Well drilling abnormal vibration working condition detection method, device, equipment and medium

Abstract

The invention relates to the field of well drilling anomaly detection, and discloses a well drilling abnormal vibration working condition detection method, device, equipment and medium, which can be realized by using a trained abnormal vibration working condition detection model, and the model comprises a multi-head self-attention layer, a feedforward network layer and a full connection layer. According to the invention, the multi-modal vibration data can be collected in the drilling operation, the multi-modal vibration data is input into the trained abnormal vibration condition detection model, and the multi-head self-attention layer efficiently captures the potential long-range dependency relationship among different modal vibration data in the multi-modal vibration data through a self-attention mechanism. And generating a multi-modal feature vector, and performing label detection by the feed-forward network layer and the full-connection layer according to the multi-modal feature vector output by the multi-head self-attention layer, thereby realizing detection of the abnormal vibration working condition of the drilling well and obtaining an abnormal vibration working condition detection result. The method can enrich the detection means of the abnormal vibration working condition in the drilling operation, and effectively improves the detection accuracy of the abnormal vibration working condition.

Description

Technical Field

[0001] The present invention relates to the field of drilling anomaly detection, and particularly to a method, device, equipment and medium for detecting abnormal vibration conditions during drilling. Background Art

[0002] With the development of science and technology, oil and gas exploration and production technologies have been continuously improved.

[0003] Drilling is a key link in the production process of oil and gas wells. With the deepening of the exploration formation depth and the complexity of formation conditions, various abnormal working conditions may occur during the drilling process, which may lead to accidents such as drill string fracture, wall collapse and blowout, threatening personnel safety and reducing production efficiency.

[0004] Related technologies generally rely on manual experience and threshold monitoring of drilling parameters to judge whether abnormal working conditions occur during drilling and their working condition types, but the accuracy rate is relatively low. Summary of the Invention

[0005] The present invention provides a method, device, equipment and medium for detecting abnormal vibration conditions during drilling, so as to solve the defect of relatively low accuracy in drilling anomaly detection in related technologies and improve the accuracy of detecting abnormal vibration conditions during drilling.

[0006] In a first aspect, the present invention provides a method for detecting abnormal vibration conditions during drilling. The method is implemented by using a trained abnormal vibration condition detection model, and the model is obtained by training a pre-trained detection model with sample data labeled with vibration condition labels; the model includes a multi-head self-attention layer, a feed-forward network layer and a fully-connected layer;

[0007] The method includes:

[0008] Preprocess multi-modal vibration data collected during drilling operations respectively to obtain processed multi-modal vibration data;

[0009] Input the processed multi-modal vibration data into the multi-head self-attention layer for local feature extraction, normalization and weight assignment to obtain a multi-modal feature vector;

[0010] Input the multi-modal feature vector into the feed-forward network layer and the fully-connected layer for label prediction to obtain the detection result of abnormal vibration conditions corresponding to the drilling operation.

[0011] Optionally, the multi-modal vibration data includes at least two of time-domain vibration data, frequency-domain vibration data and image vibration data;

[0012] When the multimodal vibration data includes time-domain vibration data, frequency-domain vibration data, and image vibration data, the processed multimodal vibration data includes processed time-domain vibration data, processed frequency-domain vibration data, and processed image vibration data, and the multimodal feature vector includes a temporal feature vector, a frequency-domain feature vector, and a spatial feature vector.

[0013] Optionally, when the multimodal vibration data includes time-domain vibration data, frequency-domain vibration data, and image vibration data, the multi-head self-attention layer includes a first local self-attention module, a second local self-attention module, a third local self-attention module, a normalization module, and a global self-attention module;

[0014] Inputting the processed multimodal vibration data into the multi-head self-attention layer for local feature extraction, normalization, and weight assignment to obtain a multimodal feature vector includes:

[0015] Inputting the processed time-domain vibration data, the processed frequency-domain vibration data, and the processed image vibration data into the first local self-attention module, the second local self-attention module, and the third local self-attention module respectively for local feature extraction to obtain an initial time feature vector, an initial frequency feature vector, and an initial spatial feature vector output by the first local self-attention module, the second local self-attention module, and the third local self-attention module respectively;

[0016] Inputting the initial time feature vector, the initial frequency feature vector, and the initial spatial feature vector into the normalization module and the global self-attention module for normalization and weight assignment to obtain the time feature vector, the frequency feature vector, and the spatial feature vector.

[0017] Optionally, inputting the initial time feature vector, the initial frequency feature vector, and the initial spatial feature vector into the normalization module and the global self-attention module for normalization and weight assignment to obtain the time feature vector, the frequency feature vector, and the spatial feature vector includes:

[0018] Inputting the initial time feature vector, the initial frequency feature vector, and the initial spatial feature vector into the normalization module respectively for normalization to obtain a normalized time feature vector, a normalized frequency feature vector, and a normalized spatial feature vector generated and output by the normalization module;

[0019] Input the normalized time feature vector, the normalized frequency feature vector, and the normalized spatial feature vector into the global self-attention module, so that the global self-attention module: perform a weight operation based on the normalized time feature vector, the normalized frequency feature vector, and the normalized spatial feature vector, determine the weight coefficients of the normalized time feature vector, the normalized frequency feature vector, and the normalized spatial feature vector, multiply the normalized time feature vector, the normalized frequency feature vector, and the normalized spatial feature vector by the corresponding weight coefficients respectively, and obtain and output the time feature vector, the frequency feature vector, and the spatial feature vector.

[0020] Optionally, the feed-forward network layer includes a feature aggregation module and an activation function; when the multi-modal vibration data includes time-domain vibration data, frequency-domain vibration data, and image vibration data, the step of inputting the multi-modal feature vector into the feed-forward network layer and the fully-connected layer for label prediction to obtain the detection result of the abnormal vibration condition corresponding to the drilling operation includes:

[0021] Input the time feature vector, the frequency feature vector, and the spatial feature vector into the feature aggregation module for feature aggregation to obtain the initial fusion feature vector generated and output by the feature aggregation module;

[0022] Input the initial fusion feature vector into the activation function for non-linear transformation to obtain the fusion feature vector generated and output by the activation function;

[0023] Input the fusion feature vector into the fully-connected layer for label prediction to obtain the detection result of the abnormal vibration condition corresponding to the drilling operation.

[0024] Optionally, when the multi-modal vibration data includes time-domain vibration data, frequency-domain vibration data, and image vibration data, the step of respectively preprocessing the multi-modal vibration data collected during the drilling operation to obtain the processed multi-modal vibration data includes:

[0025] Perform denoising on the time-domain vibration data to obtain the denoised vibration data; normalize the denoised vibration data to obtain the normalized vibration data; perform time-frequency domain feature extraction on the normalized vibration data to obtain the processed time-domain vibration data;

[0026] Eliminate the high-frequency fluctuations in the processed frequency-domain vibration data to obtain the vibration data after elimination; identify the key frequency components in the vibration data after elimination based on frequency band division and dominant frequency extraction to obtain the vibration data after identification; remove the background noise in the vibration data after identification to extract the effective spectral features in the vibration data after identification, so as to obtain the processed frequency-domain vibration data;

[0027] Perform filtering and denoising on the image vibration data to eliminate noise interference and obtain the vibration data after filtering; perform image enhancement on the vibration data after filtering to obtain the vibration data after enhancement; extract the region of interest in the vibration data after enhancement to obtain the vibration data of interest; perform motion compensation on the vibration data of interest to eliminate the influence of camera jitter or object displacement to obtain the vibration data after compensation; perform feature extraction on the vibration data after compensation to obtain the processed image vibration data.

[0028] Optionally, the vibration condition label is a normal condition or an abnormal condition, and the abnormal condition includes at least one of bit bounce, stick-slip, and whirling.

[0029] In a second aspect, the present invention provides a device for detecting abnormal vibration conditions in drilling. The device is implemented using a trained abnormal vibration condition detection model, and the model is obtained by training a pre-trained detection model using sample data labeled with vibration condition labels; the model includes a multi-head self-attention layer, a feed-forward network layer, and a fully connected layer;

[0030] The device includes:

[0031] A preprocessing unit for preprocessing the multi-modal vibration data collected during drilling operations respectively to obtain the processed multi-modal vibration data;

[0032] A first input unit for inputting the processed multi-modal vibration data into the multi-head self-attention layer for local feature extraction, normalization, and weight assignment to obtain a multi-modal feature vector;

[0033] A second input unit for inputting the multi-modal feature vector into the feed-forward network layer and the fully connected layer for label prediction to obtain the detection result of abnormal vibration conditions corresponding to the drilling operation.

[0034] In a third aspect, the present invention provides a computer device, including: a memory and a processor, which are communicatively connected to each other. The memory stores computer instructions, and the processor executes the computer instructions to execute the method for detecting abnormal vibration conditions in drilling according to the first aspect or any corresponding embodiment thereof.

[0035] Fourthly, the present invention provides a computer-readable storage medium, on which computer instructions are stored, and the computer instructions are used to cause a computer to execute the drilling abnormal vibration condition detection method according to the first aspect or any corresponding embodiment thereof described above.

[0036] The drilling abnormal vibration condition detection method, device, equipment and medium provided by the present invention can be implemented by using a trained abnormal vibration condition detection model, which includes a multi-head self-attention layer, a feed-forward network layer and a fully-connected layer. The present invention can collect multi-modal vibration data during drilling operations, input the multi-modal vibration data into the trained abnormal vibration condition detection model, and the multi-head self-attention layer efficiently captures the potential long-range dependence relationships between different modal vibration data through the self-attention mechanism in the multi-modal vibration data to generate a multi-modal feature vector, and the feed-forward network layer and the fully-connected layer perform label detection according to the multi-modal feature vector output by the multi-head self-attention layer, so as to realize the detection of drilling abnormal vibration conditions and obtain the detection result of abnormal vibration conditions. The present invention can enrich the detection means for abnormal vibration conditions during drilling operations and effectively improve the accuracy of detecting abnormal vibration conditions. Description of the Drawings

[0037] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0038] Figure 1 It is a flowchart of a drilling abnormal vibration condition detection method provided by an embodiment of the present invention;

[0039] Figure 2 It is a flowchart of another drilling abnormal vibration condition detection method provided by an embodiment of the present invention;

[0040] Figure 3 It is a schematic structural diagram of a drilling abnormal vibration condition detection device provided by an embodiment of the present invention;

[0041] Figure 4 It is a schematic structural diagram of a computer device provided by an embodiment of the present invention. Detailed Embodiments

[0042] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present invention without creative efforts fall within the protection scope of the present invention.

[0043] The following will describe Figure 1 - Figure 2 the method for detecting abnormal vibration conditions during drilling of the present invention.

[0044] As Figure 1 shown, the first method for detecting abnormal vibration conditions during drilling is proposed in this embodiment. This method is implemented using a trained abnormal vibration condition detection model, and the model is obtained by training a pre-trained detection model with sample data labeled with vibration condition labels. The model includes a multi-head self-attention layer, a feed-forward network layer, and a fully connected layer. This method includes the following steps:

[0045] S101. Preprocess the multi-modal vibration data collected during drilling operations respectively to obtain the processed multi-modal vibration data.

[0046] Optionally, the vibration condition label is a normal condition or an abnormal condition, and the abnormal condition includes at least one of bit bounce, stick-slip, and whirling.

[0047] Specifically, the multi-modal vibration data may include vibration data of different modalities.

[0048] Optionally, the multi-modal vibration data may include at least two of time-domain vibration data, frequency-domain vibration data, and image vibration data. Among them, the time-domain vibration data refers to the original data of the vibration signal changing with time, which can be represented in the form of a time series and can directly reflect the change of the amplitude of the vibration signal with time.

[0049] Among them, the frequency-domain noise data is obtained by converting the time-domain noise signal through Fourier transform, and can be composed of two columns of numerical values of frequency and amplitude, which focuses on revealing the intensity of different frequency components in the noise signal and helps to identify the main frequency range of the noise.

[0050] Among them, the image vibration data refers to converting the noise signal into a two-dimensional image through a visualization method (such as a spectrogram and a time-frequency diagram), which intuitively shows the distribution characteristics, frequency changes, or dynamic characteristics of the noise. In this embodiment, the features in the image vibration data can be extracted by plotting the amplitude and the height of the fluctuation of the image vibration data.

[0051] It should be noted that in this embodiment, multi-modal vibration data can be collected during the drilling operation of an oil and gas well, the vibration condition label corresponding to the multi-modal vibration data can be determined, and the multi-modal vibration data can be preprocessed to obtain processed multi-modal data. The processed multi-modal data is used as sample data, and the vibration condition label is marked on the sample data to obtain sample data marked with the vibration condition label, which is then input into a pre-trained detection model for detecting abnormal vibrations during drilling. The detection result of the abnormal vibration condition generated and output by the pre-trained detection model is obtained, and the parameters of the pre-trained detection model are updated according to the detection result of the abnormal vibration condition and the vibration condition label. In this embodiment, the pre-trained detection model can be iteratively updated until the model meets the performance requirements, and the model that meets the performance requirements is determined as the trained abnormal vibration condition detection model.

[0052] Specifically, the pre-trained detection model can be a Swin Transformer network structure model or other structure models, which is not limited in this embodiment.

[0053] Among them, in this embodiment, after collecting multi-modal vibration data and determining the vibration condition label corresponding to the multi-modal vibration data, the vibration condition label can be directly marked on the multi-modal vibration data to obtain multi-modal vibration data marked with the vibration condition label, and the multi-modal vibration data marked with the vibration condition label is determined as the above-mentioned sample data marked with the vibration condition label. Then, the sample data can be preprocessed to obtain preprocessed data, and then model training can be performed based on the preprocessed data.

[0054] Specifically, when this embodiment applies the trained abnormal condition detection model for detection, during the drilling operation of an oil and gas well, multi-modal vibration data can be collected in advance through multiple sensors arranged at the position of the drilling bit. The vibration data can include time-domain vibration data, frequency-domain vibration data, and image vibration data.

[0055] It should be noted that the time-domain vibration data, frequency-domain vibration data, and image vibration data are vibration data of different modes during the drilling operation. In this embodiment, multi-modal vibration data, namely time-domain vibration data, frequency-domain vibration data, and image vibration data, can be collected during the drilling operation to enrich data diversity. The multi-modal vibration data is used to train the model and applied in the actual detection of the model to realize the full utilization of the drilling collected data and improve the detection accuracy of abnormal vibration conditions.

[0056] Optionally, when the multimodal vibration data includes time-domain vibration data, frequency-domain vibration data, and image vibration data, the processed multimodal vibration data includes processed time-domain vibration data, processed frequency-domain vibration data, and processed image vibration data. Specifically, in this embodiment, the collected time-domain vibration data, frequency-domain vibration data, and image vibration data can be preprocessed respectively to obtain the corresponding processed time-domain vibration data, processed frequency-domain vibration data, and processed image vibration data.

[0057] S102. Input the processed multimodal vibration data into the multi-head self-attention layer for local feature extraction, normalization, and weight allocation to obtain a multimodal feature vector.

[0058] It should be noted that in this embodiment, a multi-head self-attention layer is introduced into the relevant network structure model for model optimization, and a trained abnormal vibration condition detection model is obtained through training. The long-range dependence relationship between multimodal data is captured through the self-attention mechanism, the feature extraction and the relationship between features are enriched, and the detection accuracy of the model is enhanced.

[0059] Specifically, in this embodiment, after obtaining the processed multimodal vibration data, the processed multimodal vibration data can be input into the multi-head self-attention layer for local feature extraction, normalization, and weight allocation in sequence to obtain a multimodal feature vector.

[0060] Optionally, when the multimodal vibration data includes time-domain vibration data, frequency-domain vibration data, and image vibration data, the processed multimodal vibration data includes processed time-domain vibration data, processed frequency-domain vibration data, and processed image vibration data, and the multimodal feature vector includes a time-series feature vector, a frequency-domain feature vector, and a spatial feature vector. At this time, the multi-head self-attention layer includes a first local self-attention module, a second local self-attention module, a third local self-attention module, a normalization module, and a global self-attention module. At this time, step S102 may include:

[0061] Input the processed time-domain vibration data, processed frequency-domain vibration data, and processed image vibration data into the first local self-attention module, the second local self-attention module, and the third local self-attention module respectively for local feature extraction to obtain the initial time feature vector, initial frequency feature vector, and initial spatial feature vector output by the first local self-attention module, the second local self-attention module, and the third local self-attention module respectively;

[0062] Input the initial time feature vector, initial frequency feature vector, and initial spatial feature vector into the normalization module and the global self-attention module for normalization and weight allocation to obtain a time feature vector, a frequency feature vector, and a spatial feature vector.

[0063] Specifically, the first local self-attention module, the second local self-attention module, and the third local self-attention module can be local window self-attention modules with a window size of 7×7 or 14×14, which are used to perform local self-attention calculations and process the spatial information of different modality data. An activation function can be set in this module, and the Gaussian Error Linear Unit (GELU) can be used to process this activation function.

[0064] It should be noted that the first local self-attention module, the second local self-attention module, and the third local self-attention module can perform local attention operations on their respective input data, extract features from the input data, and retain the important information of the corresponding modality data.

[0065] Specifically, the window size of the global self-attention module is the global scale, which is used to capture the global information interaction of different modality data. An activation function can be set in the global self-attention module to enhance the model's understanding of the global features of multimodal data.

[0066] Optionally, the above-mentioned input of the initial time feature vector, the initial frequency feature vector, and the initial spatial feature vector into the normalization module and the global self-attention module for normalization and weight assignment to obtain the time feature vector, the frequency feature vector, and the spatial feature vector includes:

[0067] Input the initial time feature vector, the initial frequency feature vector, and the initial spatial feature vector into the normalization module for normalization respectively to obtain the normalized time feature vector, the normalized frequency feature vector, and the normalized spatial feature vector generated and output by the normalization module;

[0068] Input the normalized time feature vector, the normalized frequency feature vector, and the normalized spatial feature vector into the global self-attention module, so that the global self-attention module: perform weight calculations based on the normalized time feature vector, the normalized frequency feature vector, and the normalized spatial feature vector, determine the weight coefficients of the normalized time feature vector, the normalized frequency feature vector, and the normalized spatial feature vector, and multiply the normalized time feature vector, the normalized frequency feature vector, and the normalized spatial feature vector by the corresponding weight coefficients respectively to obtain and output the time feature vector, the frequency feature vector, and the spatial feature vector.

[0069] Optionally, the normalization module may include a first normalization module, a second normalization module, and a third normalization module. The first normalization module, the second normalization module, and the third normalization module may be respectively disposed after the first local self-attention module, the second local self-attention module, and the third local self-attention module, and are respectively used to normalize the initial time feature vector, the initial frequency feature vector, and the initial spatial feature vector.

[0070] It should be noted that in this embodiment, the normalization module can standardize the output of each local self-attention module to ensure the stability of the data of each layer during the training process, and improve the training efficiency and effect of the model.

[0071] S103. Input the time feature vector, the frequency feature vector, and the spatial feature vector into a feed-forward network layer and a fully connected layer for label prediction to obtain an abnormal vibration condition detection result corresponding to the drilling operation.

[0072] Specifically, in this embodiment, the feed-forward network layer and the fully connected layer are used to process the time feature vector, the frequency feature vector, and the spatial feature vector to implement abnormal vibration label prediction and obtain an abnormal vibration condition detection result.

[0073] Optionally, the feed-forward network layer includes a feature aggregation module and an activation function. When the multi-modal vibration data includes time-domain vibration data, frequency-domain vibration data, and image vibration data, step S103 may include:

[0074] Input the time feature vector, the frequency feature vector, and the spatial feature vector into the feature aggregation module for feature aggregation to obtain an initial fusion feature vector generated and output by the feature aggregation module;

[0075] Input the initial fusion feature vector into the activation function for non-linear transformation to obtain a fusion feature vector generated and output by the activation function;

[0076] Input the fusion feature vector into the fully connected layer for label prediction to obtain an abnormal vibration condition detection result corresponding to the drilling operation.

[0077] Specifically, in this embodiment, the feature aggregation module can fuse the time feature vector, the frequency feature vector, and the spatial feature vector, that is, fuse the multi-modal feature vectors, implement cross-modal information feature fusion, and form a unified feature vector, namely the initial fusion feature vector. Then, in this embodiment, the activation function can perform non-linear transformation on the initial fusion feature vector to obtain the fusion feature vector.

[0078] Specifically, in this embodiment, the fused feature vector can be processed by a fully connected layer to obtain a high-dimensional representation of the features, which is further compressed into a low-dimensional representation related to the prediction task. Specifically, in this embodiment, a classification task or a regression task can be performed according to the type of the label through a fully connected layer, and a label prediction can be performed through an activation function, such as the Softmax function for the classification task or the linear activation function for the regression task, to generate the detection result of the abnormal vibration condition.

[0079] It should be noted that the multi-head self-attention mechanism can capture the features of the noise data and the image data through self-attention and achieve the fusion of cross-modal information. The feed-forward network layer performs feature mapping through a series of fully connected layers, thereby improving the model's ability to identify abnormal conditions.

[0080] The method for detecting abnormal vibration conditions in drilling proposed in this embodiment can be implemented using a trained abnormal vibration condition detection model, which includes a multi-head self-attention layer, a feed-forward network layer, and a fully connected layer. In this embodiment, multi-modal vibration data can be collected during the drilling operation and input into the trained abnormal vibration condition detection model. The multi-head self-attention layer can efficiently capture the potential long-range dependence relationships between different modal vibration data in the multi-modal vibration data through the self-attention mechanism, generate a multi-modal feature vector, and the feed-forward network layer and the fully connected layer can perform label detection according to the multi-modal feature vector output by the multi-head self-attention layer, so as to realize the detection of abnormal vibration conditions in drilling and obtain the detection result of abnormal vibration conditions. This embodiment can enrich the detection means for abnormal vibration conditions in drilling operations and effectively improve the accuracy of detecting abnormal vibration conditions.

[0081] Based on Figure 1 , this embodiment proposes a second method for detecting abnormal vibration conditions in drilling. When the multi-modal vibration data includes time-domain vibration data, frequency-domain vibration data, and image vibration data, the above step S101 may include:

[0082] Denoise the time-domain vibration data to obtain the denoised vibration data; normalize the denoised vibration data to obtain the normalized vibration data; extract time-frequency domain features from the normalized vibration data to obtain the processed time-domain vibration data;

[0083] Eliminate the high-frequency fluctuations in the processed frequency-domain vibration data to obtain the eliminated vibration data; identify the key frequency components in the eliminated vibration data based on frequency band division and dominant frequency extraction to obtain the identified vibration data; remove the background noise in the identified vibration data to extract the effective spectral features in the identified vibration data to obtain the processed frequency-domain vibration data;

[0084] Filter and denoise the image vibration data to eliminate noise interference and obtain the filtered vibration data; perform image enhancement on the filtered vibration data to obtain the enhanced vibration data; extract the region of interest in the enhanced vibration data to obtain the vibration data of the region of interest; perform motion compensation on the vibration data of the region of interest to eliminate the influence of camera jitter or object displacement and obtain the compensated vibration data; extract features from the compensated vibration data to obtain the processed image vibration data.

[0085] It can be understood that the processed time-domain vibration data, the processed frequency-domain vibration data, and the processed image vibration data are the data obtained by preprocessing the time-domain vibration data, the frequency-domain vibration data, and the image vibration data in this embodiment, respectively. Specifically, in this embodiment, the processed time-domain vibration data is obtained by preprocessing the time-domain vibration data, the processed frequency-domain vibration data is obtained by preprocessing the frequency-domain vibration data, and the processed image vibration data is obtained by preprocessing the image vibration data.

[0086] This embodiment specifically describes the preprocessing processes of three different modal data. For vibration data of different modalities, the preprocessing processes have different focuses.

[0087] Among them, in the process of preprocessing the time-domain vibration data in this embodiment, high-frequency noise in the time-domain vibration data can be removed by denoising methods such as low-pass filtering or wavelet transform to obtain the denoised vibration data. Subsequently, in this embodiment, the denoised vibration data can be normalized by standardization or range scaling to eliminate the dimensional difference and obtain the normalized vibration data. After that, time-frequency domain features such as mean, variance, peak value, and spectral energy can be extracted from the normalized vibration data to obtain the processed time-domain vibration data, ensuring the smoothness and consistency of the signal and providing a reliable basis for subsequent analysis.

[0088] Among them, in the process of preprocessing the frequency-domain vibration data in this embodiment, filtering methods such as moving average filtering or Savitzky-Golay filtering can be used to filter the frequency-domain vibration data to remove high-frequency fluctuations and obtain the vibration data after elimination. After that, the key frequency components in the vibration data after elimination are identified by frequency band division and dominant frequency extraction to obtain the identified vibration data. Then, the background noise in the identified vibration data is removed by noise threshold processing or background noise estimation to extract effective spectral features and obtain the processed frequency-domain vibration data, enhancing the interpretability and analysis accuracy of the data.

[0089] Among them, in the process of preprocessing the image vibration data in this embodiment, noise interference in the image vibration data can be eliminated through, for example, Gaussian filtering or median filtering to obtain the filtered vibration data. The filtered vibration data is enhanced by an image enhancement method such as contrast stretching or edge detection to highlight the vibration mode and obtain the enhanced vibration data. The region of interest in the enhanced vibration data is extracted through an image segmentation method such as threshold segmentation or region growing to obtain the vibration data of interest. And the vibration data of interest is processed through a motion compensation method such as inter-frame alignment or optical flow method to eliminate the influence of camera jitter or object displacement and obtain the compensated vibration data. Finally, the shape, texture, and motion characteristics of the vibration mode are extracted from the compensated vibration data to obtain the processed image vibration data, providing high-quality image data support for subsequent analysis. Through these targeted preprocessing steps, the quality of each data modality is ensured to meet the accuracy and reliability requirements of subsequent analysis.

[0090] The drilling abnormal vibration condition detection method proposed in this embodiment can preprocess multi-modal vibration data to ensure that the quality of each data modality meets the accuracy and reliability requirements of subsequent analysis

[0091] Such as Figure 2 As shown, the second drilling abnormal vibration condition detection method proposed in this embodiment, when the multi-modal vibration data includes time-domain vibration data, frequency-domain vibration data, and image vibration data, the multi-head self-attention layer includes a first local self-attention module, a second local self-attention module, a third local self-attention module, a normalization module, and a global self-attention module. The feed-forward network layer includes a feature aggregation module and an activation function.

[0092] Among them, the first local self-attention module, the second local self-attention module, and the third local self-attention module can be local window self-attention modules with a window size of 7×7 or 14×14. In this embodiment, the time-domain vibration data, the frequency-domain vibration data, and the image vibration data can be respectively input into the first local self-attention module, the second local self-attention module, and the third local self-attention module for processing. The third local self-attention module with a size of 14×14 can process high-dimensional vectors. The first local self-attention module, the second local self-attention module, and the third local self-attention module can respectively output an initial time feature vector, an initial frequency feature vector, and an initial spatial feature vector.

[0093] After that, in this embodiment, the initial time feature vector, the initial frequency feature vector, and the initial spatial feature vector can be respectively input into a normalization module for normalization, to obtain the normalized time feature vector, the normalized frequency feature vector, and the normalized spatial feature vector generated and output by the normalization module. The normalized time feature vector, the normalized frequency feature vector, and the normalized spatial feature vector are input into a global self-attention module to capture global connections, to obtain the time feature vector, the frequency feature vector, and the spatial feature vector.

[0094] Specifically, in this embodiment, the time feature vector, the frequency feature vector, and the spatial feature vector can be input into a feature aggregation module and an activation function of a feed-forward network layer for feature aggregation and non-linear transformation, to obtain a fused feature vector. The activation function here can be the softmax function.

[0095] After that, in this embodiment, the fused feature vector can be input into a fully-connected layer for label prediction, to obtain the abnormal vibration condition detection result output by the fully-connected layer.

[0096] In other drilling abnormal vibration condition detection methods proposed in this embodiment, the pre-trained detection model can be the Swin Transformer model. In this embodiment, the abnormal vibration condition label set by the user can be obtained, and this label is related to the distribution type of drilling data, data quality, and actual operation requirements. According to this label, the Swin Transformer model is trained. Among them, the training process of the Swin Transformer model includes using the historical multi-modal vibration data collected during drilling operations as the training set. A classification loss function is used for supervised learning training to optimize the parameters of the Swin Transformer model to ensure its high accuracy in abnormal condition detection. During the training process, a stochastic gradient descent optimizer is used for parameter update, and an early stopping strategy is adopted to avoid overfitting, so as to improve the generalization ability and stability of the model, to obtain a trained abnormal vibration condition detection model, which can process the multi-modal vibration data collected during the drilling process and identify abnormal conditions.

[0097] Specifically, in this embodiment, the drilling vibration data can be collected by sensors arranged at the position of the drilling bit, and the drilling vibration data is preprocessed. The trained Swin Transformer model is used to analyze the collected multi-modal vibration data to identify abnormal vibration conditions, such as equipment failures, bit wear, and downhole vibrations. Through the abnormal information output by the model, an alarm is provided to the ground control system in a timely manner to help the operator make real-time decisions and adjustments, so as to ensure the safety and efficiency of drilling operations.

[0098] Among them, in the process of collecting multi-modal vibration data in this embodiment, the speed, depth, vibration, and temperature data of the drill bit can be collected in real time through various types of sensors. The preprocessing process of these data can include: outlier detection and correction, missing value interpolation, and data segmentation processing. By effectively preprocessing the collected data, it is ensured that the data quality of the model input meets the requirements of abnormal working condition detection. In the data segmentation processing, this embodiment can segment the streaming data obtained by various types of sensors into data with a fixed time window length, and these segmented data will be used as inputs for the Swin Transformer to detect and analyze abnormal working conditions. This processing helps to retain the time series characteristics of the data and improve the accuracy and timeliness of abnormal detection.

[0099] Specifically, this embodiment can transmit the processed multi-modal vibration data obtained after collection and preprocessing to the Swin Transformer model based on optical fiber communication or wireless communication methods such as transport layer application layer protocols. Through the communication method, it is ensured that the data can be transmitted to the model for analysis quickly and stably, meeting the requirements of real-time detection and response. During the data transmission process, this embodiment can also use data encryption methods or data verification schemes to ensure the security and integrity of the data during transmission, prevent the data from being tampered with or lost, and ensure the reliability of abnormal vibration condition detection. This embodiment can also integrate and format the obtained high-dimensional data, thereby capturing the features therein, further facilitating subsequent data analysis, visual display, and further decision support, helping to improve the readability and usage efficiency of the data, and supporting the real-time optimization of various decisions during the operation.

[0100] In the related technology, in the process of oil and gas production, the multi-modal vibration data monitored during drilling operations is crucial. These data can not only reflect the working state of the drill bit, downhole environmental conditions, and equipment conditions, but also provide important support for formation characteristic analysis, wellbore stability monitoring, etc. In order to better ensure operation safety and operation efficiency, it is necessary to perform real-time analysis on these real-time collected data and promptly discover potential abnormal working conditions.

[0101] The drilling data analysis methods in related technologies generally rely on simple label detection or statistical anomaly detection models, which often have strong limitations. Especially when faced with complex multimodal vibration data, related technologies find it difficult to capture the complex relationships between the data, resulting in low accuracy of anomaly detection and failure to meet the needs of real-time monitoring. The application of deep learning technology in image processing and time series analysis has made significant progress, especially advanced model structures such as Transformer have demonstrated powerful capabilities in processing multimodal data. SwinTransformer can efficiently capture potential long-range dependencies in data through the self-attention mechanism, which is particularly suitable for the fusion analysis of image data and time series data. Based on these advantages, Swin Transformer is widely used in anomaly detection and pattern recognition in various fields.

[0102] However, although deep learning models have achieved certain results in multimodal data analysis, there are still many challenges in how to effectively combine noise, vibration and image data and achieve efficient and accurate abnormal condition identification. The detection methods in related technologies often have problems such as low computational efficiency and poor recognition accuracy when processing complex and multi-dimensional data. How to use deep learning models, especially Swin Transformer, to achieve efficient and accurate multimodal drilling condition identification is a technical problem that needs to be solved urgently.

[0103] The abnormal vibration condition detection method for drilling proposed in this embodiment can flexibly process and identify multimodal vibration data from multiple sensors under different drilling conditions for data analysts, and effectively identify abnormal conditions, thereby improving the overall safety and efficiency of drilling operations. This embodiment can use the self-attention mechanism to effectively capture long-range dependencies in the data, adapt to different types of multimodal data, and improve the accuracy of abnormal condition identification. This embodiment can flexibly adjust the detection label according to the drilling operation environment, thereby achieving efficient abnormal condition identification and reducing false positives and false negatives. Through the efficient Swin Transformer model structure and training process, a high abnormality detection accuracy can be maintained even in a complex data environment. It can realize multimodal fusion of time domain vibration data, frequency domain vibration data and image vibration data, ensure comprehensive analysis of different types of data, avoid information loss problems, and can be widely used in oil and gas exploration and drilling operations, significantly improving operation safety and efficiency.

[0104] Compared with the related art, this embodiment has the following significant advantages:

[0105] Accuracy: By utilizing the self-attention mechanism of Swin Transformer, it is able to capture the complex relationships between multi-modal data, enhancing the accuracy of anomaly detection.

[0106] Efficiency: Swin Transformer performs excellently in processing large-scale, multi-modal data, capable of quickly identifying abnormal working conditions and reducing the computational overhead of traditional methods.

[0107] Real-time performance: This embodiment can process and analyze data in real time, promptly detect abnormal working conditions, and ensure the safety and efficiency of operations.

[0108] This embodiment can address the limitations of related technologies under complex working conditions, effectively identify abnormal patterns in multi-modal vibration data, thereby providing reliable decision-making support for drilling operations and significantly improving operation safety and efficiency.

[0109] It should be noted that this embodiment can utilize the Swin Transformer model to process multi-modal vibration data, dynamically adjust detection labels, and combine various communication and data processing methods to accurately identify abnormal working conditions, reduce false alarms and missed detections, enhance the safety and efficiency of drilling operations, and form a complete and efficient technical process.

[0110] This embodiment can solve the limitations of related technologies regarding data requirements, flexibly adjust the compression ratio according to user needs, and achieve efficient compression of data. Meanwhile, it is not affected by the data distribution situation and can compress data at a stable compression ratio. Through the transmission of compressed data and the restoration of data at the receiving end, this embodiment can conduct in-depth analysis of downhole conditions in real time, provide strong support for decision-makers, and thus make more favorable decisions. This compression method not only improves the efficiency of drilling data processing but also makes the management of the oil and gas industry more efficient and scientific.

[0111] As Figure 3 shown, this embodiment proposes a drilling abnormal vibration condition detection device. The device is implemented using a trained abnormal vibration condition detection model, and the model is obtained by training a pre-trained detection model with sample data labeled with vibration condition labels; the model includes a multi-head self-attention layer, a feed-forward network layer, and a fully connected layer.

[0112] The device includes:

[0113] A preprocessing unit 301, configured to preprocess the multi-modal vibration data collected during drilling operations respectively to obtain processed multi-modal vibration data;

[0114] A first input unit 302, configured to input the processed multi-modal vibration data into the multi-head self-attention layer for local feature extraction, normalization, and weight assignment to obtain a multi-modal feature vector;

[0115] A second input unit 303 is configured to input the multi-modal feature vectors into a feed-forward network layer and a fully-connected layer for label prediction, so as to obtain a detection result of an abnormal vibration condition corresponding to a drilling operation.

[0116] It should be noted that for the processing procedures of the preprocessing unit 301, the first input unit 302, and the second input unit 303 and the beneficial effects brought thereby, reference can be respectively made to Figure 1 Steps S101 to S103 therein, which will not be elaborated herein.

[0117] Optionally, the multi-modal vibration data includes at least two of time-domain vibration data, frequency-domain vibration data, and image vibration data;

[0118] When the multi-modal vibration data includes time-domain vibration data, frequency-domain vibration data, and image vibration data, the processed multi-modal vibration data includes processed time-domain vibration data, processed frequency-domain vibration data, and processed image vibration data, and the multi-modal feature vectors include time-series feature vectors, frequency-domain feature vectors, and spatial feature vectors.

[0119] Optionally, when the multi-modal vibration data includes time-domain vibration data, frequency-domain vibration data, and image vibration data, the multi-head self-attention layer includes a first local self-attention module, a second local self-attention module, a third local self-attention module, a normalization module, and a global self-attention module;

[0120] The first input unit 302 is further configured to:

[0121] Input the processed time-domain vibration data, the processed frequency-domain vibration data, and the processed image vibration data into the first local self-attention module, the second local self-attention module, and the third local self-attention module respectively for local feature extraction, so as to obtain an initial time feature vector, an initial frequency feature vector, and an initial spatial feature vector output by the first local self-attention module, the second local self-attention module, and the third local self-attention module respectively;

[0122] Input the initial time feature vector, the initial frequency feature vector, and the initial spatial feature vector into the normalization module and the global self-attention module for normalization and weight assignment, so as to obtain a time feature vector, a frequency feature vector, and a spatial feature vector.

[0123] Optionally, the first input unit 302 is further configured to:

[0124] Input the initial time feature vector, the initial frequency feature vector, and the initial spatial feature vector into the normalization module respectively for normalization, so as to obtain a normalized time feature vector, a normalized frequency feature vector, and a normalized spatial feature vector generated and output by the normalization module;

[0125] Input the normalized time feature vector, the normalized frequency feature vector, and the normalized spatial feature vector into the global self-attention module, so that the global self-attention module: perform a weight operation based on the normalized time feature vector, the normalized frequency feature vector, and the normalized spatial feature vector to determine the weight coefficients of the normalized time feature vector, the normalized frequency feature vector, and the normalized spatial feature vector, and multiply the normalized time feature vector, the normalized frequency feature vector, and the normalized spatial feature vector by the corresponding weight coefficients respectively to obtain and output the time feature vector, the frequency feature vector, and the spatial feature vector.

[0126] Optionally, the feed-forward network layer includes a feature aggregation module and an activation function; when the multi-modal vibration data includes time-domain vibration data, frequency-domain vibration data, and image vibration data, the second input unit 303 is further configured to:

[0127] Input the time feature vector, the frequency feature vector, and the spatial feature vector into the feature aggregation module for feature aggregation to obtain the initial fusion feature vector generated and output by the feature aggregation module;

[0128] Input the initial fusion feature vector into the activation function for non-linear transformation to obtain the fusion feature vector generated and output by the activation function;

[0129] Input the fusion feature vector into the fully-connected layer for label prediction to obtain the detection result of the abnormal vibration condition corresponding to the drilling operation.

[0130] Optionally, when the multi-modal vibration data includes time-domain vibration data, frequency-domain vibration data, and image vibration data, the preprocessing unit 301 is further configured to:

[0131] Perform denoising on the time-domain vibration data to obtain the denoised vibration data; perform normalization on the denoised vibration data to obtain the normalized vibration data; perform time-frequency domain feature extraction on the normalized vibration data to obtain the processed time-domain vibration data;

[0132] Eliminate the high-frequency fluctuations in the processed frequency-domain vibration data to obtain the vibration data after elimination; identify the key frequency components in the vibration data after elimination based on frequency band division and dominant frequency extraction to obtain the vibration data after identification; remove the background noise in the vibration data after identification to extract the effective spectral features in the vibration data after identification to obtain the processed frequency-domain vibration data;

[0133] Filter and denoise the image vibration data to eliminate noise interference and obtain the filtered vibration data; perform image enhancement on the filtered vibration data to obtain the enhanced vibration data; extract the region of interest in the enhanced vibration data to obtain the vibration data of interest; perform motion compensation on the vibration data of interest to eliminate the influence of camera jitter or object displacement and obtain the compensated vibration data; perform feature extraction on the compensated vibration data to obtain the processed image vibration data.

[0134] Optionally, the vibration working condition label is a normal working condition or an abnormal working condition, and the abnormal working condition includes at least one of bit bouncing, stick-slip, and whirling.

[0135] The drilling abnormal vibration working condition detection device proposed in this embodiment can be implemented using a trained abnormal vibration working condition detection model. The model includes a multi-head self-attention layer, a feed-forward network layer, and a fully-connected layer. In this embodiment, multi-modal vibration data can be collected during drilling operations and input into the trained abnormal vibration working condition detection model. The multi-head self-attention layer efficiently captures the potential long-range dependence relationships between different modal vibration data in the multi-modal vibration data through the self-attention mechanism, generates a multi-modal feature vector, and the feed-forward network layer and the fully-connected layer perform label detection based on the multi-modal feature vector output by the multi-head self-attention layer, thereby realizing the detection of drilling abnormal vibration working conditions and obtaining the detection result of abnormal vibration working conditions. This embodiment can enrich the detection means for abnormal vibration working conditions during drilling operations and effectively improve the accuracy of detecting abnormal vibration working conditions.

[0136] The drilling abnormal vibration working condition detection device in this embodiment is presented in the form of functional units. Here, the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and a memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0137] The embodiment of the present invention also provides a computer device having the above Figure 3 shown drilling abnormal vibration working condition detection device.

[0138] Please refer to Figure 4, A schematic structural diagram of a computer device provided by an alternative embodiment of the present invention. The computer device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Each component communicates with each other using different buses and can be installed on a common motherboard or installed in other ways as needed. The processor can process instructions executed within the computer device, including instructions stored in the memory or on the memory to display graphical information of the GUI on an external input / output device (such as a display device coupled to the interface). In some alternative embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories. Similarly, multiple computer devices can be connected, and each device provides some necessary operations (for example, as a server array, a set of blade servers, or a multi-processor system). Figure 4 Taking one processor 10 as an example.

[0139] The processor 10 can be a central processing unit, a network processor, or a combination thereof. Among them, the processor 10 can further include a hardware chip. The above hardware chip can be an application-specific integrated circuit, a programmable logic device, or a combination thereof. The above programmable logic device can be a complex programmable logic device, a field programmable gate array, a general array logic, or any combination thereof.

[0140] Among them, the memory 20 stores instructions executable by at least one processor 10, so that at least one processor 10 executes the method shown in the above embodiments.

[0141] The memory 20 can include a program storage area and a data storage area. Among them, the program storage area can store an operating system and application programs required for at least one function. The data storage area can store data created according to the use of the computer device. In addition, the memory 20 can include a high-speed random access memory, and can also include a non-transitory memory, such as at least one disk storage device, a flash memory device, or other non-transitory solid-state storage devices. In some alternative embodiments, the memory 20 can optionally include a memory remotely set relative to the processor 10, and these remote memories can be connected to the computer device through a network. Examples of the above network include but are not limited to the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof.

[0142] The memory 20 can include a volatile memory, such as a random access memory. The memory can also include a non-volatile memory, such as a flash memory, a hard disk, or a solid-state drive. The memory 20 can also include a combination of the above types of memories.

[0143] The computer device further includes a communication interface 30 for the computer device to communicate with other devices or communication networks.

[0144] Embodiments of the present invention also provide a computer-readable storage medium. The method according to the embodiments of the present invention can be implemented in hardware, firmware, or be implemented as computer code that can be recorded on a storage medium, or be implemented as computer code that is originally stored in a remote storage medium or a non-transitory machine-readable storage medium and downloaded through a network and will be stored in a local storage medium, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only memory, a random access memory, a flash memory, a hard disk, or a solid-state drive, etc.; further, the storage medium can also include a combination of the above-mentioned types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by the computer, the processor, or the hardware, the method shown in the above embodiments is implemented.

[0145] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting abnormal vibration conditions in drilling, characterized in that: The method is implemented by using a trained abnormal vibration condition detection model, wherein the model is obtained by training a pre-trained detection model using sample data annotated with vibration condition labels; the model includes a multi-head self-attention layer, a feedforward network layer and a fully connected layer; The method comprises: Preprocessing the multi-modal vibration data collected during the drilling operation to obtain processed multi-modal vibration data; Inputting the processed multimodal vibration data into the multi-head self-attention layer for local feature extraction, normalization and weight assignment to obtain a multimodal feature vector; The multimodal feature vector is input into the feedforward network layer and the fully connected layer for label prediction to obtain the abnormal vibration condition detection result corresponding to the drilling operation.

2. The method according to claim 1, characterized in that The multimodal vibration data includes at least two of time domain vibration data, frequency domain vibration data and image vibration data; When the multimodal vibration data includes time domain vibration data, frequency domain vibration data and image vibration data, the processed multimodal vibration data includes processed time domain vibration data, processed frequency domain vibration data and processed image vibration data, and the multimodal feature vector includes a time series feature vector, a frequency domain feature vector and a spatial feature vector.

3. The method according to claim 2, characterized in that When the multimodal vibration data includes time domain vibration data, frequency domain vibration data and image vibration data, the multi-head self-attention layer includes a first local self-attention module, a second local self-attention module, a third local self-attention module, a normalization module and a global self-attention module; The processed multimodal vibration data is input into the multi-head self-attention layer for local feature extraction, normalization and weight allocation to obtain a multimodal feature vector, including: Inputting the processed time-domain vibration data, the processed frequency-domain vibration data, and the processed image vibration data into the first local self-attention module, the second local self-attention module, and the third local self-attention module, respectively, to perform local feature extraction, and obtaining initial time feature vectors, initial frequency feature vectors, and initial space feature vectors outputted by the first local self-attention module, the second local self-attention module, and the third local self-attention module, respectively; The initial time feature vector, the initial frequency feature vector and the initial space feature vector are input into the normalization module and the global self-attention module for normalization and weight allocation to obtain the time feature vector, the frequency feature vector and the space feature vector.

4. The method according to claim 3, characterized in that The step of inputting the initial time feature vector, the initial frequency feature vector, and the initial space feature vector into the normalization module and the global self-attention module for normalization and weight allocation to obtain the time feature vector, the frequency feature vector, and the space feature vector includes: Inputting the initial time feature vector, the initial frequency feature vector and the initial space feature vector into the normalization module for normalization respectively, to obtain a normalized time feature vector, a normalized frequency feature vector and a normalized space feature vector generated and output by the normalization module; The normalized time feature vector, the normalized frequency feature vector and the normalized space feature vector are input into the global self-attention module, so that the global self-attention module: performs weight operation based on the normalized time feature vector, the normalized frequency feature vector and the normalized space feature vector, determines the weight coefficients of the normalized time feature vector, the normalized frequency feature vector and the normalized space feature vector, multiplies the normalized time feature vector, the normalized frequency feature vector and the normalized space feature vector by the corresponding weight coefficients, respectively, obtains and outputs the time feature vector, the frequency feature vector and the space feature vector.

5. The method according to claim 2, characterized in that: The feedforward network layer includes a feature aggregation module and an activation function; when the multimodal vibration data includes time domain vibration data, frequency domain vibration data and image vibration data, the multimodal feature vector is input into the feedforward network layer and the fully connected layer for label prediction to obtain the abnormal vibration condition detection result corresponding to the drilling operation, including: Inputting the time feature vector, the frequency feature vector and the space feature vector into the feature aggregation module for feature aggregation to obtain an initial fusion feature vector generated and output by the feature aggregation module; Inputting the initial fused feature vector into the activation function for nonlinear transformation to obtain the fused feature vector generated and output by the activation function; The fused feature vector is input into the label prediction in the fully connected layer to obtain the abnormal vibration condition detection result corresponding to the drilling operation.

6. The method according to claim 2, characterized in that When the multimodal vibration data includes time domain vibration data, frequency domain vibration data and image vibration data, the multimodal vibration data collected during the drilling operation are preprocessed to obtain processed multimodal vibration data, including: Denoising the time-domain vibration data to obtain denoised vibration data; normalizing the denoised vibration data to obtain normalized vibration data; extracting time-frequency domain features from the normalized vibration data to obtain the processed time-domain vibration data; Eliminating high-frequency fluctuations in the processed frequency-domain vibration data to obtain eliminated vibration data; identifying key frequency components in the eliminated vibration data based on frequency band division and main frequency extraction to obtain identified vibration data; removing background noise in the identified vibration data to extract effective frequency spectrum features in the identified vibration data to obtain the processed frequency-domain vibration data; The image vibration data is filtered and denoised to eliminate noise interference to obtain filtered vibration data; the filtered vibration data is image enhanced to obtain enhanced vibration data; a region of interest is extracted from the enhanced vibration data to obtain vibration data of interest; motion compensation is performed on the vibration data of interest to eliminate the influence of camera jitter or object displacement to obtain compensated vibration data; and feature extraction is performed on the compensated vibration data to obtain the processed image vibration data.

7. The method according to any one of claims 1 to 6, characterized in that The vibration operating condition label is a normal operating condition or an abnormal operating condition, and the abnormal operating condition includes at least one of drill jumping, stick-slip and vortex.

8. A drilling abnormal vibration condition detection device, characterized in that: The device is implemented using a trained abnormal vibration condition detection model, which is obtained by training a pre-trained detection model using sample data marked with vibration condition labels; the model includes a multi-head self-attention layer, a feedforward network layer and a fully connected layer; The device comprises: A preprocessing unit, used to preprocess the multi-modal vibration data collected during the drilling operation to obtain processed multi-modal vibration data; A first input unit, used for inputting the processed multimodal vibration data into the multi-head self-attention layer for local feature extraction, normalization and weight allocation to obtain a multimodal feature vector; The second input unit is used to input the multimodal feature vector into the feedforward network layer and the fully connected layer for label prediction to obtain the abnormal vibration condition detection result corresponding to the drilling operation.

9. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the abnormal vibration condition detection method for drilling according to any one of claims 1 to 7 by executing the computer instructions.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the abnormal vibration condition detection method for drilling according to any one of claims 1 to 7.

Images