A method for fault diagnosis for drilling drilling process monitoring and related devices

Abstract

This invention belongs to the field of geological drilling engineering and discloses a fault diagnosis method and related device for monitoring the drilling process. The fault diagnosis method includes the following steps: acquiring drilling process data to be diagnosed; classifying faults using a trained fault diagnosis model based on the drilling process data to obtain fault diagnosis classification results; during the training process, the fault diagnosis model employs a hybrid resampling technique to effectively balance the sample distribution and combines simplified attention entropy as a quantitative screening index to identify and select samples with significant differences in feature complexity from the majority class samples, thereby improving the quality of the generated samples. The technical solution disclosed in this invention can solve the technical problems of insufficient model interpretability, poor diagnostic performance, and low operating efficiency in existing fault diagnosis methods.

Description

Technical Field

[0001] This invention belongs to the field of geological drilling engineering, and specifically relates to a fault diagnosis method and related device for monitoring the drilling process. Background Technology

[0002] As a core method in geological exploration, deep and complex geological exploration plays a crucial role in resource exploration, national resource security, and economic and social development. The high temperature, high ground stress, high permeability pressure environment, and complex geological structures of deep rock masses lead to frequent drilling failures. Furthermore, wear and natural aging of drilling equipment such as drilling rigs, drill rods, and drill bits can also induce failures. Once a failure occurs, it can range from interrupting the drilling process and slowing down the project progress to, in severe cases, causing the borehole to be scrapped. Therefore, researching methods for diagnosing drilling failures is of great significance for preventing drilling accidents and ensuring the safety of life and property.

[0003] Most existing fault diagnosis schemes in the drilling process use pre-trained fault diagnosis models for fault diagnosis and monitoring. However, in actual drilling operations, the drilling platform usually operates stably under normal conditions, and the failure rate is low. This results in a much smaller sample size under fault conditions than under normal conditions, which leads to class imbalance and seriously affects the performance of the fault diagnosis model. To address these issues, academia has conducted extensive research to improve the performance of fault diagnosis models under imbalanced conditions. The proposed methods can be broadly categorized into two types. The first type focuses on developing advanced learning algorithms to directly enhance the model's diagnostic capabilities on imbalanced data. However, these methods require high-quality data, and model bias can easily arise if the training data fails to accurately reflect the distribution of real-world data. The second type uses resampling techniques or generative adversarial networks (GANs) for data augmentation to balance class distributions and improve the model's generalization performance. In this context, GAN-based methods have shown excellent results in rotating machinery fault diagnosis; however, they face significant challenges such as model collapse and training instability, severely hindering further improvements in classification accuracy. In contrast, resampling-based methods are computationally simpler and can effectively rebalance the distribution of samples from different classes through resampling, undersampling, or a combination of both. However, existing resampling-based methods typically employ a single resampling strategy, making it difficult to generate high-quality sample data.

[0004] Furthermore, after sample balancing, effective feature extraction and fault classification are still required. Due to the complex characteristics of nonlinearity and non-stationarity of the collected drilling signals, traditional time-frequency domain-based analysis methods often suffer from insufficient diagnostic accuracy and poor robustness in intelligent fault diagnosis applications. In recent years, deep learning and entropy-based methods have received widespread attention in the field of fault diagnosis, as they can achieve high diagnostic accuracy. However, they still have problems to be solved, such as insufficient model interpretability, poor diagnostic performance, and low operating efficiency (interpretably, such as long training cycles). Summary of the Invention

[0005] The purpose of this invention is to provide a fault diagnosis method and related device for monitoring the drilling process, so as to solve the technical problems of insufficient model interpretability, poor diagnostic performance and low operating efficiency in existing fault diagnosis methods.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] The first aspect of this invention discloses a fault diagnosis method for monitoring the drilling process, comprising the following steps:

[0008] Acquire drilling process data to be diagnosed;

[0009] Based on the drilling process data to be diagnosed, the trained fault diagnosis model is used to classify the fault diagnosis and obtain the fault diagnosis classification results.

[0010] The training steps of the fault diagnosis model include:

[0011] Obtain the original training sample set and the original test sample set;

[0012] The SMOTE and ADASYN algorithms are used to perform mixed resampling on the minority class fault samples in the original training sample set to generate a synthetic sample set; the synthetic sample set is then merged with the original training sample set to form a training sample set with balanced class distribution.

[0013] Based on the original test sample set and the training sample set with balanced class distribution, fault features are extracted using simplified attention entropy or enhanced simplified attention entropy to obtain the feature vector test sample set and the feature vector training sample set.

[0014] Based on the feature vector test sample set and the feature vector training sample set, the selected ensemble learning classifier is trained to obtain a trained fault diagnosis model.

[0015] A further improvement of the technical solution of the present invention is that the data of the drilling process to be diagnosed is of the same type as the data of the original training sample set and the original test sample set.

[0016] A further improvement of the technical solution of the present invention is that the drilling process data to be diagnosed includes one or more of the following: vibration signal, torque signal and pressure signal of the selected component.

[0017] A further improvement to the technical solution of this invention lies in employing the SMOTE and ADASYN algorithms to perform mixed resampling processing on the minority class fault samples in the original training sample set, generating a synthetic sample set, including:

[0018] For each minority class sample set X = {x1, ..., x...} n The SMOTE and ADASYN algorithms were used for resampling to generate synthetic samples that met the preset quantity requirements, thus balancing the number of fault samples of different categories. n This represents the nth minority class sample;

[0019] The generation of synthetic samples is achieved through the following interpolation formula:

[0020] y new =x i +λ(x j -x i );

[0021] In the formula, y new Indicates a synthetic sample; x i Let x represent the current minority class sample. j This represents the selected nearest neighbor sample, where λ is a random number in the interval [0,1].

[0022] In addition, the difference between the mean simple attention entropy of the minority class samples generated by the SMOTE and ADASYN algorithms and the original majority class samples is calculated separately. The expressions for calculating the difference between the two classes are as follows:

[0023]

[0024] In the formula, E SMOTE and E ADASYN E represents the mean of the simplified attention entropy of minority class samples generated by the SMOTE and ADASYN algorithms, respectively; Majority The mean of the simplified attention entropy of the original majority class samples;

[0025] Based on the difference calculation results, a synthetic sample set is formed according to the principle of entropy difference comparison; among them, the greater the entropy difference, the more significant the difference between samples, and synthetic samples with larger entropy differences are selected first.

[0026] A further improvement of the technical solution of the present invention lies in that, when using the simple attention entropy to extract fault features based on the original test sample set and the training sample set with balanced class distribution to obtain the feature vector test sample set and the feature vector training sample set, the specific calculation steps of the simple attention entropy are as follows:

[0027] According to the following rules, for the time series X = {x1, x2,..., x N}, define the local maximum and local minimum. x N represents the Nth point of the time series X; where, for any points x i and x j in the time series X, 1 < i < N, 1 < j < N, there are: local maximum LMax: if x i-1 < x i and x i > x i+1 , then the point x i is regarded as the local maximum; local minimum LMin: if x j-1 > x j and x j < x j+1 , then the point x j is regarded as the local minimum;

[0028] Given interval points according to two different strategies of {LMin - LMax} and {LMax - LMin}, and solve the Shannon entropy of the adjacent interval point intervals. The formula is as follows:

[0029]

[0030] In the formula, H represents the Shannon entropy value; P(i) represents the probability of the interval point i appearing, and b represents the number of types of interval points;

[0031] Define the mean value of the Shannon entropy obtained by the two different strategies as the simple attention entropy. The calculation formula is as follows:

[0032]

[0033] In the formula, AtEn(X) represents the simple attention entropy value calculated for the time series X; H(j) represents the entropy value obtained from the jth strategy.

[0034] A further improvement of the technical solution of the present invention lies in that, when using the enhanced simple attention entropy to extract fault features based on the original test sample set and the training sample set with balanced class distribution to obtain the feature vector test sample set and the feature vector training sample set, the specific calculation steps of the enhanced simple attention entropy are as follows:

[0035] Given the original time series X = {x1, x2,..., x NThe sequence is divided into a series of continuous coarse-grained sequence segments; where, for a scale factor τ, starting from each point in the interval [1,τ], the k-th coarse-grained sequence... From elements Composition; based on the relationship between sequence length N and τ, two cases are handled:

[0036] When N is an integer multiple of τ, It can be represented as:

[0037]

[0038] When N is not an integer multiple of τ, there are residual sample points. It can be represented as:

[0039]

[0040] In the formula, residual sample points are processed separately to ensure data integrity;

[0041] For each coarse-grained sequence Calculate the simplified attention entropy value for each coarse-grained sequence, then average the τ entropy values ​​to obtain the final enhanced simplified attention entropy:

[0042]

[0043] In the formula, ESAtEn(X,τ) represents the enhanced simplified attention entropy value obtained by calculating the time series X.

[0044] A further improvement of the technical solution of the present invention is that, in the step of training the selected ensemble learning classifier based on the feature vector test sample set and the feature vector training sample set to obtain the trained fault diagnosis model, the selected ensemble learning classifier is an ELM-AdaBoost classifier or a BP neural network-AdaBoost classifier.

[0045] The second aspect of this invention discloses a fault diagnosis system for monitoring the drilling process, comprising:

[0046] The data acquisition module is used to acquire drilling process data to be diagnosed.

[0047] The diagnostic classification module is used to perform fault diagnosis classification based on the drilling process data to be diagnosed, using a trained fault diagnosis model, and to obtain the fault diagnosis classification result.

[0048] The training steps of the fault diagnosis model include:

[0049] Obtain the original training sample set and the original test sample set;

[0050] The SMOTE and ADASYN algorithms are used to perform mixed resampling on the minority class fault samples in the original training sample set to generate a synthetic sample set; the synthetic sample set is then merged with the original training sample set to form a training sample set with balanced class distribution.

[0051] Based on the original test sample set and the training sample set with balanced class distribution, fault features are extracted using simplified attention entropy or enhanced simplified attention entropy to obtain the feature vector test sample set and the feature vector training sample set.

[0052] Based on the feature vector test sample set and the feature vector training sample set, the selected ensemble learning classifier is trained to obtain a trained fault diagnosis model.

[0053] The third aspect of the present invention discloses an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement a fault diagnosis method for monitoring the drilling process as described in any one of the first aspects of the present invention.

[0054] The fourth aspect of this invention discloses a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements a fault diagnosis method for monitoring the drilling process as described in any one of the first aspects of this invention.

[0055] Compared with the prior art, the present invention has the following beneficial effects:

[0056] This invention discloses a fault diagnosis method for monitoring the drilling process. A trained fault diagnosis model is used to diagnose and monitor the drilling process. During training, the model employs a hybrid resampling technique to effectively balance the sample distribution and combines simplified attention entropy as a quantitative screening index to identify and select samples with significant differences in feature complexity from the majority class, thereby improving the quality of the generated samples. Furthermore, an enhanced simplified attention entropy method is used, fully leveraging the advantages of low entropy computational complexity and strong feature extraction capability. This enhances model interpretability while achieving comprehensive characterization and feature extraction of fault signal complexity at multiple scales, effectively improving the accuracy of fault diagnosis. In summary, this invention effectively solves the problem of imbalanced fault samples through resampling technology and fully utilizes the advantages of enhanced simplified attention entropy, thus improving the accuracy of fault diagnosis.

[0057] Specifically, addressing the problem that "existing resampling-based methods typically employ a single resampling strategy, making it difficult to generate high-quality sample data," this invention proposes a hybrid resampling technology solution. This technology synergistically integrates the advantages of synthetic minority oversampling technique (SMOTE) and adaptive synthetic sampling (ADASYN). SMOTE increases the number of minority class samples through linear interpolation in the feature space, while ADASYN adaptively focuses on regions that are difficult to learn based on the sample distribution density. The synergistic integration of the two methods ensures a balance in the number of samples while improving the generation quality of samples near the decision boundary.

[0058] In a further preferred embodiment of the present invention, to improve the quality of generated samples, simplified attention entropy is introduced as a quantitative indicator. Utilizing the high sensitivity of entropy to signal complexity, samples with significant differences in feature complexity are selected from the majority class. Simultaneously, the Edit Nearest Neighbor (ENN) technique is combined to accurately remove noisy and redundant samples based on the nearest neighbor distribution in the feature space, eliminating their interference with the classification decision boundary. This method significantly improves the purity and reliability of the generated sample set while enhancing the differences between samples, thus building a high-quality data foundation for subsequent resampling. Furthermore, addressing the problems of "poor diagnostic performance, low operating efficiency, and insufficient interpretability in existing fault diagnosis methods," this invention proposes a fault diagnosis method based on enhanced simplified attention entropy. This method fully leverages the advantages of low complexity in entropy calculation and strong feature extraction capabilities. By quantifying the differences in entropy values ​​of signals under different fault states, it reflects the characteristic changes in signal complexity, thereby achieving effective differentiation of different fault types. Specifically, enhanced simple attention entropy, as an interpretable physical quantity, can clearly characterize the uncertainty and randomness of signals: when a fault occurs, the signal complexity often shifts, causing the entropy value to change accordingly. This method monitors the explicit differences in entropy values ​​to achieve precise identification of various faults, and has strong interpretability and computational efficiency. Attached Figure Description

[0059] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0060] Figure 1 This is a flowchart illustrating a fault diagnosis method for monitoring the drilling process, as described in an embodiment of the present invention.

[0061] Figure 2 This is a flowchart illustrating a drilling process imbalance fault diagnosis method based on enhanced simplified attention entropy in a specific embodiment of the present invention.

[0062] Figure 3 This is a schematic diagram of the time-domain waveforms of five types of bearing signals in a specific embodiment of the present invention;

[0063] Figure 4 This is a schematic diagram of the visualization results after T-SNE dimensionality reduction in a specific embodiment of the present invention;

[0064] Figure 5 This is a schematic diagram of a fault diagnosis system for monitoring the drilling process, as described in an embodiment of the present invention. Detailed Implementation

[0065] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention; obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0066] Based on the technical solutions disclosed in the embodiments of this invention, all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of this invention. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products, or devices.

[0067] Please see Figure 1 The present invention provides a fault diagnosis method for monitoring the drilling process, which specifically includes the following steps:

[0068] Step 1: Acquire drilling process data to be diagnosed; specifically, the drilling process data to be diagnosed may include vibration signals, torque signals and pressure signals of components such as bearings and gears, which can be acquired in real time through various sensors.

[0069] Step 2: Based on the drilling process data to be diagnosed obtained in Step 1, use the trained fault diagnosis model to perform fault diagnosis and obtain the fault diagnosis classification results.

[0070] The training steps of the fault diagnosis model include:

[0071] Obtain the original training sample set and the original test sample set; in a specific exemplary technical solution, the original drilling signal can be divided by the sliding window technique based on the drilling process data collected in real time by the drilling rig sensors to construct the training sample set and the test sample set; in a further specific exemplary technical solution, the drilling process data collected in real time can be one or more of the vibration signal, torque signal and pressure signal of the pre-selected component.

[0072] The SMOTE and ADASYN algorithms are used to perform mixed resampling on the minority class fault samples in the original training sample set to generate a synthetic sample set; the synthetic sample set is then merged with the original training sample set to form a training sample set with balanced class distribution.

[0073] Based on the original test sample set and the training sample set with balanced class distribution, fault features are extracted using simplified attention entropy or enhanced simplified attention entropy to obtain the feature vector test sample set and the feature vector training sample set.

[0074] Based on the feature vector test sample set and the feature vector training sample set, the selected ensemble learning classifier is trained to obtain a trained fault diagnosis model. Specifically, the obtained feature vector training sample set can be input into the ELM-AdaBoost classifier (i.e., Extreme Learning Machine-AdaBoost classifier) ​​or the BP neural network-AdaBoost classifier for model training. Then, the feature vector test sample set is introduced into the trained classifier model to finally achieve accurate identification and classification of different fault types.

[0075] This invention discloses a computationally efficient and easily implemented lightweight diagnostic scheme that can significantly improve the accuracy and efficiency of fault diagnosis and classification, and has important engineering application value for ensuring drilling operation safety.

[0076] In a specific embodiment of the technical solution of the present invention, the SMOTE and ADASYN algorithms are used to perform mixed resampling processing on the minority class fault samples in the original training sample set to generate a synthetic sample set. The specific process is as follows:

[0077] For each minority class sample set X = {x1, ..., x...} n}, SMOTE and ADASYN algorithms are used for resampling to generate a sufficient number of synthetic samples (enough to meet the preset requirements) to balance the number of fault samples of different categories; where x n This represents the nth minority class sample;

[0078] The generation of synthetic samples is achieved through the following interpolation formula:

[0079] y new =xi +λ(x j -x i );

[0080] In the formula, y new Indicates a synthetic sample; x i Let x represent the current minority class sample. j This represents the selected nearest neighbor sample, where λ is a random number in the interval [0,1].

[0081] In addition, the difference between the mean simple attention entropy of the minority class samples generated by the SMOTE and ADASYN algorithms and the original majority class samples is calculated separately. The specific formula for calculating the difference is as follows:

[0082]

[0083] In the formula, E SMOTE and E ADASYN E represents the mean of the simplified attention entropy of minority class samples generated by the SMOTE and ADASYN algorithms, respectively; Majority The mean of the simplified attention entropy of the original majority class samples;

[0084] Based on the difference calculation results, a synthetic sample set is formed according to the principle of entropy difference comparison; among them, the greater the entropy difference, the more significant the difference between samples, and synthetic samples with larger entropy differences are selected first.

[0085] In a further preferred exemplary technical solution, based on the above technical solution, the edit nearest neighbor technique is subsequently used to post-process the synthetic sample set to eliminate potential noise samples and ensure that the final balanced dataset has better classification performance.

[0086] In the hybrid resampling technique proposed in this embodiment of the invention, oversampling is performed by synergistically integrating SMOTE and ADASYN, and simplified attention entropy is used as a quantitative indicator to select samples with significant differences in feature complexity from the majority class. Furthermore, in the preferred embodiment of the invention, the edit nearest neighbor technique is used to eliminate noise and duplicate samples, which effectively improves the quality of the generated samples.

[0087] In a specific embodiment of the technical solution of the present invention, based on the original test sample set and the training sample set with balanced class distribution, fault features are extracted using simplified attention entropy or enhanced simplified attention entropy to obtain the feature vector test sample set and the feature vector training sample set (fault features).

[0088] When using simplified attention entropy for fault feature extraction, the specific calculation steps for simplified attention entropy are as follows:

[0089] According to the following rules, the time series X = {x1, x2, ..., x...}N Define the local maximum and local minimum, x N represents the Nth point of the time series X; where,

[0090] For any point x in the time series X i and x j (1 < i < N, 1 < j < N) exist:

[0091] Local maximum (LMax): If x i-1 < x i and x i > x i+1 , then the point x i is regarded as a local maximum;

[0092] Local minimum (LMin): If x j-1 > x j and x j < x j+1 , then the point x j is regarded as a local minimum;

[0093] Given the interval points according to two different strategies of {LMin - LMax} and {LMax - LMin}, solve the Shannon entropy of the adjacent interval point intervals. The formula is as follows:

[0094]

[0095] In the formula, H represents the Shannon entropy value; P(i) represents the probability of the interval point i appearing, and b represents the number of types of interval points.

[0096] Define the mean value of the Shannon entropy obtained by the two different strategies as the simple attention entropy. The calculation formula is as follows:

[0097]

[0098] In the formula, AtEn(X) represents the simple attention entropy value calculated for the time series X; H(j) represents the entropy value obtained from the jth strategy.

[0099] The present invention adopts the simple attention entropy. The advantages compared with the prior art are as follows: The simple attention entropy is a method improved on the basis of the attention entropy. While retaining good separability, this method simplifies the calculation process, significantly improves the operation efficiency, and completely inherits the excellent anti-interference ability and robustness of the attention entropy, and does not require artificial parameter setting. In addition, compared with the deep learning model, this method has a more concise structure, a simple and clear calculation process, stronger interpretability, and does not depend on large-scale training data.

[0100] Furthermore, when using the enhanced simple attention entropy for fault feature extraction, the enhanced simple attention entropy algorithm includes the following steps:

[0101] Given an original time series X = {x1, x2, ..., x...} N}, dividing it into a series of continuous coarse-grained sequence segments; for the scale factor τ, starting from each point in the interval [1,τ], the k-th coarse-grained sequence From elements Composition; based on the relationship between sequence length N and τ, two cases are handled:

[0102] When N is an integer multiple of τ, It can be represented as:

[0103]

[0104] When N is not an integer multiple of τ, there are residual sample points. It can be represented as:

[0105]

[0106] In the formula, residual sample points are processed separately to ensure data integrity.

[0107] For each coarse-grained sequence Calculate the simplified attention entropy value for each coarse-grained sequence, and then average the τ entropy values. The enhanced simplified attention entropy can then be expressed as:

[0108]

[0109] In the formula, ESAtEn(X,τ) represents the enhanced simplified attention entropy value obtained by calculating the time series X.

[0110] The enhanced simplified attention entropy used in this embodiment of the invention is a further optimization of the simplified attention entropy. This method effectively improves the multi-scale analysis process by introducing an incomplete window processing mechanism. Compared with the simplified attention entropy, it can capture effective information in the signal more comprehensively at multiple sufficient scales, significantly enhancing the separability of signal features, while fully retaining the original advantages of the simplified attention entropy, such as computational efficiency, strong anti-interference ability, and good robustness.

[0111] In a specific exemplary technical solution, the training process of the ELM-AdaBoost classifier includes the following steps:

[0112] Given a training dataset T = {(x1,y1),(x2,y2),…,(x N ,y N )};where, x i ∈R n Represents the eigenvector, y i ∈{1,2,..k} represents the corresponding fault type label, where k is the total number of fault types;

[0113] The weight distribution of the initial training data is as follows:

[0114]

[0115] In the formula, N is the size of the training dataset;

[0116] For each iteration t = 1, 2, ..., M (where M represents the number of base classifiers), perform the following steps:

[0117] Using the current weight distribution D i (i) Train the base classifier to obtain the prediction function g of the base classifier. t (x) and its corresponding prediction error ε t The calculation method is as follows:

[0118]

[0119] Based on prediction error ε t Calculate the weight α of the current base classifier. t :

[0120]

[0121] Update the base classifier weight distribution D for the next iteration using the following formula. t+1 (i):

[0122]

[0123] In the formula, B t It is the normalization factor.

[0124] After M iterations, the prediction functions g of the M base classifiers are obtained. t (x) and its corresponding weight α t These basic classifiers are combined into a strong classifier h(x):

[0125]

[0126] In the formula, sisgn(·) is the symbolic function.

[0127] In a specific exemplary embodiment of the present invention, a method for diagnosing imbalance faults in the drilling process based on enhanced simplified attention entropy is provided, comprising the following steps:

[0128] Step 1: Based on the drilling process data collected in real time by the drilling rig's sensors, the original drilling signals are divided using the sliding window technique to construct a training sample set and a test sample set;

[0129] Step 2: Resample the minority class of fault samples in the training sample set using a hybrid resampling technique to generate a synthetic sample set;

[0130] Step 3: Merge the synthetic sample set with the original real samples to form a training sample set with balanced class distribution, and use enhanced simple attention entropy to extract fault features to obtain a feature vector training sample set and a test sample set;

[0131] Step 4: Input the acquired feature vector training sample set into the ELM-AdaBoost classifier for model training. Subsequently, input the feature vector test sample set into the trained classifier model to ultimately achieve accurate identification and classification of different fault types.

[0132] Please see Figures 2 to 4 In a specific exemplary embodiment of the present invention, a test dataset from the intelligent drilling condition simulation test bench for drilling robots provided by China University of Mining and Technology is selected to verify the diagnostic accuracy of the method of the present invention. The flowchart is as follows: Figure 2 As shown, please follow these steps:

[0133] Step 1: The dataset contains bearing vibration signals acquired at a sampling frequency of 10 kHz, covering five different operating states: Normal (NOR), Outer Ring Fault (ORF), Inner Ring Fault (IRF), Rolling Element Fault (REF), and Cage Fault (CF). A sliding window technique was used to extract 1024 sampling points from the original signal to form a single sample, with a uniform test sample size of 200 for each category, and a training sample size of 200 for the Normal state. For each fault category, three imbalanced distribution patterns were constructed, with normal sample to fault sample ratios of 20:1, 10:1, and 5:1, respectively, to reflect common skewed data distributions during actual drilling. To reduce the influence of randomness, all experiments were repeated five times under the same conditions, and the final result was the average of the five independent experiments. Figure 3 The time-domain waveforms of five types of bearing signals are shown, and Table 1 details the specific experimental parameter settings.

[0134] Table 1. Experimental parameter settings

[0135]

[0136]

[0137] Step 2: Resample the outer ring fault, inner ring fault, rolling element fault, and cage fault signals separately using hybrid resampling technology to generate a composite fault sample;

[0138] Step 3: Merge the synthetic fault samples with the real fault samples to form a training sample set with balanced class distribution, and use enhanced simple attention entropy to extract fault features to obtain a feature vector training sample set and a test sample set;

[0139] Step 4: Input the acquired feature vector training sample set into the ELM-AdaBoost classifier for model training. Then, input the feature vector test sample set into the trained classifier model to achieve accurate identification and classification of different fault types. The diagnostic results under the three imbalance ratios are shown in Table 2. As can be seen from the table, the diagnostic accuracy of the method of this invention is above 97% under all three imbalance ratios, demonstrating excellent classification performance. To further analyze the feature extraction capability of the method, taking an imbalance ratio of 20:1 as an example, T-SNE dimensionality reduction technology is further used to visualize the extracted fault features. The results are as follows... Figure 4 As shown in the figure, it is evident that similar fault samples are tightly clustered in the feature space, with clear separation boundaries between different fault categories and only a small number of overlapping edge samples. This further verifies its effectiveness and reliability in bearing fault diagnosis. Finally, to further analyze the method's operating efficiency, the algorithm's running time under three imbalance rates was statistically analyzed, and the results are shown in Table 3. It is clear that the running time in all cases is less than 90 seconds, indicating that the method has high computational efficiency.

[0140] Table 2. Diagnostic results under three imbalance rates

[0141]

[0142]

[0143] Table 3. Running time under three imbalance rates

[0144]

[0145] In summary, this invention discloses a method for diagnosing imbalanced faults in the drilling process based on enhanced simplified attention entropy. The method includes the following steps: based on drilling process data collected in real-time by drilling rig sensors, the original drilling signals are divided using a sliding window technique to construct a training sample set and a test sample set; minority class fault samples in the training sample set are resampled using a hybrid resampling technique to generate a synthetic sample set; the synthetic sample set is merged with the original real sample set to form a training sample set with balanced class distribution; fault features are extracted using enhanced simplified attention entropy to obtain a feature vector set; and the drilling process fault diagnosis result is obtained using an ELM-AdaBoost classifier. The technical solution disclosed in this invention effectively solves the problem of imbalanced fault samples in the drilling process through a resampling method, while leveraging the advantages of low computational complexity and strong separability of enhanced simplified attention entropy, achieving high fault diagnosis accuracy while ensuring computational efficiency.

[0146] The following are embodiments of the apparatus of the present invention, which can be used to execute embodiments of the method of the present invention. For details not disclosed in the apparatus embodiments, please refer to the embodiments of the method of the present invention.

[0147] Please see Figure 5 In this embodiment of the invention, a fault diagnosis system for monitoring the drilling process is provided, comprising:

[0148] The data acquisition module is used to acquire drilling process data to be diagnosed.

[0149] The diagnostic classification module is used to perform fault diagnosis classification based on the drilling process data to be diagnosed, using a trained fault diagnosis model, and to obtain the fault diagnosis classification result.

[0150] The training steps of the fault diagnosis model include:

[0151] Obtain the original training sample set and the original test sample set;

[0152] The SMOTE and ADASYN algorithms are used to perform mixed resampling on the minority class fault samples in the original training sample set to generate a synthetic sample set; the synthetic sample set is then merged with the original training sample set to form a training sample set with balanced class distribution.

[0153] Based on the original test sample set and the training sample set with balanced class distribution, fault features are extracted using simplified attention entropy or enhanced simplified attention entropy to obtain the feature vector test sample set and the feature vector training sample set.

[0154] Based on the feature vector test sample set and the feature vector training sample set, the selected ensemble learning classifier is trained to obtain a trained fault diagnosis model.

[0155] In one embodiment of the present invention, a computer device is provided, comprising a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions in the computer storage medium to achieve a corresponding method flow or corresponding function. The processor described in this embodiment of the present invention can be used to execute operations of a fault diagnosis method for monitoring the drilling process.

[0156] In one embodiment of the present invention, a storage medium is provided, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space that stores the operating system of the terminal. Furthermore, the storage space also stores one or more instructions suitable for loading and execution by a processor, which can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM (Random Access Memory) or non-volatile memory, such as at least one disk storage device. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of the fault diagnosis method for monitoring the drilling process in the above embodiments.

[0157] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, optical storage, etc.) containing computer-usable program code.

[0158] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0159] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0160] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0161] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A fault diagnosis method for monitoring the drilling process, characterized in that, Includes the following steps: Acquire drilling process data to be diagnosed; Based on the drilling process data to be diagnosed, the trained fault diagnosis model is used to classify the fault diagnosis and obtain the fault diagnosis classification results. The training steps of the fault diagnosis model include: Obtain the original training sample set and the original test sample set; The SMOTE and ADASYN algorithms are used to perform mixed resampling on the minority class fault samples in the original training sample set to generate a synthetic sample set; the synthetic sample set is then merged with the original training sample set to form a training sample set with balanced class distribution. Based on the original test sample set and the training sample set with balanced class distribution, fault features are extracted using simplified attention entropy or enhanced simplified attention entropy to obtain the feature vector test sample set and the feature vector training sample set. Based on the feature vector test sample set and the feature vector training sample set, the selected ensemble learning classifier is trained to obtain a trained fault diagnosis model. Based on the original test sample set and the training sample set with balanced class distribution, simplified attention entropy is used to extract fault features and obtain feature vector test sample set and feature vector training sample set. The specific calculation steps of simplified attention entropy are as follows: According to the following rules, LMax and LMin are time series. Define local maxima and local minima. Representing time series The Points; where, for time series any point in the middle and , There exists a local maximum LMax: if and Then point Considered a local maximum; Local minimum LMin: if and Then point It is considered a local minimum; Given interval points using two different strategies, {LMin-LMax} and {LMax-LMin}, the Shannon entropy of the interval between adjacent interval points is calculated using the following formula: ； In the formula, H Represents the Shannon entropy value; Representing interval points The probability of occurrence Indicates the number of types of points in the interval; The mean of the Shannon entropy obtained from the two different strategies is defined as the simplified attention entropy, and the calculation formula is as follows: ； In the formula, Represents the calculation of time series The obtained simplified attention entropy value; Indicates from the first The entropy value obtained from each strategy; Based on the original test sample set and the training sample set with balanced class distribution, enhanced simplified attention entropy is used to extract fault features and obtain feature vector test sample set and feature vector training sample set. The specific calculation steps of enhanced simplified attention entropy are as follows: Given the original time series It is divided into a series of continuous coarse-grained sequence segments; among which, for the scale factor From the interval [1, Starting from each point in ], the first coarse-grained sequences From elements Composition; based on sequence length and The relationship can be handled in two ways: when yes When it is an integer multiple of, It can be represented as: ； when no When the value is an integer multiple of , there are residual sample points. It can be represented as: ； In the formula, residual sample points are processed separately to ensure data integrity; For each coarse-grained sequence Calculate the simplified attention entropy value for each coarse-grained sequence, and then... The average of the entropy values ​​is taken, and the final enhanced simple attention entropy is expressed as: ； In the formula, Represents the calculation of time series The resulting enhanced simple attention entropy value.

2. The fault diagnosis method for monitoring the drilling process according to claim 1, characterized in that, The data types of the drilling process data to be diagnosed are the same as those of the original training sample set and the original test sample set.

3. The fault diagnosis method for monitoring the drilling process according to claim 1, characterized in that, The drilling process data to be diagnosed includes one or more of the following: vibration signals, torque signals, and pressure signals of the selected component.

4. The fault diagnosis method for monitoring the drilling process according to claim 1, characterized in that, The SMOTE and ADASYN algorithms are used to perform mixed resampling on minority class fault samples in the original training sample set to generate a synthetic sample set, including: For each minority class sample set The SMOTE and ADASYN algorithms were used for resampling to generate synthetic samples that met the preset quantity requirements in order to balance the number of fault samples of different categories. x n Indicates the first A minority class of samples; The generation of synthetic samples is achieved through the following interpolation formula: ； In the formula, Indicates a synthetic sample; This represents the current minority class sample. This represents the selected nearest neighbor sample. It is a random number within the interval [0,1]. In addition, the difference between the mean simple attention entropy of the minority class samples generated by the SMOTE and ADASYN algorithms and the original majority class samples is calculated separately. The expressions for calculating the difference between the two classes are as follows: ； In the formula, and represents the mean of the simplified attention entropy of minority class samples generated by the SMOTE and ADASYN algorithms, respectively; The mean of the simplified attention entropy of the original majority class samples; Based on the difference calculation results, a synthetic sample set is formed according to the principle of entropy difference comparison; among them, the greater the entropy difference, the more significant the difference between samples, and synthetic samples with larger entropy differences are selected first.

5. A fault diagnosis method for monitoring the drilling process according to claim 1, characterized in that, In the step of training the selected ensemble learning classifier based on the feature vector test sample set and the feature vector training sample set to obtain the trained fault diagnosis model, the selected ensemble learning classifier is either the ELM-AdaBoost classifier or the BP neural network-AdaBoost classifier.

6. A fault diagnosis system for monitoring the drilling process, characterized in that, For implementing the fault diagnosis method according to claim 1, the fault diagnosis system includes: The data acquisition module is used to acquire drilling process data to be diagnosed. The diagnostic classification module is used to perform fault diagnosis classification based on the drilling process data to be diagnosed, using a trained fault diagnosis model, and to obtain the fault diagnosis classification result. The training steps of the fault diagnosis model include: Obtain the original training sample set and the original test sample set; The SMOTE and ADASYN algorithms are used to perform mixed resampling on the minority class fault samples in the original training sample set to generate a synthetic sample set; the synthetic sample set is then merged with the original training sample set to form a training sample set with balanced class distribution. Based on the original test sample set and the training sample set with balanced class distribution, fault features are extracted using simplified attention entropy or enhanced simplified attention entropy to obtain the feature vector test sample set and the feature vector training sample set. Based on the feature vector test sample set and the feature vector training sample set, the selected ensemble learning classifier is trained to obtain a trained fault diagnosis model.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the fault diagnosis method for monitoring the drilling process as described in any one of claims 1 to 5.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the fault diagnosis method for monitoring the drilling process as described in any one of claims 1 to 5.

Images