Safety adaptive intelligent system and method for external delivery pump based on multi-sensor fusion

Abstract

This invention discloses an adaptive intelligent system and method for the safety of an export pump based on multi-sensor fusion, relating to the field of intelligent equipment operation and maintenance technology. The adaptive method includes: constructing a first state feature encoding based on multiple deployed sensors; extracting the local first-order change rate of the corresponding state feature at each time moment; if the corresponding state feature is determined to be in an abnormal trigger state at the current time moment, extracting the main control and adjustment path; determining the jump coefficient of the corresponding adjustment action of each control node based on the current value of the operating status of each control node in the main control and adjustment path and the average response time in historical samples, and encoding it into a control strategy instruction set for issuing control commands; determining the feedback response coefficient of the corresponding adjustment action of each control node by collecting the set of operating parameter feedback values ​​of each control node; if there is a weak response action, regenerating the adjusted control strategy instruction set to complete the adaptive control process for the safety of the export pump.

Description

Technical Field

[0001] This invention relates to the field of intelligent operation and maintenance technology for equipment, specifically to an adaptive intelligent system and method for the safety of external pumps based on multi-sensor fusion. Background Technology

[0002] As a key power equipment in oil and gas, chemical and water conservancy transportation systems, external pumps operate continuously under high pressure differentials, large flow rates and complex loads, and are easily affected by operating condition disturbances and equipment aging, which can lead to operational fluctuations.

[0003] In existing technologies, the operational safety of export pumps largely relies on the monitoring results of single-point sensors, using only pressure and flow rate as the basis for alarms or start-stop control. However, due to the complex structure of the equipment and the strong coupling of its operating states, different abnormal symptoms often exhibit abrupt changes in multiple parameters. Furthermore, existing monitoring systems generally lack data fusion and causal modeling capabilities, resulting in insensitivity to early anomaly identification, insufficient analysis of anomaly development paths, and difficulty in effectively matching the response of adjustment nodes. This can even lead to unpredictable state transitions during the operation of export pumps, resulting in typical abnormal phenomena such as severe pressure fluctuations, flow rate imbalances, and abnormally high motor loads. Ultimately, this can cause serious consequences such as pump damage, reduced efficiency, or even shutdown. To ensure its long-term stable, safe, and efficient operation, it is urgent to construct an adaptive intelligent safety assurance mechanism for export pumps that can integrate multiple operating parameters, possess causal identification capabilities, and have a closed-loop adjustment response. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a safety adaptive intelligent system and method for external pumps based on multi-sensor fusion, which solves the problems mentioned in the background technology.

[0005] To achieve the above objectives, the present invention provides a safety adaptive method for export pumps based on multi-sensor fusion, comprising the following steps:

[0006] S1. Based on the deployment of multiple sets of sensors, the first operating parameter information of the external pump is collected in real time. After time synchronization processing, the operating parameter frames at each moment are formed to construct the first state feature code.

[0007] S2. Based on the constructed first state feature encoding, extract the local first-order change rate of the corresponding state feature at each time. If it is determined that the corresponding state feature is an abnormal trigger state at the current time, calculate the trigger causal weight index of each abnormal node according to the recorded abnormal trigger vector, and construct an abnormal causal chain graph.

[0008] S3. Based on the constructed abnormal causal chain graph, generate a set of candidate adjustment paths to extract the main control adjustment path. According to the current value of the running status of each control node in the main control adjustment path and the average response time in the historical samples, determine the jump coefficient of the corresponding adjustment action of each control node and encode it into a control strategy instruction set.

[0009] S4. Using the control strategy instruction set as the control command to collect the set of operating parameter feedback values ​​of each control node, determine the feedback response coefficient of the corresponding adjustment action of each control node, and if there is a corresponding adjustment action that is a weak response action, regenerate the adjusted control strategy instruction set.

[0010] Preferably, step S1 specifically includes:

[0011] S11. Multiple sets of sensors are deployed in the inlet pipe section, outlet pipe section, pump body shell, motor connection area, mounting base and support structure of the external pump, and connected to the edge fusion data platform through an industrial bus with time synchronization function. The multiple sets of sensors include pressure sensors, flow rate sensors, temperature sensors, acceleration sensors, current sensors and displacement sensors.

[0012] S12. Based on the deployment of multiple sets of sensors, the operating status of the external pump is monitored in real time to obtain the first operating parameter information. The first operating parameter information is time-aligned according to the acquisition channel and measurement frequency to construct a unified parameter sampling time axis. Based on the sampling time axis, the first operating parameter information is time-synchronized to form operating parameter frames at each moment. The first operating parameter information includes inlet and outlet pressure, pump chamber temperature rise, motor current value, shell vibration intensity and axial displacement change.

[0013] Preferably, in step S13, for the operating parameter frames at each time point, the parameters are classified based on their physical properties. Pressure parameters are classified into the first parameter subset, temperature and current parameters into the second parameter subset, and vibration and displacement parameters into the third parameter subset, thus constructing a related coupled operating parameter subset. Standardization processing is performed on each parameter subset within the related coupled operating parameter subset. The standardization processing includes: performing normalization transformation on each operating parameter based on historical extreme values, and removing and completing parameter points with abnormal abrupt changes within the continuous sampling window. The numerical completion process uses the interpolation averaging method of samples from previous and subsequent time points for smooth repair.

[0014] S14. After standardization, based on the temporal change rate of each operating parameter, the incremental amplitude between adjacent samples, and the relative disturbance level characteristic index, a first state feature code is constructed. The first state feature code includes instantaneous pressure, pump body temperature, input current, vibration frequency, and displacement.

[0015] Preferably, step S2 specifically includes:

[0016] S21. Based on the formed first state feature encoding, perform dynamic sliding processing of multi-source state parameters to construct a multi-dimensional state feature trajectory group, wherein the multi-dimensional state feature trajectory group includes pressure gradient curve, vibration change curve, electrothermal response trajectory and displacement offset trend line.

[0017] S22. Perform first-order derivative calculations on each trajectory curve in the multidimensional state feature trajectory group to extract the local first-order rate of change of the corresponding state feature at each time step, specifically: In the formula, This represents the first-order rate of change of the corresponding state characteristic at time t. This represents the observed parameter value of the corresponding state feature at time t. Indicates the corresponding state feature at time t. The observed parameter values, Indicates the observation interval;

[0018] The local first-order rate of change of the corresponding state feature at each time step is compared with a preset mutation threshold. When the local first-order rate of change is greater than the mutation threshold, the corresponding state feature is determined to be in an abnormal trigger state at the current time step, and the corresponding abnormal trigger vector is recorded. Otherwise, it indicates that the operation status of the external pump is normal at the current time step.

[0019] Preferably, in step S23, an anomaly triggering cause-effect graph is constructed according to the generation order and the parameter category of the corresponding anomaly triggering vector. The anomaly triggering cause-effect graph includes multiple anomaly nodes and directed edges. An anomaly node represents the corresponding parameter anomaly type, and a directed edge represents the evolution path from the previous node triggering the anomaly of the next node.

[0020] S24. Based on the anomaly triggering cause-effect graph, according to the set time sliding window, calculate the average triggering delay time, total anomaly frequency, and subsequent associated anomaly frequency for each anomaly node, and calculate the triggering causality weight index for each anomaly node. Specifically: In the formula, This represents the triggering causality weight index of the corresponding abnormal node. This indicates the number of subsequent associated anomalies for the corresponding abnormal node. This indicates the total number of times the anomaly occurred for the corresponding abnormal node. This indicates the average trigger delay time for the corresponding abnormal node;

[0021] S25. Perform weighted enhancement processing on the triggering causal weight index of each abnormal node in the abnormal triggering causal graph with directed edge weights to generate an enhanced and corrected abnormal causal chain graph. The abnormal causal chain graph represents the set of abnormal causal paths that trigger safety risks under the current operating state of the external pump.

[0022] Preferably, step S3 specifically includes:

[0023] S31. Based on the generated abnormal causal chain graph, extract the corresponding causal chain path where the sum of the triggering causal weight indices of all abnormal nodes in the corresponding causal chain path exceeds a preset threshold, and record it as a candidate adjustment path. Construct a candidate adjustment path set, wherein each candidate adjustment path in the candidate adjustment path set consists of several consecutive abnormal nodes.

[0024] S32. Based on the operating parameter types associated with each abnormal node in the candidate adjustment path, retrieve the controllable adjustment execution units associated with the corresponding operating parameter types and establish a parameter-execution unit mapping table; through statistical analysis of the controllability success rate and adjustment amplitude response sensitivity of the controllable adjustment execution units corresponding to each abnormal node, and after normalization, obtain the controllability coefficient of the controllable adjustment execution units corresponding to each abnormal node; based on the candidate adjustment path set, correlate the causal influence index of each abnormal node in each candidate adjustment path with the controllability coefficient of the corresponding controllable adjustment execution unit to determine the control action priority coefficient of each candidate adjustment path, specifically: In the formula, This represents the priority coefficient of the control action of the corresponding candidate regulation path. This represents the triggering causality weight index of the corresponding abnormal node. This represents the controllability coefficient of the controllable adjustment execution unit corresponding to the abnormal node. This represents the sequence of abnormal nodes contained in the corresponding candidate adjustment path.

[0025] Preferably, in step S33, based on the control action priority coefficients of each candidate control path, the candidate control path corresponding to the maximum value of the control action priority coefficient is extracted as the main control control path. The target parameter values ​​of the state parameters of each control node covered by the starting and ending control nodes of the main control control path are recorded, and each control node in the main control control path is numbered according to the time sequence.

[0026] S34. For the target parameter values ​​of the corresponding state parameters of each control node in the main control and regulation path, combined with the current values ​​of the operating status of each control node and the average response time in historical samples, analyze the control intensity that should be applied for the corresponding regulation action of each control node in the main control and regulation path, and determine the jump coefficient of the corresponding regulation action of each control node, specifically: In the formula, This represents the jump coefficient corresponding to the adjustment action of the corresponding control node. This indicates the target parameter value that the corresponding control node should achieve for the adjustment action. This indicates the current parameter value of the corresponding control node. This represents the average response time of the corresponding control node in historical samples;

[0027] S35. Encode the jump coefficients of the corresponding adjustment actions of each control node in the main control path into a control strategy instruction set, wherein the control strategy instruction set includes the control field of the main control path number and the control parameter name, jump factor value and response time corresponding to each control node.

[0028] Preferably, step S4 specifically includes:

[0029] S41. The control strategy instruction set formed based on S35 is used as input. It is indexed according to the number of the main adjustment path and control commands are issued to the corresponding control execution components according to the content of the control field. The control execution components include the frequency converter, proportional valve, motor control unit, cooling fan start / stop unit and fluid bypass controller in the external pump.

[0030] S42. After the control command is issued, according to the set feedback cycle. The set of operating parameter feedback values ​​of each control node is periodically collected. The set of operating parameter feedback values ​​includes the feedback parameter values, instantaneous rate of change, and average response gain duration of the corresponding control node within the feedback period.

[0031] S43. Based on the set of operating parameter feedback values ​​for each control node, determine the feedback response coefficient for the corresponding adjustment action of each control node, specifically as follows: In the formula, This represents the feedback response coefficient corresponding to the adjustment action of the corresponding control node. This indicates the target parameter value that the corresponding control node should achieve for the adjustment action. This indicates the current parameter value of the corresponding control node. This represents the feedback parameter value of the corresponding control node during the feedback period T;

[0032] S44. Compare the feedback response coefficients of the corresponding adjustment actions of each control node determined in S43 with the preset response thresholds. When the feedback response coefficient of the corresponding adjustment action of a control node is less than the response threshold, mark the corresponding adjustment action as a weak response action and record the number of the corresponding control node, the jump coefficient and the name of the corresponding adjustment parameter to form an adjustment failure record table. Otherwise, it indicates that the operation status of the external pump after the control command is executed is normal.

[0033] Preferably, in step S45, based on the control parameters recorded in the adjustment failure record table, the post-adjustment jump coefficient of each control node corresponding to the adjustment action is determined, specifically as follows: In the formula, This represents the adjustment jump coefficient after adjustment corresponding to the adjustment action of the corresponding control node. This represents the gain adjustment correction factor;

[0034] S46. Based on the adjusted jump coefficients of the corresponding adjustment actions of each control node determined in S45, regenerate the adjusted control strategy instruction set and execute the operation procedures in S41-S44 until the external pump is in normal operating condition.

[0035] The external pump safety adaptive intelligent system based on multi-sensor fusion includes a coding construction module, a causal analysis module, an intensity analysis module, and an adaptive control module.

[0036] The coding construction module is based on multiple deployed sensors to collect the first operating parameter information of the external pump in real time. After time synchronization processing, it forms the operating parameter frames at each moment to construct the first state feature code.

[0037] The causal analysis module extracts the local first-order change rate of the corresponding state feature at each time step based on the constructed first state feature encoding. If the corresponding state feature is determined to be an abnormal trigger state at the current time step, the trigger causal weight index of each abnormal node is calculated according to the recorded abnormal trigger vector, and an abnormal causal chain graph is constructed.

[0038] The intensity analysis module generates a set of candidate regulation paths based on the constructed abnormal causal chain graph, in order to extract the main control regulation path. Based on the current value of the operating status of each control node in the main control regulation path and the average response time in historical samples, it determines the jump coefficient of the corresponding regulation action of each control node and encodes it into a control strategy instruction set.

[0039] The adaptive control module is used to issue control commands based on the control strategy instruction set, collect the set of operating parameter feedback values ​​of each control node, determine the feedback response coefficient of the corresponding adjustment action of each control node, and if there is a corresponding adjustment action that is a weak response action, regenerate the adjusted control strategy instruction set.

[0040] This invention provides a safety adaptive intelligent system and method for export pumps based on multi-sensor fusion, which has the following beneficial effects:

[0041] (1) The safety adaptive intelligent system and method for export pumps based on multi-sensor fusion provided by the present invention can realize the synchronous acquisition, deep fusion and abnormal causal chain modeling of multiple key operating parameters. Under the actual working conditions of strong heterogeneity of multi-source state parameters and complex coupling mechanism, it has high sensitivity of abnormal state identification and adjustment response accuracy. Compared with the existing start-stop control technology based only on single-point monitoring, the present invention can realize early detection of early fault trends including typical abnormal phenomena such as pressure fluctuation, unstable flow rate, abnormal temperature rise and sudden change of motor load. By constructing an abnormal causal chain map, it can identify the key driving path that causes state transition. Combined with control reachability and node response rate, it can dynamically generate a closed-loop adjustment strategy to effectively prevent abnormal propagation and system instability. The overall solution has the technical advantages of reasonable structural deployment, clear causal inference and closed-loop strategy generation. It can significantly reduce the downtime rate and operation and maintenance intervention frequency while ensuring the stability of export pumps under extreme operating conditions of high pressure difference, large flow rate and high frequency start-stop, and improve the operating efficiency and service life of export pumps. It has high engineering practical value and compatibility.

[0042] (2) By deploying multi-type sensor arrays in the key structural areas of the external pump and introducing an industrial communication bus with high time accuracy for synchronous acquisition, the physical quantities of pressure, flow rate, temperature, vibration, current and displacement can be uniformly incorporated into the standardized data channel for processing, thereby significantly improving the system's comprehensive coverage of different operating conditions. On this basis, the present invention constructs a multi-dimensional state feature set by performing time alignment, normalization and missing data completion on the acquired data, and further extracts high-dimensional coupling features such as instantaneous pressure difference change rate, temperature-flow coupling index, current fluctuation gradient and mechanical displacement offset, thereby improving the ability to express multi-parameter linkage mutations. At the same time, the use of sliding window analysis to construct the state evolution trajectory can realize the identification of early weak abnormal changes, effectively overcoming the lag defect of existing systems that only alarm when the abnormal critical threshold is broken, significantly enhancing the early identification capability and precursor perception level of multi-type composite anomalies during the operation of the external pump, and laying a high-quality data foundation for subsequent causal inference and response control.

[0043] (3) By constructing an abnormal causal trigger graph and causal chain graph, the triggering relationship between each state node is modeled and quantified. Combined with the triggering frequency and response characteristics of each abnormal node within the sliding time window, its causal weight index is calculated, thereby forming a set of critical paths that can be adjusted. This mechanism avoids the modeling distortion problem caused by the causal relationship relying on empirical rules or template matching in the prior art, making the abnormal development process more traceable and logically consistent. On this basis, the controllability factor of the adjustment path and the adjustment jump coefficient are introduced to dynamically calculate the priority of the control action of each path. Combined with the actuator control capability and the current running state of the node, the adjustment instruction set is constructed, realizing the closed-loop process of "abnormal path identification - control channel matching - adjustment strategy generation - feedback correction". By incorporating the feedback response coefficient into the adjustment mechanism, path jump and strategy correction are performed when adjustment fails, effectively avoiding the static defect of no alternative path after control failure in the traditional system, and improving the dynamic response and adaptive control capability in the abnormal evolution process. Attached Figure Description

[0044] Figure 1 This is a schematic diagram of the safety adaptive method for export pumps based on multi-sensor fusion according to the present invention;

[0045] Figure 2 This is a block diagram of the safety adaptive intelligent system for external pumps based on multi-sensor fusion, as described in this invention.

[0046] Figure 3 This is a logic diagram of the safety adaptive method for external pumps based on multi-sensor fusion, as described in this invention.

[0047] Figure 4 This is a schematic diagram illustrating the connection relationship of the multi-sensor fusion in this invention. Detailed Implementation

[0048] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] Example 1

[0050] Please see Figure 1 and Figure 3 This invention provides a safety adaptive method for export pumps based on multi-sensor fusion, comprising the following steps:

[0051] S1. Based on the deployment of multiple sets of sensors, the first operating parameter information of the external pump is collected in real time. After time synchronization processing, the operating parameter frames at each moment are formed to construct the first state feature code.

[0052] S2. Based on the constructed first state feature encoding, extract the local first-order change rate of the corresponding state feature at each time. If it is determined that the corresponding state feature is an abnormal trigger state at the current time, calculate the trigger causal weight index of each abnormal node according to the recorded abnormal trigger vector, and construct an abnormal causal chain graph.

[0053] S3. Based on the constructed abnormal causal chain graph, generate a set of candidate adjustment paths to extract the main control adjustment path. According to the current value of the running status of each control node in the main control adjustment path and the average response time in the historical samples, determine the jump coefficient of the corresponding adjustment action of each control node and encode it into a control strategy instruction set.

[0054] S4. Using the control strategy instruction set as the control command to collect the set of operating parameter feedback values ​​of each control node, determine the feedback response coefficient of the corresponding adjustment action of each control node, and if there is a corresponding adjustment action that is a weak response action, regenerate the adjusted control strategy instruction set.

[0055] In this embodiment, by constructing a complete closed-loop process of acquisition, analysis, modeling, control, and feedback, the fault early warning capability and intelligent adjustment response efficiency of the external pump under complex operating conditions are significantly improved. Through the collaborative deployment and time synchronization mechanism of multiple heterogeneous sensors, comprehensive real-time perception of the multi-physical domain operating status of the external pump is achieved, avoiding the anomaly omission and response lag problems caused by single-point monitoring. By calculating the local first-order change rate based on state feature encoding, early-stage linked abrupt changes can be accurately identified. Furthermore, by combining anomaly trigger vectors to construct a causal chain graph, the anomaly development path and its triggering mechanism can be analyzed from the source, enabling the location and tracing of key anomaly nodes. Simultaneously… The control strategy generation mechanism is based on the state trend and response law of the main control path. It constructs a control instruction set with precise matching of jump coefficients and introduces a feedback response coefficient mechanism to dynamically evaluate the control effect. This ensures that the control path and control parameters are automatically corrected in the event of control failure or response weakening, thereby improving the adaptability and robustness. This method not only realizes the integrated closed-loop control of anomaly prediction, path identification, control matching and response correction, but also continuously optimizes the control path and control accuracy. It is particularly suitable for oil and gas transportation and high-pressure pumping applications where the equipment undergoes drastic state changes, complex anomaly chains and extremely high operational safety requirements during long-term operation. It has strong engineering practicality and industry promotion value.

[0056] Example 2

[0057] Please refer to Figure 1 , Figure 3 and Figure 4 Specifically: S1 includes the following steps:

[0058] S11. Multiple sets of sensors are deployed in the inlet pipe section, outlet pipe section, pump body shell, motor connection area, mounting base and support structure of the external pump, and connected to the edge fusion data platform through an industrial bus with time synchronization function. The multiple sets of sensors include pressure sensors, flow rate sensors, temperature sensors, acceleration sensors, current sensors and displacement sensors.

[0059] S12. Based on the deployment of multiple sets of sensors, the operating status of the external pump is monitored in real time to obtain the first operating parameter information. The first operating parameter information is time-aligned according to the acquisition channel and measurement frequency to construct a unified parameter sampling time axis. Based on the sampling time axis, the first operating parameter information is time-synchronized to form operating parameter frames at each moment. The first operating parameter information includes inlet and outlet pressure, pump chamber temperature rise, motor current value, shell vibration intensity and axial displacement change.

[0060] It should be noted that the operating parameter frame at each time point refers to the complete set of parameters formed at the same time after time alignment and synchronization of different types of operating parameters collected by multiple sensors under a unified parameter sampling time axis. It reflects the overall operating status of the external pump at that time. Its function is to unify the heterogeneous, multi-source, and different frequency collected parameters into a structured time-series data basis, which facilitates subsequent state feature extraction and anomaly identification. The operating parameter frame is obtained by interpolating, aligning, and synchronizing the raw data collected by each sensor according to the acquisition channel identifier and measurement frequency, ensuring that the data in each time slice is an effective representation with cross-sensor consistency and high temporal resolution.

[0061] Specifically, in S13, for the operating parameter frames at each time point, the parameters are classified based on their physical properties. Pressure parameters are classified into the first parameter subset, temperature and current parameters into the second parameter subset, and vibration and displacement parameters into the third parameter subset, together constructing a related coupled operating parameter subset. Standardization processing is performed on each parameter subset within the related coupled operating parameter subset. Standardization processing includes: performing normalization transformation on each operating parameter based on historical extreme values, and removing and completing the values ​​of parameter points that show abnormal changes within the continuous sampling window. The value completion process uses the interpolation averaging method of samples from previous and subsequent time points for smooth repair.

[0062] Among these methods, parameter classification based on physical properties involves categorizing and organizing operating parameters according to the physical behavior they reflect, enabling more targeted subsequent data fusion and feature modeling operations. Its purpose is to distinguish the response types described by different parameters, thereby constructing logically clear and physically consistent parameter subsets, improving the accuracy and robustness of state analysis. In terms of processing, inlet and outlet pressures reflecting changes in conveying load and liquid resistance are classified into the first parameter subset; temperature and current reflecting pump body heating and motor load characteristics are classified into the second parameter subset; and vibration and displacement parameters characterizing structural dynamic stability are classified into the third parameter subset. For example, if the operating parameter frame at a given moment contains an outlet pressure of 4.2 MPa, a motor current of 3.6 A, and a shell vibration intensity of 0.45 g, after classification, this data is categorized into three physical feature subsets, facilitating normalization, mutation removal, and completion operations respectively, forming a standardized input structure with clear physical coupling and less noise interference.

[0063] S14. After standardization, based on the temporal change rate of each operating parameter, the incremental amplitude between adjacent samples, and the relative disturbance level characteristic index, a first state feature code is constructed. The first state feature code includes instantaneous pressure, pump body temperature, input current, vibration frequency, and displacement.

[0064] It should be noted that the first state feature encoding refers to extracting core feature indicators that can sensitively reflect changes in the operating state of the external pump based on the standardized multi-class operating parameters, and combining them in a structured form into a unified feature expression vector to support subsequent anomaly identification and causal modeling analysis. Its role is to transform the original multi-source, heterogeneous data into a state representation with physical meaning and high sensitivity to change, so as to understand the current operating state and its potential anomaly trends as a whole. The processing methods include: calculating the instantaneous differential pressure change rate based on pressure parameters, constructing electrothermal coupling indicators based on temperature and current parameters, calculating the vibration change rate based on vibration data, and obtaining the displacement offset value by combining displacement parameters.

[0065] In this embodiment, the multi-source state perception and feature extraction mechanism constructed in step S1 significantly enhances the multi-dimensional perception capability and parameter timing consistency control capability of the external pump's operating status. It solves the problems of state information distortion and analysis bias caused by single acquisition channels, scattered physical quantity distribution, and asynchronous timestamps in existing technologies. By deploying multiple sets of sensors in key areas such as the external pump's inlet pipe section, pump shell, and motor connection area, and connecting them to an industrial bus system with time synchronization capabilities, the key operating parameters of pressure, temperature, current, vibration, and displacement collected by various types of sensors can be aligned and synchronized on a unified sampling time axis. This constructs a multi-dimensional parameter frame that accurately corresponds to the actual operating behavior, providing a highly timely and consistent perception foundation for subsequent feature extraction and anomaly modeling. Operating parameters are divided into multiple coupled subsets based on their physical properties and standardized separately. In particular, the use of historical extreme value normalization and outlier removal-interpolation repair mechanism not only effectively offsets the interference of different parameter dimensions on the overall modeling, but also significantly enhances the robustness and fault tolerance to instantaneous abrupt changes. The final constructed first-state feature encoding covers highly sensitive feature indicators such as instantaneous pressure difference rate, electrothermal coupling index, vibration change rate, and displacement offset value, realizing the efficient identification capability of potential fault symptoms such as pump load changes, structural loosening, abnormal motor energy consumption, and displacement trend drift. Therefore, the implementation of this step not only realizes multi-parameter fusion perception in the physical domain, but also has the effect of forward-looking expression of the trend of operating state disturbance, which is a key foundation for subsequent anomaly identification, causal chain modeling, and regulation strategy formulation.

[0066] Example 3

[0067] Please refer to Figure 1 and Figure 3 Specifically: The specific steps of S2 include:

[0068] The specific steps in S2 include:

[0069] S21. Based on the formed first state feature encoding, perform dynamic sliding processing of multi-source state parameters to construct a multi-dimensional state feature trajectory group, wherein the multi-dimensional state feature trajectory group includes pressure gradient curve, vibration change curve, electrothermal response trajectory and displacement offset trend line.

[0070] It should be noted that the multidimensional state feature trajectory set refers to a set of multidimensional time-series change curves constructed on a continuous time axis by indexing various key feature parameters extracted from the first state feature encoding. This set reflects the dynamic evolution trend of the export pump during operation. Its function is to reveal the change process of the operating state and the development path of potential anomalies, providing an intuitive and structured data foundation for subsequent identification of abrupt change nodes and analysis of anomaly sources. The processing method is as follows: based on a sliding time window mechanism, the first state feature encoding at multiple consecutive moments is processed, and gradient change curves such as instantaneous pressure difference are plotted respectively. (Monitoring fluid shock fluctuations), vibration change curves (identifying mechanical resonance and wear signs), electrothermal response trajectories (reflecting the coupling relationship between load and heat generation), and displacement offset trend lines (monitoring structural instability behavior); for example, if within a 10-second sliding window, the pressure difference change curve shows a continuous upward trend, the electrothermal response trajectory shows a lagging increase, and the displacement trend line gradually shifts away from the central axis, it indicates that the pump body is experiencing an early abnormal situation of load accumulation and structural offset resonance; this multidimensional trajectory set can not only serve as the input basis for subsequent first-order derivative analysis, but also visually display the state evolution trajectory to support the construction of causal graphs;

[0071] S22. Perform first-order derivative calculations on each trajectory curve in the multidimensional state feature trajectory group to extract the local first-order rate of change of the corresponding state feature at each time step, specifically: In the formula, This represents the first-order rate of change of the corresponding state characteristic at time t. This represents the observed parameter value of the corresponding state feature at time t. Indicates the corresponding state feature at time t. The observed parameter values, Indicates the observation interval;

[0072] The formula in S22 quantifies the degree of change of specific state characteristics over continuous observation time. Its core purpose is to capture the abrupt change trend of the operating state within a small time scale. This calculation method directly addresses the technical bottlenecks mentioned in the background technology, such as the insensitivity of existing methods to early anomaly identification and the difficulty in real-time capture of parameter linkage abrupt changes. By analyzing the rate of change of various state parameters over continuous time, the nonlinear evolution tendency caused by micro-amplitude disturbances can be effectively identified, and it can be determined in a timely manner whether the external pump has entered a potential risk state. The key role of the local first-order rate of change in this method is to provide a basic quantitative indicator of the dynamic evolution trend. It is the starting basis for triggering anomaly causal modeling, adjustment path reasoning, and strategy generation, which helps to achieve rapid identification of state transition precursors, thereby improving the perception sensitivity and response accuracy of the entire system to complex coupled operating states.

[0073] The local first-order rate of change of the corresponding state feature at each time step is compared with a preset mutation threshold. When the local first-order rate of change is greater than the mutation threshold, the corresponding state feature is determined to be in an abnormal trigger state at the current time step, and the corresponding abnormal trigger vector is recorded. Otherwise, it indicates that the operation status of the external pump is normal at the current time step.

[0074] Among them, the mutation threshold refers to the first-order rate of change limit value used to judge whether the state feature has undergone abnormal mutation. It is an empirical boundary with discrimination sensitivity calculated based on historical operating data, and is set by combining sample distribution characteristics, adaptive quantile algorithm or operating safety margin strategy. The anomaly trigger vector is an ordered set composed of multiple state features judged as abnormal at a certain moment. It is used to record the multi-parameter triggering situation of the current anomaly, and its dimension is consistent with the first state feature encoding. By normalizing and differentiating a large amount of historical operating data, the first-order rate distribution of each feature under normal and abnormal states is analyzed, and the mutation threshold of each feature is set by the P95 or P99 quantile method. Then, in real-time monitoring, if the first-order rate of a feature such as vibration change rate exceeds its mutation threshold at time t, it is marked with 1 at the corresponding position to form an anomaly trigger vector. For example, Atr(t)={1,0,1,0} indicates that the vibration and pressure difference features trigger anomalies at that moment, forming the basis for anomaly identification.

[0075] Specifically, S23, according to the corresponding abnormal trigger vectors, construct an abnormal trigger cause graph according to the generation order and the parameter category to which it belongs. The abnormal trigger cause graph includes multiple abnormal nodes and directed edges; the abnormal node represents the corresponding parameter abnormal type, and the directed edge represents the evolution path from the previous node triggering the abnormality of the next node.

[0076] S24. Based on the anomaly triggering cause-effect graph, according to the set time sliding window, calculate the average triggering delay time, total anomaly frequency, and subsequent associated anomaly frequency for each anomaly node, and calculate the triggering causality weight index for each anomaly node. Specifically: In the formula, This represents the triggering causality weight index of the corresponding abnormal node. This indicates the number of subsequent associated anomalies for the corresponding abnormal node. This indicates the total number of times the anomaly occurred for the corresponding abnormal node. This indicates the average trigger delay time for the corresponding abnormal node. Indicates the relative frequency of occurrence of post-association anomalies;

[0077] The formula in S24 reflects the degree of causal influence of the abnormal node in the operation of the external pump. It comprehensively considers the frequency of the abnormal node subsequently triggering other abnormalities, its total number of occurrences, and its average trigger delay time, thereby accurately measuring the criticality of the node in the abnormal evolution chain. This formula is highly related to the lack of causal modeling ability and the difficulty in clearly analyzing the abnormal development path pointed out in the background technology. Through the quantitative analysis of the trigger causal weight index, the core abnormal source in multi-parameter linkage mutation can be located and identified, which is used to support the construction of abnormal causal map and the screening of regulation path. It is an important intermediate variable for building intelligent response chain and realizing precise control closed loop, which significantly improves the ability to identify and intervene in complex state coupling evolution mechanism.

[0078] S25. Perform weighted enhancement processing on the triggering causal weight index of each abnormal node in the abnormal triggering causal graph with directed edge weights to generate an enhanced and corrected abnormal causal chain graph. The abnormal causal chain graph represents the set of abnormal causal paths that trigger safety risks under the current operating state of the external pump.

[0079] It should be noted that the weighted enhancement processing of directed edge weights refers to, in the anomaly triggering causal graph, assigning stronger distinguishing weights to the directed edges connecting nodes based on the triggering causal weight index of each anomaly node to improve the clarity of the anomaly propagation path and the decision-making response capability. This generates an enhanced and corrected anomaly causal chain graph that better reflects the actual anomaly evolution logic under the current operating state. Specifically, if an anomaly in node A triggers an anomaly in node B, and node A has a high triggering causal weight index, for example, Cyg=0.78, the edge weight from A to B is increased from the base value to a weight enhancement value, W. A→B =a*Cyg+b*frequency factor, where a and b are enhancement factors (0.6 and 0.4 respectively), and the frequency factor refers to the historical statistical frequency of node A triggering node B (specifically 0.65); when the original edge weight of A→B is 0.4, after enhancement it becomes: W A→B =0.6×0.78+0.4×0.65=0.468+0.26=0.728. Thus, A→B, as a high-priority propagation path, will be selected first in subsequent adjustment strategies, thereby significantly improving the ability to identify and respond to high-risk abnormal paths.

[0080] In this embodiment, by introducing processing steps such as multi-dimensional state feature trajectory construction, local first-order change rate extraction, and causal weight index quantification calculation, dynamic identification and causal relationship modeling of potential abnormal triggering behaviors in the operation state of the external pump are realized, effectively overcoming the bottleneck of traditional methods in identifying multi-parameter linkage changes and tracing abnormal propagation paths. By performing time series processing under a sliding window on the first state feature encoding, a multi-dimensional state feature trajectory group is formed, such as pressure gradient curve, vibration change curve, electrothermal response trajectory, and displacement offset trend line, enabling dynamic tracking of the evolution trend of the operating state over time. By extracting the local change rate in the form of the first derivative, the sudden behavior occurring during the operation of the external pump can be accurately captured, and by comparing it with the set sudden change threshold, it can be timely determined whether an abnormal triggering state has been formed, significantly improving efficiency. The system enhances the sensitivity of identifying early perturbation anomalies. After anomaly triggering, it further constructs an anomaly trigger causal graph based on parameter categories and triggering order, expressing the propagation path of the anomaly trigger through directed edges between nodes, thus establishing a temporal logical chain in the anomaly development process. Based on this, a quantitative calculation model for the trigger causal weight index is proposed, incorporating the subsequent triggering capability, frequency density, and triggering delay time of the anomaly node into the evaluation dimensions and using them for causal edge weight enhancement and correction. Ultimately, an enhanced anomaly causal chain graph with significant differentiation and risk guidance capabilities is formed. This not only achieves prominent identification of potential high-risk paths but also possesses a structural foundation for highly adaptable regulation strategy generation. It is a key link in promoting the transformation of anomaly status identification of external pumps from signal monitoring to evolutionary understanding, providing accurate, efficient, and visualized decision support for regulatory intervention.

[0081] Example 4

[0082] Please refer to Figure 1 and Figure 3 Specifically: The specific steps of S3 include:

[0083] S31. Based on the generated abnormal causal chain graph, extract the corresponding causal chain path where the sum of the triggering causal weight indices of all abnormal nodes in the corresponding causal chain path exceeds a preset threshold, and record it as a candidate adjustment path. Construct a candidate adjustment path set, wherein each candidate adjustment path in the candidate adjustment path set consists of several consecutive abnormal nodes.

[0084] S32. Based on the operating parameter types associated with each abnormal node in the candidate adjustment path, retrieve the controllable adjustment execution units associated with the corresponding operating parameter types and establish a parameter-execution unit mapping table; through statistical analysis of the controllability success rate and adjustment amplitude response sensitivity of the controllable adjustment execution units corresponding to each abnormal node, and after normalization, obtain the controllability coefficient of the controllable adjustment execution units corresponding to each abnormal node; based on the candidate adjustment path set, correlate the causal influence index of each abnormal node in each candidate adjustment path with the controllability coefficient of the corresponding controllable adjustment execution unit to determine the control action priority coefficient of each candidate adjustment path, specifically: In the formula, This represents the priority coefficient of the control action of the corresponding candidate regulation path. This represents the triggering causality weight index of the corresponding abnormal node. This represents the controllability coefficient of the controllable adjustment execution unit corresponding to the abnormal node. This represents the sequence of anomalous nodes contained in the corresponding candidate adjustment path. This represents the single-point contribution value of the corresponding abnormal node to the priority of control action in the adjustment path.

[0085] The formula in S32 effectively addresses the shortcomings of the background technology in selecting abnormal response paths and formulating control strategies by introducing a control action priority coefficient and a controllability coefficient. These shortcomings include blind path selection, coarse control commands, and low control matching. The formula multiplies the trigger causality weight index of each abnormal node in the candidate control path with the controllability coefficient of its corresponding control execution unit node by node, and sums the results for all nodes in the entire path to calculate the overall control priority of the path. The controllability coefficient quantifies the response capability of each execution unit to abnormal parameters, including control success rate and sensitivity of adjustment amplitude, and is a key indicator for measuring the effectiveness of control. The control action priority coefficient integrates the two dimensions of abnormal severity and control capability, and is a core indicator for path selection and strategy deployment. This method achieves optimal ranking of multiple complex abnormal paths, improving the response accuracy and adjustment efficiency when facing high-frequency anomalies.

[0086] Specifically, S33, based on the control action priority coefficients of each candidate control path, extract the candidate control path corresponding to the maximum value of the control action priority coefficient as the main control path, record the target parameter values ​​of the state parameters of each control node covered by the starting control node and the ending control node of the main control path, and number each control node in the main control path according to the time sequence.

[0087] S34. For the target parameter values ​​of the corresponding state parameters of each control node in the main control and regulation path, combined with the current values ​​of the operating status of each control node and the average response time in historical samples, analyze the control intensity that should be applied for the corresponding regulation action of each control node in the main control and regulation path, and determine the jump coefficient of the corresponding regulation action of each control node, specifically: In the formula, This represents the jump coefficient corresponding to the adjustment action of the corresponding control node. This indicates the target parameter value that the corresponding control node should achieve for the adjustment action. This indicates the current parameter value of the corresponding control node. This represents the average response time of the corresponding control node in historical samples. This represents the target deviation value that the corresponding control node needs to be adjusted, that is, the difference between the current operating state of the node and the target state;

[0088] The formula in S34 is used to quantify the adjustment intensity required for each control node of the external pump under specific operating conditions, aiming to solve the problems of coarse adjustment amplitude setting and response lag in traditional control systems. The formula uses the difference between the target state value and the current measured value as the adjustment driving force and introduces the historical average response time as a time constraint. The output jump coefficient can reflect the control intensity to be applied under different dynamic response characteristics, thereby realizing the quantitative design of control actions. This mechanism significantly improves the adaptability and execution accuracy of the control strategy. Especially under multiple abnormal path interference, the control amplitude can be quickly adjusted to ensure that the adjustment effect and safety and stability are given equal importance, effectively solving the problems of blind adjustment and execution delay in the background technology.

[0089] S35. Encode the jump coefficients of the corresponding adjustment actions of each control node in the main control path into a control strategy instruction set, wherein the control strategy instruction set includes the control field of the main control path number and the control parameter name, jump factor value and response time corresponding to each control node.

[0090] It should be noted that the coded control strategy instruction set refers to organizing the adjustment action information to be executed by each control node in the main control path into a data instruction with a recognizable structure according to a preset format, which is convenient for identification, parsing and execution. The encoding process is based on the main path number and the adjustment parameters of each control node, and the adjustment target is encoded item by item in the form of fields. For example, if the main path number is P03, the node number is N2, the corresponding parameter is outlet pressure, the jump factor is 1.25, and the response time is 4.5 seconds, then the instruction encoding is: P03-N2-Pout-1.25-4.5. The control strategy instruction set constructed in this way not only has a clear execution order, but is also compatible with the protocol parsing format of multiple control components in the system, realizing rapid distribution of instructions and closed-loop control.

[0091] In this embodiment, through in-depth analysis of the enhanced abnormal causal chain graph and the optimal construction of adjustment paths, an effective transition from identifying abnormal propagation relationships to decision-making on adjustment paths and generating quantitative control commands is achieved. A strategy generation mechanism with a clear structure, high response efficiency, and high control matching degree is constructed, significantly improving the target focus and controllability of abnormal adjustment of the external pump. Specifically, by calculating the sum of the triggering causal weight indices of each abnormal node in the causal chain path and selecting paths exceeding a preset threshold to form a candidate adjustment path set, high-risk, high-impact adjustment target areas can be accurately extracted from a large number of potential abnormal paths. Furthermore, by establishing a mapping relationship between operating parameter types and controllable adjustment execution units, and combining the historical adjustment success rate and response sensitivity of each execution unit, the controllability coefficient of each abnormal node is calculated. This allows for a comprehensive evaluation of the possibility and strength of adjustment effects in the candidate paths, constructing a control action priority. The coefficient model, by introducing a priority coefficient, ensures that the selection of the adjustment path no longer depends solely on the intensity of the anomaly's impact, but also considers both the feasibility of adjustment and the matching of control resources. This significantly optimizes the strategy selection process from both the perspectives of control strategy rationality and resource efficiency. Furthermore, based on the current state and historical response behavior of each control node in the main control path, a jump coefficient calculation model is introduced to quantify the intensity of the applied control action into a implementable control factor. This ensures that the generated control strategy instruction set is targeted, adjustable, and operable, and fully labels the adjustment path number, parameter name, jump factor, and response time control fields in the strategy, giving the control instruction set a highly structured characteristic. Therefore, this step not only achieves a crucial leap from anomaly identification to control strategy implementation but also completes the precise selection of the adjustment path and the quantitative decision-making of action intensity in a model-driven manner, laying a solid foundation for subsequent control feedback and closed-loop optimization.

[0092] Example 5

[0093] Please refer to Figure 1 and Figure 3 Specifically: The specific steps of S4 include:

[0094] The specific steps of S4 include:

[0095] S41. The control strategy instruction set formed based on S35 is used as input. It is indexed according to the number of the main adjustment path and control commands are issued to the corresponding control execution components according to the content of the control field. The control execution components include the frequency converter, proportional valve, motor control unit, cooling fan start / stop unit and fluid bypass controller in the external pump.

[0096] S42. After the control command is issued, according to the set feedback cycle. The set of operating parameter feedback values ​​of each control node is periodically collected. The set of operating parameter feedback values ​​includes the feedback parameter values, instantaneous rate of change, and average response gain duration of the corresponding control node within the feedback period.

[0097] S43. Based on the set of operating parameter feedback values ​​for each control node, determine the feedback response coefficient for the corresponding adjustment action of each control node, specifically as follows: In the formula, This represents the feedback response coefficient corresponding to the adjustment action of the corresponding control node. This indicates the target parameter value that the corresponding control node should achieve for the adjustment action. This indicates the current parameter value of the corresponding control node. This represents the feedback parameter value of the corresponding control node during the feedback period T. This represents the actual change after the control action is executed. Indicates the expected change in the target;

[0098] The formula in S43 addresses the key issue of the lack of a feedback evaluation mechanism for the execution results of regulatory actions in the background technology. Traditional control strategies often struggle to verify the execution effect in real time when faced with regulatory actions under complex operating conditions, resulting in a lack of self-correction capabilities. This formula, however, constructs a standardized feedback response coefficient by comparing the quantitative differences between the regulation target, the state before regulation, and the feedback value after regulation. This coefficient is used to measure the actual execution effect of the control action. This process provides a quantitative feedback basis, supporting closed-loop control optimization, regulation failure identification, and dynamic adaptive readjustment mechanisms, thereby significantly improving the control accuracy and robustness of the external pump under nonlinear disturbances and multivariable fluctuations.

[0099] S44. Compare the feedback response coefficients of the corresponding adjustment actions of each control node determined in S43 with the preset response thresholds. When the feedback response coefficient of the corresponding adjustment action of a control node is less than the response threshold, mark the corresponding adjustment action as a weak response action and record the number of the corresponding control node, the jump coefficient and the name of the corresponding adjustment parameter to form an adjustment failure record table. Otherwise, it indicates that the operation status of the external pump after the control command is executed is normal.

[0100] The response threshold is a critical reference value used to determine whether the adjustment action is effective. It is usually obtained by statistical modeling of successful and unsuccessful adjustment cases in historical samples. Specifically, based on a large amount of adjustment action execution data of external pumps under different operating conditions, the gain ratio between the jump target value of each control node and the feedback parameter response is extracted, a feedback response coefficient distribution model is constructed, and the minimum gain ratio corresponding to the qualified adjustment sample in the distribution is selected as the preset response threshold. This ensures that the threshold is representative and robust, and is used to distinguish between normal adjustment response and weak response state, thereby guiding subsequent adjustment correction and strategy optimization.

[0101] Specifically, S45, based on the control parameters recorded in the adjustment failure record table, determine the post-adjustment jump coefficient for each control node corresponding to the adjustment action, specifically as follows: In the formula, This represents the adjustment jump coefficient after adjustment corresponding to the adjustment action of the corresponding control node. This represents the gain adjustment correction factor;

[0102] The formula in S45 explicitly expresses a secondary correction mechanism when the adjustment action is ineffective, effectively compensating for the problem in the background technology where the initial adjustment control action is incomplete, leading to difficulty in restoring the safe state. Traditional methods often lack automatic gain correction strategies based on response feedback, making it difficult to adapt to the risk of control failure caused by dynamic fluctuations in the operating environment. This formula, by directly coupling the initial jump coefficient with the feedback response error and introducing a gain adjustment correction factor, achieves precise amplification and iterative optimization of the adjustment intensity, significantly enhancing its adaptability and adjustment robustness. The adjusted jump coefficient represents the amplified jump value adjusted for weak response control nodes to enable the target state to reach faster or compensate for insufficient response, and is the core parameter for generating a new round of control strategy instruction set. The gain adjustment correction factor is usually obtained by learning from a historical adjustment behavior database and dynamically set in combination with the following factors: response lag rate in previous adjustments, inertia coefficient of the control unit, and disturbance level of the operating environment; it is generated through linear fitting or gradient adaptive adjustment methods.

[0103] S46. Based on the adjusted jump coefficients of the corresponding adjustment actions of each control node determined in S45, regenerate the adjusted control strategy instruction set and execute the operation procedures in S41-S44 until the external pump is in normal operating condition.

[0104] In this embodiment, by constructing a complete closed-loop system for control command execution and feedback correction, the dynamic response capability and adaptability of the external pump during abnormal adjustment processes are significantly improved. Specifically, it solves the control blind spot problem caused by the lack of feedback verification and dynamic correction mechanisms in traditional methods where adjustment commands take effect immediately upon issuance. Specifically, the control strategy instruction set constructed in S35 serves as the execution entry point, achieving precise indexing according to the main adjustment path number, and issuing control commands containing jump factors and response time fields to specific control execution components. This gives the control action multi-dimensional control characteristics of controllable precision, identifiable path, and targetable object. Next, a feedback period is set, and key parameters such as feedback parameter values, rate of change, and average response gain duration of each control node are periodically collected to form a complete set of operating parameter feedback values. The deviation between actual parameters and target values ​​is calculated... The differential trend is analyzed to derive the feedback response coefficient, which is then compared with the response threshold to determine if there is a weak response action. This generates a regulation failure record table, accurately marking control nodes that fail to achieve the expected regulation effect. This mechanism effectively avoids control failures caused by individual node response lag, equipment aging, or coupling interference. This step introduces a post-regulation jump coefficient calculation model and combines it with a gain regulation correction factor to achieve adaptive revision and optimization of the original control strategy, ultimately constructing a control command system with self-evolution capabilities. This process can be iterated until the external pump's operating state returns to normal, exhibiting significant advantages such as dynamic feedback, rapid correction, and high robustness. Therefore, this step not only achieves closed-loop adaptive control of the regulation process but also establishes a mechanism for continuous feedback from the operating state to the regulation command, ensuring the long-term effectiveness and operational stability of the abnormal regulation control system under varying operating conditions.

[0105] Example 6

[0106] Please refer to Figure 1 and Figure 2 Specifically: a safety adaptive intelligent system for external pumps based on multi-sensor fusion, including a coding construction module, a causal analysis module, an intensity analysis module, and an adaptive control module;

[0107] The coding construction module is based on multiple deployed sensors to collect the first operating parameter information of the external pump in real time. After time synchronization processing, it forms the operating parameter frames at each moment to construct the first state feature code.

[0108] The causal analysis module extracts the local first-order change rate of the corresponding state feature at each time step based on the constructed first state feature encoding. If the corresponding state feature is determined to be an abnormal trigger state at the current time step, the trigger causal weight index of each abnormal node is calculated according to the recorded abnormal trigger vector, and an abnormal causal chain graph is constructed.

[0109] The intensity analysis module generates a set of candidate regulation paths based on the constructed abnormal causal chain graph, in order to extract the main control regulation path. Based on the current value of the operating status of each control node in the main control regulation path and the average response time in historical samples, it determines the jump coefficient of the corresponding regulation action of each control node and encodes it into a control strategy instruction set.

[0110] The adaptive control module is used to issue control commands based on the control strategy instruction set, collect the set of operating parameter feedback values ​​of each control node, determine the feedback response coefficient of the corresponding adjustment action of each control node, and if there is a corresponding adjustment action that is a weak response action, regenerate the adjusted control strategy instruction set.

[0111] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A safety adaptive method for export pumps based on multi-sensor fusion, characterized in that: Includes the following steps: S1. Based on the deployment of multiple sets of sensors, the first operating parameter information of the external pump is collected in real time. After time synchronization processing, the operating parameter frames at each moment are formed to construct the first state feature code. S2. Based on the constructed first state feature encoding, extract the local first-order change rate of the corresponding state feature at each time. If it is determined that the corresponding state feature is an abnormal trigger state at the current time, calculate the trigger causal weight index of each abnormal node according to the recorded abnormal trigger vector, and construct an abnormal causal chain graph. S3. Based on the constructed abnormal causal chain graph, generate a set of candidate adjustment paths to extract the main control adjustment path. According to the current value of the running status of each control node in the main control adjustment path and the average response time in the historical samples, determine the jump coefficient of the corresponding adjustment action of each control node and encode it into a control strategy instruction set. S4. Using the control strategy instruction set as the control command to collect the set of operating parameter feedback values ​​of each control node, determine the feedback response coefficient of the corresponding adjustment action of each control node, and if there is a corresponding adjustment action that is a weak response action, regenerate the adjusted control strategy instruction set. The specific steps in S1 include: S11. Multiple sets of sensors are deployed in the inlet pipe section, outlet pipe section, pump body shell, motor connection area, mounting base and support structure of the external pump, and connected to the edge fusion data platform through an industrial bus with time synchronization function. The multiple sets of sensors include pressure sensors, flow rate sensors, temperature sensors, acceleration sensors, current sensors and displacement sensors. S12. Based on the deployment of multiple sets of sensors, the operating status of the external pump is monitored in real time to obtain the first operating parameter information. The first operating parameter information is time-aligned according to the acquisition channel and measurement frequency to construct a unified parameter sampling time axis. Based on the sampling time axis, the first operating parameter information is time-synchronized to form operating parameter frames at each moment. The first operating parameter information includes inlet and outlet pressure, pump chamber temperature rise, motor current value, shell vibration intensity and axial displacement change. S13. For the operating parameter frames at each time point, and based on their respective physical properties, the parameters are classified and processed. Pressure parameters are classified into the first parameter subset, temperature and current parameters into the second parameter subset, and vibration and displacement parameters into the third parameter subset, together constructing a related coupled operating parameter subset. Standardization processing is performed on each parameter subset within the related coupled operating parameter subset. Standardization processing includes: performing normalization transformation on each operating parameter based on historical extreme values, and removing and completing the values ​​of parameter points that show abnormal changes within the continuous sampling window. The value completion process uses the interpolation averaging method of samples from previous and subsequent time points for smooth repair. S14. After standardization, based on the temporal change rate of each operating parameter, the incremental amplitude between adjacent samples, and the relative disturbance level characteristic index, a first state feature code is constructed, wherein the first state feature code includes instantaneous pressure, pump body temperature, input current, vibration frequency, and displacement. The specific steps in S2 include: S21. Based on the formed first state feature encoding, perform dynamic sliding processing of multi-source state parameters to construct a multi-dimensional state feature trajectory group, wherein the multi-dimensional state feature trajectory group includes pressure gradient curve, vibration change curve, electrothermal response trajectory and displacement offset trend line. S22. Perform first-order derivative calculations on each trajectory curve in the multidimensional state feature trajectory group to extract the local first-order rate of change of the corresponding state feature at each time step, specifically: In the formula, This represents the first-order rate of change of the corresponding state characteristic at time t. This represents the observed parameter value of the corresponding state feature at time t. Indicates the corresponding state feature at time t. The observed parameter values, Indicates the observation interval; The local first-order rate of change of the corresponding state feature at each time step is compared with the preset mutation threshold. When the local first-order rate of change is greater than the mutation threshold, the corresponding state feature is determined to be in an abnormal trigger state at the current time step, and the corresponding abnormal trigger vector is recorded. Otherwise, it indicates that the operation state of the external pump is normal at the current time step. S23. Based on the corresponding abnormal trigger vectors, construct an abnormal trigger cause-effect graph according to the generation order and the parameter category to which it belongs. The abnormal trigger cause-effect graph includes multiple abnormal nodes and directed edges. The abnormal node represents the corresponding parameter abnormal type, and the directed edge represents the evolution path from the previous node triggering the abnormality of the next node. S24. Based on the anomaly triggering cause-effect graph, according to the set time sliding window, calculate the average triggering delay time, total anomaly frequency, and subsequent associated anomaly frequency for each anomaly node, and calculate the triggering causality weight index for each anomaly node. Specifically: In the formula, This represents the triggering causality weight index of the corresponding abnormal node. This indicates the number of subsequent associated anomalies for the corresponding abnormal node. This indicates the total number of times the anomaly occurred for the corresponding abnormal node. This indicates the average trigger delay time for the corresponding abnormal node; S25. Perform weighted enhancement processing on the triggering causal weight index of each abnormal node in the abnormal triggering causal graph with directed edge weights to generate an enhanced and corrected abnormal causal chain graph. The abnormal causal chain graph represents the set of abnormal causal paths that trigger safety risks under the current operating state of the external pump.

2. The adaptive safety method for export pumps based on multi-sensor fusion according to claim 1, characterized in that: The specific steps of S3 include: S31. Based on the generated abnormal causal chain graph, extract the corresponding causal chain path where the sum of the triggering causal weight indices of all abnormal nodes in the corresponding causal chain path exceeds a preset threshold, and record it as a candidate adjustment path. Construct a candidate adjustment path set, wherein each candidate adjustment path in the candidate adjustment path set consists of several consecutive abnormal nodes. S32. Based on the operating parameter types associated with each abnormal node in the candidate adjustment path, retrieve the controllable adjustment execution units associated with the corresponding operating parameter types and establish a parameter-execution unit mapping table; through statistical analysis of the controllability success rate and adjustment amplitude response sensitivity of the controllable adjustment execution units corresponding to each abnormal node, and after normalization, obtain the controllability coefficient of the controllable adjustment execution units corresponding to each abnormal node; based on the candidate adjustment path set, correlate the causal influence index of each abnormal node in each candidate adjustment path with the controllability coefficient of the corresponding controllable adjustment execution unit to determine the control action priority coefficient of each candidate adjustment path, specifically: In the formula, This represents the priority coefficient of the control action of the corresponding candidate regulation path. This represents the triggering causality weight index of the corresponding abnormal node. This represents the controllability coefficient of the controllable adjustment execution unit corresponding to the abnormal node. This represents the sequence of abnormal nodes contained in the corresponding candidate adjustment path.

3. The adaptive safety method for export pumps based on multi-sensor fusion according to claim 2, characterized in that: S33. Based on the control action priority coefficients of each candidate control path, extract the candidate control path corresponding to the maximum value of the control action priority coefficient as the main control path, record the target parameter values ​​of the state parameters of each control node covered by the starting control node and the ending control node of the main control path, and number each control node in the main control path according to the time sequence. S34. For the target parameter values ​​of the corresponding state parameters of each control node in the main control and regulation path, combined with the current values ​​of the operating status of each control node and the average response time in historical samples, analyze the control intensity that should be applied for the corresponding regulation action of each control node in the main control and regulation path, and determine the jump coefficient of the corresponding regulation action of each control node, specifically: In the formula, This represents the jump coefficient corresponding to the adjustment action of the corresponding control node. This indicates the target parameter value that the corresponding control node should achieve for the adjustment action. This indicates the current parameter value of the corresponding control node. This represents the average response time of the corresponding control node in historical samples; S35. Encode the jump coefficients of the corresponding adjustment actions of each control node in the main control path into a control strategy instruction set, wherein the control strategy instruction set includes the control field of the main control path number and the control parameter name, jump factor value and response time corresponding to each control node.

4. The adaptive safety method for export pumps based on multi-sensor fusion according to claim 3, characterized in that: The specific steps of S4 include: S41. The control strategy instruction set formed based on S35 is used as input. It is indexed according to the number of the main adjustment path and control commands are issued to the corresponding control execution components according to the content of the control field. The control execution components include the frequency converter, proportional valve, motor control unit, cooling fan start / stop unit and fluid bypass controller in the external pump. S42. After the control command is issued, according to the set feedback cycle. The set of operating parameter feedback values ​​of each control node is periodically collected. The set of operating parameter feedback values ​​includes the feedback parameter values, instantaneous rate of change, and average response gain duration of the corresponding control node within the feedback period. S43. Based on the set of operating parameter feedback values ​​for each control node, determine the feedback response coefficient for the corresponding adjustment action of each control node, specifically as follows: In the formula, This represents the feedback response coefficient corresponding to the adjustment action of the corresponding control node. This indicates the target parameter value that the corresponding control node should achieve for the adjustment action. This indicates the current parameter value of the corresponding control node. This represents the feedback parameter value of the corresponding control node during the feedback period T; S44. Compare the feedback response coefficients of the corresponding adjustment actions of each control node determined in S43 with the preset response thresholds. When the feedback response coefficient of the corresponding adjustment action of a control node is less than the response threshold, mark the corresponding adjustment action as a weak response action and record the number of the corresponding control node, the jump coefficient and the name of the corresponding adjustment parameter to form an adjustment failure record table. Otherwise, it indicates that the operation status of the external pump after the control command is executed is normal.

5. The adaptive safety method for export pumps based on multi-sensor fusion according to claim 4, characterized in that: S45. Based on the control parameters recorded in the adjustment failure record table, determine the post-adjustment jump coefficient of each control node corresponding to the adjustment action, specifically as follows: In the formula, This represents the adjustment jump coefficient after adjustment corresponding to the adjustment action of the corresponding control node. This represents the gain adjustment correction factor; S46. Based on the adjusted jump coefficients of the corresponding adjustment actions of each control node determined in S45, regenerate the adjusted control strategy instruction set and execute the operation procedures in S41-S44 until the external pump is in normal operating condition.

6. A multi-sensor fusion-based adaptive intelligent system for the safety of an export pump, used to implement the multi-sensor fusion-based adaptive safety method for export pumps as described in any one of claims 1 to 5, characterized in that: It includes a coding construction module, a causal analysis module, an intensity analysis module, and an adaptive control module; The coding construction module is based on multiple deployed sensors to collect the first operating parameter information of the external pump in real time. After time synchronization processing, it forms the operating parameter frames at each moment to construct the first state feature code. The causal analysis module extracts the local first-order change rate of the corresponding state feature at each time step based on the constructed first state feature encoding. If the corresponding state feature is determined to be an abnormal trigger state at the current time step, the trigger causal weight index of each abnormal node is calculated according to the recorded abnormal trigger vector, and an abnormal causal chain graph is constructed. The intensity analysis module generates a set of candidate regulation paths based on the constructed abnormal causal chain graph, in order to extract the main control regulation path. Based on the current value of the operating status of each control node in the main control regulation path and the average response time in historical samples, it determines the jump coefficient of the corresponding regulation action of each control node and encodes it into a control strategy instruction set. The adaptive control module is used to issue control commands based on the control strategy instruction set, collect the set of operating parameter feedback values ​​of each control node, determine the feedback response coefficient of the corresponding adjustment action of each control node, and if there is a corresponding adjustment action that is a weak response action, regenerate the adjusted control strategy instruction set.

Images