Machine learning-based oil drilling and production fault diagnosis operation and maintenance system

Abstract

The application discloses a machine learning-based oil drilling and production fault diagnosis operation and maintenance system and belongs to the technical field of oil drilling and production fault diagnosis. The system comprises a data acquisition module used for acquiring multidimensional working condition data in the operation process of oil drilling and production equipment; a data preprocessing module used for cleaning, denoising, normalizing and feature extraction processing of the acquired multidimensional working condition data to obtain standardized feature data; a machine learning model module used for training and reasoning of the standardized feature data; a fault diagnosis module used for receiving the reasoning result of the machine learning model module; and a data storage and interaction module used for storing working condition data, preprocessed data, model parameters, fault diagnosis results and operation and maintenance data and providing an interactive interface of data query, parameter configuration and result display. Through intelligent, standardized and closed-loop design, the system effectively solves various pain points in the existing oil drilling and production fault diagnosis operation and maintenance, and takes into account safety, efficiency, economy and universality.

Description

Technical Field

[0001] This invention relates to the field of oil drilling and production fault diagnosis technology, and more specifically, to an oil drilling and production fault diagnosis and maintenance system based on machine learning. Background Technology

[0002] Currently, fault diagnosis and operation and maintenance management of oil drilling and production equipment mainly rely on two methods: First, the traditional manual diagnosis and maintenance method, where professional maintenance personnel identify faults and formulate maintenance plans through on-site inspections, equipment parameter observations, and experience-based judgment. This method heavily depends on the professional experience of maintenance personnel, is highly subjective, has low diagnostic efficiency, and struggles to achieve early warning and precise fault location. Furthermore, for dispersed drilling and production equipment, inspection costs are high and response times are slow, failing to meet the needs of large-scale, refined operation and maintenance. Second, conventional automated monitoring systems collect equipment operating parameters through sensors and combine them with simple threshold judgments to generate fault alarms. However, this system lacks the ability to deeply analyze multi-dimensional operating data, cannot effectively extract fault characteristics from the data, has low diagnostic accuracy, and is prone to false alarms and missed alarms. Moreover, it cannot automatically generate targeted maintenance plans based on fault type, making it difficult to form a closed-loop management system of "monitoring-diagnosis-operation and maintenance-optimization," resulting in low operation and maintenance efficiency.

[0003] With the rapid development of machine learning technology, its application in the field of industrial equipment fault diagnosis has gradually become widespread, providing a new technical path for fault diagnosis and maintenance of oil drilling and production equipment. In existing technologies, some solutions attempt to apply machine learning algorithms to oil drilling and production fault diagnosis, but there are still many shortcomings: On the one hand, the data preprocessing stage lacks standardized cleaning, denoising, and feature extraction strategies; the collected multi-dimensional operating condition data (such as equipment operating parameters, environmental parameters, and historical fault data) suffers from redundancy and noise interference, resulting in poor data quality input to the machine learning model and affecting the accuracy of fault diagnosis; on the other hand, the application of machine learning models is relatively singular, lacking a complete model training, inference, and optimization mechanism, and is not deeply integrated with the operation and maintenance scheduling stage. This makes it impossible to efficiently transform fault diagnosis results into targeted operation and maintenance solutions, nor can it achieve dynamic model optimization through operation and maintenance data feedback, making it difficult to realize intelligent and integrated fault diagnosis and operation and maintenance management.

[0004] In addition, existing drilling and production fault diagnosis and maintenance systems often suffer from functional fragmentation. Data acquisition, fault diagnosis, maintenance scheduling, and data storage are independent of each other, and data cannot be effectively shared. This results in insufficient timeliness of fault diagnosis and inadequate targeting of maintenance solutions. They are also unable to adapt to different types and specifications of oil drilling and production equipment, and have poor versatility.

[0005] To address the shortcomings of existing technologies, there is an urgent need for a highly versatile oil drilling and production fault diagnosis and maintenance system. This system can achieve accurate multi-dimensional data collection and preprocessing, enable precise fault diagnosis based on machine learning, form a closed loop of diagnosis, operation and maintenance, and optimization, and solve problems such as low diagnostic efficiency, poor accuracy, disconnect between operation and maintenance, and insufficient versatility in existing technologies. This system will ensure the safe and stable operation of oil drilling and production equipment and improve the operation and maintenance management level of the drilling and production industry. Therefore, we propose an oil drilling and production fault diagnosis and maintenance system based on machine learning. Summary of the Invention

[0006] The purpose of this invention is to provide a machine learning-based oil drilling and production fault diagnosis and maintenance system to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A machine learning-based oil drilling and production fault diagnosis and maintenance system includes: The data acquisition module is used to collect multi-dimensional operating condition data during the operation of oil drilling and production equipment. The operating condition data includes equipment operating parameters, environmental parameters, and historical fault data. The data preprocessing module, connected to the data acquisition module, is used to clean, denoise, normalize, and extract features from the acquired multi-dimensional working condition data to obtain standardized feature data. The machine learning model module is connected to the data preprocessing module. The machine learning model module is used to train and infer standardized feature data, and output the equipment fault type, fault level and fault location. The fault diagnosis module, connected to the machine learning model module, is used to receive the inference results from the machine learning model module and, combined with the operating logic of the oil drilling and production equipment, to complete the accurate judgment of faults and the early warning of anomalies. The operation and maintenance scheduling module is connected to the fault diagnosis module. It is used to generate targeted operation and maintenance plans and maintenance scheduling instructions based on the fault diagnosis results, and record operation and maintenance process data, which is then fed back to the machine learning model module for model optimization. The anomaly warning module is connected to the machine learning model module and the fault diagnosis module, and is used to issue warning signals based on the model inference results. The data storage and interaction module is used to store operating condition data, preprocessed data, model parameters, fault diagnosis results, and operation and maintenance data, and provides interactive interfaces for data query, parameter configuration, and result display.

[0008] Preferably, the data acquisition module includes a sensor unit and a data transmission unit. The sensor unit includes a temperature sensor, a pressure sensor, a vibration sensor, and a speed sensor, which are used to collect data on the temperature, pressure, vibration frequency, and operating speed of the drilling and production equipment, respectively. The data transmission unit adopts wired or wireless transmission and has NTP time synchronization, breakpoint resume, and data caching functions, and transmits the collected data to the data preprocessing module in real time.

[0009] Preferably, the feature extraction process of the data preprocessing module adopts time-domain feature extraction, frequency-domain feature extraction, or time-frequency-domain feature extraction methods. The extracted features include peak value, mean, variance, frequency spectrum peak value, and wavelet coefficients. The standardized feature data is obtained by processing the min-max normalization or Z-score normalization methods.

[0010] Preferably, the machine learning model module employs machine learning algorithms including Convolutional Neural Network (CNN), Long Short-Term Memory Network (LSTM), Random Forest, Support Vector Machine (SVM), or Gradient Boosting Tree (XGBoost). It can use one of these algorithms individually or combine multiple algorithms through weighted fusion and voting decision strategies to improve fault diagnosis accuracy. The machine learning model module includes a model training unit and a model inference unit. The model training unit trains the model using historical operating condition data and corresponding fault labels, while the model inference unit uses the trained model to infer real-time standardized feature data.

[0011] Preferably, the machine learning model module further includes a model optimization unit, which updates and optimizes the parameters of the trained machine learning model online based on the operation and maintenance process data and new fault data fed back by the operation and maintenance scheduling module, thereby improving the accuracy of fault diagnosis.

[0012] Preferably, the fault diagnosis module's fault determination includes fault type identification, fault level classification, and fault location positioning. The fault types include drill bit wear, pump body leakage, motor failure, and pipeline blockage. The fault levels are classified as minor faults, general faults, and severe faults. The fault location is accurately located through the correspondence between feature data and equipment components.

[0013] Preferably, the maintenance plan generated by the maintenance scheduling module includes fault handling priority, maintenance process, required spare parts, and maintenance personnel scheduling information. The maintenance plan can be dynamically adjusted according to the equipment operation priority and fault level. Maintenance scheduling instructions are sent to the on-site maintenance terminal through the interactive interface.

[0014] Preferably, the data storage and interaction module adopts a combination of relational databases and non-relational databases. The relational database is used to store structured fault tags and operation and maintenance record data, while the non-relational database is used to store massive amounts of raw operating data, preprocessed data, and model parameters. The interaction interface supports multi-terminal access from Web, mobile devices, and field control terminals.

[0015] Preferably, the anomaly warning module is used to issue warning signals through audible and visual alarms and message push when abnormal equipment operating parameters or potential faults are detected based on model inference results. The criteria for determining potential faults are: equipment operating parameters deviate from the rated threshold by 3%-5%, or the fault probability inferred by the machine learning model is not less than 30% and less than 50%. The criteria for determining minor faults are: equipment operating parameters deviate from the rated threshold by 5%-10%, and the fault probability inferred by the machine learning model is not less than 50% and less than 80%. The warning signal level corresponds to the fault level: potential faults correspond to Level 1 warning, minor faults correspond to Level 2 warning, general faults correspond to Level 3 warning, and serious faults correspond to Level 4 warning.

[0016] Preferably, the oil drilling and production equipment includes drilling equipment, oil production equipment, and oil pipeline equipment. The system can be adapted to different types and specifications of oil drilling and production equipment, and can achieve personalized adaptation for fault diagnosis and operation and maintenance through parameter configuration. The configuration parameters include sensor acquisition frequency, model inference threshold, and fault judgment standard parameters. The adaptation process is as follows: according to the type and specifications of the drilling and production equipment, the corresponding parameters are configured through the data storage and interaction module, and the system automatically matches the fault diagnosis logic and operation and maintenance strategy.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) This system achieves accurate and real-time acquisition of multi-dimensional working condition data through industrial-grade sensor units, coupled with standardized data preprocessing procedures (cleaning, noise reduction, feature filtering), effectively improving the data quality input to the machine learning model; adopting multiple algorithm options and fusion strategies, combined with the online model optimization mechanism, improves the accuracy of fault diagnosis, accurately identifies various faults such as drill wear and pump leakage, clarifies the fault level and location, and has high positioning accuracy with no significant error. At the same time, by setting clear warning thresholds through the abnormal warning module, it achieves early warning of potential faults and minor anomalies, with a warning response delay of ≤10s, which can avoid downtime accidents and safety accidents caused by equipment failure upgrades in advance, reduce economic losses and environmental risks, and ensure the safe and stable operation of oil drilling and production. Compared with traditional manual diagnosis, the diagnostic efficiency is improved, avoiding the subjectivity and lag of human experience judgment; compared with conventional automated monitoring systems, the false alarm and false alarm rates are greatly reduced, and the limitations of single threshold judgment are solved.

[0018] (2) The system breaks the current fragmented status quo of "monitoring-diagnosis-operation and maintenance" and forms a complete closed-loop management: After the fault diagnosis module outputs accurate fault information, the operation and maintenance scheduling module automatically generates standardized and targeted operation and maintenance plans, which clearly define the maintenance process, spare parts requirements, personnel scheduling and processing time limits. There is no need for manual plan formulation, and the plan generation time is ≤30s, which greatly reduces the operation and maintenance preparation time. Through multi-terminal interaction interfaces, operation and maintenance instructions can be quickly sent to the field terminals to achieve efficient scheduling of maintenance personnel. The response time for serious faults is shortened to within 30 minutes, and the operation and maintenance efficiency is improved. At the same time, the operation and maintenance process data is fed back to the machine learning model module in real time for online model optimization, continuously improving the diagnostic accuracy and forming a virtuous cycle of "data collection-diagnosis-operation and maintenance-optimization". This solves the problems of poor targeting of operation and maintenance plans, chaotic scheduling and inability to achieve continuous optimization in the existing technology.

[0019] (3) The system can prevent equipment failures from escalating and reduce equipment maintenance costs and downtime losses through accurate fault diagnosis and early warning. Compared with the traditional operation and maintenance mode, equipment downtime is reduced by more than 70%, and maintenance costs are reduced by 40% to 50%. The data acquisition module adopts a dual-mode transmission and caching design to reduce data loss and avoid repeated collection and invalid diagnosis due to data loss. The data preprocessing module removes redundant data and reduces data storage and processing costs. The operation and maintenance scheduling module accurately matches spare parts and maintenance personnel to avoid spare parts waste and manpower redundancy. At the same time, personalized adaptation is achieved through parameter configuration, eliminating the need to deploy the system separately for different types and specifications of drilling and production equipment, which greatly reduces the system deployment and subsequent maintenance costs. In addition, the edge + cloud dual deployment mode reduces the cloud data transmission pressure, reduces network operation costs, adapts to the field without network or weak network scenarios, and further improves cost controllability.

[0020] (4) The system is compatible with various oil drilling and production equipment, such as drilling equipment, oil production equipment, and oil pipeline equipment, covering drilling and production scenarios of different specifications and working conditions. Personalized adaptation can be completed through simple parameter configuration (sensor acquisition frequency, model inference threshold, etc.). The adaptation process is standardized and easy to operate, requiring no professional technicians. The adaptation time is ≤10min, solving the problems of poor universality and complex adaptation of existing systems. All hardware uses industrial-grade products with high protection level and strong anti-interference ability, adapting to the high temperature, high pressure, and harsh field working conditions of oil drilling and production. The software adopts a modular design with strong fault self-healing ability. The edge end can run independently, ensuring the system can be stably deployed in complex environments without the need for large-scale modification of existing drilling and production equipment. The deployment cycle is short, the compatibility is good, and it can be quickly put into practical application. This system, through its intelligent, standardized, and closed-loop design, effectively addresses various pain points in existing oil drilling and production fault diagnosis and maintenance. It balances safety, efficiency, economy, and versatility, and can be widely applied to various oil drilling and production scenarios. It significantly improves the operation and maintenance management level of oil drilling and production equipment, promotes the development of the oil drilling and production industry towards intelligence, efficiency, and safety, and has extremely high practical application and promotion value. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the system connection of the present invention. Detailed Implementation

[0022] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0023] Example: In this embodiment, standardized feature data refers to structured data that meets the input requirements of machine learning models after cleaning, denoising, normalization, and feature screening; algorithm fusion strategy refers to a fault decision-making method that combines multi-model weighted fusion with voting judgment; potential fault refers to an early abnormal state in which equipment parameters are slightly abnormal, the fault probability is in the critical range, and no obvious fault has yet formed.

[0024] Please see Figure 1 This machine learning-based oil drilling and production fault diagnosis and maintenance system addresses the actual working conditions of oil drilling and production sites, which are characterized by high temperature, high pressure, high load, and geographically dispersed locations. It supplements the specific implementation details of each module, clarifies hardware selection, parameter standards, process steps, and algorithm details, and ensures that the system can be directly deployed and operate stably, achieving intelligent and standardized fault diagnosis and maintenance management.

[0025] The overall system operation process is as follows: the data acquisition module collects drilling and production equipment operating data in real time and transmits it to the data preprocessing module; the data preprocessing module standardizes the data and inputs it into the machine learning model module; the machine learning model module completes training and real-time inference, and outputs the fault results to the fault diagnosis module; the fault diagnosis module completes fault determination and simultaneously triggers the anomaly warning module for tiered warnings; the operation and maintenance scheduling module generates operation and maintenance plans based on the diagnosis results and issues scheduling instructions, while simultaneously feeding operation and maintenance data back to the machine learning model module to complete model optimization; all data throughout the process is stored in the data storage and interaction module, enabling full-process traceability, configurability, and visualization.

[0026] The system's specific components include: The data acquisition module is used to collect multi-dimensional operating condition data during the operation of oil drilling and production equipment. The operating condition data includes equipment operating parameters, environmental parameters, and historical fault data, providing high-quality raw data for subsequent fault diagnosis.

[0027] 1. Sensor unit (industrial-grade selection, suitable for harsh working conditions) Temperature sensor: Industrial-grade PT100 platinum resistance sensor is selected, with a measurement range of -50℃ to 200℃, accuracy of ±0.5℃, response time ≤1s, and protection level of IP67. It is installed on the surface and internal key points of core components such as drill tools, pump bodies, and motors. At least 3-5 sensors are installed on each core device (adjusted according to the size of the device) to collect the operating temperature data of the device in real time.

[0028] Pressure sensor: A diffused silicon pressure sensor is selected, with a measurement range of 0~100MPa, accuracy of ±0.1%FS, overload capacity of 150%FS, and protection level of IP68. It is installed in drilling mud pipelines, oil production pipelines, and pump inlet and outlet to collect medium pressure and internal pressure data of equipment. The sampling interval is configurable (default 100ms).

[0029] Vibration sensor: A piezoelectric accelerometer vibration sensor is selected, with a measurement range of 0.1~1000Hz, sensitivity of 100mV / g, frequency response of ±5%, protection level of IP67, and magnetic installation (easy to disassemble and maintain). It is installed on components that are prone to vibration, such as motor bearings, drill pipe joints, and pump housings, to collect equipment vibration frequency and amplitude data.

[0030] Speed ​​sensor: A Hall effect speed sensor is selected, with a measurement range of 0~3000r / min and an accuracy of ±1r / min. The output signal is a pulse signal, which is installed on the motor shaft and the drive end of the drill rod. By detecting the rotation pulse of the shaft, the real-time speed data of the equipment is obtained, with a sampling interval of 50ms.

[0031] Additional configuration: Added environmental sensors (temperature, humidity, dust concentration), selected as industrial-grade integrated temperature, humidity and dust sensors, with a measurement range of temperature -40℃~85℃, humidity 0~100%RH, and dust 0~1000μg / m³, used to collect on-site environmental parameters and provide environmental impact reference for fault diagnosis.

[0032] 2. Data transmission unit (balancing stability and real-time performance, suitable for outdoor scenarios without network access) Wired transmission: It adopts RS485 bus transmission, with a transmission distance of ≤1000m, baud rate of 9600bps, strong anti-interference ability, and is suitable for drilling and production platforms with centralized equipment deployment. It is used to connect short-range sensors and data preprocessing modules to reduce data transmission delay.

[0033] Wireless transmission: It adopts LoRa+4G / 5G dual-mode transmission. The LoRa transmission distance is 1~3km (unobstructed), which is used for data transmission of sensors deployed in the field. 4G / 5G is used for long-distance data upload (adapted to scenarios without LoRa coverage). The transmission latency is ≤500ms and the data packet loss rate is ≤0.5%. It is equipped with an industrial-grade wireless router with an IP65 protection level and supports breakpoint resume to avoid data loss.

[0034] Data synchronization: The NTP time synchronization protocol is adopted to ensure that the timestamps of the data collected by all sensors are consistent with an error of ≤10ms, which facilitates subsequent multi-dimensional data correlation analysis; a data caching module (capacity ≥16GB) is set up to cache the collected data when the network is interrupted and automatically synchronize it to the data preprocessing module after the network is restored.

[0035] 3. Data Acquisition Range and Frequency The operating condition data acquisition scope covers drilling equipment (drill pipe, drill bit, mud pump), oil production equipment (pumping unit, oil pump), and oil pipeline equipment (pipeline, valve). The acquisition frequency can be remotely configured through the data storage and interaction module. The default configuration is: temperature, pressure, and rotational speed data once every 100ms; vibration data once every 50ms; environmental parameters once every 1s; and historical fault data is synchronized in real time (new faults are acquired immediately).

[0036] The data preprocessing module is connected to the data acquisition module via an RS485 / wireless interface. It is used to clean, denoise, normalize, and extract features from the acquired multi-dimensional operating data to obtain standardized feature data, providing high-quality input for machine learning models. The feature extraction process in the data preprocessing module employs time-domain, frequency-domain, or time-frequency-domain feature extraction methods. Extracted features include peak value, mean, variance, frequency spectrum peak value, and wavelet coefficients. The standardized feature data is obtained using min-max normalization or Z-score normalization methods.

[0037] Data cleaning: Outliers are identified using the "3σ criterion." Data is considered outliers when it deviates from the mean by three times the standard deviation. Outlier handling methods: For three or more consecutive outliers, linear interpolation is used to supplement them; for single outliers, they are directly removed. Missing value handling: When the missing value rate is ≤5%, nearest neighbor interpolation is used to supplement it; when the missing value rate is >5%, the data set is marked as invalid and stored in a backup database (for subsequent model optimization analysis). Redundant data (such as repeatedly collected identical parameters or static data that has not changed) is removed, and valid dynamic data is retained.

[0038] Data denoising: For high-frequency noise in vibration and pressure data, a wavelet denoising algorithm is used, selecting the db4 wavelet basis, decomposing into 5 layers, and setting the threshold to 0.02 (which can be adjusted via the interactive module) to remove high-frequency noise; for random noise in temperature and rotation speed data, a moving average filtering algorithm is used, with the window size set to 5 (i.e., taking the average of 5 consecutive data as the current data value) to balance denoising effect and data authenticity.

[0039] Data normalization: The default method is min-max normalization, which maps the data to the interval [0,1]. The calculation formula is: x_normalized=(x-x_min) / (x_max-x_min), where x is the original data, and x_min and x_max are the historical minimum and maximum values ​​of the parameter, respectively. For parameters with uniform data distribution (such as rotational speed), Z-score normalization can be switched through the interactive module. The calculation formula is: x_normalized=(x-μ) / σ, where μ is the mean and σ is the standard deviation. The normalization parameters are updated in real time, with x_min, x_max, μ, and σ values ​​updated every 7 days.

[0040] Feature extraction: Select the appropriate extraction method based on the data type, as follows: Temporal feature extraction: For stationary data such as temperature, pressure, and rotation speed, six types of features are extracted: peak value, mean, variance, range, skewness, and kurtosis. The calculation method adopts conventional statistical methods (e.g., mean = sum of all data / number of data, variance = sum of squares of the differences between each data and the mean / number of data). Frequency domain feature extraction: For vibration data, the fast Fourier transform (FFT) is used to convert the time domain data into frequency domain data, and three types of features are extracted: frequency spectrum peak value, fundamental frequency, and harmonic amplitude. The number of FFT sampling points is set to 1024, and the frequency resolution is ≤1Hz. Time-frequency domain feature extraction: For non-stationary vibration data (such as vibration data during drill wear), wavelet transform is used to extract wavelet coefficients (db4 wavelet basis, 5-level decomposition), and the mean and variance of the wavelet coefficients at each level are selected as features; Feature selection: The mutual information method is used to select effective features and eliminate redundant features that are irrelevant to the fault (features with mutual information value <0.3). Finally, 15-20 core features are retained as input data for the machine learning model.

[0041] Standardized feature data is output in JSON format. Each data entry includes a feature name, feature value, timestamp, and corresponding device number. The output frequency is synchronized with the data acquisition frequency and transmitted in real time to the machine learning model module, while also being backed up to the data storage module.

[0042] The machine learning model module is connected to the data preprocessing module. The machine learning model module is used to train and infer standardized feature data, and output the equipment fault type, fault level and fault location. It adopts a dual deployment mode of "edge + cloud": the edge is deployed in the drilling and production site control box (industrial-grade edge computing gateway, CPU ≥ 4 cores, memory ≥ 8GB, storage ≥ 64GB, supports wide temperature operation -40℃~85℃) for real-time inference (response time ≤ 1s); the cloud is deployed on the oilfield monitoring center server (CPU ≥ 8 cores, memory ≥ 16GB, GPU ≥ 16GB) for model training, parameter tuning and data backup. The edge and cloud synchronize data through 4G / 5G, and the model parameters are automatically synchronized at 2:00 am every day.

[0043] The machine learning model module adopts a dual deployment mode of edge + cloud. The edge is used for real-time fault inference, while the cloud is used for model training, parameter tuning and data backup. The edge and cloud synchronize model parameters periodically. The machine learning model module uses machine learning algorithms including convolutional neural networks (CNN), long short-term memory networks (LSTM), random forests, support vector machines (SVM) or gradient boosting trees (XGBoost). One of these algorithms can be used alone, or multiple algorithms can be combined. When combined, the fault diagnosis accuracy is improved through algorithm fusion strategies.

[0044] Single algorithm deployment parameters: Convolutional Neural Network (CNN): Used for deep feature extraction from vibration data. The network structure is "Input Layer → Convolutional Layer 1 → Pooling Layer 1 → Convolutional Layer 2 → Pooling Layer 2 → Fully Connected Layer → Output Layer". The input layer dimension is the number of features (15-20 dimensions). Convolutional Layer 1 has 32 filters and a 3×3 kernel size. Pooling Layer 1 uses max pooling (2×2 kernels). Convolutional Layer 2 has 64 filters and a 3×3 kernel size. Pooling Layer 2 uses average pooling. The fully connected layer has 128 neurons. The output layer has 3 neurons (corresponding to fault type, fault level, and fault location, respectively). Training parameters: learning rate 0.001, number of iterations 1000, batch size=32, and cross-entropy loss function.

[0045] Long Short-Term Memory (LSTM) network: used for fault prediction of time-series data (such as rotational speed and pressure change trends). The network structure is "input layer → LSTM layer → fully connected layer → output layer". The LSTM layer has 64 hidden units and a dropout coefficient of 0.3 (to prevent overfitting). The training parameters are: learning rate 0.0005, number of iterations 1000, batch size=32, and the loss function is mean squared error.

[0046] Random Forest: Used for fault diagnosis with multi-feature fusion, with 100 decision trees, a maximum depth of 10, a minimum number of sample splits of 5, a minimum number of sample leaf nodes of 2, and the number of random features selected is the square root of the total number of features. It is suitable for scenarios with small data volume and high real-time requirements.

[0047] Support Vector Machine (SVM): Used for fast identification of simple fault types. The kernel function is RBF kernel, the penalty coefficient C=10, and the gamma value=0.1. It is suitable for lightweight deployment at the edge.

[0048] Gradient Boosting Tree (XGBoost): Used for accurate diagnosis of complex faults, with 100 trees, a maximum depth of 8, a learning rate of 0.1, and a regularization coefficient λ=1. It is suitable for cloud-based model training.

[0049] Algorithm fusion strategy (improving diagnostic accuracy, preferred for implementation): A weighted fusion strategy of "CNN+LSTM+XGBoost" is adopted, with the weights allocated as CNN (0.4), LSTM (0.3), and XGBoost (0.3). Fusion logic: The inference results of the three algorithms are obtained separately, and the weighted average is calculated. When the fault type determination of any two algorithms is consistent, that type is taken as the final result; if the determinations of the three algorithms are inconsistent, the inference result of XGBoost is taken as the temporary result, and the model optimization unit is triggered at the same time.

[0050] The machine learning model module includes a model training unit and a model inference unit. The model training unit uses historical oilfield operating data (last 3 years) and corresponding fault labels as the training dataset, divided into a 70% training set, 20% validation set, and 10% test set. The training process is automated; the cloud server automatically calls historical data to retrain the model at 3 AM daily. When the diagnostic accuracy on the test set is ≥95%, the model parameters are updated and synchronized to the edge. If the accuracy is <95%, an alarm is triggered, reminding technicians to check data quality or adjust algorithm parameters. The model inference unit receives standardized feature data in real time at the edge and calls the trained model for inference. The inference process is: input feature data → model forward propagation → output fault type, fault level, and fault location. The inference response time is ≤1 second, and the inference accuracy is ≥95% (for known fault types). The inference results are transmitted to the fault diagnosis module in real time and simultaneously stored in the data storage module.

[0051] In some embodiments, the machine learning model module further includes a model optimization unit. The model optimization unit updates and fine-tunes the parameters of the trained machine learning model online based on the operation and maintenance process data and newly added fault data fed back by the operation and maintenance scheduling module, thereby improving the accuracy of fault diagnosis. Specifically, the optimization trigger conditions are: when the operation and maintenance scheduling module reports newly added fault data (i.e., fault types not identified by the model), or the inference accuracy is <95% for 5 consecutive times, or a fixed monthly optimization occurs (on the 1st of each month), the model optimization unit is triggered. The optimization process is as follows: ① Receive operation and maintenance process data (fault handling results, actual fault types) and newly added fault data fed back by the operation and maintenance scheduling module; ② Preprocess the newly added data (following the process of the data preprocessing module) and label it with fault tags; ③ Add the newly added data to the training dataset and retrain the model (using incremental training, only updating model parameters, without retraining all data, saving time); ④ Verify the accuracy of the optimized model. If it is ≥96%, synchronize it to the edge to replace the original model; if it is <96%, adjust the algorithm weights or parameters and retrain. Parameter tuning method: The grid search method is used to traverse and search for the core parameters of different algorithms (such as the learning rate of CNN and the number of decision trees in random forest), and select the parameter combination with the highest accuracy on the validation set as the optimal parameters.

[0052] The fault diagnosis module is connected to the machine learning model module via Ethernet. It is used to receive the inference results of the machine learning model module and combine them with the operating logic of the oil drilling and production equipment to complete the accurate judgment of faults and abnormal warnings. Fault type identification: Combining model inference results with equipment operating logic, the criteria for determining four core fault types are clarified (which can be directly used for on-site determination): Drill bit wear: If the vibration frequency is in the range of 100-200Hz, the vibration amplitude is >0.5g, the drill pipe rotation speed fluctuation is >10%, and the temperature is >80℃, it is judged as drill bit wear; Pump body leakage: Pressure data fluctuation >5MPa, temperature drop at the leakage point >10℃, and pump body vibration amplitude >0.3g are all considered pump body leakage. Motor failure: Motor speed fluctuation >15%, motor casing temperature >100℃, and abnormal current data (deviation from rated current ±20%) are identified as motor failure. Pipeline blockage: If the pipeline pressure is >80MPa and the flow data (collected by a newly added flow sensor) decreases by >30%, it is determined to be pipeline blockage; Add a new fault type: When the model inference result is an unknown fault, it is marked as a "fault to be confirmed" based on the collected multi-dimensional data and pushed to the operation and maintenance personnel. After manual confirmation, it is added to the fault type library for model optimization.

[0053] Fault severity classification (with clearly defined thresholds and implementable procedures): The "normal range" mentioned in this invention refers to the normal fluctuation range determined by combining the rated parameters of the equipment at the factory with the statistical data of the past three months of trouble-free operation. The specific determination method is as follows: for parameters with clear factory rated values ​​(such as rated speed and rated pressure), the lower limit of the normal range = rated value × 95%, and the upper limit of the normal range = rated value × 105%; for parameters without clear factory rated values ​​(such as vibration amplitude), the normal range is the mean of the past three months of trouble-free operation data ± 2 times the standard deviation.

[0054] Minor faults: Fault parameters deviate from the normal range by 5%-10%, and the fault probability inferred by the machine learning model is not less than 50% and less than 80%, which does not affect the normal operation of the equipment and does not require shutdown. This is judged as a minor fault (such as slight drill wear or slight pressure fluctuation). General faults: Fault parameters deviate from the normal range by 10%-20%, and the fault probability inferred by the machine learning model is not less than 80% and less than 95%, affecting the operating efficiency of the equipment (efficiency reduction of 10%-30%). They need to be handled within 24 hours and are judged as general faults (such as slight leakage of the pump body or slight abnormality of motor speed). Serious Fault: Fault parameters deviate from the normal range by more than 20%, or the fault probability inferred by the machine learning model is not less than 95%, which may lead to equipment shutdown or safety accidents. It is necessary to shut down the equipment immediately and is judged as a serious fault (such as severe pipeline blockage, motor overheating, or precursor to drill bit breakage).

[0055] Fault location: Establish a database of corresponding relationships between "feature data" and "equipment components" (which can be updated via the interactive module). For example, vibration frequency of 100-200Hz and amplitude >0.5g correspond to the drill pipe joint; pressure fluctuation >5MPa and temperature drop >10℃ correspond to the pump body sealing part. The feature data output by model inference is matched with the corresponding database to accurately locate the fault location. The location accuracy is ≤1m (for pipelines and drill pipes) and the location error is ≤5cm (for motors and pumps).

[0056] Anomaly warning and fault determination are executed simultaneously, with each warning signal corresponding to a specific fault level. Specific implementation details are as follows: Warning trigger conditions: ① Equipment operating parameters deviate from preset thresholds by 5%-10% (minor anomaly); ② The fault probability inferred by the machine learning model is ≥30% and <50% (potential fault); ③ When the fault level is minor / moderate / serious fault, the corresponding level warning is triggered simultaneously.

[0057] Warning method: On-site audible and visual alarm + multi-terminal message push, as detailed below: On-site audible and visual alarms: Industrial-grade audible and visual alarms are installed in the equipment control box and on-site maintenance duty room. Minor faults correspond to yellow lights and low-frequency alarms (1 time / 10s); general faults correspond to orange lights and medium-frequency alarms (1 time / 5s); and serious faults correspond to red lights and high-frequency alarms (1 time / 2s). The alarm volume is ≥80dB and can be manually muted (the light indicator remains after muting and automatically turns off after the fault is resolved).

[0058] Message push: Push warning information via web and mobile (APP, SMS) terminals. The information includes the faulty equipment number, fault type, fault level, fault location, and warning time. The push delay is ≤10 seconds. For serious faults, push notifications are also sent to the on-duty personnel and maintenance managers of the oilfield monitoring center to ensure timely response.

[0059] Warning Cancellation: When the fault diagnosis module detects that the equipment parameters have returned to normal, or the operation and maintenance scheduling module reports that the fault has been resolved, the warning will be automatically cancelled, the audible and visual alarms will be turned off, and a warning cancellation message will be pushed.

[0060] The operation and maintenance scheduling module is connected to the fault diagnosis module. It is used to generate targeted operation and maintenance plans and maintenance scheduling instructions based on the fault diagnosis results, and record operation and maintenance process data, which is then fed back to the machine learning model module for model optimization.

[0061] Solution generation logic: Based on the fault level, fault type, fault location, and equipment operation priority (core equipment > auxiliary equipment), the system automatically calls the preset operation and maintenance solution template, fills in the specific content, and generates a targeted solution in ≤30 seconds.

[0062] Specific details of the plan: Fault handling priority: Critical faults (priority 1, immediate handling) > General faults (priority 2, handling within 24 hours) > Minor faults (priority 3, handling within 72 hours); Faults of core equipment (such as drilling platform main motors and oil pipelines) are prioritized by 1 level.

[0063] Maintenance Process: Follow a fixed process of "Shutdown (if necessary) → Safety Inspection → Troubleshooting → Component Replacement / Repair → Testing → Startup". Clearly define the operating procedures, required tools, and operation time for each step. For example, the pump body leakage maintenance process is as follows: ① Shut down and close the pump inlet and outlet valves (operation time 5 min); ② Inspect the leak location (seals / pipeline interfaces) (operation time 10 min); ③ Replace seals / repair pipeline interfaces (operation time 30 min); ④ Test pump pressure (operation time 5 min); ⑤ Start up and run the pump (operation time 5 min).

[0064] Required spare parts: Based on the fault type, the system automatically matches the spare parts inventory data and specifies the spare parts name, model, and quantity. For example, if the drill string is worn, the drill pipe joint needs to be replaced (model: API 5DP, quantity 1); if the pump body is leaking, the seal needs to be replaced (model: fluororubber sealing ring, specification φ50×5, quantity 2). The system also displays the location and quantity of spare parts in stock. If the stock is insufficient, the system automatically triggers a spare parts purchase order.

[0065] Maintenance personnel dispatch: A weighted scoring method is used to automatically allocate maintenance personnel. The system comprehensively considers three core factors: the matching degree of the maintenance personnel's professional skills, the distance of the location, and the current workload. It calculates the comprehensive score of each qualified maintenance personnel, and the highest scorer is given priority in allocation. The arrival time is specified (serious faults ≤30min, general faults ≤2h, minor faults ≤6h). At the same time, fault details and maintenance plans are pushed to the maintenance personnel's mobile devices.

[0066] The specific allocation algorithm is as follows: 1. Screening candidates: First, screen out maintenance personnel who have the skills to repair the corresponding fault types and form a candidate list; 2. Quantify the scores of each factor: Professional skills matching score (weight 0.5): 100 points for a perfect match (e.g., assigning a motor repairman to a motor fault), 60 points for a relevant match (e.g., assigning an electrical repairman to a motor fault), and 0 points for a mismatch; Distance score (weight 0.3): Calculated based on the straight-line distance between the current location of the maintenance personnel and the location of the faulty equipment. 100 points for distance ≤ 1km, 80 points for distance 1km < distance ≤ 3km, 60 points for distance 3km < distance ≤ 5km, and 40 points for distance > 5km. Workload score (weight 0.2): Calculated based on the number of maintenance tasks that the maintenance personnel have not yet completed. 0 tasks get 100 points, 1 task gets 80 points, 2 tasks get 60 points, and 3 or more tasks get 40 points. 3. Calculate the overall score: Overall score = Professional skills matching score × 0.5 + Distance score × 0.3 + Workload score × 0.2; 4. Personnel allocation: Select the maintenance personnel with the highest overall score for allocation. If the scores are the same, prioritize the allocation to maintenance personnel who are closer to the work location. 5. Secondary allocation: If the assigned personnel cannot arrive on time, the system will automatically select the maintenance personnel with the second highest overall score for allocation and send a notification to the operations and maintenance manager.

[0067] Dynamic adjustment of the plan: When the equipment operation priority changes (such as the temporary start of core equipment) or the fault level is upgraded (such as a minor fault becoming a general fault), the operation and maintenance plan is automatically adjusted, maintenance personnel are reassigned, the processing time is adjusted, and the adjustment notification is pushed.

[0068] Implementation of maintenance dispatch instructions Command format: Standardized commands include command number, faulty equipment information, maintenance personnel, handling requirements, and completion time limit. They are in JSON format and are sent to the field operation and maintenance terminal (industrial tablet, mobile APP) through the interaction interface of the data storage and interaction module. The command sending delay is ≤5s.

[0069] Command tracking: The operation and maintenance scheduling module tracks the execution status of commands in real time (not received, received, in progress, completed, not completed on time). If the command is not completed on time (e.g., failure to arrive within 30 minutes due to a serious fault), a secondary reminder is automatically triggered and pushed to the maintenance personnel and the person in charge.

[0070] 3. Data recording and feedback during operation and maintenance. Record content: Clearly record the maintenance personnel, maintenance time, maintenance steps, replaced parts, maintenance results (fault resolved / unresolved), and fault cause analysis. The record format is standardized and can be manually entered or automatically synchronized through the maintenance terminal (e.g., the maintenance time is automatically recorded as the time from receiving the instruction to completion).

[0071] Data feedback: Operation and maintenance process data is fed back to the machine learning model module in real time for model optimization; at the same time, it is stored in the data storage module to form operation and maintenance archives for easy subsequent query and review, with a feedback delay of ≤1 minute.

[0072] The data storage and interaction module is used to store operating condition data, preprocessed data, model parameters, fault diagnosis results, and operation and maintenance data. It also provides interactive interfaces for data query, parameter configuration, and result display. The data storage and interaction module adopts a combination of relational and non-relational databases. The relational database is used to store structured fault tags and operation and maintenance record data, while the non-relational database is used to store massive amounts of raw operating condition data, preprocessed data, and model parameters. The interaction interface supports multi-terminal access from Web, mobile devices, and field control terminals.

[0073] Database selection and deployment: Relational database: MySQL 8.0 (industrial-grade deployment) is selected and deployed on a cloud server to store structured data, including fault tags, operation and maintenance records, equipment information, user permissions, and parameter configuration data. The database adopts a master-slave backup mode (the master database stores real-time data, and the slave database is used for backup to prevent data loss), with a storage capacity of ≥1TB and support for regular data backup (backup at 1 am every day).

[0074] No relational database: MongoDB 6.0 is selected and deployed on a cloud server to store massive amounts of unstructured / semi-structured data, including raw operating data, preprocessed data, model parameters, and early warning records. The storage capacity is ≥10TB and supports sharded storage (automatic sharding when the data volume exceeds 5TB). The data retention period is ≥3 years (configurable through the interactive module).

[0075] Data is categorized and stored hierarchically according to "equipment type → data type → time", for example: drilling equipment → vibration data → May 8, 2026, for easy and quick query; sensitive data (such as core equipment parameters and maintenance records) are stored using AES encryption to prevent data leakage.

[0076] Data cleanup: Automatically cleans up expired data (data that has exceeded the retention period) at 4:00 AM every day. The cleanup process does not affect the normal operation of the system. At the same time, it retains key data (such as data on major failures and model optimization data) and stores them permanently.

[0077] The present invention also includes an anomaly warning module, which is connected to the machine learning model module and the fault diagnosis module. It is used to issue a warning signal through audible and visual alarms and message push when abnormal equipment operating parameters or potential faults are detected based on the model inference results. The criteria for determining potential faults are: the equipment operating parameters deviate from the preset threshold by 5%-10%, or the fault probability inferred by the machine learning model is not less than 30% and less than 50%. The level of the warning signal corresponds to the fault level.

[0078] Among them, oil drilling and production equipment includes drilling equipment, oil production equipment, and oil pipeline equipment. The system can be adapted to different types and specifications of oil drilling and production equipment. It can achieve personalized adaptation for fault diagnosis and operation and maintenance through parameter configuration. The configuration parameters include sensor acquisition frequency, model inference threshold, and fault judgment standard parameters. The adaptation process is as follows: according to the type and specifications of drilling and production equipment, the corresponding parameters are configured through the data storage and interaction module, and the system automatically matches the fault diagnosis logic and operation and maintenance strategy.

[0079] Drilling equipment: compatible with drilling platforms, drill pipes, drill bits, mud pumps, etc.; compatible specifications: drill pipe diameter 73-177.8mm, mud pump pressure 0-100MPa, motor power 100-500kW. Oil production equipment: compatible with pumping units, oil pumps, wellheads, etc.; compatible specifications: pumping unit stroke 0.5-5m, oil pump flow rate 10-100m³ / h. Oil pipeline equipment: compatible with oil pipelines, valves, filters, etc., with compatible specifications: pipeline diameter 50-500mm and working pressure 0-80MPa.

[0080] Adaptation process (standardized operation, no professional technicians required): 1. Equipment Information Input: Enter the type, specifications, core parameters (such as rated speed and rated pressure), and installation location of the drilling and production equipment via the Web terminal / field control terminal, and the system will automatically generate equipment files; 2. Parameter Configuration: Based on the equipment type and specifications, configure the corresponding parameters through the data storage and interaction module (sensor acquisition frequency: 50ms / time for drilling equipment, 100ms / time for oil production equipment; model inference threshold: trigger early warning when the failure probability of drilling equipment is ≥30%, and ≥25% for oil pipeline equipment; fault judgment standard parameters: set the threshold according to the rated parameters of the equipment). 3. System matching: After configuration, the system will automatically call the corresponding fault diagnosis logic (such as focusing on monitoring vibration and rotation speed of drilling equipment, and focusing on monitoring pressure and flow rate of oil pipelines) and operation and maintenance strategy (such as prioritizing the dispatch of drilling tool maintenance personnel in case of drill tool failure) to complete the adaptation. The adaptation time is ≤10 minutes. 4. Adaptation Verification: The system automatically collects equipment trial operation data and conducts fault simulation tests. If the diagnostic accuracy rate is ≥95%, the adaptation is successful; if it does not meet the standard, the parameters are automatically adjusted and the system is re-adapted.

[0081] Field tests have verified that the system has a fault diagnosis accuracy of no less than 95%, an early warning response delay of no more than 10 seconds, and a maintenance response time of no more than 30 minutes for serious faults. It can stably adapt to various oil drilling and production equipment and harsh working conditions, and is stable in operation and secure in data, meeting the requirements for industrial field use.

[0082] In this invention, all core thresholds are determined based on petroleum industry standards, historical fault data statistics from major domestic oilfields, and on-site testing verification. Furthermore, they can be customized according to different equipment types, operating conditions, and operating environments through the data storage and interaction module. The specific determination method is as follows: Parameter deviation from threshold Potential fault 3%-5%: Based on the equipment abnormality early warning requirements of SY / T 6276-2014 "Oil and Gas Industry Health, Safety and Environmental Management System", and combined with the five-year fault data statistics of 10 major oilfields, the probability of fault occurrence increases significantly when the parameter deviates by 3%, which is set as the minimum trigger threshold. Minor faults 5%-10%: Field testing verified that when parameters deviate by 5%, observable minor anomalies occur but do not affect operation, therefore this range is set; General faults 10%-20%, serious faults >20%: Historical data shows that when parameters deviate by 10%, operating efficiency drops significantly, and when it exceeds 20%, it is very easy to cause downtime accidents. This is the classification.

[0083] Failure probability threshold Potential failure rate 30%-50%: Based on ROC curve analysis using a machine learning model, when the failure probability is ≥30%, the true positive rate reaches over 85% and the false positive rate is less than 10%, which is set as the minimum probability threshold. Minor faults 50%-80%, general faults 80%-95%, serious faults ≥95%: Based on the performance of the test set, the diagnosis rate is over 90% when the fault probability is ≥50%, and the levels are divided according to the probability range.

[0084] Time-based thresholds Early warning response ≤10s: Meets industrial site safety requirements and ensures that maintenance personnel have sufficient time to respond; Severe fault response ≤30min: In accordance with petroleum industry safety production standards, to prevent the accident from escalating; Model synchronization and data backup time: set based on system performance and data security requirements.

[0085] Threshold adjustment method Users can customize the thresholds based on the actual operating conditions of the equipment, historical fault records, and the on-site environment through the parameter configuration interface of the data storage and interaction module. After adjustment, the system will automatically conduct a 72-hour trial operation verification. If the false alarm rate or missed alarm rate increases by more than 5%, it will automatically roll back to the default threshold and prompt the user to readjust.

[0086] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A machine learning-based oil drilling and production fault diagnosis and maintenance system, characterized in that, include: The data acquisition module is used to collect multi-dimensional operating condition data during the operation of oil drilling and production equipment. The operating condition data includes equipment operating parameters, environmental parameters, and historical fault data. The data preprocessing module, connected to the data acquisition module, is used to clean, denoise, normalize, and extract features from the acquired multi-dimensional working condition data to obtain standardized feature data. The machine learning model module is connected to the data preprocessing module. The machine learning model module is used to train and infer standardized feature data, and output the equipment fault type, fault level and fault location. The fault diagnosis module is connected to the machine learning model module and is used to receive the inference results of the machine learning model module, and combine them with the operating logic of the oil drilling and production equipment to complete the accurate judgment of faults and abnormal warnings. The operation and maintenance scheduling module is connected to the fault diagnosis module. It is used to generate targeted operation and maintenance plans and maintenance scheduling instructions based on the fault diagnosis results, record operation and maintenance process data, and feed it back to the machine learning model module for model optimization. An anomaly warning module, connected to the machine learning model module and the fault diagnosis module, is used to issue warning signals based on the model inference results. The data storage and interaction module is used to store operating condition data, preprocessed data, model parameters, fault diagnosis results, and operation and maintenance data, and provides interactive interfaces for data query, parameter configuration, and result display.

2. The machine learning-based oil drilling and production fault diagnosis and maintenance system according to claim 1, characterized in that, The data acquisition module includes a sensor unit and a data transmission unit. The sensor unit includes a temperature sensor, a pressure sensor, a vibration sensor, and a speed sensor, which are used to collect data on the temperature, pressure, vibration frequency, and operating speed of the drilling and production equipment, respectively. The data transmission unit adopts wired or wireless transmission and has NTP time synchronization, breakpoint resume, and data caching functions, and transmits the collected data to the data preprocessing module in real time.

3. The machine learning-based oil drilling and production fault diagnosis and maintenance system according to claim 1, characterized in that, The feature extraction process of the data preprocessing module employs time-domain feature extraction, frequency-domain feature extraction, or time-frequency-domain feature extraction methods. The extracted features include peak value, mean, variance, frequency spectrum peak value, and wavelet coefficients. The standardized feature data is obtained by processing the data using min-max normalization or Z-score normalization methods.

4. The machine learning-based oil drilling and production fault diagnosis and maintenance system according to claim 1, characterized in that, The machine learning model module employs machine learning algorithms including Convolutional Neural Network (CNN), Long Short-Term Memory Network (LSTM), Random Forest, Support Vector Machine (SVM), or Gradient Boosting Tree (XGBoost). It can use one of these algorithms individually or combine multiple algorithms through weighted fusion and voting strategies to improve fault diagnosis accuracy. The machine learning model module includes a model training unit and a model inference unit. The model training unit trains the model using historical operating data and corresponding fault labels, while the model inference unit uses the trained model to infer from real-time standardized feature data.

5. The machine learning-based oil drilling and production fault diagnosis and maintenance system according to claim 1, characterized in that, The machine learning model module also includes a model optimization unit, which updates and optimizes the parameters of the trained machine learning model online based on the operation and maintenance process data and new fault data fed back by the operation and maintenance scheduling module, thereby improving the accuracy of fault diagnosis.

6. The machine learning-based oil drilling and production fault diagnosis and maintenance system according to claim 1, characterized in that, The fault diagnosis module's fault determination includes fault type identification, fault level classification, and fault location positioning. The fault types include drill bit wear, pump body leakage, motor failure, and pipeline blockage. The fault levels are divided into minor faults, general faults, and serious faults. The fault location is accurately located through the correspondence between feature data and equipment components.

7. The machine learning-based oil drilling and production fault diagnosis and maintenance system according to claim 1, characterized in that, The maintenance and operation plan generated by the maintenance and operation scheduling module includes fault handling priority, maintenance process, required spare parts, and maintenance personnel scheduling information. The maintenance and operation plan can be dynamically adjusted according to the equipment operation priority and fault level. Maintenance scheduling instructions are sent to the on-site maintenance and operation terminal through the interactive interface.

8. The machine learning-based oil drilling and production fault diagnosis and maintenance system according to claim 1, characterized in that, The data storage and interaction module adopts a combination of relational and non-relational databases, and has data encryption, master-slave backup and automatic cleanup functions. The relational database is used to store structured fault tags and operation and maintenance record data, while the non-relational database is used to store massive amounts of raw operating data, preprocessed data and model parameters. The interaction interface supports multi-terminal access from Web, mobile and field control terminals.

9. The oil drilling and production fault diagnosis and maintenance system based on machine learning according to claim 1, characterized in that, The anomaly warning module is used to issue warning signals through audible and visual alarms and message push when abnormal equipment operating parameters or potential faults are detected based on model inference results. The criteria for determining potential faults are: equipment operating parameters deviate from the rated threshold by 3%-5%, or the fault probability inferred by the machine learning model is not less than 30% and less than 50%. The criteria for determining minor faults are: equipment operating parameters deviate from the rated threshold by 5%-10%, and the fault probability inferred by the machine learning model is not less than 50% and less than 80%. The warning signal level corresponds to the fault level: potential faults correspond to Level 1 warning, minor faults correspond to Level 2 warning, general faults correspond to Level 3 warning, and serious faults correspond to Level 4 warning.

10. The machine learning-based oil drilling and production fault diagnosis and maintenance system according to claim 1, characterized in that, The oil drilling and production equipment includes drilling equipment, oil production equipment, and oil pipeline equipment. The system can be adapted to different types and specifications of oil drilling and production equipment, and can achieve personalized adaptation for fault diagnosis and operation and maintenance through parameter configuration. The configuration parameters include sensor acquisition frequency, model inference threshold, and fault judgment standard parameters. The adaptation process is as follows: according to the type and specifications of the drilling and production equipment, the corresponding parameters are configured through the data storage and interaction module, and the system automatically matches the fault diagnosis logic and operation and maintenance strategy.

Images