A swab supply automatic control system based on edge computing

Abstract

The application discloses a swab feeding automatic control system based on edge calculation, belongs to the automatic control technical field, and comprises a data acquisition module, a system state evaluation module, a self-adaptive risk hedging decision module and a control instruction execution module; the data acquisition module is used for acquiring optical state data, pressure state data and motor state data of a feeding mechanism in real time, and combines the acquired optical state data, the pressure state data and the motor state data into state data and sends the state data to the system state evaluation module; the application provides a solid and reliable data basis for subsequent intelligent decision-making and changes the status that a traditional control system lacks self-cognition ability.

Description

Technical Field

[0001] This invention relates to the field of automatic control, specifically to an automatic control system for swab feeding based on edge computing. Background Technology

[0002] In extreme working environments such as high-level biosafety laboratories, automated equipment is crucial for improving testing throughput and ensuring personnel safety. Among these, the automated feeding of sample materials, such as swabs, is a key upstream link in the entire automated process. Existing feeding systems typically employ control methods based on fixed parameters or simple feedback logic. However, these methods exhibit poor robustness and are prone to serious malfunctions such as jamming when faced with complex dynamic interference caused by physical differences in multiple material sources and progressive sensor contamination. In particular, traditional control strategies cannot quantify or address the deterministic erosion of control logic caused by sensor data distortion; that is, interventions based on erroneous information can actually accelerate the system's own descent into failure. Therefore, there is an urgent need in this field for an intelligent control system capable of proactively assessing and managing the health of its own control logic, thereby achieving long-term stable operation while ensuring basic performance.

[0003] The information disclosed in the background section above is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0004] The purpose of this invention is to provide an automatic control system for swab feeding based on edge computing, so as to solve the problems mentioned in the background art.

[0005] The technical solution of the present invention includes a data acquisition module, a system status assessment module, an adaptive risk hedging decision module, and a control command execution module;

[0006] The data acquisition module is used to acquire optical status data, pressure status data and motor status data of the feeding mechanism in real time, and merge the acquired optical status data, pressure status data and motor status data into status data and send it to the system status evaluation module.

[0007] The system status assessment module is used to receive the status data, calculate and generate the material supply throughput, predicted intervention success rate and system health based on the status data, and send the generated material supply throughput, predicted intervention success rate and system health to the adaptive risk hedging decision module.

[0008] The adaptive risk hedging decision module is used to generate decision instructions based on the system health status, combined with the material supply throughput and the predicted intervention success rate.

[0009] The control instruction execution module is used to receive the decision instruction and translate the decision instruction into specific control signals for the physical actuator.

[0010] Preferably, the system status assessment module calculates the system health status as follows:

[0011] Step 1: After each intervention event occurs, based on the preset forgetting factor, the deterministic erosion degree of the control logic after the previous intervention event is attenuated to obtain the attenuated erosion degree.

[0012] Step 2: Calculate the new erosion amount by combining the preset forgetting factor, the preset event weight factor, the intensity of the intervention event, and the intervention success indicator;

[0013] Step 3: Add the attenuated erosion degree to the newly added erosion amount to update and obtain the current control logic deterministic erosion degree;

[0014] Step 4: Normalize the current control logic deterministic erosion degree with the preset failure boundary erosion degree threshold to obtain the normalized erosion degree;

[0015] Step 5: By performing reverse mapping and nonlinear processing on the normalized erosion degree, the system health degree is derived.

[0016] Preferably, the process by which the adaptive risk hedging decision module generates the decision instruction includes dynamically adjusting the aggressiveness of the predictive intervention to set an intervention trigger threshold; wherein the intervention trigger threshold is used to compare the potential anomaly score with the threshold to determine whether to perform a predictive intervention.

[0017] Preferably, when the system health is greater than the preset risk hedging threshold, the adaptive risk hedging decision module dynamically fine-tunes the aggressiveness of the predicted intervention based on the difference between the predicted intervention success rate and the preset target intervention success rate through a proportional controller.

[0018] Preferably, when the system health is less than or equal to the risk hedging entry threshold and greater than the preset health verification threshold, the adaptive risk hedging decision module gradually reduces the aggressiveness of the predictive intervention with a preset decay step size.

[0019] Preferably, when the system health is less than or equal to a preset health verification threshold, the adaptive risk hedging decision module first calculates the time change rate of the system health and generates the decision instruction based on the time change rate.

[0020] Preferably, if the rate of change over time is greater than or equal to a preset critical deterioration rate, a conservative mode instruction is generated; the conservative mode instruction is used to set the degree of aggressiveness of the predictive intervention to its preset minimum value.

[0021] Preferably, if the rate of change over time is less than the critical deterioration rate, an online self-recovery instruction is generated; the online self-recovery instruction is used to perform physical cleaning or model recalibration operations.

[0022] Preferably, when the system operates according to the conservative mode instruction for a duration exceeding a preset conservative observation period, and the average value of the time change rate within the conservative observation period is negative, an online self-recovery instruction is generated.

[0023] This invention provides an improved edge computing-based automatic swab feeding control system, which has the following improvements and advantages compared with the prior art:

[0024] 1. This invention constructs a complete state quantification and closed-loop adaptive control architecture. This architecture, through the collaborative work of a data acquisition module, a system state assessment module, an adaptive risk hedging decision-making module, and a control command execution module, realizes a complete link from multi-dimensional raw data perception to precise physical intervention. The system state assessment module extracts key state indicators such as feed throughput, predicted intervention success rate, and core system health from high-dimensional, noisy sensor data. This design transforms the fuzzy problem of system state deterioration into precisely quantifiable, monitorable, and manageable engineering indicators, providing a solid and reliable data foundation for subsequent intelligent decision-making and changing the current situation where traditional control systems lack self-awareness capabilities.

[0025] 2. This invention innovatively proposes and implements a system health metric model for evaluating the reliability of control logic. This model tracks the cumulative effect of system uncertainty caused by failed or drastic control interventions by calculating the deterministic erosion degree of the control logic. Furthermore, through nonlinear processing and inverse mapping, this risk indicator is transformed into a normalized system health score negatively correlated with risk. This mechanism enables the system to quantify and predict the health level of its own control logic, achieving a qualitative leap from passively responding to physical failures to proactively managing cumulative risks. It can effectively warn against and avoid catastrophic cascading failures caused by sensor data distortion.

[0026] 3. This invention establishes an adaptive risk hedging decision-making logic based on system health zoning. This logic executes differentiated control strategies according to different health zoning intervals. When the system is in good health, the proportional controller dynamically fine-tunes the predicted level of intervention to optimize operational efficiency. When health shows a downward trend, the logic proactively reduces the level of intervention with a preset decay step size to curb further deterioration of health. When health enters a critical zone, the logic chooses between a conservative mode maintaining minimum operational limits and a thorough online self-recovery operation based on the time rate of change of health. This hierarchical, progressive decision-making mechanism endows the system with the ability to intelligently balance efficiency and survival, greatly improving the long-term operational stability and overall equipment efficiency of the system in complex, unattended dynamic environments. Attached Figure Description

[0027] The present invention will be further explained below with reference to the accompanying drawings and embodiments:

[0028] Figure 1 This is a flowchart of the system of the present invention. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments. Example 1

[0030] Please see Figure 1 This invention provides an automatic control system for swab feeding based on edge computing, including a data acquisition module, a system status assessment module, an adaptive risk hedging decision module, and a control command execution module;

[0031] The data acquisition module is used to acquire optical status data, pressure status data and motor status data of the feeding mechanism in real time, and merge the acquired optical status data, pressure status data and motor status data into status data and send it to the system status evaluation module.

[0032] The system status assessment module is used to receive status data, calculate and generate material throughput, predicted intervention success rate and system health based on the status data, and send the generated material throughput, predicted intervention success rate and system health to the adaptive risk hedging decision module.

[0033] The adaptive risk hedging decision module is used to generate decision instructions based on the system health status, combined with the material supply throughput and the predicted intervention success rate.

[0034] The control instruction execution module is used to receive decision instructions and translate them into specific control signals for the physical actuators.

[0035] This invention provides an automatic control system for swab feeding based on edge computing. The system is deployed on an edge computing device and is characterized by including a data acquisition module, a system status assessment module, an adaptive risk hedging decision module, and a control command execution module.

[0036] The data acquisition module is designed to acquire real-time optical status data, pressure status data, and motor status data of the feeding mechanism. These multi-source heterogeneous data are merged into status data and then sent to the system status assessment module. The data acquisition module provides comprehensive raw data input for upper-level analysis, ensuring the timeliness and accuracy of subsequent assessments when faced with dynamic interferences such as swab physical differences and progressive contamination of sensors.

[0037] The system status assessment module has core assessment functions. It receives status data and calculates and generates feed throughput, predicted intervention success rate, and system health based on the status data. The system status assessment module extracts the high-dimensional and noisy raw sensor data into low-dimensional and highly condensed key status indicators, and sends the generated feed throughput, predicted intervention success rate, and system health to the adaptive risk hedging decision module, providing a solid foundation for realizing system self-awareness and status quantification.

[0038] The adaptive risk hedging decision module executes decision-making functions, generating decision instructions based on system health, material throughput, and predicted intervention success rate. The adaptive risk hedging decision module can dynamically adjust the system's behavior mode, making intelligent trade-offs between pursuing performance optimization and ensuring long-term survival, thus solving the problem of poor robustness of traditional fixed parameter control methods in complex environments.

[0039] The control command execution module performs specific operational functions, receives decision commands, and translates them into specific control signals for physical actuators. The control command execution module accurately converts the abstract commands from the decision layer into specific operations for physical components such as motors and valves, completing a closed loop from decision-making to physical intervention. This ensures that the entire system has the ability to proactively predict and mitigate the risk of catastrophic downtime in complex, unattended environments. Example 2

[0040] The system health assessment module calculates the system health status as follows:

[0041] Step 1: After each intervention event occurs, based on the preset forgetting factor, the deterministic erosion degree of the control logic after the previous intervention event is attenuated to obtain the attenuated erosion degree.

[0042] Step 2: Calculate the new erosion amount by combining the preset forgetting factor, preset event weight factor, intervention intensity of the intervention event, and intervention success indicator;

[0043] Step 3: Add the decayed erosion to the newly added erosion to update the current control logic deterministic erosion.

[0044] Step 4: Normalize the current control logic deterministic erosion degree with the preset failure boundary erosion degree threshold to obtain the normalized erosion degree;

[0045] Step 5: Derive the system health by performing reverse mapping and nonlinear processing on the normalized erosion degree;

[0046] This embodiment is a further explanation based on Embodiment 1. To further clarify, the process by which the system status assessment module calculates the system health is based on the quantification of the reliability of the control logic itself rather than directly assessing the physical state.

[0047] Furthermore, the system status assessment module is explained; among which, the material throughput can be calculated by the edge computing device within a preset time window by counting the number of swabs that successfully pass through the optical status data, which is an indicator of the system's short-term production efficiency; predictive intervention success rate It is a dynamic indicator for measuring the effectiveness of predictive interventions, calculated as follows: in the most recent Second-rate, For example, in 100 predictive intervention events, the number of times potential abnormalities were successfully eliminated versus the total number of interventions. The ratio. The criterion for successful intervention is that, after the intervention, subsequent sensor data shows that the potential abnormal score recovers to below the baseline threshold within a preset time.

[0048] After each intervention event, the deterministic erosion of the control logic following the previous intervention event is attenuated based on a preset forgetting factor to obtain the attenuated erosion. This attenuated erosion is then calculated by combining the preset forgetting factor, the preset event weighting factor, the severity of the intervention event, and the intervention success indicator. The preset forgetting factor works by adjusting the weight of historical data and the current event using a value between 0 and 1. For systems with long fault precursor cycles, a smaller value is used to give greater weight to historical data, while a larger value is used to increase sensitivity to recent events. The attenuated erosion and the new erosion are added together, and the deterministic erosion of the control logic at the current moment is updated using the following formula:

[0049]

[0050] in This indicates the degree of erosion after the intervention. This indicates the degree of erosion since the last intervention. Indicates the forgetting factor, Indicates the event weight factor. This indicates the control vector for this intervention, and its source is the output of the adaptive risk hedging decision module. This represents the reference control vector, which is derived from the statistical mean of intervention experiments on normal materials. This indicates that the intervention was successful, and its source is the determination of whether the abnormality has been eliminated through sensor data after the intervention; : Norm symbol, which in this formula represents the magnitude or size of a vector;

[0051] This step, through recursive updates, quantifies the degree of cognitive confusion caused by sensor data distortion or model drift, and cumulatively tracks the system uncertainty caused by failed or drastic control interventions.

[0052] In calculating erosion Previously, it was necessary to adjust the control vector. and Normalize each component to make it a dimensionless value; an effective method is to let Divide each component by The corresponding component in;

[0053] Assume the reference control vector is The control vector for this intervention is .

[0054] Then the relative deviation term in the formula This should be understood as finding the norm of a dimensionless deviation vector; the dimensionless deviation vector Each component The calculation is as follows:

[0055]

[0056] in, The k-th component of the dimensionless deviation vector; : Control vector for this intervention The kth component; Reference control vector The kth component;

[0057] Therefore, this item is actually a calculation. In this way, all components become dimensionless relative deviation values, allowing for meaningful norm calculations.

[0058] The current deterministic erosion degree of the control logic is normalized to the preset failure boundary erosion degree threshold to obtain the normalized erosion degree. By performing inverse mapping and nonlinear processing on the normalized erosion degree, the system health is derived using the following formula:

[0059]

[0060] in This indicates the current system health status. This represents the erosion level at the current moment, and its source is the calculation result of the previous steps. Indicates the threshold of erosion at the failure boundary; : Function that returns the maximum value among its arguments;

[0061] This step transforms the positively accumulated risk indicators into health indicators that are negatively correlated with risk and range from 0 to 1. The preset failure boundary erosion threshold is determined through extensive offline calibration experiments. To clarify the variable domain, a set of erosion values ​​recorded when reproducing two consecutive absolute jamming failure scenarios in the calibration experiments is defined as... The 95th percentile value of this set of values ​​is taken as... The robust setpoint. This method transforms the reliability of abstract control logic into precise, monitorable quantitative indicators, realizing a shift from passively responding to failures to proactively predicting and hedging cumulative risks;

[0062] Furthermore, regarding other preset parameters mentioned in this invention, such as event weight factors... Such as event weight factor proportional gain coefficient , benchmark threshold The calibration process includes measures such as the risk hedging entry threshold, health verification threshold, and critical deterioration rate. The specific values ​​of these parameters are determined using a unified offline calibration and optimization principle. The risk hedging entry threshold distinguishes the system's safe operating zone from the risk hedging zone. Its calibration goal is to find a balance that avoids premature performance sacrifice while ensuring timely intervention before a significant deterioration in health. The health verification threshold serves as the last line of defense between the risk hedging zone and critical states. Its set value typically corresponds to a relatively low level of health generally reached before absolute failure events such as continuous jamming in the system's history, and can be obtained through offline experimental data statistics. The calibration process is achieved by running numerous feeding cycles on the experimental platform, covering normal materials, boundary-size materials, and various known abnormal operating conditions. During the calibration process, the system health is recorded under different parameter combinations. Curve, predicting intervention success rate And the final mean time between failures (MTBF), using algorithms such as grid search or Bayesian optimization, to ensure that the success rate of predictive interventions is not lower than the target value. Under the premise of maximizing the mean time between failures (MTBF) as the optimization objective, an optimal set of parameters is automatically found. This method ensures that the setting of each parameter achieves an optimal balance at both the empirical and data levels, thereby guaranteeing the robustness and efficiency of the invention in practical applications.

[0063] The process of generating decision instructions by the adaptive risk hedging decision module includes dynamically adjusting the aggressiveness of predictive intervention to set an intervention trigger threshold; wherein, the intervention trigger threshold is used to compare the potential anomaly score with the threshold to determine whether to perform predictive intervention.

[0064] This embodiment is a further explanation based on Embodiment 1. To further clarify, the core of the adaptive risk hedging decision module's process of generating decision instructions lies in dynamically adjusting the degree of aggressiveness of the predicted intervention to set an intervention trigger threshold; the intervention trigger threshold is set using the following formula:

[0065]

[0066] in This indicates the intervention trigger threshold at the current moment. Indicates the baseline threshold. This indicates the predicted level of aggressiveness in intervention at the current moment;

[0067] The system's internal predictive model continuously outputs a potential anomaly score, which is derived from state data analysis by a pre-defined classifier model, such as a machine vision classifier. An intervention trigger threshold is used to compare the potential anomaly score with this threshold to determine whether to perform a predictive intervention. When the potential anomaly score exceeds the threshold, the system performs a predictive intervention action. Through this mechanism, adjusting the aggressiveness of the predictive intervention can directly and accurately control the sensitivity of the system's intervention behavior, providing an operable means for subsequent health-based adaptive strategies, enabling the system to exhibit different behavioral patterns under different risk levels.

[0068] To further illustrate, the preset classifier model can be a hybrid neural network model that integrates convolutional neural networks and long short-term memory networks. Optical state data from the state data, such as images of swabs in the feeding channel captured by an industrial camera, are input into the CNN branch to extract spatial features such as swab posture and stacking morphology. Pressure state data and motor state data, such as time-series readings of cylinder pressure sensors and time-series readings of current and torque of servo motors, are input into the LSTM branch to extract time-series features such as material flowability and transmission resistance. The features extracted by the two branches are concatenated and fused, and then passed through a fully connected layer. The Sigmoid activation function outputs a potential anomaly score between 0 and 1. The model is trained offline, and the training dataset contains a large amount of labeled data from normal feeding cycles as well as various typical pre-jamming and pre-blocking states collected from artificial simulations or historical data, enabling the model to accurately identify potential feeding anomalies.

[0069] Example 3

[0070] When the system health is greater than the preset risk hedging threshold, the adaptive risk hedging decision module dynamically fine-tunes the aggressiveness of the predicted intervention based on the difference between the predicted intervention success rate and the preset target intervention success rate through a proportional controller.

[0071] When the system health is less than or equal to the risk hedging entry threshold and greater than the preset health verification threshold, the adaptive risk hedging decision module gradually reduces the aggressiveness of the predictive intervention with a preset decay step size.

[0072] This embodiment is a further explanation based on embodiment 2. To further clarify, the adaptive risk hedging decision module executes differentiated control strategies according to the different intervals of the system's health status.

[0073] When the system health level exceeds the preset risk hedging threshold, the system is in a safe operating zone, and the control objective is to optimize intervention efficiency. At this time, the adaptive risk hedging decision module dynamically fine-tunes the predicted intervention aggressiveness based on the difference between the predicted intervention success rate and the preset target intervention success rate using the following formula with a proportional controller:

[0074]

[0075] in This indicates the current level of aggressiveness in predicting intervention. This indicates the predicted level of intervention aggressiveness at the previous moment. Represents the proportional gain coefficient. This indicates the current success rate of predicted interventions. Indicates the success rate of the target intervention. and These represent the minimum and maximum values ​​of the degree of radicalism, respectively. : Function, used to limit a value between a maximum and a minimum value;

[0076] This strategy enables the system to adaptively seek the optimal intervention point when it is in a healthy state. If the success rate is lower than the target, it will intervene more aggressively, and if it is higher, it will intervene more conservatively, thereby maximizing the accuracy and efficiency of short-term control without compromising long-term health.

[0077] When the system health level is less than or equal to the risk hedging entry threshold but greater than the preset health verification threshold, the system enters the risk hedging zone, and the control strategy shifts from optimizing performance to curbing health deterioration. At this point, the adaptive risk hedging decision module gradually reduces the aggressiveness of predictive interventions with a preset decay step size. This strategy is a key preventative measure that proactively raises the intervention threshold and reduces intervention behaviors that may cause erosion, directly addressing the root cause of system health deterioration and effectively curbing the trend of health deterioration, thus gaining valuable time for system stabilization or recovery.

[0078] The above uses the clip function to limit the degree of aggression. The design within the range ensures that the control system remains stable when faced with extreme success rate deviations, preventing unlimited adjustments and serving as an important measure to guarantee the robustness of the model.

[0079] When the system health is less than or equal to the preset health verification threshold, the adaptive risk hedging decision module first calculates the time change rate of the system health and generates a decision instruction based on the time change rate.

[0080] This embodiment is a further explanation based on Embodiment 2. To further clarify, when the system health is less than or equal to a preset health verification threshold, the system enters a critical state and must make a choice in the final risk response strategy. At this time, the adaptive risk hedging decision module calculates the time change rate of the system health using the following formula:

[0081]

[0082] in This indicates the rate of change in health status over time. Indicates the current health status. Indicates the health status at the previous moment. Indicates a preset calculation time interval;

[0083] The system generates decision instructions based on the rate of change over time. This step provides the system with the ability to dynamically perceive deteriorating trends. Compared to relying solely on static values ​​of health, introducing the rate of change makes decision-making more forward-looking, enabling the system to distinguish between two distinct critical states: a slow decline and a precipitous drop. This provides a crucial basis for taking appropriate final intervention measures.

[0084] Example 4

[0085] If the rate of change over time is greater than or equal to the preset critical deterioration rate, a conservative mode instruction is generated; the conservative mode instruction is used to set the degree of aggressiveness of the predictive intervention to its preset minimum value.

[0086] If the rate of change over time is less than the critical rate of deterioration, an online self-recovery command is generated; the online self-recovery command is used to perform physical cleaning or model recalibration operations.

[0087] When the system runs according to the conservative mode command for a longer period than the preset conservative observation period, and the average value of the time change rate within the conservative observation period is negative, an online self-recovery command is generated.

[0088] This embodiment is a further explanation based on the above embodiment. To further clarify, the adaptive risk hedging decision module executes the final risk response strategy based on the time rate of change.

[0089] If the rate of change over time is greater than or equal to the preset critical deterioration rate, that is, although the health is low, the rate of deterioration is still controllable, a conservative mode instruction is generated. The conservative mode instruction is used to set the degree of aggressiveness of the predictive intervention to its preset minimum value. This conservative mode is the last line of defense to avoid catastrophic failures. It maintains the operation of the system with minimal intervention and aims to observe whether the system can achieve self-stability by relying on its own robustness, thus avoiding unnecessary downtime.

[0090] If the rate of change over time is less than the critical rate of deterioration, i.e. the health status drops precipitously, an online self-recovery command is generated. The online self-recovery command is used to perform physical cleaning or model recalibration operations. This decision can decisively stop the deteriorating process and solve the problem by performing recovery measures, avoiding the occurrence of absolute failure events such as two consecutive jams.

[0091] The physical cleaning operation may specifically include a preset sequence of actions, such as: controlling the motor of the feeding mechanism to reverse a short distance to attempt to loosen any potentially stuck swabs; activating the air valve connected to the feeding channel to perform one or more high-pressure gas jets to remove debris or dust attracted by electrostatic charge from the channel; the model recalibration operation may include: adjusting the reference control vector... Updates can be performed; for example, the system can automatically select the actual control vector from the most recent N successful feeding events that did not trigger intervention. And calculate its statistical mean as the new To adapt to baseline shifts caused by batch changes in swabs or mechanical wear; to fine-tune the aforementioned classifier model online, using recently judged normal data as new training samples, and to iteratively update the weights of the last few layers of the model to adapt to slow changes such as gradual contamination of the sensor.

[0092] Furthermore, when the system runs according to the conservative mode instruction for a longer period than the preset conservative observation period, and the average value of the rate of change over time is negative within the conservative observation period, an online self-recovery instruction is also generated. This supplementary condition adds a time dimension to the decision-making logic, ensuring that the system will not remain indefinitely in an ineffective conservative strategy, thus guaranteeing the robustness and completeness of the decision and ultimately ensuring that the system can effectively recover from various complex sub-health states.

[0093] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. An automatic control system for swab feeding based on edge computing, characterized in that, It includes a data acquisition module, a system status assessment module, an adaptive risk hedging decision-making module, and a control command execution module; The data acquisition module is used to acquire optical status data, pressure status data and motor status data of the feeding mechanism in real time, and merge the acquired optical status data, pressure status data and motor status data into status data and send it to the system status evaluation module. The system status assessment module is used to receive the status data, calculate and generate the material supply throughput, predicted intervention success rate and system health based on the status data, and send the generated material supply throughput, predicted intervention success rate and system health to the adaptive risk hedging decision module. The adaptive risk hedging decision module is used to generate decision instructions based on the system health status, combined with the material supply throughput and the predicted intervention success rate. The control instruction execution module is used to receive the decision instruction and translate the decision instruction into specific control signals for the physical actuator; The process by which the system status assessment module calculates the system health is as follows: Step 1: After each intervention event occurs, based on the preset forgetting factor, the deterministic erosion degree of the control logic after the previous intervention event is attenuated to obtain the attenuated erosion degree. Step 2: Calculate the new erosion amount by combining the preset forgetting factor, the preset event weight factor, the intensity of the intervention event, and the intervention success indicator; Step 3: Add the attenuated erosion degree to the newly added erosion amount to update and obtain the current control logic deterministic erosion degree; The formula for calculating the deterministic erosion degree of the control logic at the current moment is: in This indicates the degree of erosion after the intervention. This indicates the degree of erosion since the last intervention. Indicates the forgetting factor, This represents the control vector for this intervention. Represents the reference control vector. This indicates that the intervention was successful. Represents the norm symbol; Step 4: Normalize the current control logic deterministic erosion degree with the preset failure boundary erosion degree threshold to obtain the normalized erosion degree; Step 5: By performing inverse mapping and nonlinear processing on the normalized erosion degree, the system health degree is derived; the formula for calculating the system health degree is: in This indicates the current system health status. Indicates the erosion level at the current moment. This represents the threshold for erosion at the failure boundary.

2. The automatic control system for swab feeding based on edge computing according to claim 1, characterized in that, The process by which the adaptive risk hedging decision module generates the decision instruction includes dynamically adjusting the aggressiveness of the predictive intervention to set an intervention trigger threshold; wherein the intervention trigger threshold is used to compare the potential anomaly score with the threshold to determine whether to perform a predictive intervention.

3. The automatic swab feeding control system based on edge computing according to claim 2, characterized in that, When the system health is greater than the preset risk hedging threshold, the adaptive risk hedging decision module dynamically fine-tunes the aggressiveness of the predicted intervention based on the difference between the predicted intervention success rate and the preset target intervention success rate through a proportional controller.

4. The automatic swab feeding control system based on edge computing according to claim 2, characterized in that, When the system health is less than or equal to the risk hedging entry threshold and greater than the preset health verification threshold, the adaptive risk hedging decision module gradually reduces the aggressiveness of the predictive intervention with a preset decay step size.

5. The automatic swab feeding control system based on edge computing according to claim 2, characterized in that, When the system health is less than or equal to the preset health verification threshold, the adaptive risk hedging decision module first calculates the time change rate of the system health and generates the decision instruction based on the time change rate.

6. The automatic swab feeding control system based on edge computing according to claim 5, characterized in that, If the rate of change over time is greater than or equal to a preset critical deterioration rate, a conservative mode instruction is generated; the conservative mode instruction is used to set the degree of aggressiveness of the predictive intervention to its preset minimum value.

7. The automatic swab feeding control system based on edge computing according to claim 6, characterized in that, If the rate of change over time is less than the critical deterioration rate, an online self-recovery command is generated; the online self-recovery command is used to perform physical cleaning or model recalibration operations.

8. The automatic swab feeding control system based on edge computing according to claim 6, characterized in that, When the system operates according to the conservative mode instruction for a duration exceeding the preset conservative observation period, and the average value of the time change rate within the conservative observation period is negative, an online self-recovery instruction is generated.

Images