Operational context-aware production process alarm rule evolution method and system

Abstract

The application relates to the technical field of industrial automation and intelligent manufacturing, and particularly discloses a production process alarm rule evolution method and system based on operation context awareness. By fusing multi-source real-time and historical data, a dynamic deduction model is constructed, abnormal trigger probabilities of each production unit are predicted in advance, and a candidate set of vulnerable units is generated. Based on virtual simulation and context-weighted comprehensive production loss evaluation, core intervention units are screened out. Through enumeration combination, three-dimensional priority sorting and nested process scheduling optimization verification cycles, the minimum effective intervention combination and the matching optimization strategy capable of making the remaining risk controllable are determined. The combination is verified in an online environment, and if successful, the combination is solidified as a formal alarm rule; if failed, the situation deviation is analyzed and the model is updated, forming a self-evolution mechanism of perception-evaluation-optimization-verification-evolution, so that the alarm rule can continuously adapt to the dynamic production environment.

Description

Technical Field

[0001] This invention relates to the field of industrial automation and intelligent manufacturing technology, specifically to a method and system for evolving alarm rules in a production process that is aware of the operational context. Background Technology

[0002] In modern industrial production, alarm systems are crucial tools for ensuring safe and stable operation. Traditional alarm rules are typically based on fixed thresholds or static logic settings, such as statistical process control methods or empirical rules based on historical fault data. While these methods can identify anomalies to some extent, their limitations are becoming increasingly apparent: First, fixed thresholds cannot adapt to the dynamic fluctuations of multi-dimensional operational contexts during production, such as changes in raw materials, equipment aging, and order priority adjustments, easily leading to false alarms or missed alarms. Second, existing alarm rules lack quantitative assessments of the overall impact on the production system, often resulting in alarm storms or critical alarms being overwhelmed, increasing the workload of operators and delaying the handling of genuine risks.

[0003] In recent years, with the development of industrial big data and artificial intelligence technologies, some research has begun to introduce data-driven methods for alarm optimization, such as machine learning-based anomaly detection models. However, these methods still have significant shortcomings: on the one hand, most models only focus on single equipment or process parameters, failing to fully consider the coupling effects between equipment in the production system and the synergistic impact of global production goals; on the other hand, existing methods generally lack evolution mechanisms, and once rules are deployed, they are difficult to adaptively adjust with changes in the production environment, leading to a decay of alarm effectiveness over time. Furthermore, traditional methods often remain at a simple alarm-manual handling mode in alarm response strategies, failing to optimize with process scheduling and control systems, and unable to provide system-level adaptive adjustment schemes before or simultaneously with alarm triggering, making it difficult to achieve true intelligent early warning and self-healing control.

[0004] In summary, existing alarm technologies have significant shortcomings in context awareness, system-level impact assessment, optimization, and adaptive evolution, which restricts their effective application in complex and dynamic production environments. Therefore, there is an urgent need to propose an intelligent alarm rule generation method that can perceive the operational context in real time, quantitatively assess the alarm impact, and possess self-optimization and continuous evolution capabilities. Summary of the Invention

[0005] To overcome the problems of existing alarm systems in the background art regarding context awareness, system-level impact assessment, optimization, and adaptive evolution, this invention provides a method and system for evolving alarm rules in production processes with operational context awareness. It proactively identifies risks through dynamic deduction models, selects core intervention units based on context-weighted production loss assessments, and combines multi-objective combination optimization and online verification to achieve adaptive generation and continuous evolution of alarm rules.

[0006] The specific technical solution of this application is as follows:

[0007] According to one aspect of this application, a method for evolving production process alarm rules that is aware of the runtime context is provided, comprising:

[0008] Based on the current production operation context data, a dynamic inference model is constructed to predict the probability and temporal distribution of abnormal triggering of equipment in future time windows, and to generate a candidate set of vulnerable units;

[0009] Simulation analysis is performed on the units in the candidate set of vulnerable units to evaluate the comprehensive production loss caused by the failure of each unit, and core intervention units are selected based on predefined economic impact factors and loss thresholds.

[0010] Perform a combinatorial space exploration on the core intervention unit set, enumerate all possible unit combinations and sort them by priority; through nested optimization and verification loops, re-tune the control strategy for each priority combination and evaluate the overall system loss until the minimum effective intervention combination that can make the overall system loss meet the preset requirements is selected.

[0011] The minimum effective intervention combination is validated online. If the validation is successful, the current context, unit combination, and optimization strategy are encapsulated into a formal alarm rule and stored. If the validation fails, the difference between the actual operating context and the assumptions of the dynamic inference model is analyzed, and the difference factors are fed back to update the dynamic inference model and unit set. The process of screening and optimization is then restarted until the validation is successful and a formal alarm rule is generated.

[0012] As a further option of the method of the present invention, the operating context data includes a manpower configuration matrix, order queue and order priority weight, and real-time energy cost coefficient obtained through the manufacturing execution system; an environmental parameter vector collected through an Internet of Things sensor network; equipment maintenance status and health status codes obtained through an equipment management platform; and real-time equipment sensor information streams collected through an industrial bus or Internet of Things gateway.

[0013] As a further option of the method of the present invention, the dynamic inference model adopts a hybrid architecture based on the fusion of physical mechanisms and data-driven approaches, including:

[0014] Process disturbance path simulation based on material and energy balance;

[0015] Device health status probability prediction based on survival analysis or deep time series prediction;

[0016] By integrating process disturbance simulation results, equipment health probability predictions, and context constraint weights, the comprehensive triggering probability and time sequence distribution of each unit triggering a system-level alarm within a future time window are calculated.

[0017] As a further option of the method of the present invention, the comprehensive production loss is obtained by context-weighted multi-objective loss aggregation calculation, and the multi-objective loss includes: expected delivery delay loss due to failure, expected additional energy consumption loss, and expected product quality degradation loss.

[0018] The weighting coefficients for each loss term are dynamically determined based on the current runtime context.

[0019] As a further option of the method of the present invention, the priority ranking is based on a three-dimensional scoring function, including:

[0020] Inverse indicator of size calculate: The scale inverse indicator encourages the selection of combinations with fewer intervention units, which conforms to the principle of minimum intervention.

[0021] Risk index calculate: ,in It is a combination All units in the prediction time window While maintaining normal joint probability;

[0022] Context fit metrics calculate: ,in It is an evaluation portfolio With the current running context The matching function;

[0023] Based on inverse scale indicators Risk index and context fit metrics Calculate the overall priority score for each combination. ;in, , , These are the inverse indicators of scale. Risk index and context fit metrics The assigned weights;

[0024] Sort all combinations from highest to lowest based on their overall priority score to form a priority queue.

[0025] As a further option of the method of the present invention, in the nested optimization verification loop, for each candidate combination, the control parameter set is adjusted by the process scheduling optimizer, the backup path flag and buffer strategy parameters are enabled, so as to minimize the overall production loss of the remaining unit set in the simulation environment, and re-evaluate whether the loss is lower than the dynamic tolerable loss threshold.

[0026] The system configuration after the nested optimization verification loop is simulated in a digital twin environment to simulate virtual island disconnection. The comprehensive production loss index caused by the remaining unit set is recalculated and compared with the tolerable loss threshold dynamically updated based on the current context to determine whether the verification passes.

[0027] As a further option of the method of the present invention, in the online operation verification stage, the minimum effective intervention combination and its supporting optimization strategy are deployed as temporary alarm rules to monitor actual production indicators within a preset verification period; if the indicators meet expectations, the current context, unit combination and optimization strategy are encapsulated as formal alarm rules and stored in the rule base; if the verification fails, the situation deviation analysis process is triggered.

[0028] As a further option of the method of the present invention, the context deviation analysis after verification failure includes:

[0029] Compare the actual runtime context data sequence with the assumed context of the dynamic inference model;

[0030] Identify key discrepancy factors as causes of model inaccuracy;

[0031] The discrepancy factors are fed back to the dynamic inference model for updates, and the entire process from model building to online validation is retried.

[0032] As a further option of the method of the present invention, it also includes recording the running context features, unit combination status, optimization strategy parameters and verification period performance data corresponding to the rule when generating a formal alarm rule, forming a reusable rule knowledge entry to support subsequent rule matching and scenario adaptive invocation.

[0033] This application also provides a runtime context-aware production process alarm rule evolution method system, the system comprising:

[0034] The runtime context awareness and model building module is configured to: collect multi-dimensional runtime context data of the current production system; build and run a dynamic inference model based on the data to predict the probability and timing distribution of abnormal triggering of each device or process unit in the system within a future time window; and generate a first-round candidate set of vulnerable units based on the prediction results.

[0035] The simulation evaluation and core unit screening module is configured to: perform virtual island disconnection simulation for each unit in the candidate set of vulnerable units; during the simulation, dynamically calculate the comprehensive production loss index caused by unit failure based on the current running context; and screen out core intervention units whose comprehensive production loss exceeds the threshold and require intervention based on the predefined tolerable loss threshold, forming a core intervention unit set.

[0036] The combinatorial optimization and verification decision module is configured to: explore the combinatorial space of the core intervention unit set, enumerate all possible non-empty unit subsets, and prioritize them according to multi-dimensional factors such as combinatorial size, joint risk probability, and adaptability to the current context; execute nested optimization and verification loops for each candidate combination according to priority order, where: for the current candidate combination, the system control strategy of the remaining unit set is readjusted and optimized through the process scheduling optimizer; evaluate the comprehensive production loss of the optimized system in the face of the risk of the remaining units in the digital twin environment; iteratively execute this process until the minimum effective intervention combination and its corresponding optimization strategy that can make the comprehensive system loss meet the preset requirements are selected;

[0037] The online verification and rule evolution module is configured to: deploy the minimum effective intervention combination and corresponding optimization strategies to the actual production system for online operation verification; if the verification is successful, encapsulate the current operating context features, unit combinations and optimization strategies into formal alarm rules and store them in the rule base; if the verification fails, analyze the differences between the actual operating context and the context used in the model building stage, and feed back the identified key difference factors to the operating context awareness and model building module to trigger the update of the dynamic inference model and the re-execution of subsequent modules until a formal alarm rule that has been successfully verified is generated.

[0038] The beneficial effects of this application are as follows:

[0039] This invention significantly improves the accuracy and operational efficiency of alarm systems by employing context-aware and dynamic inference mechanisms. Specifically, this method converges hundreds of static alarm rules based on fixed thresholds into a small number of dynamic rules with context adaptability. Applying this method, the system's false alarm rate decreases by an average of approximately 35%, while the missed alarm rate for critical risks decreases by over 50%. This is primarily due to the dynamic inference model's prediction accuracy of over 85% for the probability of abnormal triggering of equipment in future time windows. Combined with context-weighted production loss simulation, it effectively filters out redundant alarms with minimal impact on overall production goals, enabling maintenance personnel to focus on truly high-risk intervention points.

[0040] Furthermore, this invention achieves a fundamental shift from passive alarm to proactive prevention and closed-loop self-evolution. Through combinatorial optimization and nested verification, the system can automatically find the minimum effective intervention combination and generate corresponding process adjustment strategies, enabling the production system to adaptively adjust when facing potential risks. Practical application data shows that this method can reduce unplanned downtime caused by potential equipment failures by approximately 25% and overall production losses by 15%–30%. More importantly, the system establishes the ability for continuous rule evolution through online verification and scenario deviation analysis feedback. During a three-month trial run, all performance degradation issues in the initial rule base caused by environmental changes were updated through an automatic regeneration process, achieving stable long-term performance maintenance of the alarm system. Attached Figure Description

[0041] Figure 1 A schematic diagram of the overall evolution method for production process alarm rules that are aware of the runtime context;

[0042] Figure 2 Flowchart of S100, a method for evolving production process alarm rules that is aware of the operating context;

[0043] Figure 3 S200 flowchart for the evolution method of production process alarm rules with runtime context awareness;

[0044] Figure 4 S300 flowchart for the evolution method of production process alarm rules with runtime context awareness;

[0045] Figure 5 S400 flowchart for the evolution method of production process alarm rules that are aware of the running context. Detailed Implementation

[0046] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0047] Modern industrial production processes are highly complex, integrating numerous automated equipment, process systems, and information management modules. Production process alarm systems are crucial for ensuring safe, stable, and efficient operation. In traditional production systems, alarm rules are typically based on fixed thresholds or static logic settings, making it difficult to adapt to the variability and complexity of the production environment. As production systems expand in scale and become more coupled, traditional alarm methods often suffer from problems such as missed alarms, false alarms, and alarm storms, leading to low operational efficiency and uncontrollable production losses. While some data-driven alarm optimization methods exist, most lack dynamic awareness of the operational context and have not achieved systematic and progressive evolution of alarm rules, making it difficult to maintain high availability and adaptability in actual production.

[0048] The theoretical foundation of this invention is built upon multi-dimensional operational context fusion modeling theory, simulation-based dynamic assessment theory of production losses, and multi-objective combined optimization and verification theory. By constructing a dynamic inference model to proactively identify vulnerable links, quantifying production losses based on context weighting, and utilizing intelligent optimization search and verification to determine the optimal intervention rules, the invention ultimately achieves adaptive generation and continuous evolution of alarm rules.

[0049] The core theory is as follows:

[0050] Defined at time The multidimensional runtime context of a production system is a vector. , among which, components Representing the Class context parameters, Total number of parameters. Context vector. It is time-varying and forms the basis of all decisions.

[0051] For the first in the production system Each device or process unit defines its future time window. Within a specific context The probability of an exception occurring is . Calculations are performed using a dynamic simulation model. The input to the dynamic simulation model includes real-time sensor data streams from the device. Historical performance degradation trend and the current context The trigger probability distribution forms the basis of vulnerability assessment.

[0052] Define the comprehensive production loss index Used to quantize when the unit set The anticipated losses at the entire production system level in the event of a failure or disconnection. The comprehensive production loss index is a context-weighted aggregation of multi-objective losses: ;in: Describing the set Losses due to anticipated delivery delays caused by the malfunction. This indicates the expected additional energy consumption loss. This indicates the expected loss due to product quality degradation. , , In the current context The dynamically determined loss weight coefficients reflect the relative importance of different production objectives in the current scenario.

[0053] In the combinatorial optimization phase, for the combination of Any candidate combination consisting of core intervention units Define its three-dimensional priority scoring function. Used for sorting: ;in: It is a combination The base number, The term gives higher priority to smaller combinations, and the coefficient Its weight. It is a combination The probability that all units in the evaluation period will remain normal. The term gives higher priority to combinations with low joint survival probabilities, and the coefficient Its weight. It is a combination With the current context Fit, coefficient Its weight.

[0054] In nested optimization verification loops, for the selected combination The goal of the process scheduling optimizer is to minimize the set of remaining cells. , The overall production loss. Optimization variables include the set of control parameters. Alternate path enabled flag Buffer strategy parameters The optimization problem can be formalized as follows: ; ;in Representing the A process, equipment, or safety constraint. The total number of constraints is determined. After optimization, a new system configuration is obtained, and the system is re-evaluated. .

[0055] The above theoretical framework provides the logical basis for this invention, ensuring the accuracy, adaptability, and systematic nature of the alarm rule generation, optimization, and evolution process.

[0056] The specific embodiments of the present invention will be described in detail below.

[0057] Example 1:

[0058] Please see Figure 1 This illustrates a runtime context-aware production process alarm rule evolution method provided by an embodiment of the present invention, the method comprising:

[0059] S100: Collects multi-dimensional runtime context data, integrates real-time and historical information, constructs a dynamic inference model, and outputs the first round of vulnerability unit candidate set.

[0060] S200: Perform virtual island disconnection simulation for candidate units, calculate the comprehensive production loss based on context weighting, and select core intervention units.

[0061] S300: Perform combinatorial space exploration and nested optimization verification on the core intervention unit set to find the minimum effective intervention combination.

[0062] S400: Initiate online operation verification for the minimum effective intervention combination. If successful, it is packaged into a formal alarm rule and stored in the database. If it fails, it analyzes the deviation and feeds back to update the dynamic inference model.

[0063] The specific plan is as follows:

[0064] In a context-aware production process alarm rule evolution method, S100 constructs a dynamic inference model that can predict future risks by real-time sensing and fusing multi-source data.

[0065] Please refer to Figure 2 It illustrates a flowchart of an exemplary production process alarm rule evolution method S100 that is aware of the runtime context, which includes:

[0066] S110: In this invention, the runtime context awareness layer interfaces with each production management system through a standard industrial communication protocol to synchronously collect raw data that constitutes a multi-dimensional runtime context state vector at fixed intervals.

[0067] In one possible implementation, the data collection content and method include:

[0068] The manpower allocation matrix for the current production line can be obtained in real time through the Manufacturing Execution System (MES) interface. Order queue and priority weights of each order and real-time energy cost coefficients from the enterprise resource planning system. .

[0069] Environmental parameter vectors are collected through an IoT sensor network deployed in the workshop environment. This includes temperature, humidity, and dust concentration.

[0070] The maintenance status of critical production equipment can be obtained through the equipment management platform. This includes the last maintenance time, the planned maintenance time, and a health status code based on preliminary sensor readings.

[0071] High-speed acquisition of real-time sensor information streams from all relevant devices via industrial bus or IoT gateway. Process parameters such as motor current, pressure, flow rate, speed, and temperature.

[0072] Extract recent performance degradation trend data for each device from the historical database. Performance degradation trend data are represented as a time series of slow shifts in key parameters or a performance index calculated based on statistical process control.

[0073] All collected data is assigned a unified timestamp and undergoes alignment and interpolation processing at the data fusion layer to ensure time synchronization and create a consistent global data snapshot. ,in .

[0074] S120: In this invention, the global data snapshot obtained in S110 is used to drive a pre-trained or configured dynamic inference model.

[0075] In this invention, the dynamic extrapolation model simulates a future preset time window. Internally, it refers to the behavior of the production system under various potential disturbances, and the predicted probability of abnormal triggering of each unit.

[0076] In one possible implementation, the dynamic simulation model employs a hybrid architecture that integrates physical mechanisms and data-driven approaches. For process disturbance paths, a dynamic simulation model based on differential equations or discrete events, using material balance and energy balance as the basis, is used to simulate the propagation and evolution of process parameters under given initial and boundary conditions.

[0077] In one possible implementation, the operation of the dynamic inference model includes:

[0078] For device health probability, a dynamic extrapolation model based on survival analysis or deep time series prediction is used. The input to the dynamic extrapolation model is real-time sensor data from the device. and its historical trend of degradation Output to device In the future Conditional probability of a specific type of failure occurring within the system Context parameters As a regulating factor in dynamic inference models.

[0079] Define context constraint weight vector This is used to quantify the current context's level of attention or tolerance to different types of anomalies.

[0080] The simulation results of process disturbances, the prediction of equipment health probability, and the weights of contextual constraints are fused together. For each equipment unit... Calculate its future Internally, after comprehensively considering the process chain effect and its own health status, the overall trigger probability of ultimately triggering a system-level alarm is determined. and its possible triggering time distribution The fusion process employs weighted summation or probabilistic graphical dynamic deduction models for reasoning.

[0081] S130: In this invention, based on the output of the dynamic simulation model, potential risk units that need to enter the next stage of detailed evaluation are selected.

[0082] In one possible implementation, the logic for generating the first round of vulnerability candidate sets is as follows:

[0083] Set a trigger probability threshold For all device units If the overall trigger probability is considered If so, then the unit will be included in the candidate set.

[0084] Attach a trigger condition label to each unit in the candidate set. The label should include at least: the predicted main triggering factor, the overall trigger probability value, the most likely triggering time interval, and the relevant contextual factors.

[0085] Output a first-round set of vulnerability unit candidates with rich metadata. , as the input of S200.

[0086] In a context-aware production process alarm rule evolution method, S200, based on refined production loss simulation and evaluation, selects the core units that truly require intervention from the candidate set, and filters out units that have limited impact on the overall production target even if an anomaly occurs.

[0087] Please refer to Figure 3 It illustrates a flowchart of an exemplary production process alarm rule evolution method S200 that is aware of the operating context, which includes:

[0088] S210: In this invention, for the first round of vulnerability unit candidate set Each unit in Initiate a separate simulation.

[0089] In one possible implementation, the simulation process is a hypothetical unit. At this moment, the system disconnects or fails, but at the same time, other parts of the main production process try their best to keep running.

[0090] S220: In this invention, during and after the virtual simulation operation, the quantization is due to the unit The multi-dimensional production losses caused by disconnection are dynamically weighted according to the current operating context to form a comprehensive production loss index.

[0091] In one possible implementation, the comprehensive production loss index is calculated as follows:

[0092] After the simulation model finishes running, the following result data is extracted, including:

[0093] Delivery delay loss Specifically, this involves calculating the delay in the completion time of each order due to the failure relative to the plan, and then calculating the expected economic loss by combining the urgency of the order itself with the delay penalty rate stipulated in the contract.

[0094] Energy cost loss Specifically, this involves multiplying the difference between the total system energy consumption during the simulation and the baseline energy consumption during normal operation by the current real-time energy cost. .

[0095] Quality deviation loss Specifically, if a fault causes process parameters to deviate from the ideal range and thus affect product quality, the loss is calculated based on the defect rate or grade decline estimated by the quality inspection model, combined with the product value.

[0096] Based on the current running context The dynamic loss weight coefficients are obtained from a predefined strategy library or a real-time calculation module. , , ,in , , This represents the weight calculation function.

[0097] Calculate the first round of vulnerability unit candidate set Middle unit Comprehensive production loss index : Comprehensive Production Loss Index Quantitatively reflects the unit under the current specific production situation. The severity of the malfunction.

[0098] S230: In this invention, the calculated comprehensive loss index is compared with the loss boundary that the system can tolerate in the current context, and a screening decision is made.

[0099] In one possible implementation, the core intervention unit screening logic is as follows: setting a dynamic tolerable loss threshold. For candidate sets Each unit in Execute the judgment:

[0100] like Then remove the unit from the candidate set. ;

[0101] like Then retain the unit. They were then designated as core intervention units.

[0102] All the retained and labeled units constitute the core intervention unit set. Simultaneously, the comprehensive loss index corresponding to each core unit is recorded. .

[0103] In a context-aware production process alarm rule evolution approach, the core challenge for S300 is selecting an optimal subset of units from multiple core intervention units that need to be covered by the new alarm rules. The goal is to minimize the intervention scope and, through system optimization, reduce the remaining risk to an acceptable level.

[0104] Please refer to Figure 4 The diagram illustrates a flowchart of an exemplary runtime context-aware production process alarm rule evolution method S300, which includes:

[0105] S310: In this invention, for the inclusion A collection of core intervention units We enumerate all possible non-empty subset combinations, which represent potential coverage areas of different alarm rules. Then, we perform multi-dimensional evaluation and ranking of the non-empty subset combinations to guide the order of subsequent optimization and verification.

[0106] In one possible implementation, the non-empty subset combination enumeration is generated from combinations of single units to combinations containing all... The complete set of each unit There are 10 possible combinations. Each combination is denoted as . ,in This is a composite index.

[0107] In one possible implementation, the ranking of non-empty subset combinations is based on a comprehensive priority score, which is calculated based on three standardized indicators, including a size inverse indicator, a risk indicator, and a context fit indicator.

[0108] Specifically:

[0109] 1. Inverse indicator of scale The inverse scale indicator encourages the selection of combinations with fewer intervention units, which aligns with the principle of minimum intervention.

[0110] 2. Risk Level Indicators .in It is a combination All units in the prediction time window The normal joint probability is maintained at the same time, and is estimated based on the independent triggering probability of each unit predicted in S100, while taking into account the fault correlation between units.

[0111] 3. Context fit metrics Context fit metrics evaluation combination The strength of association with the current runtime context. A higher fit indicates that making rules for this combination is more meaningful in the current context.

[0112] Based on the above three standardized indicators, in one possible implementation, the comprehensive priority score for each combination is calculated as follows: ;in, , , Assign weights to the three indicators.

[0113] All combinations according to Sort them from highest to lowest to form a priority queue. The ranking principle is: the higher the score, the smaller the portfolio size, the higher the risk, and the more relevant it is to the current scenario, and it should be prioritized for subsequent verification.

[0114] S320: In this invention, from the priority queue We start with the highest priority combination and try each combination in turn. For each combination we try, we don't adopt it directly. Instead, we first try to compensate for the risks of not selecting other units through process scheduling optimization, and then verify whether the system is safe after compensation.

[0115] In one possible implementation, each step of the nested optimization verification loop is described in detail below:

[0116] a) From Extract the highest-scoring combination that has not yet been processed, and denote it as... .

[0117] b) Calculate the remaining set of core intervention units, i.e. , The remaining core intervention unit set is in rule coverage. Even after that, the system itself still needs to bear the risks of the unit.

[0118] c) Using the current running context With constraints on the production system state, the goal is to minimize the set of remaining units. With the potential overall production loss as the objective, the process scheduling optimizer is activated. It outputs an optimized system configuration scheme. .

[0119] As an option in this implementation, the variables that the optimizer can adjust include, but are not limited to:

[0120] After control parameters are reset, backup paths are enabled, buffer strategies are adjusted, and the optimizer is run.

[0121] d) Apply the optimized configuration scheme in the digital twin environment. Then, the virtual island disconnection simulation and loss calculation of S220 are re-executed, but this time the simulation object is the remaining core intervention unit set. Each unit in the system. Calculate the new system state after optimization. The resulting new comprehensive production loss index .

[0122] e) Reassess the resulting loss index With dynamic tolerable loss threshold If a comparison is made, If, then the verification passes; if If not, the verification will fail.

[0123] S330: In this invention, the final alarm rule coverage is determined based on the result of the nested optimization verification loop.

[0124] Specifically, if in step S320-e, for the current combination If the verification passes, the loop terminates. At this point, Determined as the minimum effective intervention combination The term "minimum effective" refers to the first successfully validated combination in the current priority ranking that, after system optimization, allows the remaining risk to meet the target. This also includes the configuration scheme output by the optimizer. The optimization strategy that complements this alarm rule has been retained.

[0125] If the verification fails in step S320-e, it means that even if it is covered... Optimizations were made, but the remaining risks remained unacceptable. At this point, the system automatically abandoned the plan. Then from the priority queue Select the next highest priority combination and repeat step S320 to start a new round of selection-optimization-verification loop.

[0126] This process iterates until a combination is found that passes the validation, or all combinations have been traversed. Theoretically, all combinations should pass because they cover all core units, the remaining set is empty, and the loss is zero. The final output is... and its corresponding It will be sent to the S400 for practical testing.

[0127] In a context-aware production process alarm rule evolution method, S400 aims to place the theoretically optimal intervention scheme derived from S300 into a real production environment for a period of trial operation and effect verification, thereby deciding whether to solidify it into a formal, reusable alarm rule.

[0128] Please refer to Figure 5 The diagram illustrates a flowchart of an exemplary runtime context-aware production process alarm rule evolution method S400, which includes:

[0129] S410: In this invention, the minimum effective combination of interventions determined in S300 is used. and supporting optimization strategies It is deployed into the actual production control and monitoring system and enters a verification period of a preset duration.

[0130] During the verification period, specifically, a new temporary rule is created in the alarm rule base. This new rule logically monitors... When the triggering condition is met, the system not only generates an alarm for the unit's status, but also automatically or suggest the execution of related optimization strategies. .

[0131] In one possible implementation, the system did not fail during the verification period. If a major anomaly occurs outside the coverage area, and the core production indicators remain stable within the preset target range or show improvement, the verification is successful, the temporary rule is converted into a formal alarm rule, and it is encapsulated and stored in the rule base.

[0132] S420: In this invention, if production indicators fail to meet standards or unexpected anomalies occur during the verification period, it means that there is a deviation between theoretical deduction and actual operation, indicating verification failure. When the verification failure conditions are met, the system automatically starts the scenario deviation analyzer to perform analysis and feedback processes.

[0133] In one possible implementation, the analysis and feedback process after verification failure includes:

[0134] Context bias analyzer compares and validates the actual operating context data sequence during the validation period. The context of inference assumptions relied upon when constructing the dynamic inference model in phase S100. The focus of the analysis is to identify the systematic differences between the two. For example: Are there new contextual dimensions that were not considered in the extrapolation? Do the actual fluctuation ranges of certain contextual parameters far exceed the model assumptions? Are the fault coupling relationships between devices more complex than the model presupposes?

[0135] The context bias analyzer outputs one or more key difference factors. Key Difference Factors This is the main reason for the inaccuracy of the projection.

[0136] Key differentiators identified As a correction signal, it is fed back into the dynamic inference model construction process of stage S100. The specific operations are: retraining the model with new data; updating the context weight mapping relationship in the model; and correcting the assumptions about the correlation of equipment failures, etc.

[0137] After the model is updated, the system automatically re-triggers the complete process starting from S100: based on the updated model and the latest running context, it re-outputs the vulnerable unit candidate set, re-executes S200 to S400 until verification is passed, and converts the temporary rules into formal alarm rules and encapsulates them into the rule base.

[0138] Example 2:

[0139] A system for evolving alarm rules in a production process that is aware of runtime context includes:

[0140] The runtime context awareness and model building module is configured to: collect multi-dimensional runtime context data of the current production system; build and run a dynamic inference model based on the data to predict the probability and timing distribution of abnormal triggering of each device or process unit in the system within a future time window; and generate a first-round candidate set of vulnerable units based on the prediction results.

[0141] The simulation evaluation and core unit screening module is configured to: perform virtual island disconnection simulation for each unit in the candidate set of vulnerable units; during the simulation, dynamically calculate the comprehensive production loss index caused by unit failure based on the current running context; and screen out core intervention units whose comprehensive production loss exceeds the threshold and require intervention based on the predefined tolerable loss threshold, forming a core intervention unit set.

[0142] The combinatorial optimization and verification decision module is configured to: explore the combinatorial space of the core intervention unit set, enumerate all possible non-empty unit subsets, and prioritize them according to multi-dimensional factors such as combinatorial size, joint risk probability, and adaptability to the current context; execute nested optimization and verification loops for each candidate combination according to priority order, where: for the current candidate combination, the system control strategy of the remaining unit set is readjusted and optimized through the process scheduling optimizer; evaluate the comprehensive production loss of the optimized system in the face of the risk of the remaining units in the digital twin environment; iteratively execute this process until the minimum effective intervention combination and its corresponding optimization strategy that can make the comprehensive system loss meet the preset requirements are selected;

[0143] The online verification and rule evolution module is configured to: deploy the minimum effective intervention combination and corresponding optimization strategies to the actual production system for online operation verification; if the verification is successful, encapsulate the current operating context features, unit combinations and optimization strategies into formal alarm rules and store them in the rule base; if the verification fails, analyze the differences between the actual operating context and the context used in the model building stage, and feed back the identified key difference factors to the operating context awareness and model building module to trigger the update of the dynamic inference model and the re-execution of subsequent modules until a formal alarm rule that has been successfully verified is generated.

[0144] Those skilled in the art will understand that the embodiments of this application are provided as methods, systems, or computer program products. Therefore, this application takes the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application takes the form of a computer program product implemented on one or more computer storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer program code. The solutions in the embodiments of this application are implemented using various computer languages, exemplified by the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0145] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, are implemented by computer program instructions. These computer program instructions are provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0146] These computer program instructions are also stored in a computer read-memory memory (CROM) that can direct a computer or other programmed data processing device to operate in a specific manner, such that the instructions stored in the CROM produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0147] These computer program instructions are also loaded onto a computer or other programmed data processing device, causing a series of operational steps to be performed on the computer or other programmed device to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmed device for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0148] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0149] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for evolving alarm rules in a production process that is aware of runtime context, characterized in that, include: Based on the current production operation context data, a dynamic inference model is constructed to predict the probability and temporal distribution of abnormal triggering of equipment in future time windows, and to generate a candidate set of vulnerable units; Simulation analysis is performed on the units in the candidate set of vulnerable units to evaluate the comprehensive production loss caused by the failure of each unit, and core intervention units are selected based on predefined economic impact factors and loss thresholds. Among them, the comprehensive production loss index defines comprehensive production loss. Used to quantize when the unit set The expected losses at the entire production system level when a failure or disconnection occurs; the comprehensive production loss index is a context-weighted aggregation of multi-objective losses: ;in: Describing the set Losses due to anticipated delivery delays caused by the malfunction. This indicates the expected additional energy consumption loss. This indicates the expected loss due to product quality degradation. , , In the current context The dynamically determined loss weight coefficients reflect the relative importance of different production objectives in the current scenario; Perform a combinatorial space exploration on the core intervention unit set, enumerate all possible unit combinations and sort them by priority; through nested optimization and verification loops, re-tune the control strategy for each priority combination and evaluate the overall system loss until the minimum effective intervention combination that can make the overall system loss meet the preset requirements is selected. In the nested optimization and verification loop, for the selected combination The goal of the process scheduling optimizer is to minimize the set of remaining cells. The overall production loss; optimization variables include the set of control parameters. Alternate path enabled flag Buffer strategy parameters The optimization problem is formalized as follows: ; ;in Representing the A process, equipment, or safety constraint. To constrain the total number, a new system configuration is obtained after optimization, and then re-evaluated. ; The minimum effective intervention combination is validated online. If the validation is successful, the current context, unit combination, and optimization strategy are encapsulated into a formal alarm rule and stored. If the validation fails, the difference between the actual operating context and the assumptions of the dynamic inference model is analyzed, and the difference factors are fed back to update the dynamic inference model and unit set. The process of screening and optimization is then restarted until the validation is successful and a formal alarm rule is generated.

2. The production process alarm rule evolution method with runtime context awareness according to claim 1, characterized in that, The operational context data includes manpower allocation matrix, order queue and order priority weight, and real-time energy cost coefficient obtained through the manufacturing execution system; environmental parameter vectors collected through the Internet of Things sensor network; equipment maintenance status and health status codes obtained through the equipment management platform; and real-time equipment sensor information streams collected through industrial bus or Internet of Things gateway.

3. The production process alarm rule evolution method based on runtime context awareness according to claim 1, characterized in that, The dynamic inference model adopts a hybrid architecture based on the fusion of physical mechanisms and data-driven approaches, including: Process disturbance path simulation based on material and energy balance; Device health status probability prediction based on survival analysis or deep time series prediction; By integrating process disturbance simulation results, equipment health probability predictions, and context constraint weights, the comprehensive triggering probability and time sequence distribution of each unit triggering a system-level alarm within a future time window are calculated.

4. The production process alarm rule evolution method with runtime context awareness according to claim 1, characterized in that, The comprehensive production loss is calculated by context-weighted multi-objective loss aggregation, which includes: expected delivery delay loss due to failure, expected additional energy consumption loss, and expected product quality degradation loss. The weighting coefficients for each loss term are dynamically determined based on the current runtime context.

5. The production process alarm rule evolution method based on runtime context awareness according to claim 1, characterized in that, The priority ranking is based on a three-dimensional scoring function, including: Inverse indicator of size calculate: The scale inverse indicator encourages the selection of combinations with fewer intervention units, which conforms to the principle of minimum intervention. Risk index calculate: ,in It is a combination All units in the prediction time window While maintaining normal joint probability; Context fit metrics calculate: ,in It is an evaluation portfolio With the current running context The matching function; Based on the inverse index of scale Risk index and context fit metrics Calculate the overall priority score for each combination. : ;in, , , These are the inverse indicators of scale. Risk index and context fit metrics The assigned weights; Sort all combinations from highest to lowest based on their overall priority score to form a priority queue.

6. The production process alarm rule evolution method based on runtime context awareness according to claim 1, characterized in that, In the nested optimization and verification loop, for each candidate combination, the control parameter set, the alternate path flag and the buffer strategy parameters are adjusted by the process scheduling optimizer to minimize the overall production loss of the remaining unit set in the simulation environment, and the loss is re-evaluated to see if it is below the dynamic tolerable loss threshold. The system configuration after the nested optimization verification loop is simulated in a digital twin environment to simulate virtual island disconnection. The comprehensive production loss index caused by the remaining unit set is recalculated and compared with the tolerable loss threshold dynamically updated based on the current context to determine whether the verification passes.

7. The production process alarm rule evolution method based on runtime context awareness according to claim 1, characterized in that, During the online operation verification phase, the minimum effective intervention combination and its corresponding optimization strategy are deployed as temporary alarm rules to monitor actual production indicators within a preset verification period. If the indicators meet expectations, the current context, unit combination, and optimization strategy are encapsulated as formal alarm rules and stored in the rule base. If the verification fails, the situation deviation analysis process is triggered.

8. The production process alarm rule evolution method based on runtime context awareness according to claim 7, characterized in that, The context bias analysis following verification failure includes: Compare the actual runtime context data sequence with the assumed context of the dynamic inference model; Identify key discrepancy factors as causes of model inaccuracy; The discrepancy factors are fed back to the dynamic inference model for updates, and the entire process from model building to online validation is retried.

9. The production process alarm rule evolution method based on runtime context awareness according to claim 1, characterized in that, It also includes recording the running context characteristics, unit combination status, optimization strategy parameters and verification period performance data corresponding to the rule when generating formal alarm rules, forming reusable rule knowledge entries to support subsequent rule matching and scenario adaptive invocation.

10. An early warning system for a production process alarm rule evolution method based on any one of claims 1-9, characterized in that, The system includes: The runtime context awareness and model building module is configured to: collect multi-dimensional runtime context data of the current production system; build and run a dynamic inference model based on the data to predict the probability and timing distribution of abnormal triggering of each device or process unit in the system within a future time window; and generate a first-round candidate set of vulnerable units based on the prediction results. The simulation evaluation and core unit screening module is configured to: perform virtual island disconnection simulation for each unit in the candidate set of vulnerable units; during the simulation, dynamically calculate the comprehensive production loss index caused by unit failure based on the current running context; and screen out core intervention units whose comprehensive production loss exceeds the threshold and require intervention based on the predefined tolerable loss threshold, forming a core intervention unit set. The combinatorial optimization and verification decision module is configured to: explore the combinatorial space of the core intervention unit set, enumerate all possible non-empty unit subsets, and prioritize them according to multi-dimensional factors such as combinatorial size, joint risk probability, and adaptability to the current context; execute nested optimization and verification loops for each candidate combination according to priority order, where: for the current candidate combination, the system control strategy of the remaining unit set is readjusted and optimized through the process scheduling optimizer; evaluate the comprehensive production loss of the optimized system in the face of the risk of the remaining units in the digital twin environment; iteratively execute this process until the minimum effective intervention combination and its corresponding optimization strategy that can make the comprehensive system loss meet the preset requirements are selected; The online verification and rule evolution module is configured to: deploy the minimum effective intervention combination and corresponding optimization strategies to the actual production system for online operation verification; if the verification is successful, encapsulate the current operating context features, unit combinations and optimization strategies into formal alarm rules and store them in the rule base; if the verification fails, analyze the differences between the actual operating context and the context used in the model building stage, and feed back the identified key difference factors to the operating context awareness and model building module to trigger the update of the dynamic inference model and the re-execution of subsequent modules until a formal alarm rule that has been successfully verified is generated.

Images