Manufacturing intelligent scheduling and dispatching method based on AI edge computing terminal

Abstract

The application relates to the technical field of intelligent manufacturing and industrial automation, in particular to a manufacturing intelligent production scheduling and dispatching method based on an AI edge computing terminal, which comprises the following steps: generating a theoretical production scheduling reference time axis based on a work order BOM and a standard process route SOP issued based on ERP / MES and equipment OEE time sequence slices, workstation flow transfer beats and edge buffer pool inventory water levels collected in real time by the edge computing terminal; performing parameterized disturbance simulation on the theoretical production scheduling reference time axis to generate a theoretical residual time sequence set; generating a real residual time sequence according to actual workstation completion times; determining a disturbance type by using dynamic time warping distance calculation and similarity calculation based on a process route topology; generating a local work order translation dispatching instruction according to the disturbance type, and performing dynamic calibration to optimize the scheduling accuracy of subsequent production plans.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing and industrial automation technology, specifically to a manufacturing intelligent scheduling and dispatching method based on AI edge computing terminals. Background Technology

[0002] Intelligent production scheduling and operation technology based on edge computing terminals is applied in discrete manufacturing workshops. Its scheduling results and operation response directly affect the efficiency of work order flow, equipment utilization level and the continuous stability of work-in-process between processes. Therefore, timely, accurate and executable dynamic adjustment of production plans is an important foundation for ensuring the stable operation of the manufacturing process.

[0003] Existing production scheduling and dispatching methods have many problems. For example, they rely heavily on static plans, historical average cycle times, or single-point timeout alarms issued by Enterprise Resource Planning (ERP) or Manufacturing Execution System (MES) for judgment. They are difficult to accurately characterize the deviation propagation relationship between the plan and the actual execution under disturbances such as material fluctuations, equipment degradation, and changes in personnel efficiency. Especially when the process route is long, the processes are closely related, and the inventory level of the buffer pool between processes is constantly changing, it is easy to have inaccurate disturbance identification, frequent global rescheduling, excessive local adjustment range, and misscheduling. They cannot balance scheduling agility and production plan stability. Summary of the Invention

[0004] The purpose of this invention is to provide a manufacturing intelligent scheduling and dispatching method based on AI edge computing terminals. It aims to improve the existing technology, which relies on static planning or single-point alarms, making it difficult to accurately characterize the propagation relationship of deviations under disturbances such as changes in materials, equipment or personnel. This can easily lead to inaccurate disturbance identification, frequent global rescheduling, and misscheduling, thus failing to balance scheduling agility and production plan stability.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a manufacturing intelligent scheduling and dispatching method based on AI edge computing power terminals, including: generating a theoretical scheduling benchmark time axis containing the theoretical completion time sequence of each process based on the work order BOM and standard process route SOP issued by ERP / MES, as well as the equipment OEE time sequence slice, workstation turnover takt time and edge buffer pool inventory level collected in real time by the edge computing power terminal.

[0006] Based on the preset set of production disturbance parameters and their disturbance pattern mapping relationship, parameterized disturbance simulation is performed on the theoretical production scheduling reference time axis to generate a theoretical residual time series set; The actual completion time sequence is determined based on the actual workstation completion time collected by the edge computing terminal, and the difference between the actual completion time sequence and the theoretical completion time sequence in the theoretical production scheduling benchmark time axis is calculated to generate the actual residual time sequence. Dynamic time warping distance calculation and process route topology-based similarity calculation are used to compare the actual residual time series with the theoretical residual time series set. Combined with a preset judgment threshold, the disturbance type judgment result is determined, and the affected process and affected time window are located. Based on the disturbance type judgment result, the affected process, the affected time window, and the edge buffer pool inventory level, a local work order translation scheduling instruction is generated that acts on the work order subset associated with the affected process. The local work order shifting and scheduling instruction is executed, and the set of production disturbance parameters and the preset judgment threshold are dynamically calibrated according to the cycle recovery index and inventory stability index within the evaluation period.

[0007] Preferably, generating the theoretical production scheduling benchmark timeline includes: determining the material constraints of the work order based on the work order BOM; determining the sequence of operations and standard operating time based on the standard operating procedure (SOP); calculating the theoretical start time and theoretical completion time of each operation under material constraints and sequence of operations constraints based on the standard operating time, the theoretical capacity limit of the equipment, and the work order requirements; and constructing the theoretical production scheduling benchmark timeline according to the theoretical start time and the theoretical completion time, as the time benchmark data for production planning management.

[0008] Preferably, the set of production disturbance parameters includes an equipment attenuation factor, a material supply fluctuation factor, and a personnel efficiency fluctuation factor; wherein, the equipment attenuation factor is used to characterize the exponential attenuation pattern of continuously weakening processing capacity over time; the material supply fluctuation factor is used to characterize the step pattern of a sudden drop in inventory at a predetermined time; and the personnel efficiency fluctuation factor is used to characterize the rhythm fluctuation pattern of changes in shifts or personnel status cycles; the parameterized disturbance simulation specifically involves: calling the parameterized simulation engine built into the edge computing terminal to parameterize and superimpose the equipment attenuation factor, the material supply fluctuation factor, and the personnel efficiency fluctuation factor to generate corresponding theoretical residual time series, which are used to establish the mapping relationship between disturbance type and plan deviation.

[0009] Preferably, the generation of the actual residual time series includes: determining the actual completion time series based on the actual workstation completion time collected by the edge computing terminal; performing a difference calculation between the actual completion time series and the theoretical completion time series in the theoretical production scheduling benchmark time axis to generate the actual residual time series; wherein, each theoretical residual time series in the theoretical residual time series set is a difference result between the simulated completion time series and the theoretical completion time series.

[0010] Preferably, when comparing the actual residual time series with the theoretical residual time series set, the process specifically includes: calculating the time pattern score and the topology propagation score respectively; weighting the time pattern score and the topology propagation score by combining preset time pattern weights and topology propagation weights to determine the normalized target matching score; when the target matching score is higher than or equal to a preset high threshold, outputting the disturbance type determination result corresponding to the theoretical residual time series that makes the target matching score meet the condition, and triggering the local work order shift scheduling instruction; when the target matching score is lower than or equal to a preset low threshold, outputting the random fluctuation determination result, and suppressing the local work order shift scheduling instruction; when the target matching score is between the high threshold and the low threshold, outputting the determination result to be confirmed, and maintaining the current production scheduling plan.

[0011] Preferably, the generation of the partial work order translation scheduling instruction includes: locating the affected process and the affected time window based on the disturbance type determination result; determining a set of work orders that can be translated based on the edge buffer pool inventory level and satisfying the material availability constraint, process sequence constraint, and buffer pool inventory level safety constraint; within the affected time window, and only for the subset of work orders associated with the affected process, performing sequence adjustment and start / complete time reallocation on the set of work orders that can be translated, and generating the partial work order translation scheduling instruction for partial updating of the production plan.

[0012] Preferably, the dynamic calibration includes: calculating a cycle time recovery index, an inventory stability index, and a misjudgment suppression index within the evaluation period; performing sensitivity analysis on the disturbance parameters in the production disturbance parameter set based on the cycle time recovery index to determine the set of disturbance parameters to be calibrated; updating the judgment threshold based on the inventory stability index and the misjudgment suppression index; and writing the updated set of disturbance parameters to be calibrated and the judgment threshold into the edge computing terminal to complete closed-loop calibration, thereby improving the adaptability of the production plan to the actual production environment.

[0013] Preferably, when the parametric simulation engine generates the theoretical residual time series set, it further includes: slicing the theoretical production scheduling baseline time axis according to a preset time window; injecting single disturbance factors and composite disturbance factors for each time window; attaching a disturbance type identifier to each injection result to form a correspondence between disturbance type and residual form; and outputting the theoretical residual time series set according to the correspondence, which is used as a reference template library for subsequent disturbance type matching.

[0014] Preferably, the equipment OEE time sequence slice includes an availability subsequence, a performance subsequence, and a quality subsequence; the workstation turnover cycle is a sequence of process-level completion time intervals; the edge buffer pool inventory level is a sequence of work-in-process quantities between processes; and the disturbance type determination result includes at least equipment attenuation disturbance, material supply disturbance, personnel efficiency disturbance, and random fluctuation.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention constructs a theoretical production scheduling benchmark timeline based on work order BOM, standard process route SOP, theoretical equipment capacity limit, and real-time data collection of equipment OEE time sequence slices, workstation turnover rate, and edge buffer pool inventory level. It can form a production plan time benchmark that excludes historical data deviations under material constraints and process sequence constraints. Combined with parameterized perturbation simulation of equipment degradation, material supply fluctuations, and personnel efficiency fluctuations, and generating a theoretical residual time sequence set, it realizes an interpretable expression of the propagation pattern of planning deviations. 2. This invention obtains the actual residual time series by differencing the actual completion time series with the theoretical completion time series, and performs a dual comparison by combining dynamic time warping distance calculation and similarity calculation based on process route topology, thereby improving the accuracy of judging equipment attenuation disturbances, material supply disturbances, personnel efficiency disturbances and random fluctuations. 3. This invention generates local work order translation scheduling instructions only for affected processes, affected time windows, and related work order subsets, avoiding the problems of frequent global rescheduling and excessively large local adjustment ranges. This ensures the stability of the production rhythm of unaffected sections. More importantly, it dynamically calibrates the disturbance parameter set and judgment threshold by combining cycle time recovery indicators, inventory stability indicators, and misjudgment suppression indicators, enabling edge terminals to continuously adapt to changes in the actual state of the workshop, thereby balancing scheduling agility, execution feasibility, and production plan stability. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be conventionally introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Figure 1 This is a flowchart of the intelligent production scheduling and production planning method for manufacturing based on AI edge computing terminals according to the present invention. Detailed Implementation

[0017] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0018] Please see Figure 1 Intelligent production scheduling and production planning methods based on AI edge computing terminals include: Based on the work order BOM and standard process route SOP issued by ERP / MES, as well as the equipment OEE time sequence slice, workstation turnover cycle and edge buffer pool inventory level collected in real time by the edge computing terminal, a theoretical production scheduling benchmark time axis containing the theoretical completion time sequence of each process is generated. Based on the preset set of production disturbance parameters and their disturbance pattern mapping relationship, parametric disturbance simulation is performed on the theoretical production scheduling benchmark time axis to generate a theoretical residual time series set; The actual completion time sequence is determined based on the actual workstation completion time collected by the edge computing terminal, and the difference between the actual completion time sequence and the theoretical completion time sequence in the theoretical production scheduling benchmark time axis is calculated to generate the actual residual time sequence. Dynamic time warping distance calculation and process route topology-based similarity calculation are used to compare the actual residual time series with the theoretical residual time series set. Combined with the preset judgment threshold, the disturbance type judgment result is determined, and the affected process and affected time window are located. Based on the disturbance type judgment result, affected process, affected time window and edge buffer pool inventory level, local work order translation scheduling instructions that act on the work order subset associated with the affected process are generated. Execute local work order shifting and scheduling instructions, and dynamically calibrate the set of production disturbance parameters and preset judgment thresholds based on the cycle time recovery index and inventory stability index within the evaluation period.

[0019] This embodiment provides a manufacturing intelligent scheduling and dispatching mechanism based on AI edge computing terminals; specifically, this embodiment takes an automotive electronic controller production workshop as the entire scenario, and the workshop includes continuous processes such as incoming material loading, surface mounting, reflow soldering, automatic optical inspection, component insertion, online testing, conformal coating and final inspection and packaging; The ERP system issues multiple controller work orders, and the MES system synchronizes the required components, semi-finished product shells, label accessories, and corresponding standard process routes for each work order. The edge computing terminal deployed on the production line side is connected to the equipment acquisition gateway to receive data from the pick-and-place machine, reflow soldering machine, automatic optical inspection (AOI) equipment, test bench, and inter-process buffer in real time, which is used to complete the production scheduling benchmark construction, disturbance identification, and local scheduling locally. To avoid confusion in terminology, the term "edge buffer pool inventory level" will be used consistently when referring to the number of inter-process buffers, safety zone judgment, and flow interruption risk judgment in the following text; and the term "theoretical residual time series" will be used consistently when referring to the reference objects generated by parametric disturbance simulation and used for subsequent comparison. Specifically, a theoretical production scheduling benchmark timeline is constructed. This step does not directly use the historical average delivery cycle as a reference, but filters out the impact of occasional disturbance data in the workshop, retaining only the inherent material constraints and process sequence constraints in the business. Its technical purpose is to pre-construct a theoretical production scheduling benchmark timeline for work order flow under standard working conditions, and use it as a reference standard for subsequent identification of actual production deviations. For example, a controller motherboard must first undergo surface mount technology (SMT), soldering, AOI (Automated Optical Inspection), and in-circuit testing. The casing assembly can only occur after the motherboard has passed testing. Therefore, the order of each node in the reference timeline is determined by the manufacturing process itself and is not affected by temporary noise. After obtaining the timeline, the edge computing terminal calls the built-in parametric simulation engine to transform common production disturbances into a set of disturbance patterns with industrial mechanisms; gradual wear and tear of equipment often manifests as a continuous lengthening of the cycle time. Upstream feeding anomalies typically manifest as a rapid decrease in work-in-process inventory after a certain point in time, causing subsequent workstations to wait; personnel shift changes and skill level differences are more akin to periodic rhythm fluctuations; the system generates multiple sets of theoretical residual time series on the theoretical time axis through disturbance patterns with industrial mechanisms, serving as reference comparison objects; Furthermore, the edge computing terminal continuously collects the actual completion time sequence; after subtracting the theoretical completion time from the actual completion time, the actual residual time sequence is obtained; its industrial significance lies in the fact that this residual directly reflects the degree and direction of the actual execution deviating from the standard working conditions; if all subsequent completion nodes of a certain process are gradually delayed, it often means that there is a trend of capacity decay in that process; if a sudden delay occurs in a certain period of time, and the cycle time returns to normal before and after the delay, it is closer to a short-term blockage after the material supply is interrupted. During the comparison phase, the system considers two types of information simultaneously: one is the degree of proximity in terms of time patterns, and the other is whether the propagation paths in the process topology are consistent. The former is used to identify whether the delay is a slow accumulation or a sudden jump, while the latter is used to identify whether the delay is propagating along the main chain from patch to detection or is limited to the test branch. Through the coupling of these two types of information, the system can reduce the false positive rate of anomalies caused by timeouts at a single workstation. If a real residual is similar in form to the theoretical residual of the equipment attenuation type in terms of timing, and the delay propagation path is consistent with the dependency chain of the subsequent process of the equipment, then the corresponding disturbance type is output; if the real residual is scattered, short, and discontinuous, then it is judged as random fluctuation. After determining the type of disturbance, the system does not directly rearrange the entire workshop, but only generates local work order translation scheduling instructions based on the affected processes, the affected time windows, and the inventory level of the edge buffer pool between processes. The translation here refers to the partial rearrangement of the start order, waiting order and succession order of related work orders, so that the unaffected sections can maintain their original production rhythm and avoid frequent fluctuations in the overall production schedule caused by single-point anomalies. After execution, the edge computing terminal observes whether the cycle time has recovered and whether the inventory level of the edge buffer pool has returned to the safe zone within the evaluation period, and fine-tunes the disturbance parameters and judgment threshold accordingly, so that the subsequent identification is closer to the current workshop status. As an exception handling mechanism, if the data issued by ERP or MES is missing, such as a work order lacking a complete BOM or standard operating time, the edge computing terminal will mark the work order as pending verification and will not include it in the automatic migration objects, only retaining the monitoring function; if the on-site data acquisition link is interrupted for a short time, the system will maintain the most recent stable production scheduling plan and suspend new disturbance judgment to prevent misscheduling caused by data breakpoints. If multiple disturbance types occur simultaneously and the matching results are not unique, the pending confirmation status will be output first, and the scheduling scope will be narrowed to the minimum impact window, without performing global work order reordering; if the inventory level of the buffer pool behind the affected process is already at a low level, the system will adopt a conservative release strategy for subsequent work orders to avoid the material shortage workstations from continuing to run idle. For example, during the day shift in the automotive electronic controller production workshop, the MES issued work orders for batch A and batch B of controllers, with batch A being prioritized for supply to the vehicle assembly line. In the first half of the morning, surface mount technology (SMT), soldering, and AOI all ran at standard pace, and the edge computing terminal established the theoretical production schedule benchmark timeline for the day based on this. As noon approached, wear and tear on the SMT nozzles caused the mounting pace to gradually slow down, and the completion times of AOI and online testing were subsequently delayed. After aligning the actual completed sequence with the reference time axis, the system obtains a gradually widening actual residual time series; it then compares it with the theoretical residual time series of equipment attenuation in the theoretical residual time series set and finds that its shape and propagation path are highly consistent, so it is determined to be the equipment attenuation disturbance of the chip mounting process. Instead of recalculating all work orders, the system shifted some B batch work orders to after B batch A within the affected window of the placement to AOI process. It also released pre-installed plug-in work orders that still had the necessary materials and workstations to ensure that there was no material shortage during the subsequent testing and packaging. After one evaluation cycle, the equipment maintenance was completed, the cycle time was restored, and the edge buffer pool inventory level returned to the safe zone. The system then wrote the disturbance parameters of this event back to the edge terminal for use by subsequent shifts. The purpose of this step is to form a complete closed loop with theoretical benchmarks, disturbance simulations, actual residuals, and local work order shift scheduling, so that the edge terminal can identify the trend disturbances that truly affect the stability of the production plan under limited computing power, and make dynamic adjustments only to the necessary range, thereby taking into account both scheduling agility and global plan stability.

[0020] As one embodiment of the present invention, generating a theoretical production scheduling benchmark timeline includes: determining the material constraints of the work order based on the work order BOM; determining the sequence of processes and standard operating time based on the standard operating procedure (SOP); calculating the theoretical start time and theoretical completion time of each process under material constraints and process sequence constraints based on the standard operating time, the theoretical capacity limit of the equipment, and the work order requirements; and constructing a theoretical production scheduling benchmark timeline according to the theoretical start time and theoretical completion time, which serves as the time benchmark data for production planning management.

[0021] This embodiment provides a step-by-step guide to constructing a theoretical production scheduling baseline timeline. Specifically, in the above-mentioned automotive electronic controller production line scenario, if the baseline is established solely based on historical average output, past anomalies may be mixed into the standard reference, leading to subsequent residual distortion. Therefore, this embodiment further decouples and models material constraints, process sequence, and theoretical equipment capabilities, so that the baseline timeline only expresses how work orders should flow when the rules are fully met. Specifically, the system reads the work order BOM to identify the material composition and dependencies of each work order. For a controller motherboard work order, the BOM may contain materials such as chips, resistors and capacitors, connectors, heat sinks and packaging labels, but these materials do not all play a binding role at the same time. For example, chips and solder paste determine whether the surface mount technology (SMT) process can begin, while housings and screws affect whether the final assembly process can be connected. By identifying these constraints in advance, the system can avoid allocating theoretical start points when the actual conditions for starting the process are not yet met. Subsequently, the system determines the sequence of processes and standard operating time based on the standard operating procedure (SOP). The standard operating time here reflects the normal time it takes for the equipment to complete the process under nominal capacity, standard fixtures, and qualified materials, rather than the experience value from a certain historical production. For controller products, surface mount technology (SMT) must precede reflow soldering, reflow soldering must precede AOI, and functional testing must be performed after component insertion and soldering are completed. This sequence is derived from the physical forming process of the manufactured object and cannot be arbitrarily reversed. Based on this, the system combines the theoretical upper limit of equipment capacity and work order demand to form theoretical start and completion nodes; for further explanation, the following simplified logical verification model is constructed: assuming work order It needs to go through a process , , ,in Corresponding patch, Corresponding AOI, Corresponding online test; like The standard operating time is one standard cycle unit. and Each is a standard beat unit, and Only in Begin after completion. Only in Once completed, the system will receive the following information sequentially. exist , , The theoretical starting and theoretical completion positions on the site; If another work order and If the placement machine is shared, but there is no competition for the testing station, then... exist The theoretical nodes should be arranged in order to After that, and The theoretical nodes can be extended according to the completion status of the previous sequence; thus, the system connects the theoretical start and completion nodes of multiple work orders into a whole timeline. As an exception handling mechanism, if the BOM shows that the critical materials are frozen, pending inspection, or the substitute materials have not been released, the corresponding work order can enter the planning pool, but will not enter the theoretical benchmark set that the materials meet. If the SOP version is inconsistent with the current process change order, the system will prioritize the effective version and mark the old version work order as a manual review object; if the theoretical capacity limit of the equipment is changed due to maintenance, such as the theoretical capacity being increased after the pick-and-place machine is replaced, the system will regenerate the baseline time axis of the corresponding process segment to avoid continuing to use outdated capacity parameters; For example, in the same workshop, the A batch of controller work orders includes three key materials: motherboard, casing, and label. The standard process route is as follows: SMT placement, reflow soldering, AOI, insertion, online testing, final assembly, and final inspection. After receiving the BOM and SOP, the edge computing terminal first determines that the motherboard components are complete and the casing has not yet arrived at the final assembly station, but this does not affect the front-end processing. Therefore, the front-end process is allowed to enter the theoretical production schedule. According to the standard cycle time, the theoretical start and finish positions from patch placement to final inspection are generated for each work order. Multiple work orders are then connected in series according to the nominal capacity of the equipment to form the theoretical production scheduling baseline timeline for the shift. This timeline does not include disturbances such as equipment failure, personnel fatigue, and material arrival delays, and only represents the expected state. The purpose of this step is to establish a unified reference for subsequent residual analysis that excludes historical data bias, so that actual deviations can be accurately interpreted as disturbances from the equipment, materials, or personnel side, rather than spurious disturbances caused by errors in the theoretical baseline data itself.

[0022] As one embodiment of the present invention, the production disturbance parameter set includes equipment attenuation factor, material supply fluctuation factor and personnel efficiency fluctuation factor; wherein, the equipment attenuation factor is used to characterize the exponential attenuation pattern of the continuous reduction of processing capacity over time; the material supply fluctuation factor is used to characterize the step pattern of a sudden drop in inventory at a predetermined time. The personnel efficiency fluctuation factor is used to characterize the rhythm fluctuation pattern that changes with shift or personnel status cycle; the parameterized disturbance simulation is specifically: calling the parameterized simulation engine built into the edge computing terminal to parameterize and superimpose the equipment attenuation factor, material supply fluctuation factor and personnel efficiency fluctuation factor to generate the corresponding theoretical residual time series, which is used to establish the mapping relationship between disturbance type and plan deviation.

[0023] This embodiment provides a mechanism for parameterized simulation of production disturbances. Specifically, when only theoretical production scheduling benchmarks are available, although the system establishes theoretical completion nodes, it is still difficult to physically trace the causes of deviations. Especially in automotive electronic controller workshops, although different disturbances all lead to delays, their propagation methods differ. If the disturbance mechanism is not explicitly introduced, normal fluctuations are easily misjudged as abnormal. Therefore, this embodiment establishes an interpretable residual template through three types of disturbance factors: equipment, materials, and personnel. Specifically, the equipment degradation factor is used to represent the state of continuous decline in processing capacity over time. This phenomenon is common in chip mounter nozzle wear, reflow soldering temperature control drift, and test fixture contact aging. The workshop performance is usually not a sudden shutdown, but rather the cycle time of the same batch of work orders gradually lengthens, and subsequent workstations will experience cumulative lag along the main process chain. Material supply fluctuation factor corresponds to the situation of intermittent material feeding, batch waiting for inspection, or rapid consumption of intermediate buffer after a certain point in time. Its performance is closer to a sudden gap: the previous cycle time was normal, but the workstations in a certain window are waiting for materials. Personnel efficiency fluctuation factor reflects the periodic cycle time changes caused by shift handover, new staff replacement, and increased rework load. Its impact often has the characteristics of shift rhythm and is not necessarily long-term. The parameterized simulation engine in the edge computing terminal does not employ high-complexity global simulation. Instead, it generates a theoretical residual template by superimposing offset patterns according to perturbation types on an existing theoretical timeline. For further explanation, the following simplified logical verification model is constructed: assuming there are four consecutive completion nodes on the theoretical timeline. , , , If the injected equipment causes attenuation disturbance, the corresponding residual template will show an offset that gradually increases from front to back. If the supply of injected materials fluctuates, it may... Afterwards, there is a sudden increase in the offset, which then levels off after refueling; if the efficiency of the injection personnel is disturbed, then... arrive The offset may exhibit alternating high and low beat swings; through this morphological mapping, the system can transform abstract disturbances into comparable time series templates. As an anomaly handling mechanism, if a certain type of disturbance lacks historical support in the current workshop, such as newly introduced equipment not yet accumulating enough attenuation samples, the system can first activate the industry-standard initial template and gradually correct it in subsequent calibration stages. If equipment degradation and material fluctuations occur simultaneously within the same time window, the system allows the generation of composite templates. However, when edge computing power is insufficient, single-factor templates and a small number of high-frequency composite templates are reserved first to avoid the template library exceeding the preset storage space threshold. If a disturbance factor is disabled by the on-site process engineer, for example, if a workstation is fully automated and the frequency of human intervention is lower than the preset lower limit of influence, the system will not generate the corresponding template for that workstation to reduce mismatches. For example, in the same production line, during continuous production in the morning, wear on the pick-and-place machine nozzles can lead to a decrease in the success rate of picking up and placing components. The equipment needs to perform compensation actions more frequently, so its theoretical residual template exhibits a gradually expanding delay pattern. In the morning, a batch of connectors arrived and were awaiting inspection, which caused the plug-in station to have no material to install for a period of time. As a result, the template showed a step delay after a certain time point. During the shift handover in the afternoon, the newly hired operator's actions were unstable in the three-proof coating replenishment process, so the template showed that the cycle time fluctuated with the shift. The edge computing terminal stores these templates locally for matching with the subsequent actual residuals. The purpose of this step is to transform the experience of on-site experts into an executable perturbation dictionary at the edge, so that the system's identification of production deviations is based on real industrial mechanisms, rather than remaining at the level of unexplainable simple timeout alarms.

[0024] As one embodiment of the present invention, generating a real residual time series includes: determining an actual completion time series based on the actual workstation completion time collected by the edge computing terminal; performing a difference calculation between the actual completion time series and the theoretical completion time series in the theoretical production scheduling benchmark time axis to generate a real residual time series; wherein, each theoretical residual time series in the theoretical residual time series set is a difference result between the simulated completion time series and the theoretical completion time series.

[0025] This embodiment provides a step for generating a real residual time series. Specifically, based on the existing theoretical benchmark and theoretical residual template, it is also necessary to map the real-time execution status of the workshop into a comparable unified data format. Otherwise, the theoretical template and the field data will be at different semantic layers and cannot be directly matched. Therefore, this embodiment extracts the real residual time series by aligning and differentiating the actual completion time and the theoretical completion time. Specifically, the actual completion time sequence originates from device events received by the edge computing terminal, workstation barcode scanning records, test completion feedback, and buffer entry and exit records. For automotive electronic controller production lines, the pick-and-place machine can provide the timestamp of the board leaving the equipment, the AOI can provide the inspection end time, the online testing station can provide the program execution completion time, and the final inspection station can form product completion nodes through barcode transit time. The system merges these scattered events according to work orders and processes to form a continuous actual completion time sequence. The system compares each actual completion node with the theoretical completion node under the same work order, same process, and same process version; this comparison is not used to show complex mathematical processes, but to reveal the deviation of the actual manufacturing process from the standard working conditions. If the difference remains close to zero for a long period, it indicates that the on-site execution is consistent with the baseline; if the difference gradually increases, it indicates that production capacity is being continuously eroded; if the difference suddenly increases after a certain point, it indicates that the production chain has experienced significant blockage during that period. To further clarify the difference calculation process, the following simplified model is constructed: If a work order is in , , The theoretical completion position on the sequence is , , The actual completion location is the sequence. , , The actual residual can then be expressed as three offsets. , , ;like Within the preset benchmark range Positive deviation occurs If the positive deviation shows an increasing trend, it indicates that the delay has a tendency to propagate backward and accumulate; Furthermore, the theoretical residual template also adopts the same expression method; that is, the simulated completion sequence and the theoretical completion sequence are differentially analyzed with the same caliber to obtain a set of theoretical residual time series; in this way, the actual residual and the theoretical residual have the same data structure and physical meaning, and can both be understood as the delay form relative to the standard working condition, thus having comparability; As an exception handling mechanism, if the actual completion time of a certain workstation is missing, for example, if the barcode scanner is temporarily offline, the system can combine the upstream and downstream process times and the entry and exit events of the buffer area to fill in the gap, but this will reduce the reliability level of the data. If the number of consecutive missing segments exceeds the preset window, the system will stop automatically judging that segment and only retain the monitoring mark; if the theoretical node changes version due to process changes, the system must realign according to the version to avoid mistaking the time difference between the old and new routes as production disturbances; if the work order is reworked midway, the rework path will be established as a separate residual chain and will not be mixed with the first-pass path. For example, in the production of controller motherboards, the edge terminal receives timestamps for completion of surface mount technology (SMT), AOI (Automated Optical Inspection) and online testing, respectively. After aligning the actual completion sequence of a batch of work orders with the theoretical timeline of the shift, the system finds that the offset of the SMT process is within the preset reference range, the offset of the AOI process begins to increase, and the offset of the online testing process continues to increase, thus forming a backward-expanding real residual time sequence. In contrast, the theoretical residual generated in the template library for the attenuation of the SMT equipment also shows the same expansion trend, thus providing a basis for subsequent matching. The purpose of this step is to compress complex, discrete, and diverse field events into a uniform expression relative to the theoretical benchmark, so that different workstations, work orders, and disturbance types can be compared and judged on a unified scale.

[0026] As one embodiment of the present invention, when comparing the actual residual time series with the theoretical residual time series set, the specific steps include: calculating the time shape score and the topology propagation score respectively; combining the preset time shape weight and topology propagation weight to perform a weighted calculation on the time shape score and the topology propagation score to determine the normalized target matching score; when the target matching score is higher than or equal to a preset high threshold, outputting the disturbance type determination result corresponding to the theoretical residual time series that makes the target matching score meet the condition, and triggering a local work order translation scheduling instruction; When the target matching score is lower than or equal to the preset low threshold, a random fluctuation judgment result is output, and the local work order translation scheduling instruction is suppressed; when the target matching score is between the high threshold and the low threshold, a judgment result to be confirmed is output, and the current production scheduling plan is maintained.

[0027] This embodiment provides a disturbance type determination and scheduling triggering mechanism. Specifically, the fact that the morphological similarity between the actual residual and a certain theoretical residual time series reaches a preset matching threshold is not sufficient as a sufficient condition for triggering automatic scheduling, because fluctuations in the workshop with amplitudes lower than the preset tolerance threshold, pauses with durations shorter than the preset judgment window, and barcode scanning delays are all possible. Especially in continuous manufacturing scenarios, the cycle time of different processes has natural scaling. If time stretching and process propagation paths are not considered, misjudgments are likely to occur. Therefore, this embodiment introduces a dual comparison based on time pattern and process topology; to ensure consistent naming, the theoretical residual timing sequence, theoretical residual timing sequence and theoretical residual timing sequence set in the following text all refer to the same type of reference object; when triggering execution is involved, the name "local work order translation scheduling instruction" is used uniformly and no other name is used; Specifically, dynamic time warping distance is used to handle the same type of disturbance but with different evolution time periods; for example, equipment degradation may gradually appear within two hours or deteriorate rapidly within half an hour. The common feature is the continuous accumulation of the delay pattern, while the difference is only the different unfolding speed. Through time warping, the system does not require the actual residual and the theoretical residual to be completely synchronized at each node, but focuses on whether they belong to the same evolution trend. At the same time, the similarity of the process route topology is used to confirm whether this trend is propagating along the correct manufacturing dependency chain; if there is a problem in the chip mounting process, its impact should mainly spread along the chip mounting-soldering-AOI-testing chain, and should not appear simultaneously on completely independent packaging branches without cause; To further illustrate, a simplified logical verification model is constructed as follows: Assume the actual residual occurs in the process... , , The above shows a gradual increase, and the theoretical residual time series A also gradually increases, while the theoretical residual time series B is at... Then it suddenly changed; If we only look at the latter part of the results, both may show delays; however, through time warping, the trends of the actual residual and the theoretical residual time series A are closer; further considering topological propagation, if the actual delay only occurs in the period between... Subsequent associated nodes, but not propagated to branches without direct dependence, indicate that they are more consistent with the device attenuation type; the system merges the two comparison results into a normalized target matching score; Furthermore, to avoid the target matching score remaining at the purely descriptive level, in this embodiment, the edge computing terminal first calculates the time morphology score and topology propagation score for each theoretical residual time series, and then performs weighted fusion; the time morphology score comes from the monotonic mapping of the dynamic time warping distance, that is, the smaller the distance, the higher the score; the topology propagation score comes from the degree of overlap between the actual residual occurrence process set and the theoretical residual time series affecting the process set, as well as whether the propagation order is consistent; Specifically, the degree of overlap is determined by calculating the ratio of the number of intersection elements to the number of union elements of the actual residual-affected process set and the theoretical residual-affected process set; the consistency of propagation order is determined by comparing the relative edit distance between the actual delayed process sequence and the theoretically affected process sequence. The system performs a weighted summation of the above overlap score and consistency score to obtain the original value of the similarity score based on the process route topology. For automotive electronic controller production lines, the main chain from surface mount to testing can be set as a high-weight chain, while the packaging branch, which is not directly coupled to the current disturbance, can be set as a low-weight chain. In this way, even if two theoretical residual timing sequences are similar in time pattern, as long as the propagation path of one of the theoretical residual timing sequences is inconsistent with the actual field, its final target matching score will be reduced. Furthermore, the calculation of the normalized target matching score follows a unified approach: first, the dynamic time-normalized distance is converted into an interval-based morphological similarity score. Then, the hit rate of the process route topology is converted into a range-based topology similarity score. and combined with satisfying Preset weights and The target matching score is obtained through weighted calculation. ; The interval conversion employs maximum-minimum value normalization, mapping the original values ​​of the dynamic time-normalized distance combined with the inverse mapping relationship and the topological similarity score to a dimensionless numerical range of 0 to 1, thereby eliminating the impact of dimensional differences on the weighted calculation results. In specific deployments, if the workshop focuses more on the consistency of the propagation chain, the weight of the topological item is increased; if the workshop's cycle stretching characteristics are more stable, the weight of the morphological item is increased. In this way, the system clarifies the business logic of comparison, synthesis, and triggering, avoiding the uninterpretability of the judgment process. In terms of output, this embodiment adopts a three-stage decision-making approach: high threshold, low threshold, and intermediate buffer. When the output is higher than the high threshold, it indicates that the current actual residual is sufficiently consistent with the time series of a certain theoretical residual in terms of both morphology and propagation. A clear disturbance type can be output and a local work order shift scheduling instruction can be triggered. When the output is lower than the low threshold, it indicates that the field deviation is closer to irregular scattered fluctuations. The system classifies it as random fluctuations and does not trigger a local work order shift scheduling instruction. When the situation falls between these two points, it indicates that the on-site status shows some signs of abnormality, but is not yet sufficient to support the automatic execution of local work order shifting and scheduling instructions. The system maintains the current production schedule and outputs a pending confirmation status so that the situation can be further observed or the team leader can be prompted to intervene. Furthermore, to prevent the target matching score from fluctuating around the threshold and causing scheduling jitter, the system uses a continuous observation window for confirmation; that is, it does not immediately execute the local work order translation scheduling instruction once a single sampling period exceeds the high threshold, but requires the same disturbance type to maintain the highest matching in multiple consecutive sampling periods, and the affected process and the propagation path to remain consistent before outputting the formal judgment result and executing the corresponding local work order translation scheduling instruction. If a sudden drop in score, a break in the propagation path, or a switch in the theoretical residual timing type occurs during a certain sampling period, the system will roll back the judgment of that round to the pending confirmation state for continued observation; this can filter out non-continuous noise such as late barcode scanning, instantaneous material blockage, and short pauses. As an anomaly handling mechanism, if the target matching scores of multiple theoretical residual time series are close to the high threshold at the same time, the system will prioritize the type of disturbance with a shorter topology propagation path and a more concentrated range of influence in order to avoid compound misjudgment. If the difference between the high threshold and the low threshold is less than the preset threshold difference, causing the intermediate buffer range to be less than the preset margin, the system will automatically widen the distance between the two during the calibration cycle to retain the necessary observation zone; if the jitter of the collected data causes the matching result to frequently cross the threshold, the system can set a minimum continuous observation window, that is, only when the conditions are met for multiple consecutive sampling cycles will a formal judgment be output and a local work order shift scheduling instruction be executed. For example, during the operation of the controller production line, the completion time of the AOI post-workstation was continuously delayed in the morning. After comparing the actual residual with the theoretical residual time series of equipment attenuation, material shortage, and personnel efficiency, the edge terminal found that the theoretical residual time series of equipment attenuation was the closest in time pattern and the delay propagation path mainly extended along the main chain after patching. Therefore, a high target matching score was obtained, which exceeded the high threshold. Thus, the system determined it to be an equipment attenuation disturbance and triggered a local work order shift scheduling instruction. In another scenario, if the three-proof coating station experiences fluctuations in several work orders that do not exceed the preset tolerance threshold at the moment of shift change, but the shifts are discontinuous and do not match any theoretical residual timing, the system will identify them as random fluctuations and suppress local work order shifting instructions. If an abnormality in material arrival initially manifests only as a shortage of work-in-process that has not triggered the alarm lower limit, and the matching score falls in the middle range, the system will not adjust the plan for the time being, but will continue to observe the changes in the inventory level of the edge buffer pool. The purpose of this step is to upgrade anomaly identification from a single-point timeout judgment to a comprehensive judgment of morphology and topology, and to avoid frequent production scheduling fluctuations through hierarchical thresholds, so that local dynamic adjustments are only triggered when there is sufficient evidence.

[0028] As one embodiment of the present invention, the generation of a partial work order shifting scheduling instruction includes: locating the affected process and the affected time window based on the disturbance type determination result; determining a set of shiftable work orders that meet the material availability constraint, process sequence constraint, and edge buffer pool inventory level constraint based on the edge buffer pool inventory level; and within the affected time window, and only for the subset of work orders associated with the affected process, performing sequence adjustment and start / end time reallocation on the set of shiftable work orders to generate a partial work order shifting scheduling instruction for partially updating the production plan.

[0029] This embodiment provides a mechanism for generating local work order shifting and scheduling instructions. Specifically, after clarifying the type of disturbance, if a global rescheduling method is still used, although a new theoretical plan can be obtained, it will disrupt the stable rhythm of unaffected sections, causing frequent equipment changes, repeated operator confirmations, and fluctuations in downstream delivery plans. Therefore, this embodiment limits the adjustment scope to the affected processes and their associated work order subsets, achieving local repair rather than a complete replanning. Specifically, the system locates the affected process and time window based on the type of disturbance; if it is determined to be attenuation of the surface mount equipment, the affected window usually covers the propagation range of the surface mount process and several subsequent processes. If the problem is determined to be a fluctuation in material supply, the key focus is often on the period of time when the work is stopped due to material shortage. The system combines the inventory level of the edge buffer pool to determine the set of work orders that can be moved. The inventory level here is not just the warehouse inventory, but the real-time quantity of work-in-process between processes, which directly reflects whether the subsequent work stations will experience a break in flow or overload. Only work orders that simultaneously meet the requirements of material availability, sequential establishment of processes, and buffer pool not falling below the safety line are allowed to enter the translation set; the safety line of the buffer pool is the minimum theoretical quantity of work-in-process required to maintain the continuous operation of downstream equipment, which is dynamically calculated by the system based on the standard flow rhythm of downstream key processes and the preset maximum allowable idling waiting time of equipment. Furthermore, the system adjusts the execution order and reallocates start and completion times for these work orders within a local window; for further explanation, the following simplified logical verification model is constructed: assuming there are work orders within the affected window. , , ,in and They all need to go through the affected pick-and-place machine. The patch installation is complete; we can proceed directly to the later-stage testing. If the water level in the buffer pool is lower than the preset safety threshold before the test, the system will release the water first. The corresponding backend tasks are implemented to keep the test station online without interruption, while also... Move the current window to the next level and allocate resources to deliveries with higher priority and complete material sets. Therefore, local scheduling does not simply postpone all work orders, but rather performs a more refined sequence reconstruction within the affected area. Furthermore, to ensure the feasibility of selecting the set of work orders that can be moved, the edge computing terminal performs constraint checks on each candidate work order. Specifically, it first checks whether the work order still needs to go through the affected process within the affected window; then it checks whether its preceding processes have been completed or can be completed sequentially after redistribution. Next, check whether the corresponding materials, fixtures and tooling are available within the time period after the translation; check whether the translation operation will cause the downstream buffer pool to fall below the safety line or the upstream buffer pool to exceed the congestion line; only when all the above checks are passed is the work order included in the translation set; in this way, local translation is not an abstract sequential adjustment, but a linkage check based on the process chain, material chain and work-in-process chain. Furthermore, the redistribution of start and finish times follows the principle of first locking the unchanged parts, and then partially extending or moving them forward; that is, the system first freezes the work order nodes outside the affected time window to ensure that the unaffected sections continue to be executed according to the original plan; only within the window are the start time, waiting time and succession time of the work orders that can be moved rearranged rearranged. If a work order is moved to the back, the theoretical nodes of its subsequent processes will be moved back synchronously; if a work order is moved to the front into the idle equipment window, the system will simultaneously verify that it will not crowd out confirmed high-priority work orders; in this way, the local scheduling results can be directly mapped into the process-level start / complete plan that can be executed by MES, without the need for full table recalculation. Furthermore, when multiple candidate work orders meet the constraints, the system sorts them according to a unified priority rule: first, it compares the delivery priority and the urgency of the production task; then, it compares whether the low-water level buffer can be quickly replenished; and finally, it compares the impact of model changeover, the continuity of the same process family, and the current completion status of the work order. Taking the automotive electronic controller scenario as an example, if batch A is responsible for supplying the whole vehicle assembly and batch B is used for regular inventory replenishment, and the pre-test buffer is close to the safety lower limit, then the system prioritizes ensuring that batch A passes through the affected processes, while prioritizing the release of work orders that have completed the front-end and can immediately replenish the pre-test work-in-process. Through unified rule sorting, the process succession reorganization under specific constraints is realized. In abnormal situations, if there are no work orders that can be moved within the affected time window that meet the constraints, the system will not generate a forced move instruction, but will maintain the original plan and output a resource-limited flag; if the buffer pool level is already below the safety line, the system will prioritize the supply guarantee strategy, that is, prioritize the release of work orders that can quickly replenish downstream work-in-process, rather than just sorting them by delivery priority. If materials are available for a work order but the preceding process has not been completed, the work order should not be inserted ahead of schedule, skipping the process sequence. If multiple work orders can be moved but have a high impact on the corresponding model change, the system will prioritize work orders within the same process family to reduce equipment changeover losses. For example, in the controller workshop, after the pick-and-place machine experiences a continuous speed reduction, the affected window is locked in the pick-and-place to AOI process chain in the latter half of the morning; the edge terminal detects that there are still some work-in-process in the AOI pre-buffer pool, while the online testing pre-buffer pool is close to the safety lower limit; the system then selects several B batch work orders from the current work order pool that have completed the front-end and can directly enter the testing and final assembly, and releases them to the back-end first to ensure continuous operation of the testing station and assembly station; At the same time, low-priority work orders that still need to go through the affected pick-and-place machine will be moved to the back, leaving the limited pick-and-place capacity for high-priority work orders in batch A; after the generated local work order shift scheduling instructions are sent to MES, only a small number of work order sequences within the affected window are modified, while other sections remain unchanged from the original plan. The purpose of this step is to offset the chain reaction caused by single-point disturbances by adjusting the order of the smallest possible range, reduce the plan jitter and execution complexity caused by global reordering, and improve continuous production capacity under limited resource conditions.

[0030] As one embodiment of the present invention, dynamic calibration includes: calculating cycle recovery index, inventory stability index, and misjudgment suppression index within the evaluation period; performing sensitivity analysis on the disturbance parameters in the production disturbance parameter set based on the cycle recovery index to determine the set of disturbance parameters to be calibrated; updating the judgment threshold based on the inventory stability index and the misjudgment suppression index; and writing the updated set of disturbance parameters to be calibrated and the judgment threshold into the edge computing terminal to complete closed-loop calibration, thereby improving the adaptability of the production plan to the actual production environment.

[0031] This embodiment provides a dynamic calibration mechanism. Specifically, if the system uses the initial disturbance template and fixed threshold for a long time, as the equipment ages, the process is optimized, and the team structure changes, the original template will gradually deviate from the actual on-site state, resulting in problems such as recognition lag or false triggering. Therefore, this embodiment performs closed-loop correction of the disturbance parameters and judgment threshold within the evaluation cycle after scheduling execution. Specifically, the cycle time recovery index is used to observe whether the target process returns to a state close to the standard cycle time after local scheduling and on-site handling; its industrial meaning is that if an event judged as equipment degradation recovers quickly after maintenance, it indicates that this type of template has a good fit to the real problem that meets the preset standard. If the recovery is not obvious, it may mean that the original judgment was biased, or that the disturbance parameter setting deviates from the magnitude requirement of the actual working condition; the inventory stability index is used to evaluate whether the inter-process buffer pool has recovered to the safe zone after local scheduling, and whether the inventory fluctuation range is kept within the preset safety inventory variance threshold; if the inventory fluctuation range continues to exceed the preset safety inventory variance threshold after scheduling, it indicates that the current scheduling strategy and threshold setting may still be too sensitive; the misjudgment suppression index is used to measure the situation where it is proven that no adjustment is needed after triggering or that no significant blockage is ultimately formed without triggering, thus constraining the judgment logic from the result side; Based on this, the system performs sensitivity analysis on the disturbance parameters to identify which parameters have the greatest impact on the recovery results. Specifically, the sensitivity analysis process is as follows: the system introduces parameter perturbations of a preset step size sequentially on the base values ​​of the current production disturbance parameters, calculates the change in the beat recovery index before and after the perturbation, and obtains the ratio of the change to the corresponding parameter perturbation. The disturbance parameters whose absolute value of the ratio exceeds the preset sensitivity threshold are extracted and together constitute the set of disturbance parameters to be calibrated. To further illustrate, the following simplified logical verification model is constructed: If the equipment attenuation parameters in the template library include three levels: mild, moderate, and severe, and recent events have shown that only moderate templates can better explain the on-site recovery process, then the system will include parameters near the moderate level as key calibration targets. If the personnel efficiency template is frequently triggered but ultimately recovers naturally after a short period of time, the system should reduce the sensitivity of this type of template. For threshold updates, if the inventory is stable but there are many false alarms, the high threshold can be appropriately increased or the low threshold trigger probability can be reduced. If missed alarms cause the edge buffer pool inventory level to frequently fall below the preset safety lower limit, the trigger conditions can be appropriately relaxed so that the system can intervene earlier. Furthermore, to avoid dynamic calibration remaining at the level of empirical description, this embodiment provides a structured definition of the calculation methods for the three evaluation indicators; the cycle time recovery indicator targets the affected process and its subsequent key processes, comparing the actual cycle time after local scheduling execution with the standard cycle time. If the deviation continues to converge within the evaluation period, the recovery is considered good. The inventory stability indicator targets the buffer pool related to the affected window, examining whether it has returned to the safe zone and whether there are still continuous low water levels or continuous high accumulation; the misjudgment suppression indicator targets the judgment result itself, counting the number of times that scheduling was triggered but no substantial improvement was observed, and the number of times that scheduling was not triggered but significant blockages occurred subsequently; through a unified standard, the system precipitates the scheduling effectiveness assessment into reusable edge-end assessment evidence. In specific quantification, the cycle time recovery index is calculated as the root mean square error of the difference between the actual cycle time of the target process and the standard cycle time within the evaluation period; the inventory stability index is calculated as the variance of the actual inventory level of each process buffer pool deviating from the preset benchmark water level target value within the evaluation period; the misjudgment suppression index is calculated as the sum of the weighted penalty values ​​of the number of mis-triggered scheduling and the number of missed-triggered scheduling, wherein the weight of the mis-triggered penalty and the weight of the missed-triggered penalty are preset based on the proportion of downtime loss cost generated by historical scheduling. Furthermore, the set of disturbance parameters to be calibrated is not modified simultaneously for all parameters, but rather first located by event type and then filtered by impact degree; for equipment degradation events, parameters related to degradation start time, degradation rate and propagation depth are checked first; for material supply events, parameters related to step occurrence time, gap duration and buffer pool response threshold are checked first; for personnel efficiency events, parameters related to shift cycle, fluctuation amplitude and duration window are checked first. If a certain type of parameter is highly correlated with the recovery results in multiple evaluation periods, it will be included in the calibration set; if a certain type of parameter is not sensitive to the results for a long time, it will not be adjusted for the time being to avoid excessive drift of edge parameters. Furthermore, the threshold update adopts an iterative correction with a limited step size to suppress discontinuous jumps in parameters; that is, when the inventory stability index indicates that the buffer pool has stabilized after scheduling, but the misjudgment suppression index shows that there are too many false triggers, the system adjusts the high threshold upward according to the preset first compensation step size and adjusts the low threshold downward according to the preset second compensation step size. When the beat recovery indicator is poor and the inventory stability indicator shows that the buffer pool inventory level frequently falls below the lower limit of the safety line, the system prioritizes lowering the high threshold or extending the confirmation window so that the system can identify trend disturbances earlier. When the three indicators give inconsistent directions, the system only allows a single fine adjustment of one direction, and then continues to correct it after verification in the next evaluation cycle. This can avoid the threshold frequently crossing the judgment boundary between different shifts, which would lead to inaccurate judgment of the boundary of similar events. Furthermore, version management is adopted when writing back to the edge computing terminal; that is, a new parameter version number is generated for each calibration, and the corresponding evaluation cycle, applicable product family, applicable process version and effective time are recorded; if the new version leads to a significant increase in misjudgments in subsequent cycles, the system can revert to the previous stable version; this process does not change the write-back action in this embodiment, but further clarifies the traceable and reversible mechanism after write-back, making the long-term operation of the edge terminal more stable. As an exception handling mechanism, if the data sample is insufficient during the evaluation period, such as when a new product is launched or a certain workstation is shut down for a long time, the system will not automatically modify the core parameters, but only retain the observation records; if the calibration directions given by different indicators conflict with each other, such as when the cycle time is well recovered but the inventory is still unstable, the system will prioritize keeping the threshold unchanged and only make minor adjustments to the local template parameters; if writing back to the edge terminal fails, the system will retain the original parameters and continue to run, and cache the version to be written back to the local non-volatile memory, and write it again after communication is restored; For example, after the controller production line has been running for several shifts, the edge terminal found that in most of the recent events that were identified as chip placement equipment degradation, the cycle time quickly recovered after the nozzle was replaced, and the water level in the AOI pre-buffer pool returned to the stable zone synchronously, indicating that this type of template is effective. However, after multiple triggers of the personnel efficiency template triggered by the afternoon shift handover, the system found that the inventory had not actually become unstable and was therefore overly sensitive. Therefore, at the end of the evaluation period, the system lowered the sensitivity parameters of the personnel efficiency fluctuation template and appropriately increased the corresponding trigger threshold, while maintaining or even refining the attenuation template for patch devices. The updated template and threshold were written to the edge terminal for continued use on the next working day. The purpose of this step is to enable the production scheduling identification system to continuously adjust its boundary judgment parameters as the actual state of the workshop evolves, avoiding long-term static execution of the initial template, thereby improving the consistency between the scheduling strategy and the production site.

[0032] As one embodiment of the present invention, when the parametric simulation engine generates the theoretical residual time series set, it further includes: slicing the theoretical production scheduling baseline time axis according to a preset time window; injecting single disturbance factors and composite disturbance factors for each time window; attaching a disturbance type identifier to each injection result to form a correspondence between disturbance type and residual form; and outputting the theoretical residual time series set according to the correspondence, which is used as a reference template library for subsequent disturbance type matching.

[0033] This embodiment provides a time window slicing construction mechanism for a theoretical residual template library. Specifically, although the previous embodiment was able to generate templates for different disturbances, if the templates were only constructed according to the whole shift or the whole day, it would be difficult to adapt to the occurrence and recovery of such local events in a specific period of time in the workshop. Especially in the production of automotive electronic controllers, shift handover, material change, first piece confirmation, and temporary material replenishment all have obvious time window characteristics. Therefore, this embodiment further slices the theoretical production scheduling benchmark time axis and injects disturbances into different windows. Specifically, the system first segments the theoretical production scheduling baseline time axis according to preset time windows. The time windows can be divided according to shift nodes, equipment maintenance nodes, material delivery cycles, or fixed-length windows. The industrial significance of doing this is that the impact background of the same type of disturbance is not the same in different time periods. For example, the fluctuations that do not exceed the preset tolerance threshold when the equipment is first started in the morning and the capacity decay after continuous operation in the afternoon may both cause changes in the cycle time, but their residual shape and propagation depth are often different. The system injects single perturbation factors and composite perturbation factors in each time window; the single perturbation factor is used to characterize the deviation pattern dominated by a certain type of cause, while the composite perturbation factor is used to describe the superposition state that is more common in real workshops. To further illustrate, a simplified logical verification model is constructed as follows: If a baseline timeline is divided into windows... , , ,exist If only device attenuation is injected, then the template is obtained. ;exist If both equipment attenuation and material fluctuations are simultaneously injected, a template is obtained. ;exist If the efficiency of the injected personnel fluctuates, then a template is obtained. Each template has a corresponding disturbance type identifier and time window identifier, thus not only identifying the disturbance type, but also the time interval in which the disturbance occurs; After attaching a disturbance type identifier to each injection result, the system forms a correspondence between time window, disturbance type, and residual form, and outputs it as a template library. During subsequent matching, the edge terminal can not only match the template type to which the actual residual belongs, but also locate the corresponding time window, thereby reducing cross-scenario mismatches. For example, the same time cycle fluctuation template should not be directly applied to the early morning maintenance period, but the personnel efficiency fluctuation template before and after lunch should be applied to the same time cycle fluctuation. As an exception handling mechanism, if the time window segmentation length is less than the preset lower limit of the time window, causing the template library data volume to exceed the preset storage threshold at the edge, the system retains the core window according to product family and high-frequency scenario, and uses shared templates for low-frequency windows; if there are too many composite disturbance templates, the system prioritizes retaining the combined templates that have appeared frequently in history and are easily confused with single disturbances; if the actual event falls at the intersection of two time windows, the system can call the templates of adjacent windows at the same time to participate in the comparison, avoiding the boundary error caused by fixed time threshold truncation; For example, in the daily production scheduling of the controller workshop, the system divides the theoretical time axis into the early shift start window, the stable production window, and the shift handover window. For the stable production window, the system generates templates for chip mounting equipment attenuation, connector supply interruption, and a combination of both. For the shift handover window, the system focuses on generating templates for personnel efficiency fluctuations and templates for personnel fluctuations combined with material replenishment delays. Each template is marked with its window and type. In this way, when there are fluctuations in the clock speed during the afternoon shift handover period, the system will prioritize calling the relevant templates in the shift handover window, instead of mistakenly matching the early shift start template. The purpose of this step is to further refine disturbance identification from type matching to type + time period matching, improve the template library's coverage of the phased characteristics of the real workshop, and take into account the lightweight deployment requirements of the edge.

[0034] As one embodiment of the present invention, the equipment OEE time sequence slice includes an availability subsequence, a performance subsequence, and a quality subsequence; the workstation turnover cycle is a sequence of time intervals between process-level completions; the edge buffer pool inventory level is a sequence of work-in-process quantities between processes; and the disturbance type determination result includes at least equipment attenuation disturbance, material supply disturbance, personnel efficiency disturbance, and random fluctuation.

[0035] This embodiment provides a definition method for edge acquisition data and disturbance judgment categories. Specifically, if the edge terminal only receives single-cycle information, although it can measure the cycle length of a certain workstation, it is difficult to distinguish different reasons such as the decline in the equipment's own capacity, unstable upstream supply, or increased quality rework. Therefore, this embodiment defines the physical meaning of the input data in layers to make subsequent judgments interpretable. To ensure consistent terminology, the term "edge buffer pool inventory level" will be used uniformly when describing the quantity of inter-process buffers, flow interruption risk, safety zone judgment, or accumulation status. The data body corresponds to the inter-process work-in-process quantity sequence. Specifically, the equipment OEE time sequence slice is further divided into availability subsequence, performance subsequence, and quality subsequence; availability subsequence reflects whether the equipment is in a productive state over a period of time, and is suitable for identifying the impact of downtime, standby, and maintenance; performance subsequence reflects whether the equipment reaches the nominal speed under operating conditions, and is suitable for identifying tool wear, mechanism aging, and cycle time decline. The quality rate subsequence reflects whether the output is stable and qualified, and is suitable for identifying batch quality anomalies and increased rework burden; the workstation turnover takt time is defined as the process-level completion time interval sequence, used to describe the completion rhythm of adjacent products or adjacent work orders in the same process; the edge buffer pool inventory level is defined as the inter-process work-in-process quantity sequence, used to reveal whether a certain connection in the production chain is accumulating or depleting. These data combinations can support more detailed disturbance differentiation; for example, a decrease in availability and a simultaneous decrease in performance rate are more likely to correspond to equipment deterioration; a normal performance rate but a sudden drop in the edge buffer inventory level is closer to upstream material shortages or fluctuations in incoming materials; a decrease in quality rate and an increase in the fluctuation of rework process cycle time indicate that quality problems are dragging down the production line rhythm in the opposite direction; based on these different observations, the system will classify the judgment results into at least four categories: equipment attenuation disturbance, material supply disturbance, personnel efficiency disturbance, and random fluctuation; here, random fluctuation is used to accommodate short-term deviations that do not have continuity and propagation patterns, preventing all small fluctuations from being forcibly attributed to causes; To further illustrate this, a simplified logical verification model is constructed as follows: If the availability of the equipment remains stable within a certain time window, but the performance rate declines continuously, and the interval between the placement and AOI workstations gradually lengthens, then it is closer to equipment degradation. If availability and performance are relatively stable, but the inventory level in the front edge buffer pool of the plug-in module suddenly drops and there is a wait at the plug-in station, it is closer to a material supply disturbance; if the equipment indicators are generally normal, but the cycle time at the shift handover window fluctuates periodically, it can be classified as a personnel efficiency disturbance; if there is no consistent change in the three types of indicators, and only a few work orders are slightly delayed, it is classified as random fluctuation. As an exception handling mechanism, if a certain OEE subsequence cannot be obtained temporarily, such as when the quality inspection equipment is offline for a short period of time, the system can reduce the time series weight of the quality-related theoretical residuals within that time window, but will not stop the entire judgment process. If the inventory level of the edge buffer pool is abnormally collected, such as the counter jamming causing the work-in-process quantity to be distorted between processes, the system will use the station pass-through record as an auxiliary verification; if multiple signals are contradictory at the same time, such as the performance rate decreasing but the cycle time not increasing, the system will put the window into the pending confirmation state to avoid hastily outputting the clear disturbance type. For example, in the controller production line, the edge terminal continuously receives the availability, performance rate and quality rate slices of the pick-and-place machine, as well as the pick-and-place completion interval, AOI completion interval and test completion interval, and synchronously reads the edge buffer pool inventory level before AOI, before test and before final assembly. One afternoon, the system observed that the availability of the pick-and-place machine remained basically unchanged, but the performance rate continued to decline. The cycle time before and after AOI gradually slowed down, and the inventory level of the test front edge buffer pool began to drop. Therefore, it was classified as equipment degradation disturbance. On another occasion, it was found that the inventory level of the insertion front edge buffer pool suddenly approached zero, but the OEE on the equipment side did not decline significantly. Therefore, it was classified as material supply disturbance. Small fluctuations that occurred briefly during shift handover were classified as random fluctuations if they did not show continuous propagation characteristics. The purpose of this step is to provide multi-perspective evidence to support disturbance identification by defining the edge acquisition data in an industrially meaningful hierarchical manner, so that the judgment results can not only be used for automatic scheduling, but also be easy for manufacturing engineers to understand and verify.

[0036] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any conventional modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A manufacturing intelligent scheduling and production planning method based on AI edge computing terminals, characterized in that, include: Based on the work order BOM and standard process route SOP issued by ERP / MES, as well as the equipment OEE time sequence slice, workstation turnover cycle and edge buffer pool inventory level collected in real time by the edge computing terminal, a theoretical production scheduling benchmark time axis containing the theoretical completion time sequence of each process is generated. Based on the preset set of production disturbance parameters and their disturbance pattern mapping relationship, parameterized disturbance simulation is performed on the theoretical production scheduling reference time axis to generate a theoretical residual time series set; The actual completion time sequence is determined based on the actual workstation completion time collected by the edge computing terminal, and the difference between the actual completion time sequence and the theoretical completion time sequence in the theoretical production scheduling benchmark time axis is calculated to generate the actual residual time sequence. Dynamic time warping distance calculation and process route topology-based similarity calculation are used to compare the actual residual time series with the theoretical residual time series set. Combined with a preset judgment threshold, the disturbance type judgment result is determined, and the affected process and affected time window are located. Based on the disturbance type judgment result, the affected process, the affected time window, and the edge buffer pool inventory level, a local work order translation scheduling instruction is generated that acts on the work order subset associated with the affected process. The local work order shifting and scheduling instruction is executed, and the set of production disturbance parameters and the preset judgment threshold are dynamically calibrated according to the cycle recovery index and inventory stability index within the evaluation period. When comparing the actual residual time series with the theoretical residual time series set, the specific steps include: calculating the time morphology score and the topology propagation score respectively; and weighting the time morphology score and the topology propagation score by combining preset time morphology weights and topology propagation weights to determine the normalized target matching score. When the target matching score is higher than or equal to a preset high threshold, the disturbance type determination result corresponding to the theoretical residual time sequence that makes the target matching score meet the condition is output, and the local work order shift scheduling instruction is triggered; when the target matching score is lower than or equal to a preset low threshold, the random fluctuation determination result is output, and the local work order shift scheduling instruction is suppressed; when the target matching score is between the high threshold and the low threshold, the determination result to be confirmed is output, and the current production scheduling plan is maintained. The generation of the local work order shifting scheduling instruction includes: locating the affected process and the affected time window based on the disturbance type determination result; determining the set of shiftable work orders that meet the material availability constraint, process sequence constraint, and edge buffer pool inventory level constraint based on the edge buffer pool inventory level; within the affected time window, and only for the subset of work orders associated with the affected process, performing sequence adjustment and start / complete time reallocation on the set of shiftable work orders, and generating the local work order shifting scheduling instruction for locally updating the production plan.

2. The method according to claim 1, characterized in that, Generating the theoretical production scheduling baseline time axis includes: determining the material constraints of the work order based on the work order BOM; determining the sequence of operations and standard operating time based on the standard operating time, the theoretical capacity limit of the equipment, and the work order requirements, under material constraints and sequence of operations constraints; and constructing the theoretical production scheduling baseline time axis according to the theoretical start time and the theoretical completion time, as the time baseline data for production planning management.

3. The method according to claim 1, characterized in that, The set of production disturbance parameters includes equipment attenuation factor, material supply fluctuation factor, and personnel efficiency fluctuation factor; wherein, the equipment attenuation factor is used to characterize the exponential attenuation pattern of continuously weakening processing capacity over time; the material supply fluctuation factor is used to characterize the step pattern of sudden inventory drop at a predetermined time; and the personnel efficiency fluctuation factor is used to characterize the rhythm fluctuation pattern of changes with shift or personnel status cycle. The parameterized disturbance simulation specifically involves calling the parameterized simulation engine built into the edge computing terminal to parameterize and superimpose the equipment attenuation factor, the material supply fluctuation factor, and the personnel efficiency fluctuation factor to generate corresponding theoretical residual time series, which are used to establish the mapping relationship between disturbance type and plan deviation.

4. The method according to claim 1, characterized in that, The generation of the actual residual time series includes: determining the actual completion time series based on the actual workstation completion time collected by the edge computing terminal; performing difference calculation between the actual completion time series and the theoretical completion time series in the theoretical production scheduling benchmark time axis to generate the actual residual time series; wherein, each theoretical residual time series in the theoretical residual time series set is the difference result between the simulated completion time series and the theoretical completion time series.

5. The method according to claim 1, characterized in that, The dynamic calibration includes: calculating the cycle recovery index, inventory stability index, and misjudgment suppression index within the evaluation period; performing sensitivity analysis on the disturbance parameters in the production disturbance parameter set based on the cycle recovery index to determine the set of disturbance parameters to be calibrated; updating the judgment threshold based on the inventory stability index and the misjudgment suppression index; and writing the updated set of disturbance parameters to be calibrated and the judgment threshold into the edge computing terminal to complete the closed-loop calibration, thereby improving the adaptability of the production plan to the actual production environment.

6. The method according to claim 3, characterized in that, When the parametric simulation engine generates the theoretical residual time series set, it further includes: slicing the theoretical production scheduling baseline time axis according to a preset time window; injecting single disturbance factors and composite disturbance factors for each time window; attaching a disturbance type identifier to each injection result to form a correspondence between disturbance type and residual form; and outputting the theoretical residual time series set according to the correspondence, which is used as a reference template library for subsequent disturbance type matching.

7. The method according to claim 1, characterized in that, The equipment OEE time sequence slice includes availability subsequence, performance subsequence, and quality subsequence; the workstation turnover takt time is the process-level completion time interval sequence; the edge buffer pool inventory level is the inter-process work-in-process quantity sequence; the disturbance type determination result includes at least equipment attenuation disturbance, material supply disturbance, personnel efficiency disturbance, and random fluctuation.

Images