Oil-gas separation control method and system based on spiral flow channel

Abstract

The invention discloses an oil-gas separation control method and system based on a spiral runner, and belongs to the technical field of fluid separation control, and the method comprises the steps: collecting external circulation parameters of a lubricating oil circulation loop and internal process parameters of a spiral runner separator, and generating multi-source synchronous data; performing feature extraction operation on the internal process parameters to generate separation process feature vectors; analyzing by using a data and state mapping rule to generate an oil bubble depth diagnosis state; on the basis of the oil bubble depth diagnosis state and the separation process feature vector, online optimization calculation is carried out, and a dynamic control instruction is generated; and adjusting an execution mechanism of the spiral flow channel separator, collecting adjusted internal process parameters, and generating an updated separation process feature vector. By adopting the technical scheme of combining multi-source synchronous data acquisition, separation process feature extraction, oil bubble depth diagnosis and online optimization calculation, fine diagnosis and adaptive optimization control of the separation process can be realized, and the separation efficiency and the system stability are improved.

Description

Technical Field

[0001] This invention relates to the field of fluid separation control technology, and in particular to an oil-gas separation control method and system based on a spiral flow channel. Background Technology

[0002] Spiral flow channel oil-gas separators are core components widely used in critical fields such as lubrication systems for large rotating machinery and lubrication systems for aero engines. Their main function is to effectively separate gas mixed in with lubricating oil using the centrifugal force field generated by the fluid movement within the spiral channel. This ensures the quality of the lubricating oil, prevents emulsification and oxidation, and protects critical components from damage caused by cavitation and other phenomena. The stability of its operation directly affects the safety and reliability of the entire machine. Therefore, effective control of the spiral flow channel oil-gas separation process is crucial.

[0003] In related technologies, Chinese invention patent CN119195880A discloses an oil-gas separator control method, device, equipment, and storage medium. This includes real-time acquisition of engine oil temperature and crankcase pressure values ​​during engine operation; when the engine oil temperature is not lower than a set temperature value and the crankcase pressure is greater than a set pressure value, dynamically controlling the motor speed of the oil-gas separator based on a first control parameter associated with the crankcase pressure value and the maximum motor speed of the oil-gas separator; and when the engine oil temperature is not lower than the set temperature value and the crankcase pressure is not greater than the set pressure value, dynamically controlling the motor speed of the oil-gas separator based on a second control parameter associated with engine speed and torque and the maximum motor speed of the oil-gas separator.

[0004] Regarding the aforementioned technologies, the inventors believe they have technical defects in practical applications. Their control logic is coarse and simplistic, relying solely on external macroscopic indicators such as temperature, pressure, or mileage for threshold-based triggering. They lack a deep understanding of the dynamic flow field inside the separator, the microscopic evolution of bubbles, and the cleanliness of the oil. This results in an inability to precisely adjust for different forms of oil mist particles. Furthermore, in complex operating conditions, they can only forcibly intervene by switching the maximum rotation speed. This not only makes it difficult to achieve an optimal balance between separation efficiency and energy consumption, but also, due to the lack of fluid stability assessment and model self-correction mechanisms, limits the system's adjustment accuracy, stability, and adaptability to different operating conditions. Summary of the Invention

[0005] To address the aforementioned issues, this invention provides an oil-gas separation control method and system based on a spiral flow channel. It employs a technical solution that combines multi-source synchronous data acquisition, separation process feature extraction, oil bubble depth diagnosis, and online optimization calculation. This enables refined diagnosis and adaptive optimization control of the separation process, thereby improving separation efficiency and system stability.

[0006] The above objectives can be achieved through the following approach: A spiral flow channel-based oil-gas separation control method includes: collecting external circulation parameters of the lubricating oil circulation loop and internal process parameters of the spiral flow channel separator to generate multi-source synchronous data; performing feature extraction operations on the internal process parameters in the multi-source synchronous data to generate a separation process feature vector; parsing the external circulation parameters and the separation process feature vector using data-state mapping rules to generate an oil bubble depth diagnostic state; performing online optimization calculations based on the oil bubble depth diagnostic state and the separation process feature vector to generate dynamic control commands; adjusting the actuator of the spiral flow channel separator according to the dynamic control commands and collecting the adjusted internal process parameters to generate an updated separation process feature vector.

[0007] Optionally, the generation of multi-source synchronous data includes: acquiring the oil tank level signal, main circuit pressure signal, and oil turbidity signal to generate external circulation parameters; acquiring the dynamic differential pressure signal and pipe wall vibration signal at key locations inside the spiral flow separator to generate internal process parameters; and performing time-scale alignment processing on the external circulation parameters and the internal process parameters to generate multi-source synchronous data.

[0008] Optionally, generating the separation process feature vector includes: performing time-frequency analysis on the dynamic differential pressure signal to generate differential pressure fluctuation feature parameters; performing spectral energy analysis on the pipe wall vibration signal to generate vibration energy distribution feature parameters; and performing vector fusion on the differential pressure fluctuation feature parameters and the vibration energy distribution feature parameters to generate the separation process feature vector.

[0009] Optionally, generating the oil bubble depth diagnostic status includes: constructing data and state mapping rules, parsing the feature vector of the separation process, and generating an internal bubble movement pattern; using the liquid level signal change trend in the external circulation parameters to verify and correct the internal bubble movement pattern, and generating a corrected bubble movement pattern; performing morphological classification and load assessment on the corrected bubble movement pattern, and generating an oil bubble depth diagnostic status that includes bubble morphological classification and separation load level.

[0010] Optionally, the construction of data and state mapping rules includes: collecting historical multi-source synchronous data under various operating conditions and calculating the corresponding historical separation process feature vectors, while obtaining the internal bubble motion pattern labels determined by observation methods to generate a labeled training dataset; and using machine learning algorithms to perform feature learning and fitting on the labeled training dataset to generate data and state mapping rules.

[0011] Optionally, generating dynamic control instructions includes: importing the oil bubble depth diagnostic status and the separation process feature vector into the separation process health evaluation function to generate expected improvement effect values ​​under different combinations of control parameters; selecting a combination of control parameters that makes the expected improvement effect value meet preset preferred conditions, and generating dynamic control instructions.

[0012] Optionally, the method further includes: adjusting the ideal target interval parameter of the separation process health evaluation function according to the bubble morphology classification in the bubble depth diagnostic state; when the bubble morphology is classified as small-diameter bubble enrichment, adjusting the ideal target interval parameter to focus on reducing the pressure difference fluctuation frequency; when the bubble morphology is classified as large-diameter bubble clusters, adjusting the ideal target interval parameter to focus on optimizing the vibration energy distribution in a specific frequency band.

[0013] Optionally, generating the updated separation process feature vector includes: parsing the dynamic control command to obtain the flow regulation gain and pressure control threshold, and driving the actuator to generate the changed flow field operating state; after a preset fluid inertial stabilization time, acquiring the dynamic differential pressure signal and pipe wall vibration signal under the changed flow field operating state to generate the adjusted internal process parameters; and performing feature extraction operations on the adjusted internal process parameters to generate the updated separation process feature vector.

[0014] Optionally, the method further includes: comparing the updated separation process feature vector with the expected improvement value to generate a control effect feedback signal; and using the control effect feedback signal to perform online correction of the data-state mapping rule.

[0015] Based on the same inventive concept, this invention also provides an oil-gas separation control system based on a spiral flow channel, including a multi-source data acquisition module for acquiring external circulation parameters of the lubricating oil circulation loop and internal process parameters of the spiral flow channel separator, generating multi-source synchronous data; a separation feature extraction module for performing feature extraction operations on the internal process parameters in the multi-source synchronous data, generating a separation process feature vector; an oil bubble depth diagnosis module for analyzing the external circulation parameters and the separation process feature vector using data and state mapping rules, generating an oil bubble depth diagnosis state; a control strategy optimization module for performing online optimization calculations based on the oil bubble depth diagnosis state and the separation process feature vector, generating dynamic control commands; and a closed-loop adjustment module for adjusting the actuator of the spiral flow channel separator according to the dynamic control commands and acquiring the adjusted internal process parameters, generating an updated separation process feature vector.

[0016] Compared with the prior art, the present invention has the following advantages: This invention uses multi-source data fusion sensing to simultaneously collect and analyze external circulation parameters reflecting the macroscopic state of the system and microscopic process parameters characterizing the internal dynamics of the separator. This constructs a comprehensive and in-depth cognitive foundation for the oil and gas separation process, enabling control decisions to no longer rely on single, lagging external indicators, thereby improving the sensing accuracy and decision reliability of the control system.

[0017] This invention proposes a method for diagnosing oil bubble depth. By extracting features from internal process signals and analyzing them using data and state mapping rules, abstract sensor data is transformed into an intuitive diagnosis of physical phenomena such as internal bubble morphology, motion patterns, and separation load levels. This enables effective insight into the internal black box state of the separation process and provides a prerequisite for achieving mechanism-based precise control.

[0018] This invention establishes a forward-looking online optimization and adaptive control strategy that can pre-evaluate the potential improvement effects of different combinations of control parameters before executing specific adjustment actions, and then select the optimal one. Furthermore, the system can dynamically adjust the optimization target based on real-time diagnosed bubble morphology, enabling the control strategy to flexibly address different types of separation challenges and achieving intelligent and personalized control.

[0019] This invention constructs a complete closed-loop self-learning and correction mechanism. By comparing the expected and actual effects before and after control execution, a feedback signal is generated, and this signal is used to correct the core diagnostic model, namely the data-state mapping rules, online. This adaptive learning capability enables the system to continuously track and adapt to operating condition drift caused by factors such as equipment aging and oil deterioration, ensuring the long-term stability and effectiveness of the control system.

[0020] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures pointed out in the description, claims, and drawings. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a schematic flowchart of an oil-gas separation control method based on a spiral flow channel according to an embodiment of the present invention.

[0023] Figure 2 This is a schematic diagram illustrating the principle of multi-source data acquisition and time-scale alignment in an embodiment of the present invention.

[0024] Figure 3 This is a control instruction optimization decision matrix diagram based on a health evaluation function according to an embodiment of the present invention.

[0025] Figure 4 This is a schematic diagram of the structure of an oil-gas separation control system based on a spiral flow channel according to an embodiment of the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, 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.

[0027] Reference Figure 1 One embodiment of the present invention proposes an oil-gas separation control method based on a spiral flow channel. It adopts a technical solution that combines multi-source synchronous data acquisition, separation process feature extraction, oil bubble depth diagnosis and online optimization calculation, which can realize refined diagnosis and adaptive optimization control of the separation process, and improve separation efficiency and system stability.

[0028] The method described in this embodiment specifically includes: S1. Collect external circulation parameters of the lubricating oil circulation loop and internal process parameters of the spiral flow channel separator to generate multi-source synchronous data; Optionally, generating multi-source synchronized data includes: The system acquires the oil tank level signal, main circuit pressure signal, and oil turbidity signal to generate external circulation parameters. The dynamic differential pressure signal and pipe wall vibration signal at key locations inside the spiral flow channel separator are acquired to generate internal process parameters; The external loop parameters and the internal process parameters are time-scaled and aligned to generate multi-source synchronized data.

[0029] Specifically, a magnetostrictive level sensor with an accuracy better than 0.5% is deployed in the oil tank, continuously acquiring the oil tank level signal at a sampling frequency of no less than 10 Hz. This signal directly reflects the increase or decrease in oil volume caused by changes in the gas content of the oil. A piezoresistive pressure sensor with a range of 0 to 2 MPa is installed on the high-pressure section after the main oil circuit pump, acquiring the main circuit pressure signal at a sampling frequency of 10 Hz. This signal characterizes the oil supply capacity and flow resistance of the circulation system. Simultaneously, an online oil turbidity sensor based on the near-infrared light scattering principle is deployed on the separator outlet pipeline. Its measurement range is typically 0 to 1000 NTU, also acquiring the oil turbidity signal at a frequency of 10 Hz to quantify the content of microbubbles remaining in the lubricating oil after separation, thereby evaluating the separation effect. These three sets of time-series signals together constitute the external circulation parameters reflecting the system-level operating status.

[0030] A dynamic pressure sensor with a response frequency of up to 5 kHz is installed near the inlet and outlet flanges of the helical flow channel separator. The pressure at both points is simultaneously acquired via a high-speed data acquisition card, and the difference is calculated in real time to generate a dynamic differential pressure signal. The fluctuation pattern and amplitude of this signal can reflect the drastic changes in the local pressure field caused by bubble coalescence, breakup, and flow pattern transformation within the flow channel. Simultaneously, a piezoelectric accelerometer with a sensitivity of approximately 100 mV / acceleration unit is installed on the exterior of the helical section pipe wall via magnetic adsorption or bolt fastening to acquire pipe wall vibration signals. The sensor's installation location needs to be predetermined through finite element modal analysis, typically selected in the region most sensitive to internal fluid pulsation responses. The acquisition frequency of the pipe wall vibration signal is set above 10 kHz to capture the structural vibration response generated by large bubble impacts, eddy shedding, and fluid-structure interaction effects. These two sets of high-frequency signals constitute the internal process parameters characterizing the physical processes in the core region of the separator.

[0031] A unified high-precision master clock source is employed, such as synchronizing the time of all data acquisition modules to microsecond-level accuracy via the Network Time Protocol (NTP). During processing, the highest sampling frequency, 10 kHz, is used as the system's main time axis. For external loop parameters with lower sampling frequencies, such as the 10 kHz liquid level signal, upsampling is performed using zero-order hold or linear interpolation algorithms. The zero-order hold algorithm assigns the value of the previous original sampling point to all high-frequency timestamps within the time interval between two original sampling points. This process ensures that each high-frequency data point's timestamp corresponds to a unique, temporally closest external loop parameter value, thus generating a data matrix. Each row of the matrix represents a single instant, and the columns contain the liquid level signal, main loop pressure signal, oil turbidity signal, dynamic differential pressure signal, and pipe wall vibration signal at that instant, forming the final multi-source synchronized data used for analysis. Figure 2As shown in the figure, the 10Hz low-frequency signal points are filled into the 10kHz high-frequency time axis using the zero-order hold algorithm.

[0032] For example, in the scenario of high-altitude, high-cruise power operation of the lubricating oil system of an aircraft engine, At any given time, a magnetostrictive level sensor with an accuracy better than 0.5% acquires a level signal of 450.5 mm at a frequency of 10 Hz; a piezoresistive pressure sensor with a range of 0-2 MPa acquires a main circuit pressure signal of 1.20 MPa at a frequency of 10 Hz; and an online oil turbidity sensor acquires a turbidity signal of 150 NTU at a frequency of 10 Hz. These signals constitute the external circulation parameters. Simultaneously, a dynamic pressure sensor with a response frequency of 5 kHz at the separator inlet and outlet generates a dynamic differential pressure signal at the current moment by performing differential sampling via a high-speed acquisition card. A piezoelectric accelerometer with a sensitivity of 100 mV / g acquires pipe wall vibration signals at a frequency of 10 kHz. These two sets of signals constitute the internal process parameters. The system synchronizes time to the microsecond level via NTP, using a 10 kHz sampling frequency as the main time axis. For the 10 Hz external parameters, a zero-order hold algorithm is used for upsampling, i.e., at... to Within the 1000 high-frequency sampling points, all values ​​are assigned as... The original values ​​at each time point. Taking the 500th high-frequency point as an example, the acquired dynamic differential pressure signal value is 0.015 MPa, the pipe wall vibration acceleration value is 1.2 m / s², and the generated synchronous data matrix row vector is [450.5, 1.20, 150, 0.015, 1.2]. This processing ensures that each high-frequency data point at each time point has a corresponding external loop parameter value, forming the final multi-source synchronous data used for analysis. This method utilizes a high-precision sensor array to construct a multi-dimensional data matrix and solves the data misalignment problem caused by inconsistent sampling frequencies of heterogeneous sensors through time-scale alignment processing. It provides the system with a high-confidence real-time data benchmark, ensuring the accuracy of subsequent flow regime identification in capturing transient fluctuations.

[0033] S2. Perform feature extraction operations on the internal process parameters in the multi-source synchronization data to generate a separation process feature vector; Optionally, the generation of the separation process feature vector includes: Time-frequency analysis is performed on the dynamic differential pressure signal to generate differential pressure fluctuation characteristic parameters; The vibration signal of the pipe wall is subjected to spectral energy analysis to generate vibration energy distribution characteristic parameters; The pressure difference fluctuation characteristic parameters and the vibration energy distribution characteristic parameters are fused into a vector to generate a separation process characteristic vector.

[0034] Specifically, a dynamic differential pressure signal data window with a duration of, for example, 0.2 seconds is extracted. This duration contains sufficient fluctuation information while ensuring real-time diagnostics. Continuous Wavelet Transform (CWT) is applied to this data window, selecting, for example, the Morlet mother wavelet, due to its good localization properties in the time-frequency domain, making it suitable for capturing transient events in fluids. The transform yields a two-dimensional time-frequency energy spectrum. Instead of directly using the entire spectrum, a set of differential pressure fluctuation characteristic parameters are calculated from it. For example, the integrated energy in a specific low-frequency band (0-50 Hz) is calculated to characterize the slow pressure pulsations caused by the passage of large bubbles or gas masses; simultaneously, the integrated energy in the mid-to-high frequency band (50-500 Hz) is calculated to reflect the aggregation and breakup of small bubble groups. Furthermore, the energy entropy across the entire frequency band is calculated. This index measures the dispersion of pressure fluctuation energy in the frequency domain; energy concentration indicates the presence of a dominant flow mechanism.

[0035] The pipe wall vibration signal was captured using the same 0.2-second time window, and a Fast Fourier Transform (FFT) was applied to obtain the power spectral density map of the vibration signal. Vibration energy distribution characteristic parameters were extracted from this power spectrum. These parameters include the root mean square (RMS) value of the vibration energy in the low-frequency band (20-100 Hz), where the energy is mainly related to the mechanical impact of large air masses striking the pipe wall. Simultaneously, the RMS value of the energy in the high-frequency band (1-3 kHz) was calculated, where vibrations are typically associated with the collapse of microbubbles or acoustic emissions from highly turbulent flow fields. By calculating the energy proportions of different frequency bands, the distribution pattern of vibration energy can be further identified.

[0036] The two sets of feature parameters are fused to generate the final separation process feature vector. This is achieved through feature concatenation, where the pressure difference fluctuation feature parameter and the vibration energy distribution feature parameter are linked end-to-end in a predetermined order to form a higher-dimensional combined vector. This vector is the separation process feature vector, and its mathematical expression is as follows: , Here, This represents the final generated feature vector of the separation process. It is the low-frequency integrated energy calculated from the dynamic differential pressure signal. It is the integrated energy in the mid-to-high frequency band calculated from the dynamic differential pressure signal. It is the time-frequency energy entropy calculated from the dynamic pressure difference signal. It is the root mean square value of low-frequency vibration energy calculated from the pipe wall vibration signal. It is the root mean square value of high-frequency vibration energy calculated from the pipe wall vibration signal. It is the energy percentage of a specific frequency band calculated from the pipe wall vibration signal.

[0037] For example, a data window of 0.2s is extracted from the synchronization signal stream, corresponding to 2000 sampling points. A continuous wavelet transform (CWT) is performed on the dynamic differential pressure signal using the Morlet mother wavelet, and the integrated energy in the 0-50Hz low-frequency band is calculated. and integrated energy in the mid-to-high frequency band of 50-500Hz And calculate the full-band energy entropy. Subsequently, a Fast Fourier Transform (FFT) was applied to the pipe wall vibration signal to extract the root mean square (RMS) value of the vibration energy in the low-frequency band of 20-100Hz. Given the sum of squares of the amplitudes of the vibration acceleration sequence within this frequency band. According to the root mean square calculation process: Combined with the calculated root mean square value of vibration energy in the 1-3kHz high-frequency band and the proportion of energy in a specific frequency band The features are spliced ​​in a predetermined order to generate the final separation process feature vector V=[12.5,8.2,0.65,1.2,0.4,0.25]. This method transforms the original waveform signal into a feature vector with physical meaning by performing wavelet transform on the dynamic pressure difference and spectral analysis on the vibration signal, and can identify micro-fluid behaviors such as bubble coalescence, breakup and atmospheric mass impact.

[0038] S3. Analyze the external loop parameters and the feature vector of the separation process using the data and state mapping rules to generate the oil bubble depth diagnostic status; Optionally, the generation of the oil bubble depth diagnostic status includes: Construct data and state mapping rules, parse the feature vector of the separation process, and generate the internal bubble motion pattern; The internal bubble motion pattern is verified and corrected by using the liquid level signal change trend in the external circulation parameters, and a corrected bubble motion pattern is generated. The modified bubble motion pattern is morphologically classified and load assessed to generate an oil bubble depth diagnostic status that includes bubble morphology classification and separation load level.

[0039] Specifically, a pre-trained classification model is invoked, which is the data-state mapping rule. The separation process feature vector, calculated in real time, is then used as the aforementioned vector. The input vector is used as the model's input. The model's internal weight matrix and nonlinear activation function operate on the input vector, mapping it to a predefined class space. The model's output is a discrete label representing the most likely internal bubble motion pattern, such as "stable bubbly flow," "pulsating clump flow," or "violently turbulent flow." This pattern provides a preliminary assessment of the macroscopic dynamic behavior of the fluid inside the separator.

[0040] The slope of the linear regression of the oil tank level signal over a past period is calculated to obtain a quantified value of the trend of the level signal change. Subsequently, a logic-based verification process is executed. For example, if the internal bubble movement pattern is diagnosed as "stable bubbly flow" with good separation effect, but at the same time there is no oil replenishment or discharge in the tank and the temperature change is within the allowable range, and detection is performed... If the value is significantly greater than a positive threshold, such as ±0.5 mm / min, it indicates that the oil volume has expanded due to the increased gas content, and a large amount of gas remains trapped in the circulating oil, contradicting the diagnostic result of efficient separation. At this point, a correction logic is triggered to modify the internal bubble movement pattern to a pattern that better matches the macroscopic performance, such as "microbubble bypass" or "emulsion state," thereby generating a corrected bubble movement pattern.

[0041] The corrected macroscopic motion pattern is further refined and quantified. Morphological classification is based on the corrected bubble motion pattern, combined with specific component values ​​in the separation process feature vector, to achieve a more refined division. For example, when the corrected pattern is "pulsating clump flow," if the parameters representing low-frequency pressure fluctuations in the separation process feature vector... Parameters of low-frequency vibration energy If all exceed their high-level threshold, the bubble morphology is classified as "large-diameter bubble clusters". If the pattern is "stable bubbly flow", but the high-frequency vibration energy... If the value is abnormally high, the morphology is classified as "enrichment of small-diameter bubbles". Load assessment compares the overall amplitude of the separation process characteristic vector with a preset health baseline to calculate a separation load level. This level can be divided into four categories: "low load", "healthy load", "critical load", and "overload". Finally, the bubble morphology classification result is combined with the separation load level to form a structured data object, namely the oil bubble depth diagnostic status, such as {bubble morphology classification: large-diameter bubble clusters, separation load level: overload}, and this is output to the next control stage.

[0042] For example, real-time vector The input model initially identifies the bubble movement pattern as "steady bubbly flow." To verify accuracy, the system calculates the slope of the linear regression of the liquid level signal over a past period. mm / min, while the preset positive threshold is 0.5 mm / min. Because A value significantly greater than the threshold with no oil replenishment indicates that the oil volume has expanded due to increased gas content, causing a conflict in the diagnostic logic. The system triggers correction logic, changing the bubble movement mode to "microbubble bypass." This is combined with vector... Mid-to-high frequency vibration energy The numerical values ​​are refined into a morphological classification of "small-diameter bubble enrichment," and the load level is assessed by comparing it with a healthy baseline, generating a final diagnostic status: {bubble morphology classification: small-diameter bubble enrichment, separation load level: healthy load}. This method employs a dual diagnostic mechanism, utilizing the macroscopic liquid level expansion trend to logically verify the microscopic pattern recognition results, reducing the misdiagnosis rate that might arise from relying solely on signal features and ensuring the physical and logical rigor of the oil bubble diagnostic conclusions.

[0043] Optionally, the rules for constructing the data and state mapping include: Collect historical multi-source synchronous data under various operating conditions and calculate the corresponding historical separation process feature vectors. At the same time, obtain the internal bubble motion pattern labels determined by observation methods and generate a labeled training dataset. Machine learning algorithms are used to learn and fit features to the labeled training dataset, generating data-state mapping rules.

[0044] Specifically, on a dedicated oil-gas separation test bench, operating conditions were varied, covering multiple operating conditions from the design point to deviation points. These changes were achieved by adjusting the main oil pump speed to alter the flow rate, for example, from 50% to 120% of the rated flow, and by injecting different amounts of compressed air to change the inlet gas content, for example, from 1% to 15%. At each stable operating point, external circulation parameters and internal process parameters were simultaneously recorded for at least several minutes. Simultaneously, a high-speed camera system installed at key observation locations on the spiral flow separator captured images of the gas-liquid two-phase flow within the flow channel at a rate of thousands of frames per second. Experienced fluid mechanics engineers analyzed these high-speed recordings frame by frame, manually classifying the physical phenomena corresponding to each data segment based on the size, shape, distribution, and trajectory of the bubbles, and assigning clear labels to the internal bubble movement patterns, such as "stable bubbly flow," "pulsating clump flow," "large-diameter bubble clumps," or "emulsified and mixed flow." Subsequently, the same feature extraction operations as online diagnostics were performed on the collected historical multi-source synchronous data to calculate the historical separation process feature vector corresponding to each data segment. Finally, the feature vector of each historical separation process is paired with its corresponding manually labeled internal bubble motion pattern to form a huge labeled training dataset containing tens of thousands of samples.

[0045] Machine learning algorithms are used to learn and fit features to a labeled training dataset. A suitable machine learning algorithm for handling nonlinear classification problems is selected, such as a Support Vector Machine (SVM) with a radial basis function kernel. The labeled training dataset is input into the algorithm. During training, the algorithm iteratively optimizes to find an optimal hyperplane that can separate feature vector sample points represented by different labels in the multidimensional feature space with the maximum margin. To ensure the model's generalization ability, K-fold cross-validation is typically used to optimize the model's hyperparameters, such as the penalty coefficient C and the kernel parameter gamma. After training, the algorithm's final output is a set of fixed model parameters, including support vectors, weights, and bias terms. This set of parameters collectively defines the nonlinear mapping relationship from the input separation process feature vectors to the output internal bubble motion pattern labels, which itself constitutes the data and state mapping rules for final deployment.

[0046] For example, during the offline calibration phase, the flow rate was adjusted to fluctuate between 50% and 120% of the rated flow rate on the experimental bench by regulating the main oil pump speed, and 1% to 15% compressed air was injected. The system collected 1000 sets of historical feature vectors V and corresponding internal bubble motion pattern labels. The mapping rule was fitted using a Support Vector Machine (SVM) algorithm with radial basis function kernels, and a penalty coefficient was set. nuclear parameters The training process iteratively finds the optimal hyperplane. When the real-time vector V of the cruise scenario is input into the model, if the classification projection value obtained from the internal weight matrix operation is 1.55, the output discrete label is determined as "pulsating clumping flow" after nonlinear mapping. This rule establishes the mapping logic between the sensor feature space and the physical flow pattern. This method utilizes historical big data and machine learning algorithms to construct mapping rules, enabling the control system to learn from expert experience and identify complex nonlinear flow states. By continuously iterating and optimizing hyperparameters, the model can adapt to the characteristics of separators of different specifications, improving the versatility and intelligence of the control scheme.

[0047] S4. Based on the oil bubble depth diagnostic status and the feature vector of the separation process, perform online optimization calculation to generate dynamic control commands; Optionally, the generation of dynamic control instructions includes: The oil bubble depth diagnostic status and the separation process feature vector are imported into the separation process health evaluation function to generate the expected improvement effect value under different combinations of control parameters. Select a combination of control parameters that makes the expected improvement value meet the preset preferred conditions, and generate dynamic control commands.

[0048] Specifically, a set of candidate control parameter combinations is generated within a preset discrete control parameter space. These parameters directly correspond to the adjustable actuators of the helical flow separator, such as flow regulation gain and pressure control threshold. A typical parameter combination space might include flow gain adjustments of {-5%, -2%, 0%, +2%, +5%} and pressure threshold adjustments of {-10kPa, 0, +10kPa} relative to the current setpoint, thus forming 15 candidate control strategies. For each candidate control parameter combination, a dynamic process response prediction model is invoked. This model is built based on a Long Short-Term Memory (LSTM) network and learns the temporal evolution relationship between historical control sequences and separation process feature vectors to characterize the nonlinear dynamic response characteristics of the flow field under different control actions. Based on the current system state and the input control adjustment, the new stable state to be reached after a fluid inertial stabilization time is predicted, and a predicted future separation process feature vector is output. Subsequently, each predicted future separation process feature vector is substituted into the separation process health evaluation function to calculate a quantified expected improvement value. The evaluation function is a mathematical model that maps feature vectors to a single health score, and its general form can be expressed as: , in, Representing the The expected improvement value corresponding to each combination of control parameters is a scalar; the larger the value, the better the expected separation effect. It is a health evaluation function for the separation process; It is in the The feature vector of the future separation process predicted under the action of a combination of control parameters; It is the first of the predicted vectors One component, such as the predicted low-frequency pressure difference fluctuation energy; These are preset weighting coefficients, reflecting the importance of different feature components to the overall health assessment; It is a normalization function that takes characteristic components with different physical dimensions and numerical ranges. Mapped to a dimensionless interval of 0 to 1, so that they can be summed by weight.

[0049] Select a combination of control parameters that makes the expected improvement value meet preset optimal conditions, and generate dynamic control commands. For example... Figure 3 As shown in the figure, a discrete control parameter search space consisting of 15 candidate strategies is presented, where the vertical axis represents the adjustment gradient of the flow regulation gain, and the horizontal axis represents the adjustment gradient of the pressure control threshold. The optimal condition here is typically set to find the control parameter combination that maximizes the expected improvement. The calculated values ​​are obtained by iterating through all candidate combinations. Find the maximum value among them. and their corresponding control parameter combinations This optimal combination of control parameters This means that the data is parsed and encapsulated into specific dynamic control instructions. For example, if the optimal combination is flow gain +2% and pressure threshold -10kPa, then a digital instruction containing the target flow and target pressure is generated and sent to the lower-level closed-loop control module via the industrial bus to drive the actuators such as frequency converters and control valves.

[0050] For example, during the cruise operation of an aircraft engine lubricating oil system, based on the diagnostic status of "small-diameter bubble enrichment," the control module needs to optimize the separation environment. The system first generates a set of candidate combinations in the discrete control parameter space, for example, the flow gain is set to... The pressure threshold is set to For a specific candidate combination, such as flow gain +2% and pressure threshold -10kPa, the system invokes a prediction model built on a Long Short-Term Memory (LSTM) network to predict the feature vector of the future separation process after the action is performed and the fluid inertia stabilizes for 15 seconds. Subsequently, this vector is substituted into the health evaluation function of the separation process. In the middle, the expected improvement value is calculated quantitatively. Set weight coefficients The range is [0.2, 0.2, 0.1, 0.2, 0.2, 0.1]. If the normalization function... The processed scores are [0.9, 0.85, 0.95, 0.9, 0.8, 0.85], so the expected improvement value is calculated as follows: The system iterates through 15 candidate combinations and calculates their... Value. If 0.87 is the maximum value. Then the corresponding optimal combination of control parameters The flow gain of +2% and the pressure threshold of -10 kPa are parsed and encapsulated into digital commands containing the target flow rate and target pressure, which drive the frequency converter and regulating valve via the industrial bus. This method performs online optimization based on a predictive model and a health evaluation function, enabling advance evaluation and selection of the optimal control action. This feedforward predictive control mode avoids drastic flow field fluctuations caused by blind adjustment, ensuring that the system always evolves smoothly towards the operating point with the optimal separation effect.

[0051] Optionally, the method further includes: Based on the bubble morphology classification in the oil bubble depth diagnostic status, adjust the ideal target interval parameter of the separation process health evaluation function; When the bubble morphology is classified as small-diameter bubble enrichment, the parameters of the ideal target range are adjusted to focus on reducing the frequency of differential pressure fluctuations. When the bubble morphology is classified as a large-diameter bubble cluster, the parameters of the ideal target interval are adjusted to focus on optimizing the vibration energy distribution in a specific frequency band.

[0052] Specifically, this method assumes that the oil bubble depth diagnosis module has already output a clear diagnostic conclusion including bubble morphology classification. Before performing online optimization calculations, the separation process health evaluation function is first adjusted based on this classification result. The internal parameters, namely, the parameters for adjusting its ideal target interval. The ideal target interval parameter refers to the parameter within the normalized function of the health assessment function. The target value or target range used. Normalization function. A common implementation is the Gaussian membership function, which is mathematically expressed as: , in, For characteristic components The normalized evaluation value, It is a specific component in the feature vector of the separation process, such as the frequency of pressure difference fluctuations or the vibration energy in a specific frequency band. and Together they constitute the characteristic components The ideal target interval parameters, where This represents the central value of the feature under ideal conditions, while This represents its acceptable range of fluctuation or standard deviation. When the eigenvalue... The closer to the center value , The closer the value is to 1, the healthier the indicator; conversely, the closer it is to 0. Dynamically adjusting the ideal target range parameter involves modifying it in real time based on the bubble shape classification. and The value of .

[0053] When the bubble morphology classification in the oil bubble depth diagnostic status is determined to be "enrichment of small-diameter bubbles," the primary problem identified is the large number of difficult-to-separate microbubbles. These microbubbles typically cause high-frequency pressure pulsations. Therefore, an automatic adjustment logic is triggered to focus on reducing the frequency of differential pressure fluctuations. In the separation process health evaluation function, for characteristic components related to the frequency of differential pressure fluctuations, such as those characterizing high-frequency fluctuations... Its ideal target center value This could be significantly reduced, potentially decreasing its acceptable range. In this way, in subsequent optimization calculations, any future that can be predicted will be considered. Control strategies that lower the value will achieve a higher expected improvement value in the health assessment. Therefore, they are more likely to be selected in the end.

[0054] Conversely, when the bubble morphology is classified as "large-diameter bubble clusters," the main problem shifts to the intense mechanical impacts and low-frequency pulsations caused by the impact, deformation, and breakup of these large bubble clusters. The control objective then becomes stabilizing the flow field and avoiding this destructive flow pattern. This triggers another set of adjustment logic to focus on optimizing the vibrational energy distribution within a specific frequency band. For low-frequency vibrational energy... and low-frequency pressure pulsation The relevant feature components, and their ideal target center values and acceptable range Adjust to an ideal numerical range corresponding to stable flow. Simultaneously, this may increase the weight of these characteristic components in the overall health function score. In this way, the optimization process will prioritize combinations of control parameters that can effectively suppress low-frequency vibrations and pressure fluctuations, prompting the system to quickly escape the unstable state of large-diameter bubble clusters.

[0055] For example, when the diagnostic conclusion changes to "large-diameter bubble clusters," the primary issue is the unstable pulsation caused by the impaction of large bubbles. The evaluation module adjusts the normalization function. The parameters. For low-frequency vibration components. Its evaluation function is The system assigns the ideal center value based on the classification conclusion. Reduced from 1.0 to 0.5, and within the acceptable range. Reduced from 0.5 to 0.2. If the currently measured eigenvalue... The score is 1.2, the original score was approximately 0.923, and the adjusted score is 1.2. The step-like decrease in the score prompts the system to prioritize control parameter combinations that effectively suppress low-frequency vibrations and pressure fluctuations during subsequent optimization. This method achieves directional control guidance for different fault modes by dynamically adjusting the ideal target interval parameters of the evaluation function. Focusing on frequency suppression or energy optimization for small-diameter and large-diameter bubbles respectively, the control strategy is highly targeted and can quickly resolve the core separation contradictions under the current operating conditions.

[0056] S5. Adjust the actuator of the spiral flow channel separator according to the dynamic control command and collect the adjusted internal process parameters to generate an updated separation process feature vector.

[0057] Optionally, the generation of the updated separation process feature vector includes: The dynamic control command is analyzed to obtain the flow regulation gain and pressure control threshold, and the actuator is driven to generate the changed flow field operating state. After a preset fluid inertial stabilization time, dynamic differential pressure signals and pipe wall vibration signals are collected under the changed flow field operating state to generate adjusted internal process parameters. Feature extraction operations are performed on the adjusted internal process parameters to generate an updated separation process feature vector.

[0058] Specifically, a dynamic control command is a data packet containing a specific target value. This packet is parsed to extract key control parameters, namely flow regulation gain and pressure control threshold. For example, a command might be parsed as "increase the main oil pump frequency by 3 Hz" and "decrease the setpoint of the return oil line pressure control valve by 15 kPa". These parsed values ​​are converted into a communication protocol format that the corresponding actuators can recognize, such as Modbus RTU or CANopen, and sent to the main oil pump's frequency converter and the pressure control valve's PID controller via fieldbus. Upon receiving the command, the actuator immediately acts, changing the hydrodynamic conditions of the lubricating oil circulation loop, thereby generating the altered flow field operating state.

[0059] After the adjustment action is performed, a preset waiting period, known as the fluid inertial settling time, is entered. To ensure that the acquired data accurately reflects the stable characteristics under the new operating conditions, it is necessary to wait for the flow field to transition from a transient response to a new steady or quasi-steady state. The length of this settling time is predetermined through experimental calibration or fluid dynamics simulation, depending on factors such as the specific system's pipeline length, fluid viscosity, and pump / valve response characteristics. Its magnitude is typically between several seconds and tens of seconds, for example, set to 15 seconds. Immediately after the fluid inertial settling time ends, the data acquisition program is started to collect the internal process parameters under the changed flow field operating state. At the same high sampling frequency as the initial data acquisition, the dynamic pressure difference signal between the inlet and outlet of the spiral flow separator and the vibration signal of the pipe wall are collected again. The data collected this time is called the adjusted internal process parameters because it directly reflects the actual impact of the control action on the physical processes in the core region of the separator.

[0060] The same feature extraction operations as in the initial stage are performed on the adjusted internal process parameters to generate an updated separation process feature vector. Upon receiving newly acquired dynamic differential pressure signals and pipe wall vibration signals, time-frequency analysis and spectral energy analysis are performed to calculate new differential pressure fluctuation characteristic parameters and vibration energy distribution characteristic parameters, which are then fused into a new vector. This vector is the updated separation process feature vector, which digitally describes the system's latest operating state after completing one control adjustment. It serves as the input for the next control cycle and is passed to the oil bubble depth diagnostic module, thus initiating a new round of diagnostic-optimization-control loop.

[0061] For example, after the control command execution mechanism is activated, the system enters the feedback monitoring phase. First, the system parses the dynamic control command data packet and extracts key control parameters, such as increasing the main oil pump frequency by 3Hz and decreasing the pressure control valve setpoint by 15kPa. These values ​​are converted to Modbus RTU protocol format and used to drive the frequency converter and PID controller via the fieldbus, generating the changed flow field operating state. To avoid transient shocks during adjustment, the system enters a preset 15s fluid inertial stabilization time. After the flow field transitions to the new steady state, the acquisition module recaptures the dynamic differential pressure signals and pipe wall vibration signals at the inlet and outlet of the spiral flow separator at a sampling rate of 10kHz, generating the adjusted internal process parameters. Taking the dynamic differential pressure signal as an example, continuous wavelet transform is applied under the new steady state to recalculate the updated mid-to-high frequency integrated energy. The value was 7.5, a significant decrease from the unadjusted 8.2, indicating that the intense activity of the small bubble clusters was suppressed. The system then performed the same feature extraction operation on all adjusted parameters to calculate the updated value. , , , , , The components are then fused into an updated feature vector representing the separation process. This vector digitally characterizes the latest physical state of the separator after control intervention and is passed as the input to the diagnostic module for the next cycle, achieving closed-loop adaptive control. This method incorporates fluid inertial settling time, ensuring that the feedback data is acquired in the adjusted steady-state phase and eliminating the influence of transient disturbances during adjustment on feedback accuracy. This provides a realistic physical feedback basis for evaluating control quality and ensures the iterative stability of the control loop.

[0062] Optionally, the method further includes: The updated separation process feature vector is compared with the expected improvement value to generate a control effect feedback signal; The data and state mapping rules are corrected online using the control effect feedback signal.

[0063] Specifically, the updated separation process feature vector, measured after adjustment, is compared with the expected improvement value generated during the control decision-making stage to generate a quantified control effect feedback signal. This comparison is not a direct comparison of a vector and a scalar, but rather a comparison of the updated separation process feature vector... Substitute it again into the health evaluation function of the separation process. In this process, the actual health score resulting from this control and adjustment is calculated. Then, this actual score is compared with the highest expected improvement value that was anticipated during optimization. Comparison. Control effect feedback signal. The difference between the two can be defined as: , Here, This is the control effect feedback signal, which is a scalar. If A value close to zero or positive indicates that the actual improvement meets or exceeds expectations, and the control model is accurate. A significant negative value indicates that the prediction is seriously inconsistent with the actual situation, suggesting that the data and state mapping relationship within the system may have drifted and needs to be corrected.

[0064] The data-state mapping rules are corrected online using control effect feedback signals. This correction process consists of a trigger condition and an update algorithm. The trigger condition is the control effect feedback signal. If the absolute value exceeds a preset error tolerance threshold, for example, a threshold set to 0.1, then a correction procedure is initiated when the predicted health improvement differs from the actual improvement by more than 10%. When correction is triggered, it is assumed that before implementing control, the initial separation process feature vector is mapped by the data and state rules. The parsed internal bubble motion pattern labels There is uncertainty. The complete data chain recorded in this control iteration, i.e. This serves as a new learning sample. It is the feature vector before control. It is the optimal control instruction executed. This is the actual feature vector after control. This sample is fed into an online learning module. This module employs an incremental learning algorithm, such as gradient-based online learning or unsupervised self-organizing map networks. The goal of the algorithm is to fine-tune the model parameters of the data-state mapping rules, such as the weights of the neural network, so that, given... In this case, the classification results output by the model can better explain why control is applied. Later it will produce This is the result. By continuously accumulating these corrective samples triggered by significant prediction errors, the data and state mapping rules can gradually adapt to the slow changes in the system caused by factors such as wear and oil aging, thus achieving self-evolution of the model.

[0065] For example, the updated feature vector Substitute into the evaluation function to get the actual score The expected improvement value in the prediction phase is Calculate the feedback signal: The system sets an error tolerance threshold of 0.1. Although the current single-error absolute value of 0.04 did not trigger correction, the system records this data. If in subsequent iterations... If the threshold is exceeded, an incremental learning process is initiated. This utilizes the control data chain before and after the process. , , As new samples, fine-tune model parameters and correct Existing model analysis deviation This method enables the rules to adapt to nonlinear drift caused by oil aging or wear. By constructing a control effect feedback signal for online correction, the system acquires self-evolution capabilities, automatically compensating for model drift caused by long-term factors such as oil aging and mechanical wear. This mechanism ensures that the system maintains a high level of diagnostic accuracy and control reliability throughout its entire lifecycle.

[0066] Based on the same inventive concept, such as Figure 4 As shown, the present invention also provides an oil-gas separation control system based on a spiral flow channel, comprising: The multi-source data acquisition module is used to collect external circulation parameters of the lubricating oil circulation loop and internal process parameters of the spiral flow channel separator, and generate multi-source synchronous data. The feature extraction module is used to perform feature extraction operations on the internal process parameters in the multi-source synchronous data to generate a separation process feature vector. The oil bubble depth diagnosis module is used to analyze the external loop parameters and the feature vector of the separation process using data and state mapping rules to generate the oil bubble depth diagnosis status. The control strategy optimization module is used to perform online optimization calculations based on the oil bubble depth diagnostic status and the feature vector of the separation process, and generate dynamic control commands. The closed-loop adjustment module is used to adjust the actuator of the spiral flow channel separator according to the dynamic control command and collect the adjusted internal process parameters to generate an updated separation process feature vector.

[0067] It should be noted that the electrical connections between the various units described above do not necessarily represent direct or indirect connections. Any indirect connection method can be applied to the embodiments of the present invention as long as it achieves the purpose of the present invention. The above descriptions are merely exemplary embodiments of the present invention and should not be construed as limiting the scope of the present invention.

[0068] All equivalent changes and modifications made in accordance with the teachings of this invention are still within the scope of this invention. Those skilled in the art will readily conceive of other embodiments of this invention upon considering the specification and the disclosure of practical truth. This application is intended to cover any variations, uses, or adaptations of this invention that follow the general principles of this invention and include common knowledge or conventional techniques in the art not described herein.

Claims

1. A method for controlling oil-gas separation based on a spiral flow channel, characterized in that, The method includes: Collect external circulation parameters of the lubricating oil circulation loop and internal process parameters of the spiral flow channel separator to generate multi-source synchronous data; Feature extraction operations are performed on the internal process parameters in the multi-source synchronous data to generate a separation process feature vector; The external loop parameters and the feature vector of the separation process are analyzed using data and state mapping rules to generate the oil bubble depth diagnostic status; Based on the oil bubble depth diagnostic status and the feature vector of the separation process, online optimization calculation is performed to generate dynamic control commands; The actuator of the spiral flow channel separator is adjusted according to the dynamic control command, and the adjusted internal process parameters are collected to generate an updated separation process feature vector.

2. The oil-gas separation control method based on a spiral flow channel according to claim 1, characterized in that, The generation of multi-source synchronization data includes: The system acquires the oil tank level signal, main circuit pressure signal, and oil turbidity signal to generate external circulation parameters. The dynamic differential pressure signal and pipe wall vibration signal at key locations inside the spiral flow channel separator are acquired to generate internal process parameters; The external loop parameters and the internal process parameters are time-scaled and aligned to generate multi-source synchronized data.

3. The oil-gas separation control method based on a spiral flow channel according to claim 2, characterized in that, The feature vector generated during the separation process includes: Time-frequency analysis is performed on the dynamic differential pressure signal to generate differential pressure fluctuation characteristic parameters; Spectral energy analysis is performed on the pipe wall vibration signal to generate vibration energy distribution characteristic parameters; The pressure difference fluctuation characteristic parameters and the vibration energy distribution characteristic parameters are fused into a vector to generate a separation process characteristic vector.

4. The oil-gas separation control method based on a spiral flow channel according to claim 3, characterized in that, The generated oil bubble depth diagnostic status includes: Construct data and state mapping rules, parse the feature vector of the separation process, and generate the internal bubble motion pattern; The internal bubble motion pattern is verified and corrected by using the liquid level signal change trend in the external circulation parameters, and a corrected bubble motion pattern is generated. The modified bubble motion pattern is morphologically classified and load assessed to generate an oil bubble depth diagnostic status that includes bubble morphology classification and separation load level.

5. The oil-gas separation control method based on a spiral flow channel according to claim 4, characterized in that, The rules for constructing the data and state mapping include: Collect historical multi-source synchronous data under various operating conditions and calculate the corresponding historical separation process feature vectors. At the same time, obtain the internal bubble motion pattern labels determined by observation methods and generate a labeled training dataset. Machine learning algorithms are used to learn and fit features to the labeled training dataset, generating data-state mapping rules.

6. The oil-gas separation control method based on a spiral flow channel according to claim 4, characterized in that, The generated dynamic control commands include: The oil bubble depth diagnostic status and the separation process feature vector are imported into the separation process health evaluation function to generate the expected improvement effect value under different combinations of control parameters. Select a combination of control parameters that makes the expected improvement value meet the preset preferred conditions, and generate dynamic control commands.

7. The oil-gas separation control method based on a spiral flow channel according to claim 6, characterized in that, The method further includes: Based on the bubble morphology classification in the oil bubble depth diagnostic status, adjust the ideal target interval parameter of the separation process health evaluation function; When the bubble morphology is classified as small-diameter bubble enrichment, the parameters of the ideal target range are adjusted to focus on reducing the frequency of differential pressure fluctuations. When the bubble morphology is classified as a large-diameter bubble cluster, the parameters of the ideal target interval are adjusted to focus on optimizing the vibration energy distribution in a specific frequency band.

8. The oil-gas separation control method based on a spiral flow channel according to claim 6, characterized in that, The generated updated separation process feature vector includes: The dynamic control command is analyzed to obtain the flow regulation gain and pressure control threshold, and the actuator is driven to generate the changed flow field operating state. After a preset fluid inertial stabilization time, dynamic differential pressure signals and pipe wall vibration signals are collected under the changed flow field operating state to generate adjusted internal process parameters. Feature extraction operations are performed on the adjusted internal process parameters to generate an updated separation process feature vector.

9. The oil-gas separation control method based on a spiral flow channel according to claim 8, characterized in that, The method further includes: The updated separation process feature vector is compared with the expected improvement value to generate a control effect feedback signal; The data and state mapping rules are corrected online using the control effect feedback signal.

10. An oil-gas separation control system based on a spiral flow channel, characterized in that, The system includes: The multi-source data acquisition module is used to collect external circulation parameters of the lubricating oil circulation loop and internal process parameters of the spiral flow channel separator, and generate multi-source synchronous data. The feature extraction module is used to perform feature extraction operations on the internal process parameters in the multi-source synchronous data to generate a separation process feature vector. The oil bubble depth diagnosis module is used to analyze the external loop parameters and the feature vector of the separation process using data and state mapping rules to generate the oil bubble depth diagnosis status. The control strategy optimization module is used to perform online optimization calculations based on the oil bubble depth diagnostic status and the feature vector of the separation process, and generate dynamic control commands. The closed-loop adjustment module is used to adjust the actuator of the spiral flow channel separator according to the dynamic control command and collect the adjusted internal process parameters to generate an updated separation process feature vector.

Images