Drilling tool fault real-time prediction method and device based on deep learning

Abstract

The invention provides a drilling tool fault real-time prediction method and device based on deep learning, and relates to the technical field of drilling tool fault real-time prediction.Real-time parameter data of drilling tool work is monitored in real time, a deep learning network model used for drilling tool prediction is established, fault time and fault features are predicted through the deep learning network model, and the drilling tool fault real-time prediction method and device based on deep learning are obtained. The method can find the sign that the drilling tool may have a fault in advance, thereby taking maintenance measures in time, reducing the safety risk of the drilling tool, predicting the residual life of the drilling tool by predicting the fault time, improving the prediction precision, helping an engineer to formulate a more effective drilling tool maintenance plan, avoiding the condition of excessive maintenance or untimely maintenance, and improving the maintenance efficiency of the drilling tool. Therefore, the resource utilization efficiency is optimized, intelligent decision support can be provided based on a deep learning network model and real-time data analysis, and a management layer is helped to make a more accurate and effective decision.

Description

Technical Field

[0001] This invention relates to the field of real-time drill string failure prediction technology, and specifically to a method and apparatus for real-time drill string failure prediction based on deep learning. Background Technology

[0002] Drilling is a crucial step in the exploration and development of oil and gas resources. Statistics show that drilling costs account for 50% to 80% of the total cost of oil and gas exploration and development. With the increasing demand for oil and gas resources, drilling projects are increasingly focused on deep and ultra-deep oil and gas resources, and drilling costs will continue to rise. Therefore, reducing drilling costs is a perpetual theme in modern drilling engineering. However, downhole accidents significantly reduce drilling efficiency, prolong drilling cycles, and increase drilling costs, with losses from drill string accidents being particularly prominent.

[0003] One of the main functions of drilling tools is to transmit drilling pressure and torque. Therefore, drilling tools are subjected to complex compressive, tensile, bending, and torsional loads. These loads are dynamic, causing the downhole drilling tools to vibrate in the longitudinal, lateral, and torsional directions. The combined stress under the dynamic and periodic loads causes fatigue cracks to initiate at stress concentration points. As the load continues, these cracks propagate radially and circumferentially along the axis perpendicular to the drill string, leading to drill string fatigue failure. Once the drilling tools break and fall into the well, due to their offset from the wellbore and difficulty in centering, retrieval is extremely difficult, greatly hindering drilling operations and causing huge losses to oilfield companies.

[0004] Existing technologies construct lifespan models using data from monitored equipment and determine the equipment's condition and lifespan based on the model results.

[0005] For example, Chinese patent document, publication number CN114462662A, publication date May 10, 2022, entitled "An Invention Patent for a Big Data Prediction and Analysis Method for Drilling Tool Life," utilizes existing drilling reports to build a database, employs a fully interconnected feedforward hidden layer network (i.e., backpropagation learning rules), and establishes a big data prediction model for drilling tool life based on drilling parameters. This model automatically tracks parameters of interest to the drilling tool, fits existing theoretical models and existing field experience data, and predicts tool life based on reliable actual data.

[0006] However, the existing technology disclosed above, with publication number CN114462662A, still has shortcomings. This existing technology uses big data to predict information and obtain the approximate lifespan of tools under different conditions. However, the accuracy of the lifespan prediction is insufficient, and it cannot achieve real-time prediction of drill string failures and lifespan. Summary of the Invention

[0007] To address the problems in the prior art, this invention monitors real-time parameter data of the drilling tool and establishes a deep learning network model for drilling tool prediction. By predicting failure time and characteristics through the deep learning network model, it can detect signs of potential drilling tool failure in advance, thereby enabling timely maintenance measures and reducing drilling tool safety risks. Furthermore, by predicting the failure time, the remaining lifespan of the drilling tool is predicted, resulting in improved prediction accuracy. This helps engineers develop more effective drilling tool maintenance plans, avoiding over-maintenance or untimely maintenance, thus optimizing resource utilization efficiency. Based on the deep learning network model and real-time data analysis, it can provide intelligent decision support, helping management make more accurate and effective decisions.

[0008] This invention is achieved through the following technical solution:

[0009] This invention provides a method for real-time prediction of drill string faults based on deep learning, comprising the following steps:

[0010] S1. Obtain the attribute data of the target drill string and collect the environmental data when the target drill string is working;

[0011] S2. Based on the attribute data of the target drilling tool, obtain historical failure data and historical environmental data of drilling tools that are the same as or similar to the attribute data of the target drilling tool, and conduct failure feature data analysis based on the historical failure data and historical environmental data to establish a deep learning network model.

[0012] S3. Monitor the real-time parameter data of the target drilling tool, then perform fault correlation analysis on the real-time parameter data, extract the time history data of the changes from the real-time parameter data based on the results of the fault correlation analysis, and preprocess the time history data of the changes.

[0013] S4. Input the target drill string attribute data and environmental data obtained in step S1 into the deep learning network model established in step S2 for offline training to obtain the predicted failure time and failure feature data for drill string failure simulation. Then, input the change time history data preprocessed in step S3 into the deep learning network model obtained from offline training for data correction to obtain new predicted failure time and failure feature data.

[0014] S5. Based on step S4, establish a fault hazard threshold according to the new fault characteristic data, compare the fault hazard threshold with the new fault characteristic data, and make a fault hazard judgment.

[0015] S6. Based on the change time history data after preprocessing in step S3 and the new predicted failure time in step S4, predict the remaining life of the target drill bit.

[0016] Furthermore, in step S1, an information collection module is established at the working end of the drill string. The information collection module collects the type and size of the target drill string as attribute data. Geological data of the drill string operation is input at the working end of the drill string as environmental data.

[0017] Furthermore, the drill string types include drill pipe, weighted drill pipe, drill collars, and drill bits; the geological data include geological blocks, wellbore dimensions, and mud properties.

[0018] Furthermore, step S2 is as follows:

[0019] S2.1. Collect working data of the same attribute of the drill bit in the working end of the drill bit according to the attribute data, then extract historical failure data from its working data, and then extract historical environmental data recorded at the same time according to the historical failure data.

[0020] S2.2. Combine historical failure data with historical environmental data to perform failure feature data analysis, obtain failure feature data that represents the drilling tool's inability to continue to be used under each type of environmental data, and then use historical failure data, historical environmental data, and failure feature data as basic data to establish a deep learning network model for drilling tool prediction.

[0021] Furthermore, in step S2.1, working data of drilling tools with the same attributes are collected, including the model and material of the drilling tool.

[0022] Furthermore, historical failure data of the drill bit is extracted from the collected working data, namely failure events that render the drill bit unusable. Failures include breakage, wear, or plastic deformation of the drill bit.

[0023] Furthermore, the historical environmental data is the environmental data extracted at the time of each failure event that caused the drill bit to become unusable.

[0024] Furthermore, the failure characteristic data includes failure mode characteristics, environmental factor characteristics, operating parameter characteristics, and usage history characteristics; failure mode characteristics include failure type and failure mechanism; environmental factor characteristics include temperature, pressure, wellbore size, and geological characteristics; operating parameter characteristics include rotational speed, drilling pressure, displacement, and vibration characteristics; and usage history characteristics include operation cycle and maintenance records.

[0025] Furthermore, step S2.2, establishing the deep learning network model, includes: combining historical environmental data and failure feature data to construct feature vectors or feature tensors for inputting the deep learning network model; using labeled historical failure data as the target and the feature vectors or feature tensors as input to train the deep learning network model, as shown in the following formula:

[0026] X i =[Ei ,F i ];

[0027] Among them, X i Including historical environmental data and failure characteristic data, E i F is the environmental data at the time of the i-th failure event. i These are failure characteristic data related to the failure event;

[0028]

[0029] in, The model is the predicted value, and the deep learning function receives the input X. i And process it.

[0030] Furthermore, in step S3, a drill bit sensor is installed at the working end of the drill bit to monitor the real-time parameter data of the target drill bit during operation, and the real-time parameter data fed back by the drill bit sensor is extracted, including drilling pressure, rotation speed, stand pressure, torque, displacement, inlet and outlet density and gas measurement value. Through fault correlation analysis, only the drilling pressure, rotation speed, stand pressure and torque parameters are retained as change time history data.

[0031] Furthermore, fault correlation analysis specifically refers to performing fault correlation analysis on each parameter in the real-time parameter data to determine its degree of influence on key indicators during the drilling process. The formula is as follows:

[0032]

[0033] Where, ρ X,Y It is the covariance of X and Y, σ X and σ Y It represents the standard deviation of X and Y; key indicators include drill bit operating status, wellbore condition, and ρ. X,Y The value of ρ is in the range of [-1, 1]. X,Y =1 indicates a perfect positive correlation, where X and Y change in perfect synchronization, ρ X,Y =-1 indicates a perfect negative correlation; as X increases, Y decreases. ρ X,Y =0 indicates no linear dependence, when ρ X,Y If the absolute value is greater than 0.7, the parameter is determined to be closely related to the key indicators. Based on the results of the fault correlation analysis, the parameter closely related to the changes in the key indicators is selected as the important time history data of the change.

[0034] Furthermore, step S3 involves preprocessing the time-history data, specifically by processing the time-history data using a continuous wavelet transform method, and then labeling and segmenting the processed time-history data as data samples.

[0035] Furthermore, the time-history data undergoes preprocessing, specifically including the following steps:

[0036] First, for each parameter in the changing time history data, apply continuous wavelet transform and extract features from the results of the continuous wavelet transform;

[0037]

[0038] Where x(t) is the original signal, ψ is the complex conjugate of the wavelet function, a is the scale parameter, b is the time offset, and t is the time variable; for each chosen a and b, the wavelet function is scaled and translated, and then the inner product operation is performed with the signal x(t). This integration operation represents how the wavelet function matches the signal x(t) at different times b and scales a.

[0039] Then, the processed time-history data is labeled as data samples. Based on the labels, the data samples are divided into training set, validation set and test set for subsequent model training and evaluation.

[0040] Furthermore, in step S4, the formula is as follows:

[0041]

[0042] Where, x t This represents the attribute data and environmental data at time t. It is data that predicts failure time and failure characteristics;

[0043]

[0044] Where x't represents the real-time time history data at time t, It is the predicted fault time and fault characteristic data from the previous moment. It is a new predicted fault time and fault characteristic data corrected based on real-time changing time history data x't.

[0045] Furthermore, step S5, the fault hazard assessment, includes:

[0046] Drill string failure characteristic data includes information from multiple dimensions, providing... This represents the fault characteristic data of the i-th dimension, and the fault hazard threshold is Threshold = Threshold (1) Threshold (2) Threshold (i) The formula is as follows:

[0047]

[0048] in, This represents the relative deviation between the i-th dimension fault feature and the threshold, when If the i-th dimension of the fault characteristic exceeds the set fault danger threshold, then the drill string is determined to be a dangerous drill string.

[0049] Furthermore, in step S6, the remaining life of the target drill string is predicted using the following formula:

[0050]

[0051] Where, x t This is the current time-history data, T remaining The remaining lifetime is f(t'|x) = t / (t'|x) * ... t () is the change in time history data x at a given current moment. t Under the given conditions, the probability density function of the drill string failure time can be used to obtain the predicted remaining life T by integrating over t'. remaining .

[0052] A second aspect of the present invention provides a real-time drilling tool fault prediction device based on deep learning, comprising:

[0053] The first module is used to acquire the attribute data of the target drilling tool and collect the environmental data when the target drilling tool is working; based on the attribute data of the target drilling tool, it acquires the historical failure data and historical environmental data of the drilling tool that are the same as or similar to the attribute data of the target drilling tool.

[0054] The second module is used to monitor real-time parameter data of the target drilling tool's operation;

[0055] The third module is used to extract real-time parameter data from the second module, perform fault correlation analysis on the real-time parameter data, extract change time history data from the real-time parameter data based on the fault correlation analysis results, and preprocess the change time history data.

[0056] The fourth module is used to analyze failure characteristic data based on historical failure data and historical environmental data, thereby establishing a deep learning network model. The target drill string attribute data and environmental data obtained from the first module are input into the established deep learning network model for offline training to obtain predicted failure time and failure characteristic data for drill string failure simulation. Then, the change time history data preprocessed from the third module is input into the deep learning network model obtained from offline training for data correction to obtain new predicted failure time and failure characteristic data. A failure risk threshold is set based on the new failure characteristic data, and the failure risk threshold is compared with the new failure characteristic data to determine the failure risk. The remaining life of the target drill string is predicted based on the change time history data preprocessed from the third module combined with the new predicted failure time.

[0057] A third aspect of the present invention provides a computer device including a processor, an input device, an output device, and a memory, wherein the processor, the input device, the output device, and the memory are interconnected, wherein the memory is used to store a computer program, the computer program including program instructions, and the processor is configured to invoke the program instructions to perform some or all of the steps described in the first aspect of the present invention.

[0058] A fourth aspect of the present invention provides a computer-readable storage medium, characterized in that: the computer-readable storage medium stores a computer program, the computer program including program instructions, which, when executed by a processor, cause the processor to perform some or all of the steps described in the first aspect of the present invention.

[0059] The beneficial effects of this invention are as follows:

[0060] 1. This invention extracts historical failure data and historical environmental data from drilling tool operating data to obtain failure characteristic data representing the unusability of the drilling tool under each environmental condition. Then, using the historical failure data, historical environmental data, and failure characteristic data as foundational data, a deep learning network model for drilling tool prediction is established. This model monitors real-time parameters of the drilling tool's operation and predicts failure times and characteristics through the deep learning network model. This allows for early detection of potential drilling tool failures, enabling timely maintenance measures and reducing drilling tool safety risks.

[0061] 2. This invention compares the fault hazard threshold with new fault characteristic data to determine the fault hazard of the target drilling tool and accurately determine whether the target drilling tool is a dangerous drilling tool.

[0062] 3. This invention predicts the remaining life of the target drilling tool by predicting the failure time, which improves the prediction accuracy. Predicting the remaining life of the drilling tool can help engineers develop more effective drilling tool maintenance plans, avoid over-maintenance or untimely maintenance, thereby optimizing resource utilization efficiency. Based on deep learning models and real-time data analysis, it can provide intelligent decision support to help management make more accurate and effective decisions. Attached Figure Description

[0063] Figure 1 This is an overall flowchart of the present invention;

[0064] Figure 2 This is an example of an implementation of the present invention. Detailed Implementation

[0065] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0066] Example 1

[0067] like Figure 1 As shown, a real-time prediction method for drill string faults based on deep learning includes the following steps:

[0068] S1. Obtain the attribute data of the target drill string and collect the environmental data when the target drill string is working;

[0069] S2. Based on the attribute data of the target drilling tool, obtain historical failure data and historical environmental data of drilling tools that are the same as or similar to the attribute data of the target drilling tool, and conduct failure feature data analysis based on the historical failure data and historical environmental data to establish a deep learning network model.

[0070] S3. Monitor the real-time parameter data of the target drilling tool, then perform fault correlation analysis on the real-time parameter data, extract the time history data of the changes from the real-time parameter data based on the results of the fault correlation analysis, and preprocess the time history data of the changes.

[0071] S4. Input the target drill string attribute data and environmental data obtained in step S1 into the deep learning network model established in step S2 for offline training to obtain the predicted failure time and failure feature data for drill string failure simulation. Then, input the change time history data preprocessed in step S3 into the deep learning network model obtained from offline training for data correction to obtain new predicted failure time and failure feature data.

[0072] S5. Based on step S4, establish a fault hazard threshold according to the new fault characteristic data, compare the fault hazard threshold with the new fault characteristic data, and make a fault hazard judgment.

[0073] S6. Based on the change time history data after preprocessing in step S3 and the new predicted failure time in step S4, predict the remaining life of the target drill bit.

[0074] Example 2

[0075] This embodiment further elaborates and supplements the implementation of the present invention based on Embodiment 1.

[0076] S1. Obtain the attribute data of the target drill string and collect the environmental data when the target drill string is working;

[0077] Step S1 involves establishing an information collection module at the drill string's working end, which collects the target drill string's type and size as attribute data; and inputting geological data from the drill string's operation at the drill string's working end as environmental data.

[0078] As an example, an information collection module is established at the working end of the drill string. The information collection module uses sensors to identify and record the type and size information of the drill string. This information can be codes, RFID tags, or other readable identifiers. The information collection module is not limited to using sensors; any other device that can identify and record the type and size information of the drill string can be used as an information collection module.

[0079] As an example, the drilling tool working end is designed with an input module, which is a touch screen input, allowing users or operators to input the geological data of the current drilling tool operation as environmental data; the input module can also adopt voice input, allowing users or operators to input the current geological data of the drilling tool as environmental data into the drilling tool by voice.

[0080] Drilling tool types include drill pipe, weighted drill pipe, drill collars, and drill bits, each with different strength and stress characteristics; geological data includes geological blocks, wellbore size, and mud properties; due to different formation conditions, drilling operation parameters vary considerably, therefore early warning work must be carried out based on different geological blocks; in different wellbore sizes, the clearance between the drill tool and the wellbore wall is different, and the stress characteristics of the drill tool vary considerably, therefore it is also necessary to distinguish between different wellbore sizes; mud properties affect the suspension gravity of the drill tool and also affect its mechanical properties, therefore it is also necessary to determine the mud properties.

[0081] Example 3

[0082] This embodiment further elaborates and supplements the implementation of the present invention based on Embodiment 1 or Embodiment 2.

[0083] S2. Based on the attribute data of the target drilling tool, obtain historical failure data and historical environmental data of drilling tools that are the same as or similar to the attribute data of the target drilling tool, and conduct failure feature data analysis based on the historical failure data and historical environmental data to establish a deep learning network model.

[0084] The steps in S2 are as follows:

[0085] S2.1. Collect working data of the same attribute of the drill bit in the working end of the drill bit according to the attribute data, then extract historical failure data from its working data, and then extract historical environmental data recorded at the same time according to the historical failure data.

[0086] In step S2.1, collect working data of drilling tools with the same attributes, including the model and material of the drilling tool.

[0087] S2.2. Combine historical failure data with historical environmental data to perform failure feature data analysis, obtain failure feature data that represents the drilling tool's inability to continue to be used under each type of environmental data, and then use historical failure data, historical environmental data, and failure feature data as basic data to establish a deep learning network model for drilling tool prediction.

[0088] Historical failure data of the drilling tools are extracted from the collected working data. These failures are events that render the drilling tools unusable, including breakage, wear, or plastic deformation of the drilling tools.

[0089] Historical environmental data is extracted for each failure event that renders the drill string unusable, showing the environmental data at the time of its occurrence.

[0090] Failure characteristic data includes failure mode characteristics, environmental factor characteristics, operating parameter characteristics, and usage history characteristics; failure mode characteristics include failure type and failure mechanism; environmental factor characteristics include temperature, pressure, wellbore size, and geological characteristics; operating parameter characteristics include rotational speed, drilling pressure, displacement, and vibration characteristics; and usage history characteristics include operation cycle and maintenance records.

[0091] Furthermore, step S2.2, establishing the deep learning network model, includes: combining historical environmental data and failure feature data to construct feature vectors or feature tensors for inputting the deep learning network model; using labeled historical failure data as the target and the feature vectors or feature tensors as input to train the deep learning network model, as shown in the following formula:

[0092] X i =[E i ,F i ];

[0093] Among them, X i Including historical environmental data and failure characteristic data, E i F is the environmental data at the time of the i-th failure event. i These are failure characteristic data related to the failure event;

[0094]

[0095] in, The model is the predicted value, and the deep learning function receives the input X. i And process it. is the label of the failure event (e.g., a binary label indicating whether a failure has occurred), and the goal of the deep learning network model is to learn to make the predicted value as close as possible to the true label.

[0096] Deep learning network models receive input and adjust their internal parameters through a learning process (training) so that they can predict an output. By learning the relationships between features, they can make accurate predictions.

[0097] In summary, the structure of the input data is provided, and the learning capabilities of the input and deep learning network model are used to generate the predicted output.

[0098] Example 4

[0099] This embodiment further elaborates and supplements the implementation of the present invention based on Embodiment 1, Embodiment 2 or Embodiment 3.

[0100] S3. Monitor the real-time parameter data of the target drilling tool, then perform fault correlation analysis on the real-time parameter data, extract the time history data of the changes from the real-time parameter data based on the results of the fault correlation analysis, and preprocess the time history data of the changes.

[0101] Step S3 involves monitoring real-time parameter data of the target drill bit during operation by installing drill bit sensors at the working end of the drill bit, and extracting real-time parameter data fed back by the drill bit sensors, including drill pressure, rotation speed, stand pressure, torque, displacement, inlet and outlet density, and gas measurement values. Through fault correlation analysis, only drill pressure, rotation speed, stand pressure, and torque parameters are retained as time history data of changes.

[0102] Fault correlation analysis specifically refers to performing fault correlation analysis on each parameter in the real-time parameter data to determine its degree of influence on key indicators during the drilling process. The formula is as follows:

[0103]

[0104] Where, ρ X,Y It is the covariance of X and Y, σ X and σ Y It represents the standard deviation of X and Y; key indicators include drill bit operating status, wellbore condition, and ρ. X,Y The value of ρ is in the range of [-1, 1]. X,Y =1 indicates a perfect positive correlation, where X and Y change in perfect synchronization, ρ X,Y =-1 indicates a perfect negative correlation; as X increases, Y decreases. ρ X,Y =0 indicates no linear dependence, when ρ X,Y If the absolute value is greater than 0.7, the parameter is determined to be closely related to the key indicators. Based on the results of the fault correlation analysis, the parameter closely related to the changes in the key indicators is selected as the important time history data of the change.

[0105] X and Y represent different variables, and their specific roles depend on the actual application scenario. For example, X is the machine's operating temperature, while Y is the machine's failure rate.

[0106] Step S3 involves preprocessing the time-history data. Specifically, this means processing the time-history data using the continuous wavelet transform method, and then labeling and segmenting the processed time-history data as data samples.

[0107] Furthermore, the time-history data undergoes preprocessing, specifically including the following steps:

[0108] First, for each parameter in the changing time history data, apply continuous wavelet transform and extract features from the results of the continuous wavelet transform. These features can be statistics of the wavelet coefficients (such as mean, variance), frequency band energy distribution, etc.

[0109]

[0110] Where x(t) is the original signal, ψ is the complex conjugate of the wavelet function, a is the scale parameter, b is the time offset, and t is the time variable; for each chosen a and b, the wavelet function is scaled (through a) and translated (through b) and then its inner product is performed with the signal x(t). This integration operation represents how the wavelet function matches the signal x(t) at different times b and scales a.

[0111] Then, the processed time-history data is labeled as data samples. Based on the labels, the data samples are divided into training set, validation set and test set for subsequent model training and evaluation.

[0112] The label can be a category (such as normal state, abnormal state) or other related tags.

[0113] Example 5

[0114] This embodiment further elaborates and supplements the implementation of the present invention based on Embodiment 1, Embodiment 2, Embodiment 3 or Embodiment 4.

[0115] S4. Input the target drill string attribute data and environmental data obtained in step S1 into the deep learning network model established in step S2 for offline training to obtain the predicted failure time and failure feature data for drill string failure simulation. Then, input the change time history data preprocessed in step S3 into the deep learning network model obtained from offline training for data correction to obtain new predicted failure time and failure feature data.

[0116] In step S4, the formula is as follows:

[0117]

[0118] Where, x t This represents the attribute data and environmental data at time t. It is data that predicts failure time and failure characteristics;

[0119] Deep learning network models learn patterns and features from data, such as the timing of drill string failures under different properties and environmental conditions. When new data is fed into the deep learning network model, it calculates and predicts the likely timing and characteristics of drill string failures within a future period. This prediction helps in taking preventative measures to reduce the impact of failures on operations.

[0120]

[0121] Where x't represents the real-time time history data at time t, It is the predicted fault time and fault characteristic data from the previous moment. It is a new predicted fault time and fault characteristic data corrected based on real-time changing time history data x't.

[0122] Real-time acquisition and preprocessing of x't data ensures data quality and integrity. Modelonline uses real-time x't data and previous prediction results... Update the current status information of the drill string, including adjusting the predicted failure time and updating possible failure characteristics, and output the updated status information. Right now Used to guide operations or make real-time decisions.

[0123] Example 6

[0124] This embodiment further elaborates and supplements the implementation of the present invention based on Embodiment 1, Embodiment 2, Embodiment 3, Embodiment 4 or Embodiment 5.

[0125] S5. Based on step S4, establish a fault hazard threshold according to the new fault characteristic data, compare the fault hazard threshold with the new fault characteristic data, and make a fault hazard judgment.

[0126] In this embodiment, a fault risk threshold is set for each type of fault characteristic data. This fault risk threshold can be based on statistical analysis of historical data or a fixed value set according to the experience of domain experts. For example, the maximum allowable time interval or the maximum allowable numerical range of a certain type of fault in the drilling tool can be set.

[0127] Step S5, fault hazard assessment includes:

[0128] Drill string failure characteristic data includes information from multiple dimensions, providing... This represents the fault characteristic data of the i-th dimension, and the fault hazard threshold is Threshold = Threshold (1) Threshold (2) Threshold (i) The formula is as follows:

[0129]

[0130] in, This represents the relative deviation between the i-th dimension fault feature and the threshold, when If the i-th dimension of the fault characteristic exceeds the set fault danger threshold, then the drill string is determined to be a dangerous drill string.

[0131] S6. Based on the change time history data after preprocessing in step S3 and the new predicted failure time in step S4, predict the remaining life of the target drill bit.

[0132] Step S6 predicts the remaining life of the target drill string using the following formula:

[0133]

[0134] Where, x t This is the current time-history data, T remaining The remaining lifetime is f(t'|x) = t / (t'|x) * ... t () is the change in time history data x at a given current moment. t Under the given conditions, the probability density function of the drill string failure time can be used to obtain the predicted remaining life T by integrating over t'. remaining The current time-history data serves as input to the deep learning network model, reflecting the current state of the drilling tool. The deep learning network model uses this data to estimate possible future failure times, thereby calculating the remaining life of the drilling tool.

[0135] To demonstrate the effectiveness of this method, this embodiment utilizes data from two wells in the Shuangyushi block of Sichuan province that experienced 8-inch drill collar fractures during drilling in a 444.5mm borehole. Both wells were drilled in the same block and geological formation, using the same drilling fluid properties and drill string type; therefore, their drill string properties and working environment were consistent. The drill string combination used in the two wells with a 444.5mm borehole was as follows: Φ444.5mm PDC drill bit * 0.6m + Φ228mm drill collar * 45.53m + Φ203mm drill collar * 26.76m + Φ178mm drill collar * 18.89m + Φ139.7mm drill pipe. First, this model was used to train a deep learning network model for the 444.5mm wellbore in the block using the service data of the 8-inch drill collar in Well #1. The training data included the time history data of drilling pressure, rotation speed, stand pressure, and torque throughout the entire process from drill collar entry to breakage. After obtaining the neural network model with fault time prediction and fault feature judgment, the time history data of drilling pressure, rotation speed, stand pressure, and torque of the broken drill collar in Well #2 were input into the model for tracking. Fault risk judgment was performed on the broken 8-inch drill collar, and the real-time monitoring parameter output graph is shown below. Figure 2 As shown.

[0136] Example 7

[0137] This embodiment discloses a real-time drill string fault prediction device based on deep learning, comprising:

[0138] The first module is used to acquire the attribute data of the target drilling tool and collect the environmental data when the target drilling tool is working; based on the attribute data of the target drilling tool, it acquires the historical failure data and historical environmental data of the drilling tool that are the same as or similar to the attribute data of the target drilling tool.

[0139] The second module is used to monitor real-time parameter data of the target drilling tool's operation;

[0140] The third module is used to extract real-time parameter data from the second module, perform fault correlation analysis on the real-time parameter data, extract change time history data from the real-time parameter data based on the fault correlation analysis results, and preprocess the change time history data.

[0141] The fourth module is used to analyze failure characteristic data based on historical failure data and historical environmental data, thereby establishing a deep learning network model. The target drill string attribute data and environmental data obtained from the first module are input into the established deep learning network model for offline training to obtain predicted failure time and failure characteristic data for drill string failure simulation. Then, the change time history data preprocessed from the third module is input into the deep learning network model obtained from offline training for data correction to obtain new predicted failure time and failure characteristic data. A failure risk threshold is set based on the new failure characteristic data, and the failure risk threshold is compared with the new failure characteristic data to determine the failure risk. The remaining life of the target drill string is predicted based on the change time history data preprocessed from the third module combined with the new predicted failure time.

[0142] Example 8

[0143] To achieve the above objectives, according to another aspect of this application, a computer device is also provided, including a processor, an input device, an output device, and a memory, wherein the processor, input device, output device, and memory are interconnected, wherein the memory is used to store a computer program, the computer program including program instructions, and the processor is configured to invoke the program instructions to perform some or all of the steps described in the first aspect of the present invention.

[0144] In this embodiment, the processor can be a central processing unit (CPU). The processor can also be 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, or combinations of the above types of chips.

[0145] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs, non-transitory computer-executable programs, and units, such as the program units corresponding to the above-described method embodiments of the present invention. The processor executes various functional applications and data processing of the processor by running the non-transitory software programs, instructions, and modules stored in the memory, thereby implementing the methods described in the above-described method embodiments.

[0146] The memory may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created by the processor, etc. Furthermore, the memory may include high-speed random access memory and non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, the memory may optionally include memory remotely located relative to the processor, which can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0147] The one or more units are stored in the memory, and when executed by the processor, they perform the methods in Embodiment 1, Embodiment 2, Embodiment 3, Embodiment 4, Embodiment 5 or Embodiment 6 described above.

[0148] Example 9

[0149] This embodiment discloses a computer-readable storage medium storing a computer program. The computer program includes program instructions, which, when executed by a processor, cause the processor to perform the steps in Embodiments 1, 2, 3, 4, 5, or 6 described above.

Claims

1. A real-time prediction method for drill string faults based on deep learning, characterized in that: Includes the following steps: S1. Obtain the attribute data of the target drill string and collect the environmental data when the target drill string is working; S2. Based on the attribute data of the target drilling tool, obtain historical failure data and historical environmental data of drilling tools that are the same as or similar to the attribute data of the target drilling tool, and conduct failure feature data analysis based on the historical failure data and historical environmental data to establish a deep learning network model. S3. Monitor the real-time parameter data of the target drilling tool, then perform fault correlation analysis on the real-time parameter data, extract the time history data of the changes from the real-time parameter data based on the results of the fault correlation analysis, and preprocess the time history data of the changes. S4. Input the target drill string attribute data and environmental data obtained in step S1 into the deep learning network model established in step S2 for offline training to obtain the predicted failure time and failure feature data for drill string failure simulation. Then, input the change time history data preprocessed in step S3 into the deep learning network model obtained from offline training for data correction to obtain new predicted failure time and failure feature data. S5. Based on step S4, establish a fault hazard threshold according to the new fault characteristic data, compare the fault hazard threshold with the new fault characteristic data, and make a fault hazard judgment. S6. Based on the change time history data after preprocessing in step S3 and the new predicted failure time in step S4, predict the remaining life of the target drill bit.

2. The real-time prediction method for drill string faults based on deep learning as described in claim 1, characterized in that: Step S1 involves establishing an information collection module at the working end of the drill string, collecting the type and size of the target drill string as attribute data, and inputting geological data of the drill string operation at the working end as environmental data.

3. The real-time prediction method for drill string faults based on deep learning as described in claim 2, characterized in that: The drilling tool types include drill pipe, weighted drill pipe, drill collars, and drill bits; the geological data include geological blocks, wellbore dimensions, and mud properties.

4. A real-time drilling tool fault prediction method based on deep learning as described in any one of claims 2, characterized in that: The S2 step is as follows: S2.

1. Collect working data of the same attribute of the drill bit in the working end of the drill bit according to the attribute data, then extract historical failure data from its working data, and then extract historical environmental data recorded at the same time according to the historical failure data. S2.

2. Combine historical failure data with historical environmental data to perform failure feature data analysis, obtain failure feature data that represents the drilling tool's inability to continue to be used under each type of environmental data, and then use historical failure data, historical environmental data, and failure feature data as basic data to establish a deep learning network model for drilling tool prediction.

5. The real-time prediction method for drill string faults based on deep learning as described in claim 4, characterized in that: In step S2.1, working data of drilling tools with the same attributes are collected, including the model and material of the drilling tool.

6. The real-time prediction method for drill string faults based on deep learning as described in claim 5, characterized in that: Historical failure data of the drilling tools are extracted from the collected working data. These failures are events that render the drilling tools unusable, including breakage, wear, or plastic deformation of the drilling tools.

7. The real-time prediction method for drill string faults based on deep learning as described in claim 6, characterized in that: The historical environmental data is the environmental data extracted at the time of each failure event that caused the drill string to become unusable.

8. The real-time prediction method for drill string faults based on deep learning as described in claim 7, characterized in that: Failure characteristic data includes failure mode characteristics, environmental factor characteristics, operating parameter characteristics, and usage history characteristics; failure mode characteristics include failure type and failure mechanism; environmental factor characteristics include temperature, pressure, wellbore size, and geological characteristics; operating parameter characteristics include rotational speed, drilling pressure, displacement, and vibration characteristics; and usage history characteristics include operation cycle and maintenance records.

9. The real-time prediction method for drill string faults based on deep learning as described in claim 8, characterized in that: Step S2.2, establishing the deep learning network model, includes: combining historical environmental data and failure feature data to construct feature vectors or feature tensors for inputting the deep learning network model; using labeled historical failure data as the target and the feature vectors or feature tensors as input to train the deep learning network model, as shown in the following formula: X i =[E i ,F i ]; Among them, X i Including historical environmental data and failure characteristic data, E i F is the environmental data at the time of the i-th failure event. i These are failure characteristic data related to the failure event; in, The model is the predicted value, and the deep learning function receives the input X. i And process it.

10. The method for real-time prediction of drill string faults based on deep learning as described in claim 1, characterized in that: In step S3, drill string sensors are installed at the working end of the drill string to monitor the real-time parameter data of the target drill string during operation, and the real-time parameter data fed back by the drill string sensors are extracted, including drilling pressure, rotation speed, stand pressure, torque, displacement, inlet and outlet density and gas measurement value. Through fault correlation analysis, only the parameters of drilling pressure, rotation speed, stand pressure and torque are retained as time history data of change.

11. The method for real-time prediction of drill string faults based on deep learning as described in claim 10, characterized in that: Fault correlation analysis specifically refers to performing fault correlation analysis on each parameter in the real-time parameter data to determine its degree of influence on key indicators during the drilling process. The formula is as follows: Where, ρ X,Y It is the covariance of X and Y, σ X and σ Y It represents the standard deviation of X and Y; key indicators include drill bit operating status, wellbore condition, and ρ. X,Y The value of ρ is in the range of [-1, 1]. X,Y =1 indicates a perfect positive correlation, where X and Y change in perfect synchronization, ρ X,Y =-1 indicates a perfect negative correlation; as X increases, Y decreases. ρ X,Y =0 indicates no linear dependence, when ρ X,Y If the absolute value is greater than 0.7, the parameter is determined to be closely related to the key indicators. Based on the results of the fault correlation analysis, the parameter closely related to the changes in the key indicators is selected as the important time history data of the change.

12. The real-time prediction method for drill string faults based on deep learning as described in claim 1, characterized in that: Step S3 involves preprocessing the time-history data, specifically by processing the time-history data using a continuous wavelet transform method, and then labeling and segmenting the processed time-history data as data samples.

13. The method for real-time prediction of drill string faults based on deep learning as described in claim 12, characterized in that: Preprocessing of time-history data includes the following steps: First, for each parameter in the changing time history data, apply continuous wavelet transform and extract features from the results of the continuous wavelet transform; Where x(t) is the original signal, ψ is the complex conjugate of the wavelet function, a is the scale parameter, b is the time offset, and t is the time variable; for each chosen a and b, the wavelet function is scaled and translated, and then the inner product operation is performed with the signal x(t). This integration operation represents how the wavelet function matches the signal x(t) at different times b and scales a. Then, the processed time-history data is labeled as data samples. Based on the labels, the data samples are divided into training set, validation set and test set for subsequent model training and evaluation.

14. The real-time prediction method for drill string faults based on deep learning as described in claim 1, characterized in that: In step S4, the formula is as follows: Where, x t This represents the attribute data and environmental data at time t. It is data that predicts failure time and failure characteristics; Where x't represents the real-time time history data at time t, It is the predicted fault time and fault characteristic data from the previous moment. It is a new predicted fault time and fault characteristic data corrected based on real-time changing time history data x't.

15. The method for real-time prediction of drill string faults based on deep learning as described in claim 1, characterized in that: Step S5, the fault hazard assessment includes: Drill string failure characteristic data includes information from multiple dimensions, providing... This represents the fault characteristic data of the i-th dimension, and the fault hazard threshold is Threshold = Threshold (1) Threshold (2) Threshold (i) The formula is as follows: in, This represents the relative deviation between the i-th dimension fault feature and the threshold, when If the i-th dimension of the fault characteristic exceeds the set fault danger threshold, then the drill string is determined to be a dangerous drill string.

16. The method for real-time prediction of drill string faults based on deep learning as described in claim 1, characterized in that: In step S6, the remaining life of the target drill string is predicted using the following formula: Where, x t This is the current time-history data, T remaining The remaining lifetime is f(t'|x) = t / (t'|x) * ... t () is the change in time history data x at a given current moment. t Under the given conditions, the probability density function of the drill string failure time can be used to obtain the predicted remaining life T by integrating over t'. remaining .

17. A real-time drilling tool fault prediction device based on deep learning, characterized in that: include: The first module is used to acquire the attribute data of the target drill string and collect environmental data when the target drill string is working. Based on the attribute data of the target drilling tool, obtain historical failure data and historical environmental data of drilling tools that are the same as or similar to the attribute data of the target drilling tool; The second module is used to monitor real-time parameter data of the target drilling tool's operation; The third module is used to extract real-time parameter data from the second module, perform fault correlation analysis on the real-time parameter data, extract change time history data from the real-time parameter data based on the fault correlation analysis results, and preprocess the change time history data. The fourth module is used to analyze failure feature data based on historical failure data and historical environmental data, thereby establishing a deep learning network model. The target drill string attribute data and environmental data obtained from the first module are input into the established deep learning network model for offline training to obtain the predicted failure time and failure feature data for drill string failure simulation. Then, the change time history data preprocessed from the third module is input into the deep learning network model obtained from offline training for data correction to obtain new predicted failure time and failure feature data. A fault risk threshold is established based on the new fault characteristic data. The fault risk threshold is compared with the new fault characteristic data to determine the fault risk. The remaining life of the target drill bit is predicted based on the change time history data after preprocessing in the third module and the new predicted fault time.

18. A computer device, characterized in that: The system includes a processor, an input device, an output device, and a memory, which are interconnected. The memory is used to store a computer program, which includes program instructions. The processor is configured to invoke the program instructions to execute the method as described in any one of claims 1-16.

19. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, the computer program including program instructions that, when executed by a processor, cause the processor to perform the method as described in any one of claims 1-16.

Images