Intelligent manufacturing computing power collaborative regulation method and system based on dynamic performance evaluation

Abstract

The present application relates to the technical field of industrial edge computing and intelligent manufacturing task scheduling, and particularly relates to an intelligent manufacturing computing power collaborative regulation method and system based on dynamic performance evaluation. The edge node obtains production event sequences, control data update time and task attributes in a read-only manner; calculates edge side data receiving period deviation and input data age, identifies production state, and predicts available time window according to similar key event interval; calculates resource conflict degree and collaborative decay coefficient between tasks, and corrects candidate batch predicted execution time; calculates task performance value in combination with period deviation and priority, generates a scheduling decision and feeds back correction parameters accordingly; when the period deviation is abnormal, it is judged whether it is caused by edge load and a protection operation is performed. The present application provides an intelligent manufacturing computing power collaborative regulation method and system based on dynamic performance evaluation, which effectively solves the problems of non-real-time task window out-of-bound, resource contention, data expiration and deviation source confusion.

Description

Technical Field

[0001] This invention relates to the fields of industrial edge computing and intelligent manufacturing task scheduling technology, and in particular to an intelligent manufacturing computing power collaborative control method and system based on dynamic performance evaluation. Background Technology

[0002] In scenarios such as discrete manufacturing, automotive parts processing, electronic assembly, and assembly line inspection, edge computing nodes typically undertake tasks such as PLC status acquisition, online inspection data reception, alarm forwarding, equipment status analysis, quality traceability and organization, historical data compression, log archiving, and report statistics. When these tasks run on the same edge node, they share CPU, memory, disk I / O, network bandwidth, and accelerator resources.

[0003] Existing edge task scheduling methods mostly rely on CPU utilization, memory utilization, task priority, or fixed time slices for task initiation and delay processing. While this approach can meet basic needs when production cycles are stable and the number of tasks is small, it is prone to the following problems when a single node handles multiple tasks such as online detection and verification, status analysis, traceability and organization, historical data compression, and report statistics: First, the timing of non-real-time task initiation does not match critical events on the production floor. For example, initiating a disk write or network upload task just before the next detection trigger, processing begins, or fixture action is about to occur may lead to significant deviations in the data reception cycle of the edge-side acquisition service.

[0004] Second, resource contention during multi-task concurrency is difficult to predict. The estimated execution time of a single task cannot reflect the actual execution time when multiple tasks share CPU, disk I / O, or network bandwidth, and fixed-priority scheduling is prone to causing task window overflows.

[0005] Third, tasks with expired input data may still be initiated. Input data for tasks such as online detection and verification, and status analysis have collection times and allowed valid periods. If tasks are initiated solely based on queue order or resource availability, results that are no longer suitable for online judgment may be generated.

[0006] Fourth, the sources of control cycle deviations are easily confused. Data reception cycle deviations observed at the edge may be caused by increased edge load, or by equipment alarms, sensor malfunctions, network jitter, or tool change / replacement processes. Attributing all deviations directly to edge task load can easily lead to unnecessary task limitations; while completely ignoring the impact of edge load may cause non-real-time tasks to consume the resources required for control data acquisition, alarm forwarding, and online detection data reception.

[0007] Furthermore, industrial sites have high requirements for control safety. If edge scheduling schemes are understood as directly interfering with the internal logic of PLCs, motion controllers, or safety interlocking devices, their acceptability will be reduced. This invention uses a read-only approach to acquire production events and control-related data, without writing instructions to the internal logic of programmable controllers, motion controllers, or safety interlocking devices. It only schedules and controls non-real-time tasks on edge computing nodes, thus balancing edge computing power utilization and on-site control safety. Summary of the Invention

[0008] To achieve the above-mentioned objectives, this invention provides a method and system for collaborative control of computing power in intelligent manufacturing based on dynamic performance evaluation, which solves or at least alleviates the problems in the prior art such as the inability to predict production windows during edge task scheduling, multi-task resource contention, expired input data, and confusion of the sources of cycle deviation.

[0009] To achieve the above objectives, the present invention provides the following technical solution: a method for collaborative control of computing power in intelligent manufacturing based on dynamic performance evaluation, comprising: The system acquires production event sequences and control data update times in read-only mode, calculates edge-side data reception cycle deviation, and acquires task attributes, including input data timestamps, estimated execution time, resource requirements, and priority. The current production status is identified based on the production event sequence; Generate a time window available for non-real-time tasks; Calculate the resource conflict degree between tasks and construct a collaborative attenuation coefficient to correct the estimated execution time of candidate tasks; The validity of the data is assessed based on the age of the input data, and the task performance value is calculated by combining the periodic deviation and priority. Based on the time window, collaborative attenuation coefficient, efficiency value, and current production status, a scheduling decision is generated and the task is executed. When the cycle deviation exceeds the limit, determine whether the edge load is related to the deviation change. If they are related, perform protection operation. The method does not write instructions to the programmable controller or safety interlock device.

[0010] To further realize the present invention, the following technical solutions may be preferred: Preferably, identifying the current production state based on the production event sequence includes: Identify normal production status, tool change or form change status, pre-restore processing status, and alarm status; During the tool change or tool change state, the remaining time is estimated based on historical duration to concentrate on executing backlogged tasks; In the state of imminent resumption of processing, a snapshot is saved and suspended for the interruptible task.

[0011] Preferably, the generation of a non-real-time task availability window includes: Estimate production cycle time based on historical intervals of similar key events; Determine the dynamic safety margin by combining the cycle time fluctuation; The available time window is determined based on the predicted time of the next critical event, the current time, and the dynamic safety margin.

[0012] Preferably, the resource conflict degree between the computational tasks, and the construction of a collaborative attenuation coefficient to correct the estimated execution time of the candidate tasks, include: Calculate the resource conflict degree between each pair of tasks using the resource requirement vector of each task; The synergy attenuation coefficient is determined based on the average resource conflict degree of all task pairs; The total estimated execution time is obtained by multiplying the sum of the estimated execution times of each task in the candidate task set by the cooperative attenuation coefficient.

[0013] Preferably, the resource conflict degree is calculated as follows: Calculate the cosine similarity between the resource requirement vectors of the two tasks; The resource conflict degree is obtained by multiplying the cosine similarity by the minimum and maximum resource occupancy of the two tasks in each resource dimension.

[0014] Preferably, the step of evaluating data validity based on the age of the input data and calculating task performance value in combination with periodic deviation and priority includes: When the data age exceeds the allowed validity period, the data validity will be set to zero; When the data age has not exceeded the allowed validity period, the data validity is calculated in an exponential decay form. The control cycle stability is determined based on the cycle deviation. The priority normalized value and the control cycle stability are weighted and summed, divided by the operating load evaluation value, and then multiplied by the data validity to obtain the task performance value.

[0015] Preferably, the step of determining whether the edge load is correlated with the deviation change, and performing a protection operation if correlated, includes: When the edge integrated load change exceeds the limit, the cycle deviation change exceeds the limit, the load increase precedes the deviation increase, there are currently no alarms or tool change events, and the length of the critical data queue increases, it is determined to be a correlation. Perform protective operations, including stopping the startup of new tasks, suspending interruptible tasks and saving snapshots, and preserving resources related to control data acquisition and online monitoring.

[0016] Preferably, the task to be performed includes: The actual execution time of the task is obtained, and the estimated execution time is adjusted by giving more weight to recent execution records than to older execution records.

[0017] A smart manufacturing computing power collaborative control system based on dynamic performance evaluation includes: The data acquisition module is used to acquire production event sequences and control data update time in read-only mode, and to calculate the data reception cycle deviation at the edge. The task attribute module is used to obtain the attributes of the task to be executed, including the input data timestamp, estimated execution time, resource requirements, and priority. The working condition identification module is used to identify the current production status based on the production event sequence; The window calculation module is used to generate available time windows for non-real-time tasks; The collaborative evaluation module is used to calculate the resource conflict degree between tasks and construct a collaborative attenuation coefficient to correct the expected execution time of candidate tasks. The performance evaluation module is used to assess the validity of the input data based on age, and to calculate the task performance value by combining periodic deviation and priority. The scheduling decision module is used to generate scheduling decisions based on the time window, the collaborative attenuation coefficient, the efficiency value, and the current production status. An execution feedback module is used to execute tasks and obtain the actual execution time in order to correct the estimated execution time. The exception handling module is used to determine whether the edge load and the deviation change are related when the cycle deviation exceeds the limit. If they are related, protection operations are performed. The system does not write instructions to the programmable controller or safety interlock device.

[0018] An electronic device includes a processor, a memory, and a communication interface, wherein the communication interface is used to connect to production field equipment in a read-only manner, the memory stores a computer program, and the processor executes the computer program to implement the method described above.

[0019] The beneficial effects of this invention are: This invention predicts the available time window for non-real-time tasks by using production event sequences, ensuring that the start time of non-real-time tasks matches the next critical production event and reducing the risk of window overflow. It also calculates resource conflict degree and collaborative decay coefficient using resource demand vectors, making the batch execution time of concurrent multi-task operations closer to actual resource usage.

[0020] Meanwhile, this invention calculates data validity by using the age of input data and an exponential decay method, and uses data validity as a multiplicative gating term for task performance value to prevent tasks with expired input data from being mistakenly started. The edge load impact status is jointly determined by the data reception cycle deviation, edge load changes, event status, and critical data queue length observed at the edge, reducing the risk of misjudging device alarms or sensor anomalies as edge load problems. Attached Figure Description

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

[0022] Figure 2 This is a flowchart of the method of the present invention.

[0023] Figure 3 This is a schematic diagram of the multi-task collaborative evaluation and task performance assessment of the present invention.

[0024] Figure 4 This is a schematic diagram of the exception handling and execution feedback closed loop of the present invention. Detailed Implementation

[0025] In the description of this invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Example 1

[0027] See Figure 1 This embodiment provides an intelligent manufacturing computing power collaborative control system based on dynamic performance evaluation. The system is deployed in an edge computing node, an industrial computer, or a server communicatively connected to the edge computing node. The system receives production site events and control-related data in read-only mode, without writing instructions to the internal logic of the PLC, motion controller, or safety interlock device.

[0028] The system includes a data acquisition module, a task attribute module, a working condition identification module, a window calculation module, a collaborative evaluation module, a performance evaluation module, a scheduling decision module, an execution feedback module, and an exception handling module.

[0029] The data acquisition module connects to a PLC, industrial gateway, industrial camera, sensors, robot controller, and online inspection equipment. It receives events including workpiece entry into the workstation, fixture closure, machining start, machining end, inspection trigger, inspection completion, tool change, fixture switching, alarm generation, and alarm deactivation. Each event includes at least the event type, occurrence time, and source identifier. All events are written to an event buffer for subsequent operation condition identification and time window calculation.

[0030] The data acquisition module also records the arrival time of control-related data at the edge acquisition service. This control-related data includes PLC status data, online detection results, alarm forwarding data, and sensor status data. The actual reception cycle at the edge is calculated using the following formula:

[0031] in, The k-th time when the control-related data arrives at the edge acquisition service. This represents the actual reception period at the edge.

[0032] The data reception period deviation observed at the edge side is calculated using the following formula:

[0033] in, This refers to the data reception period deviation observed on the edge side. For reference update cycle.

[0034] The task attribute module manages the attributes of tasks to be executed. These attributes include task type, input data timestamp, allowed data validity period, estimated execution time, task priority, resource requirement vector, interruptibility flag, snapshot save time, and failure handling strategy. The resource requirement vector can be represented as:

[0035] in, , , , Representing tasks Normalized requirements for CPU, memory, disk I / O, and network resources.

[0036] The operating condition identification module identifies the current production status based on events in the event buffer. When a tool change start or changeover start event is detected but no corresponding end event is detected, it is marked as a tool change or changeover state. When a tool change end, changeover end, alarm reset, or manual intervention end event is detected, it is marked as a pre-recovery processing state. When an alarm is detected but no alarm clearing event is detected, it is marked as an alarm state. If none of the above states are identified and the production cycle continues, it is marked as a normal production state.

[0037] The window calculation module estimates the available time window for non-real-time tasks under normal production conditions. It selects similar key events, such as detection triggering, processing start, or workpiece leaving the workstation, and calculates the average of the intervals between the most recent events as the cycle time estimate. The dynamic safety margin is determined jointly by the basic safety margin and the cycle time fluctuation.

[0038] in, For dynamic safety margin, Basic safety margin, This refers to the beat fluctuation. This is the margin adjustment coefficient.

[0039] The predicted time for the next critical event is:

[0040] The available time window for non-real-time tasks is:

[0041] in, The most recent time that a similar key event occurred, This is the estimated beat rate. For the current moment, This is the available time window for non-real-time tasks.

[0042] The collaborative evaluation module calculates the resource conflict degree between tasks based on the resource requirement vectors of the candidate tasks. For any two tasks in the candidate set... and Resource conflict level is calculated using the following formula:

[0043] in, Resource conflict level between tasks. From a resource perspective.

[0044] The mean resource conflict degree of candidate set J is:

[0045] in, For the number of tasks.

[0046] The synergistic attenuation coefficient is calculated using the following formula:

[0047] in, For the coordinated attenuation coefficient, This is the attenuation intensity coefficient.

[0048] The total estimated execution time after the candidate set is adjusted is:

[0049] in, For the task The estimated execution time, The total estimated execution time after adjusting for the candidate set.

[0050] The performance evaluation module calculates task performance values ​​based on data age, control cycle stability, task priority, and runtime load. Data age is:

[0051] Data validity value:

[0052] in, For data validity values, The attenuation coefficient is... To allow for a valid time period.

[0053] The stability of the control period is:

[0054] in, Allowable deviation.

[0055] Task runtime load evaluation value It is obtained by weighting the impact of queuing, estimated time, communication, and resource consumption, and a basic term is set to avoid the denominator being zero.

[0056] Task performance value:

[0057] in, This represents the task performance value. This is a normalized value for task priority. and For weights.

[0058] For the candidate task set Batch performance values ​​can be calculated:

[0059] The scheduling decision module generates scheduling decisions based on the production status.

[0060] Under normal production conditions, if the sum of the revised total estimated execution time, snapshot save time, and security retention time of the candidate task set does not exceed the available time window for non-real-time tasks, and the batch performance value meets the threshold, then a start instruction is generated. If the complete task cannot meet the window requirements but the task allows for fragmented execution, then a fragmentation instruction is generated. If the task performance value is low or the window is insufficient, then a delay instruction is generated.

[0061] During tool change or tool replacement, the scheduling decision module prioritizes selecting non-real-time tasks that can be processed in batches from the delay queue, such as historical image compression, log archiving, report generation, and quality traceability processing tasks. If the remaining tool change or tool replacement time is expected to meet the task execution requirements, a centralized execution instruction is generated.

[0062] In the pre-resumption processing state, the scheduling decision module generates snapshot saving and suspension instructions for interruptible tasks. The snapshot content includes the location of processed data, fragment number, intermediate variable status, and output checksum value.

[0063] In alarm status, the scheduling decision module suspends the startup of ordinary non-real-time tasks, while retaining resources related to alarm data interception, fault diagnosis, and alarm forwarding.

[0064] The execution feedback module records the actual start time, end time, actual execution time, whether the time window has been exceeded, snapshot saving time, and batch actual execution time of the task. The task attribute module updates the estimated execution time based on the actual execution time.

[0065] in, This refers to the actual execution time. The originally scheduled execution time, This is the updated estimated execution time. To update the weights.

[0066] Reference Figure 4 The anomaly handling module continuously monitors the overall edge load and the data reception cycle deviation observed at the edge. Greater than At that time, the module calculates the overall load change at the edge. and period deviation change If the load change exceeds the load threshold, the periodic deviation change exceeds the deviation threshold, the load increase occurs before the periodic deviation increase, there are currently no alarms or tool change events, and the critical data queue length increases, then the system is considered to be in an edge load impact state. In this case, protection operations are performed, including stopping the startup of new non-real-time tasks, suspending interruptible tasks and saving snapshots, transferring low-priority tasks to a delayed queue, and preserving resources related to control data acquisition, alarm forwarding, and online detection.

[0067] When an alarm event occurs, even if the edge load and cycle deviation increase simultaneously, the anomaly handling module does not directly determine that the state is affected by the edge load. Instead, it records the alarm association information and reports the anomaly to the site. Example 2

[0068] See Figure 2 This application provides a method for collaborative control of computing power in intelligent manufacturing based on dynamic performance evaluation. This method is executed collaboratively by modules deployed on edge computing nodes or industrial computers.

[0069] Step S1: Obtain the production site event sequence and control-related data update time.

[0070] Edge computing nodes receive production site events and control-related data in read-only mode. Read-only access methods include industrial communication interfaces, industrial gateway mirroring channels, message queues, and edge acquisition services. The system only reads the status and events of field devices and does not write instructions to the internal logic of PLCs, motion controllers, or safety interlock devices.

[0071] An event must include at least the event type, occurrence time, and source identifier, and be written to the event buffer in chronological order. When relevant control data arrives at the edge acquisition service, its arrival time is recorded, and the deviation between the actual reception period at the edge and the data reception period observed at the edge is calculated.

[0072] Step S2: Obtain the attributes of the task to be executed

[0073] The task attribute module reads the attributes of the tasks to be executed in the task queue, including task type, input data timestamp, allowed valid time, estimated execution time, resource requirement vector, task priority, interruptibility flag, snapshot save time and failure handling strategy.

[0074] The resource requirement vector is obtained by normalizing the expected utilization of CPU, memory, disk I / O, and network resources. The estimated execution time is adjusted based on the task's historical execution records, ensuring that scheduling decisions reflect the actual operational situation.

[0075] Step S3: Identify the current production status

[0076] The condition recognition module reads the event buffer. If a tool change start or type change start event is detected but no corresponding end event is detected, the module enters the tool change state or type change state. If a tool change end, type change end, manual intervention end, or alarm reset event is detected, the module enters the pre-resumption machining state. If an alarm generation event is detected but no alarm clearing event is detected, the module enters the alarm state. Other continuous production scenarios are marked as normal production states.

[0077] Step S4: Generate a time window available for non-real-time tasks

[0078] Under normal production conditions, the window calculation module selects similar key events, such as processing start, detection trigger, or workpiece leaving the workstation. It estimates the production cycle time based on the time intervals of several recent similar key events and determines the dynamic safety margin by combining this with cycle time fluctuations.

[0079] The predicted time of the next critical event is obtained from the occurrence time and tick estimate of the most recent critical event of the same type. After deducting the dynamic safety margin between the current time and the predicted time of the next critical event, the available time window for non-real-time tasks is obtained.

[0080] If the tick fluctuation increases significantly, the window calculation module can close new non-real-time task startup windows and only allow already running tasks to exit or be suspended according to the snapshot strategy.

[0081] Step S5: Calculate resource conflict degree and collaborative attenuation coefficient

[0082] The collaborative evaluation module selects tasks from the task queue that have non-zero data validity, meet resource requirements, and have not been excluded by alarm status, forming a candidate set.

[0083] For each pair of tasks in the candidate set, the cosine similarity of the resource demand vectors is first calculated, and then the resource conflict degree is obtained by combining the overlap strength of each resource dimension. The resource conflict degrees of all task pairs are averaged, multiplied by the attenuation strength coefficient, and incremented by one to obtain the collaborative attenuation coefficient. The sum of the expected execution times of all tasks in the candidate set is multiplied by the collaborative attenuation coefficient to obtain the corrected total expected execution time.

[0084] Step S6: Calculate task performance value

[0085] The performance evaluation module calculates the age of the input data. If the age of the input data exceeds the allowed validity period, the data validity value is set to zero, and the task does not proceed to the normal startup process. If the age does not exceed the allowed validity period, the data validity value is calculated using an exponential decay method.

[0086] The performance evaluation module then calculates the control cycle stability based on the data reception cycle deviation observed at the edge, and calculates the task performance value by combining the task priority and runtime load evaluation value. The data validity value is used as a multiplication gating term in the calculation, so when the input data expires, the task performance value directly becomes zero.

[0087] Step S7: Generate scheduling decision

[0088] The scheduling decision module generates scheduling decisions based on production status, available time windows for non-real-time tasks, collaborative attenuation coefficient, corrected total estimated execution time, and task performance value.

[0089] Under normal production conditions, if the candidate set meets the window and performance requirements, a start instruction is generated; if the task allows sharding and the complete execution time exceeds the window, a sharding instruction is generated; if the window is insufficient or the performance value is insufficient, a delay instruction is generated.

[0090] During tool change or tool change, the system selects suitable tasks for centralized processing from the delay queue and generates batch start instructions.

[0091] In the pre-resumption state, the system generates snapshot save and suspend instructions for interruptible tasks.

[0092] In alarm mode, the system suspends normal non-real-time tasks, but retains resources related to alarm processing, fault diagnosis, and data acquisition.

[0093] Step S8: Execute the task and adjust the estimated execution time.

[0094] The task execution module starts, segments, suspends, or resumes tasks according to scheduling decisions. The execution feedback module records the actual execution time, snapshot save time, task completion status, and window out-of-bounds status. The task attribute module adjusts the estimated execution time based on the execution feedback. (See reference...) Figure 3 The collaborative evaluation module can adjust the collaborative attenuation intensity coefficient based on the actual execution time of the batch.

[0095] Step S9: Edge load impact status identification and protection operation

[0096] When the data reception cycle deviation observed at the edge exceeds the allowable range, the anomaly handling module calculates the overall edge load change and the cycle deviation change. If the load increase precedes the cycle deviation increase, and there are currently no equipment alarms or tool change / type change events, and the critical data queue length increases simultaneously, then it is determined to be an edge load-affected state.

[0097] Once the system is determined to be in an edge load impact state, it stops starting new non-real-time tasks, suspends interruptible tasks and saves snapshots, transfers low-priority tasks to a delay queue, and retains resources related to control data acquisition, alarm forwarding, and online detection.

[0098] If an alarm event or field equipment malfunction occurs, the system does not directly attribute the period deviation to the edge load, but instead records the field malfunction and reports it. Example 3 Verification

[0099] Test equipment and scenarios

[0100] The verification environment was set up on an automotive parts manufacturing production line. The production line included CNC machining centers, industrial robots, online vision inspection stations, and a PLC control system. The system under test ran on an industrial computer and was connected to the PLC, industrial cameras, and robot controller via industrial Ethernet, using OPC UA and Modbus TCP protocols for communication.

[0101] The non-real-time tasks running on the edge nodes include: online inspection result verification task A, equipment status analysis task B, quality traceability data processing task C, historical inspection image compression task D, production log archiving task E, and daily report generation task F. Among them, tasks A and B have higher priority, while tasks C through F have lower priority.

[0102] During the test, the system ran in a loop with a scheduling cycle of approximately 100ms. Each scheduling cycle reads the event cache, refreshes the task queue, calculates the available time window, evaluates task performance, and generates a new scheduling decision.

[0103] Typical operating condition 1: Normal and stable production

[0104] Under normal and stable production conditions, the production line cycle time is approximately 3 seconds, and the predicted time for the next detection trigger event is approximately 42.6 seconds from now. The system adds tasks B, C, and D to the candidate set. The estimated execution times for the three tasks are approximately 1.2s, 8.5s, and 18.0s, respectively.

[0105] The resource conflict levels between task pairs are recorded as follows: - The resource conflict degree between Task B and Task C is approximately 0.23; - The resource conflict between Task B and Task D is approximately 0.15; The resource conflict between Task C and Task D is approximately 0.45.

[0106] Based on the preceding calculation rules, the average resource conflict degree of the candidate set is:

[0107] Attenuation intensity coefficient When the value is 1.0, the cooperative attenuation coefficient is:

[0108] The revised total estimated execution time is:

[0109] Considering the approximately 1.5s snapshot save time and approximately 1.49s security retention time, the total time required is approximately:

[0110] If the available time window is less than 42.6 seconds, the scheduling decision module generates a start command. After task execution, no record of exceeding the available time window was found. In contrast, without considering co-factor decay, the system underestimates batch execution time and is prone to misjudging the executable window when resource contention is significant.

[0111] Typical working condition 2: Concentrated task execution during tool change.

[0112] Once the tool change start event occurs, the working condition identification module switches the production status to tool change status. Since machining is paused at this time, the system selects tasks such as historical detection image compression, log archiving, and report generation from the delay queue and places them into the centralized execution queue.

[0113] During tool change, the system prioritizes tasks with shorter estimated execution times, data validity that still meets requirements, and low resource conflict. If the tool change completion event occurs prematurely, the scheduling decision module immediately issues snapshot saving and suspension commands to interruptible tasks. When a task resumes, it continues execution from the snapshot record position, without needing to start from the beginning.

[0114] Typical Working Condition 3: Suspending the task in the pre-processing state.

[0115] When a tool change ends, manual intervention ends, or an alarm reset event is detected, the system determines that the system is in a state of imminent resumption of machining. At this time, machining actions, detection triggers, and fixture actions may resume quickly. The scheduling decision module stops starting new ordinary non-real-time tasks and saves a snapshot of the currently executing interruptible tasks.

[0116] The snapshot includes the processed file block number, intermediate state, output file checksum, and task recovery location. The task execution module resumes execution based on the snapshot the next time the tool change or low-load window is entered.

[0117] Typical operating condition 4: Determining the correlation between edge load changes and period deviation changes

[0118] Testers injected a set of high disk I / O tasks into the edge nodes during non-critical production periods to simulate abnormal storage bandwidth consumption by non-real-time tasks. Before the injection, the data reception cycle deviation observed at the edge was small, and the overall edge load was at a normal level. After the injection, the overall edge load increased significantly, followed by a larger data reception cycle deviation observed at the edge, and a backlog of PLC data arrival queues. Currently, there are no alarm events or tool change / type change events.

[0119] The anomaly handling module detected a load change preceding a period deviation change, along with an increase in the length of the critical data queue, classifying it as an edge load impact state. The system immediately stopped starting new non-real-time tasks, saved snapshots of interruptible tasks and suspended them, moved low-priority tasks to a delay queue, and preserved resources related to control data acquisition and online monitoring. After the protection operation was executed, the data reception cycle gradually returned to stability.

[0120] In another set of control tests, poor sensor contact caused unstable control-related data, but the overall edge load did not increase significantly, and alarm events occurred on-site. The anomaly handling module did not determine that the edge load was affecting the state; instead, it reported a sensor anomaly warning and retained the original edge task scheduling strategy.

[0121] Increased edge I / O load Significantly increased Significantly increased Edge load affects status Triggering protection, suspending interruptible tasks Sensor malfunction Significantly increased No significant changes Not determined to be an edge load-affected state Report on-site abnormalities

[0122] Typical Working Condition 5: Data Validity Index Decay and Multiplicative Gating

[0123] To verify the data validity mechanism, online detection result verification task A was selected. This task has high requirements for the freshness of the input images, and will not enter the normal startup process after the allowed valid time has expired.

[0124] When the input image is recently acquired, the data validity value is high, and the task performance value meets the startup conditions. The system starts the task when the window allows. As the age of the input data increases, the data validity value gradually decreases, and the task ranking decreases accordingly. When the age of the input data exceeds the allowed valid time, the data validity value is set to zero, the task performance value is set to zero simultaneously, and the scheduling decision module no longer starts the task but instead transfers it to the re-acquisition process.

[0125] This approach differs from simple binary validity checks. For tasks still within their validity period but with different data ages, the system can assign different scheduling priorities, allowing tasks closer to expiration to enter the execution process more cautiously.

[0126] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for collaborative control of computing power in intelligent manufacturing based on dynamic performance evaluation, characterized in that, include: The production event sequence and control data update time are obtained in read-only mode, and the data reception cycle deviation at the edge is calculated. Obtain task attributes, including input data timestamp, estimated execution time, resource requirements, and priority; The current production status is identified based on the production event sequence; Generate a time window available for non-real-time tasks; Calculate the resource conflict degree between tasks and construct a collaborative attenuation coefficient to correct the estimated execution time of candidate tasks; The validity of the data is assessed based on the age of the input data, and the task performance value is calculated by combining the periodic deviation and priority. Based on the time window, collaborative attenuation coefficient, efficiency value, and current production status, a scheduling decision is generated and the task is executed. When the cycle deviation exceeds the limit, determine whether the edge load is related to the deviation change. If they are related, perform protection operation.

2. The method according to claim 1, characterized in that, The step of identifying the current production status based on the production event sequence includes: Identify normal production status, tool change or form change status, pre-restore processing status, and alarm status; During the tool change or tool change state, the remaining time is estimated based on historical duration to concentrate on executing backlogged tasks; In the state of imminent resumption of processing, a snapshot is saved and suspended for the interruptible task.

3. The method according to claim 1, characterized in that, The generation of a non-real-time task availability window includes: Estimate production cycle time based on historical intervals of similar key events; Determine the dynamic safety margin by combining the cycle time fluctuation; The available time window is determined based on the predicted time of the next critical event, the current time, and the dynamic safety margin.

4. The method according to claim 1, characterized in that, The resource conflict degree between the computational tasks is used to construct a collaborative attenuation coefficient to correct the estimated execution time of candidate tasks, including: Calculate the resource conflict degree between each pair of tasks using the resource requirement vector of each task; The synergy attenuation coefficient is determined based on the average resource conflict degree of all task pairs; The total estimated execution time is obtained by multiplying the sum of the estimated execution times of each task in the candidate task set by the cooperative attenuation coefficient.

5. The method according to claim 4, characterized in that, The resource conflict level is calculated as follows: Calculate the cosine similarity between the resource requirement vectors of the two tasks; The resource conflict degree is obtained by multiplying the cosine similarity by the minimum and maximum resource occupancy of the two tasks in each resource dimension.

6. The method according to claim 1, characterized in that, The step of evaluating data validity based on the age of the input data and calculating task performance value by combining periodic deviation and priority includes: When the data age exceeds the allowed validity period, the data validity will be set to zero; When the data age has not exceeded the allowed validity period, the data validity is calculated in an exponential decay form. The control cycle stability is determined based on the cycle deviation. The priority normalized value and the control cycle stability are weighted and summed, divided by the operating load evaluation value, and then multiplied by the data validity to obtain the task performance value.

7. The method according to claim 1, characterized in that, The determination of whether the edge load is correlated with the deviation change, and if so, the execution of protection operations, includes: When the edge integrated load change exceeds the limit, the cycle deviation change exceeds the limit, the load increase precedes the deviation increase, there are currently no alarms or tool change events, and the length of the critical data queue increases, it is determined to be a correlation. Perform protective operations, including stopping the startup of new tasks, suspending interruptible tasks and saving snapshots, and preserving resources related to control data acquisition and online monitoring.

8. The method according to claim 1, characterized in that, The tasks to be performed include: The actual execution time of the task is obtained, and the estimated execution time is adjusted by giving more weight to recent execution records than to older execution records.

9. A smart manufacturing computing power collaborative control system based on dynamic performance evaluation, characterized in that, include: The data acquisition module is used to acquire production event sequences and control data update time in read-only mode, and to calculate the data reception cycle deviation at the edge. The task attribute module is used to obtain the attributes of the task to be executed, including the input data timestamp, estimated execution time, resource requirements, and priority. The working condition identification module is used to identify the current production status based on the production event sequence; The window calculation module is used to generate available time windows for non-real-time tasks; The collaborative evaluation module is used to calculate the resource conflict degree between tasks and construct a collaborative attenuation coefficient to correct the expected execution time of candidate tasks. The performance evaluation module is used to assess the validity of the input data based on age, and to calculate the task performance value by combining periodic deviation and priority. The scheduling decision module is used to generate scheduling decisions based on the time window, the collaborative attenuation coefficient, the efficiency value, and the current production status. An execution feedback module is used to execute tasks and obtain the actual execution time in order to correct the estimated execution time. The exception handling module is used to determine whether the edge load and the deviation change are related when the cycle deviation exceeds the limit. If they are related, protection operations are performed.

10. An electronic device, characterized in that, The device includes a processor, a memory, and a communication interface, wherein the communication interface is used to connect to production field equipment in a read-only manner, the memory stores a computer program, and the processor executes the computer program to implement the method according to any one of claims 1 to 8.

Images