementing pump, and constructing a feature importance factor based on the relative inter-class dispersion distance, so as to perform weighted processing on the historical input features of the cementing pump according to the feature importance factor;

[0106] The model training and prediction module 540 is suitable for inputting the weighted historical input features and historical operating conditions of the cementing pump into a preset weighted width learning model, training the preset weighted width learning model, and obtaining the trained preset weighted width learning model as a fault classifier, so as to predict the fault diagnosis result based on the real-time data of the cementing pump speed sensor and vibration sensor during the cementing operation.

[0107] Optionally, the multi-dimensional historical characteristics of the vibration sensor include:

[0108] Maximum value, minimum value, mean, standard deviation, peak-to-peak value, effective value, skewness, kurtosis, centroid frequency, root mean square frequency, mean frequency, frequency standard deviation.

[0109] Optionally, the weighted processing module 530 is further adapted to:

[0110] The relative dispersion analysis method is used to calculate the first and second class dispersions of the cementing pump under each historical operating condition based on the mean of the historical input characteristics of the cementing pump under each historical operating condition and the covariance matrix under the entire historical operating condition. The first class dispersion is the overall inter-class dispersion of the cementing pump; the second class dispersion is the inter-class dispersion of the cementing pump after removing any feature.

[0111] The relative inter-class dispersion distance of each feature in the historical input features of the cementing pump is calculated based on the inter-class dispersion of the first and second classes.

[0112] Based on the comparison results of the inter-class dispersion and the inter-class dispersion, the feature importance factor is determined; wherein, the feature importance factor is determined by the relative inter-class dispersion distance or a preset factor according to the comparison results;

[0113] A diagonal matrix is ​​constructed based on the feature importance factors, and the historical input features of the cementing pump are weighted using the diagonal matrix.

[0114] Optionally, the model training prediction module 540 is further adapted to:

[0115] An input matrix is ​​constructed based on the weighted historical input characteristics of the cementing pump, and the historical operating conditions are processed by one-hot encoding to obtain the output matrix.

[0116] Perform a primary mapping on the input matrix to obtain primary mapping nodes, and then perform an enhanced mapping on the primary mapping nodes to obtain enhanced mapping nodes;

[0117] The primary mapping node and the enhanced mapping node are merged, and the output weight parameters are calculated based on the output matrix to complete the training of the preset weighted width learning model. The trained preset weighted width learning model is then used as a fault classifier.

[0118] Optionally, the model training prediction module 540 is further adapted to:

[0119] Based on the preset nonlinear function of the primary mapping and the preset first weight parameter set, the input matrix is ​​subjected to primary mapping to obtain primary mapping nodes; there are multiple primary mapping nodes.

[0120] Multiple primary mapping nodes are merged and then enhanced mapping is performed according to a preset nonlinear function of enhanced mapping and a preset second set of weight parameters to obtain enhanced mapping nodes; there are multiple enhanced mapping nodes.

[0121] The primary mapping node and the enhanced mapping node are merged, and the output weight parameters are calculated based on the output matrix, which further include:

[0122] The primary mapping node and the enhanced mapping node are merged to obtain the first matrix. The generalized inverse processing result of the first matrix is ​​calculated, and the output weight parameters are calculated based on the generalized inverse processing result and the output matrix.

[0123] Optionally, the model training prediction module 540 is further adapted to:

[0124] Acquire real-time data from the cementing pump speed sensor and the vibration sensor during cementing operations;

[0125] For any vibration sensor's real-time data, multi-dimensional real-time features of the vibration sensor are extracted based on the time and frequency domains; the multi-dimensional real-time features of the vibration sensor are combined with the real-time features of the speed sensor's real-time data to obtain the real-time input features of the cementing pump.

[0126] The real-time input features of the cementing pump are weighted according to the feature importance factor. The weighted real-time input features of the cementing pump are then input into the fault classifier to predict the fault diagnosis result.

[0127] The descriptions of the above modules refer to the corresponding descriptions in the method embodiments, and will not be repeated here.

[0128] This invention also provides a non-volatile computer storage medium storing at least one executable instruction that can perform the operation corresponding to the cementing pump fault diagnosis method in any of the above method embodiments.

[0129] This application provides a computer program product, which includes at least one executable instruction or computer program that enables a processor to perform the operation corresponding to the cementing pump fault diagnosis method in any of the above method embodiments.

[0130] Figure 6 The diagram illustrates the structure of a computing device according to an embodiment of the present invention. The specific embodiments of the present invention do not limit the specific implementation of the computing device.

[0131] like Figure 6 As shown, the computing device may include: a processor 602, a communication interface 604, a memory 606, and a communication bus 608.

[0132] in:

[0133] The processor 602, communication interface 604, and memory 606 communicate with each other via communication bus 608.

[0134] Communication interface 604 is used to communicate with other network elements such as clients or other servers.

[0135] The processor 602 is used to execute program 610, which can specifically execute the relevant steps in the above-described embodiment of the cementing pump fault diagnosis method.

[0136] Specifically, program 610 may include program code that includes computer operation instructions.

[0137] Processor 602 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention. The computing device includes one or more processors, which may be processors of the same type, such as one or more CPUs; or processors of different types, such as one or more CPUs and one or more ASICs.

[0138] Memory 606 is used to store program 610. Memory 606 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0139] Specifically, program 610 can be used to cause processor 602 to execute the cementing pump fault diagnosis method in any of the above method embodiments. The specific implementation of each step in program 610 can be found in the corresponding descriptions of the steps and units in the above cementing pump fault diagnosis embodiments, and will not be repeated here. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the devices and modules described above can be referred to the corresponding process descriptions in the foregoing method embodiments, and will not be repeated here.

[0140] The algorithms or displays provided herein are not inherently related to any particular computer, virtual system, or other device. Various general-purpose systems can also be used in conjunction with the teachings herein. The required structure for constructing such systems is apparent from the above description. Furthermore, the embodiments of the present invention are not directed to any particular programming language. It should be understood that the embodiments of the present invention described herein can be implemented using various programming languages, and the above description of specific languages ​​is for the purpose of disclosing preferred embodiments of the present invention.

[0141] Numerous specific details are set forth in the specification provided herein. However, it will be understood that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures, and techniques have not been shown in detail so as not to obscure the understanding of this specification.

[0142] Similarly, it should be understood that, in order to streamline the embodiments of the invention and aid in understanding one or more of the various inventive aspects, features of the embodiments of the invention are sometimes grouped together in a single embodiment, figure, or description thereof in the above description of exemplary embodiments of the invention. However, this disclosure should not be construed as reflecting an intention that the claimed embodiments of the invention require more features than are expressly recited in each claim. Rather, as reflected in the following claims, inventive aspects lie in fewer than all features of a single foregoing disclosed embodiment. Therefore, the claims following the detailed description are hereby expressly incorporated into that detailed description, wherein each claim itself is a separate embodiment of the invention.

[0143] Those skilled in the art will understand that modules in the device of the embodiments can be adaptively changed and placed in one or more devices different from that embodiment. Modules, units, or components in the embodiments can be combined into a single module, unit, or component, and further, they can be divided into multiple sub-modules, sub-units, or sub-components. Except where at least some of such features and / or processes or units are mutually exclusive, any combination can be used to combine all features disclosed in this specification (including the accompanying claims, abstract, and drawings) and all processes or units of any method or device so disclosed. Unless expressly stated otherwise, each feature disclosed in this specification (including the accompanying claims, abstract, and drawings) may be replaced by an alternative feature that serves the same, equivalent, or similar purpose.

[0144] Furthermore, those skilled in the art will understand that although some embodiments herein include certain features included in other embodiments but not others, combinations of features from different embodiments are intended to be within the scope of the invention and form different embodiments. For example, in the following claims, any of the claimed embodiments can be used in any combination.

[0145] The various component embodiments of the present invention can be implemented in hardware, or as software modules running on one or more processors, or a combination thereof. Those skilled in the art will understand that microprocessors or digital signal processors (DSPs) can be used in practice to implement some or all of the functions of some or all of the components according to the embodiments of the present invention. The embodiments of the present invention can also be implemented as device or apparatus programs (e.g., computer programs and computer program products) for performing part or all of the methods described herein. Such programs implementing the embodiments of the present invention can be stored on a computer-readable medium, or can be in the form of one or more signals. Such signals can be downloaded from an Internet website, provided on a carrier signal, or provided in any other form.

[0146] It should be noted that the above embodiments are illustrative of the present invention and not restrictive of the invention, and that those skilled in the art can devise alternative embodiments without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses should not be construed as limiting the claims. The word "comprising" does not exclude the presence of elements or steps not listed in the claims. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. Embodiments of the present invention can be implemented by means of hardware comprising several different elements and by means of a suitably programmed computer. In the unit claims enumerating several means, several of these means may be embodied by the same item of hardware. The use of the words first, second, and third, etc., does not indicate any order. These words can be interpreted as names. The steps in the above embodiments, unless otherwise specified, should not be construed as limiting the order of execution.

Claims

1. A method for diagnosing faults in cementing pumps, characterized in that, The methods include: Acquire historical data from cementing pump speed sensor, vibration sensor, and historical operating conditions during cementing operations; the historical operating conditions include normal operating conditions and fault operating conditions. For any vibration sensor's historical data, multi-dimensional historical features of the vibration sensor are extracted based on the time and frequency domains; the multi-dimensional historical features of the vibration sensor are combined with the historical data of the rotation speed sensor to obtain the historical input features of the cementing pump. Calculate the relative inter-class dispersion distance of each historical input feature in the historical input features of the cementing pump, and construct a feature importance factor based on the relative inter-class dispersion distance, so as to weight the historical input features of the cementing pump according to the feature importance factor; The weighted historical input features and historical operating conditions of the cementing pump are input into a preset weighted width learning model. The preset weighted width learning model is trained to obtain a fault classifier, which is used to predict fault diagnosis results based on real-time data from the cementing pump speed sensor and vibration sensor during cementing operations.

2. The method according to claim 1, characterized in that, The multi-dimensional historical characteristics of the vibration sensor include: Maximum value, minimum value, mean, standard deviation, peak-to-peak value, effective value, skewness, kurtosis, centroid frequency, root mean square frequency, mean frequency, frequency standard deviation.

3. The method according to claim 1, characterized in that, The step of calculating the relative inter-class dispersion distance of each historical input feature in the historical input features of the cementing pump, and constructing a feature importance factor based on the relative inter-class dispersion distance, and then weighting the historical input features of the cementing pump according to the feature importance factor, further includes: The relative dispersion analysis method is used to calculate the first and second inter-class dispersion of the cementing pump under each historical operating condition based on the mean of the historical input characteristics of the cementing pump under each historical operating condition and the covariance matrix under all historical operating conditions. The first inter-class dispersion is the overall inter-class dispersion of the cementing pump, and the second inter-class dispersion is the inter-class dispersion of the cementing pump after removing any feature. The relative inter-class dispersion distance of each feature in the historical input features of the cementing pump is calculated based on the first inter-class dispersion and the second inter-class dispersion. Based on the comparison results of the first inter-class dispersion and the second inter-class dispersion, a feature importance factor is determined; wherein, the feature importance factor is determined by the relative inter-class dispersion distance or a preset factor based on the comparison results; A diagonal matrix is ​​constructed based on the aforementioned feature importance factors, and the historical input features of the cementing pump are weighted using the diagonal matrix.

4. The method according to claim 1, characterized in that, The step of inputting the weighted historical input features and historical operating conditions of the cementing pump into a preset weighted width learning model, and training the preset weighted width learning model, further includes: An input matrix is ​​constructed based on the weighted historical input characteristics of the cementing pump, and the historical operating conditions are processed by one-hot encoding to obtain the output matrix. The input matrix is ​​subjected to a primary mapping to obtain primary mapping nodes, and the primary mapping nodes are subjected to an enhanced mapping to obtain enhanced mapping nodes. The primary mapping node and the enhanced mapping node are merged, and the output weight parameters are calculated according to the output matrix to complete the training of the preset weighted width learning model, and the trained preset weighted width learning model is used as a fault classifier.

5. The method according to claim 4, characterized in that, The step of performing a primary mapping on the input matrix to obtain primary mapping nodes, and then performing an enhanced mapping on the primary mapping nodes to obtain enhanced mapping nodes, further includes: Based on the nonlinear function of the preset primary mapping and the preset first weight parameter set, the input matrix is ​​subjected to primary mapping to obtain primary mapping nodes; there are multiple primary mapping nodes. The multiple primary mapping nodes are merged and then enhanced mapping is performed according to a preset nonlinear function of enhanced mapping and a preset second weight parameter set to obtain enhanced mapping nodes; there are multiple enhanced mapping nodes. The step of merging the primary mapping node and the enhanced mapping node, and calculating the output weight parameters based on the output matrix, further includes: The primary mapping node and the enhanced mapping node are merged to obtain a first matrix. The generalized inverse processing result of the first matrix is ​​calculated, and the output weight parameter is calculated based on the generalized inverse processing result and the output matrix.

6. The method according to claim 1, characterized in that, The method of predicting fault diagnosis results based on real-time data from cementing pump speed and vibration sensors during cementing operations further includes: Acquire real-time data from the cementing pump speed sensor and the vibration sensor during cementing operations; For any vibration sensor's real-time data, multi-dimensional real-time features of the vibration sensor are extracted based on the time and frequency domains; the multi-dimensional real-time features of the vibration sensor are combined with the real-time data of the rotation speed sensor to obtain the real-time input features of the cementing pump. The real-time input features of the cementing pump are weighted according to the feature importance factor, and the weighted real-time input features of the cementing pump are input into the fault classifier to predict the fault diagnosis result.

7. A cementing pump fault diagnosis device, characterized in that, The device includes: The historical data acquisition module is suitable for acquiring historical data from the cementing pump speed sensor, historical data from the vibration sensor, and historical operating conditions during cementing operations; the historical operating conditions include normal operating conditions and fault operating conditions. The multi-dimensional extraction module is suitable for extracting multi-dimensional historical features of any vibration sensor based on time and frequency domains from historical data of the vibration sensor; and combining the multi-dimensional historical features of the vibration sensor with the historical data of the rotation speed sensor to obtain the historical input features of the cementing pump. The weighted processing module is adapted to calculate the relative inter-class dispersion distance of each historical input feature in the historical input features of the cementing pump, and to construct a feature importance factor based on the relative inter-class dispersion distance, so as to perform weighted processing on the historical input features of the cementing pump according to the feature importance factor; The model training and prediction module is adapted to input the weighted historical input features and historical operating conditions of the cementing pump into a preset weighted width learning model, train the preset weighted width learning model, and obtain the trained preset weighted width learning model as a fault classifier, so as to predict the fault diagnosis result based on the real-time data of the cementing pump speed sensor and vibration sensor during the cementing operation.

8. A computing device, characterized in that, include: The processor, memory, communication interface, and communication bus are provided, wherein the processor, memory, and communication interface communicate with each other via the communication bus. The memory is used to store at least one executable instruction, which causes the processor to perform the operation corresponding to the cementing pump fault diagnosis method as described in any one of claims 1-6.

9. A computer storage medium, characterized in that, The storage medium stores at least one executable instruction, which causes the processor to perform the operation corresponding to the cementing pump fault diagnosis method as described in any one of claims 1-6.

10. A computer program product, characterized in that, It includes at least one executable instruction that causes the processor to perform the operation corresponding to the cementing pump fault diagnosis method as described in any one of claims 1-6.

Images