Production planning method based on industrial big data

Abstract

The application relates to the technical field of industrial manufacturing management, and discloses a production planning method based on industrial big data, which comprises the following steps: configuring a production line station as an autonomous node; each node acquires the physical accumulation degree of a work-in-process queue in real time through a state sensing device and generates a back pressure value; periodically broadcasting a data packet containing a node identifier and a back pressure value to an upstream; and dynamically distributing workpieces to a node with the lowest load according to the received downstream back pressure value. The application realizes autonomous balancing of a production line load through a decentralized back pressure value propagation mechanism, accurately maps the future time load of the back pressure value by combining standard working hour weighted calculation, and introduces a trend code to predict the state change of a node, so that a distributed production control system with real-time response and decision-making capability is formed.

Description

TECHNICAL FIELD

[0001] The application relates to a production planning method based on industrial big data, and belongs to the technical field of industrial manufacturing management. BACKGROUND

[0002] In the technical field of industrial manufacturing management, mainstream production planning methods rely on centralized scheduling systems, such as advanced planning and scheduling systems (APS), to generate static production plans through global data modeling. When dealing with high-frequency dynamic disturbances (such as urgent orders and material micro-delays), these systems need to perform global recalculation, resulting in a response delay of tens of minutes. During this period, the production line is in a state of instruction vacuum or relies on manual intervention, and the actual throughput efficiency is much lower than the theoretical peak value.

[0003] The more fundamental limitation is that existing technologies regard worker operation cycle differences, equipment micro-fluctuations and other microscopic uncertainties as noise filtering, rather than available resources. In typical scenarios, when a high-priority order is inserted: the central system is busy with re-planning, while the experienced workstations form a semi-finished product backlog due to operation speedup, and the new workstations cause upstream waiting due to slightly slower operation - this local capacity potential difference can be optimized immediately, but it is ignored by the system. The underlying contradiction lies in: pursuing theoretical optimization requires complex calculations, sacrificing agility; the second-level state changes in the physical field are not converted into scheduling basis; disturbances cause chain shutdowns and cannot adaptively relieve production flow.

[0004] The industry has tried to improve by enhancing data collection or optimizing algorithms, but this has increased system complexity and failed to break through the shackles of centralized calculation and decentralized execution. As the demand for flexible manufacturing upgrades, how to build a non-central coordination mechanism to convert microscopic disturbances into optimization potential in real time and achieve self-balancing of the production line has become a technical problem to be solved by the application. SUMMARY

[0005] The application provides a production planning method based on industrial big data, which mainly aims to solve the problem that centralized scheduling systems cannot respond to high-frequency disturbances in real time and utilize local dynamic potential to achieve self-balancing of the production line due to calculation lag and lack of microscopic perception.

[0006] To achieve the above purpose, the application provides a production planning method based on industrial big data, which comprises the following steps:

[0007] Step a, configure multiple workstations on the production line as independent nodes, and each node is configured with a state perception device to obtain the physical accumulation degree of the in-process product queue at the inlet in real time;

[0008] Step b, each node generates a back pressure value as a time accumulation value representing its instantaneous production load according to the physical accumulation degree of the work-in-process queue and the time weighting calculation rule stored in its internal memory according to the standard time value of different workpiece types;

[0009] Step c, each node periodically broadcasts a data packet to all its directly connected upstream nodes, the data packet contains the unique identifier of the node and the back pressure value generated by the node;

[0010] Step d, an upstream node located at the production line branch intersection makes workpiece branch decision by continuously listening to and receiving data packets broadcasted from all its downstream candidate nodes according to the back pressure values carried by each downstream candidate node in the data packet; the upstream node assigns the workpiece to be branched to the downstream node with the smallest back pressure value among all downstream candidate nodes.

[0011] Preferably, in step a, the state sensing device is an optical sensor or a basic vision sensor.

[0012] Preferably, in step b, the microcontroller of each node stores a query table of different workpiece types and their corresponding standard time values; and the back pressure value of the node is calculated in real time as the sum of the standard time values of all workpieces to be processed in the work-in-process queue of the node, and the back pressure value is mapped to an integer value from zero to two hundred and fifty-five.

[0013] Preferably, in step c, each node also attaches a trend code generated according to the trend of its own back pressure value within a preset time window when broadcasting its back pressure value; and in step d, when there are multiple downstream candidate nodes with the same minimum back pressure value, the upstream node preferentially directs the workpiece to the downstream candidate node whose trend code indicates a stable or decreasing trend.

[0014] Preferably, when the upstream node detects that there are multiple downstream candidate nodes with the same minimum back pressure value and the trend codes of the downstream candidate nodes all indicate a stable or decreasing trend, the upstream node performs the following steps before assigning the workpiece: broadcast a heartbeat probe packet to all downstream candidate nodes with the same minimum back pressure value and whose trend codes indicate a stable or decreasing trend; measure and compare the round trip time RTT of the signals returned by each downstream candidate node to confirm that the probe packet has been received; and direct the real workpiece to the downstream candidate node with the shortest round trip time RTT.

[0015] Preferably, in step a, the microcontroller of each node communicates with the upstream node through a wireless local area network within the workshop.

[0016] Preferably, when a node is unable to process a workpiece according to its configured preset rules, the node issues an abnormality indication signal and indicates the abnormality by illuminating the indicator light bound to its workstation and broadcasting the workpiece's identification identifier; in response to an authorized operator using a handheld teach pendant, the operator successively reads the workpiece's identification identifier and the identification identifier of the target node that can process the workpiece; the handheld teach pendant packages the association between the workpiece's identification identifier and the target node's identification identifier and broadcasts it; after receiving the broadcast, the target node automatically adds a new rule to its configured preset rules that allows it to process this type of workpiece.

[0017] Preferably, the handheld teaching pendant integrates a wireless communication module and RFID reading and writing functions, and completes the reading of the workpiece and target node identity identifiers and the broadcast of association information through two physical scanning actions.

[0018] Preferably, when making workpiece diversion decisions, if multiple downstream candidate nodes have the same minimum back pressure value and their trend codes all indicate stability or decline, the upstream node determines the workpiece allocation according to the following rule: if the absolute value of the difference between any two of the round-trip time (RTT) of all these downstream candidate nodes is less than or equal to a preset round-trip time difference threshold, that is, the following condition is met: in, Indicates the first Round-trip time of each downstream candidate node Indicates the first Round-trip time of each downstream candidate node If the preset round-trip time difference threshold is used, the upstream node will use a polling strategy to allocate workpieces.

[0019] Compared with the prior art, the beneficial effects of the present invention are:

[0020] 1. By nodeifying workstations and giving them autonomous state awareness, each node can capture the physical stacking status of the work-in-process queue in real time. Nodes generate back pressure values ​​based on standard time weighting rules, and then spread the load information upstream through a periodic broadcast mechanism. Upstream nodes located at the diversion point dynamically allocate workpieces according to the downstream back pressure value, forming a self-organizing guidance mechanism based on local information entropy. This mechanism eliminates the dependence on the central scheduling server, enabling the production line to achieve continuous micro-adjustment through autonomous collaboration between nodes when dealing with disturbances such as order fluctuations and equipment abnormalities, avoiding the response delay caused by traditional global replanning.

[0021] 2. Back pressure value calculation is not simply a matter of counting queue lengths. Instead, it uses the standard working hours corresponding to the workpiece type as a weighting factor, transforming the physical stacking degree into an estimate of future time load. This design allows the system to distinguish the actual load differences between a small number of complex workpieces and a large number of simple workpieces, avoiding misscheduling caused by differences in task complexity. Combined with a periodic broadcast mechanism, upstream nodes can accurately identify the real capacity bottleneck, rather than the apparent congestion, when making load allocation decisions, significantly improving resource matching accuracy. A trend code is added to the back pressure value broadcast. Through simplified time-series analysis of historical load data, the node status is upgraded from an instantaneous snapshot to an information carrier carrying the trend direction. When multiple downstream nodes have the same back pressure value, the upstream prioritizes nodes with stable or decreasing loads, proactively avoiding paths that are about to become overloaded. This design injects an implicit damping effect into the distributed system, effectively suppressing chain congestion caused by fluctuations in task allocation and improving the stability of production line operation.

[0022] 3. When back pressure values ​​and trend codes still cannot determine the outcome, a heartbeat detection packet mechanism is introduced. Upstream nodes indirectly obtain comprehensive health indicators such as node computing load and communication latency by measuring the response delay of downstream nodes to the detection packets. This mechanism reuses existing communication links to predict hidden risks such as equipment aging and software lag at near-zero cost, upgrading the diversion decision from avoiding current congestion to avoiding potentially failed nodes, significantly reducing the risk of abnormal downtime. When a node cannot handle a new type of workpiece, an abnormal response is triggered by binding the physical indicator light of the workstation and RFID scanning action. After the authorized operator associates the workpiece with the target node using a teach pendant, the new rules are updated in a distributed manner in the form of broadcast. This design transforms worker experience into structured knowledge that can be executed by the machine, achieving three key breakthroughs: rule updates and production line operation are carried out concurrently; physical scanning actions replace complex programming interfaces; and new rules are synchronized to the entire system through broadcast, enabling the production line to obtain open adaptability and avoiding the lack of flexibility caused by rigid rules in traditional solutions. Attached Figure Description

[0023] Figure 1 This is a flowchart of the multi-level arbitration decision-making process for the diversion node in this invention;

[0024] Figure 2 This is a performance comparison chart between the method of this invention and a traditional APS system;

[0025] Figure 3 This is a timing diagram for the unknown workpiece processing and new rule learning of the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with specific embodiments. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] The production planning method based on industrial big data disclosed in this invention abandons the traditional centralized server scheduling model and instead constructs a distributed adaptive control system. The core of this system lies in transforming each workstation on the production line into a functionally autonomous intelligent node. Each node is given independent real-time status perception, load calculation, and information communication capabilities. The system establishes a dynamic balancing procedure based on local information interaction, enabling autonomous optimization of the production process. Specifically, each node accurately perceives the physical backlog of work-in-process at its entry point and quantifies it into a back pressure value that characterizes future time load. This load information is then propagated upwards according to a decentralized broadcast protocol. Finally, the upstream node at the production line's distribution point autonomously and instantly executes the optimal workpiece assignment decision based on the latest load status continuously received from downstream nodes. This addresses a typical situation requiring frequent switching between multiple product models and high-frequency emergency order insertion disturbances. In a flexible electronics assembly workshop, the specific procedures for deploying this invention are as follows: First, each key workstation on the production line, such as a component placement station, wave soldering oven, or functional testing station, is configured as an independently operating node. The core of the node is an embedded microcontroller, and a state sensing device is precisely installed at the physical buffer queue at its material inlet. This device is preferably a high-frequency infrared photoelectric sensor, whose installation position and detection angle are calibrated offline to enable it to accurately detect the physical stacking height or extension length of the work-in-process queue in the vertical or horizontal direction. This transforms the congestion state of the physical world in real time into a continuous analog voltage signal or discrete digital quantity that can be processed by the microcontroller, providing the most original and delay-free physical input for subsequent accurate load calculations. The conversion of the physical stacking degree of the work-in-process queue obtained by the state sensing device into the number of discrete workpieces is achieved through a multi-dimensional mapping table pre-calibrated during the node deployment phase. To achieve this, the construction procedure for this mapping table is as follows: for each type of unique identifier... The workpiece types are distinguished, and the quantities are sequentially... (from Increment to physical capacity limit The workpieces are placed in the queue, and each increment is recorded. The corresponding stable sensor raw output value This process will pair all the pairs. Populate to mapping table In this way, in actual operation, when the upstream diversion node confirms that the type of workpiece about to enter the queue is... Then, the microcontroller at the downstream node can use the real-time readings from the current sensor. By querying this table And find the conditions that are met. The entries are used to unambiguously resolve the number of workpieces in the queue. This mechanism ensures that the differences in sensor signals generated by workpieces of different physical sizes can be accurately compensated when calculating queue length.

[0028] Given that queue length alone cannot measure the differences in process complexity between different workpieces, leading to serious misjudgments of the actual load, each node's microcontroller is configured to execute a backpressure value generation algorithm based on standard time weighting. A lookup table is pre-stored in the node's non-volatile memory, establishing a precise mapping between the unique identifier of different workpiece types and their corresponding standard time values. These standard time values ​​are authoritative benchmark data determined and input by engineers during the production line process verification phase. When the state sensing device detects any change in the physical state of the work-in-process queue, the microcontroller immediately triggers a calculation process, polling all workpiece identifiers in the queue, retrieving and accumulating their respective identifiers from the lookup table. The sum of the standard working hours represents the instantaneous production load, or the original back pressure value, representing the total time required for the node to complete all current tasks. To adapt to the low bandwidth and high reliability requirements of industrial wireless networks, this cumulative time value is linearly quantized and mapped to an unsigned 8-bit integer value from zero to 255 before broadcasting. This final back pressure value accurately translates the physical stacking degree into a quantitative prediction of future time occupancy, allowing the system's decision-making basis to leap from the apparent queue length to the actual processing time. To achieve complete decentralization and eliminate the inherent communication bottlenecks and computational delays of the central scheduling system, each node follows a back pressure-driven distributed communication protocol. The microcontroller within each node communicates... Through its integrated wireless communication module, such as Wi-Fi or Zigbee, it accesses the workshop's wireless LAN and broadcasts a structured data packet at a stable, configurable frequency, for example, twice per second, to all its physically or logically directly connected upstream nodes. The payload of this data packet is designed to be extremely concise, containing only two core fields: a unique identifier for the broadcasting node and the node's latest generated 8-bit integer backvoltage value. Correspondingly, an upstream node located at a production line branch point, for example needing to decide which of three parallel test benches to send a circuit board under test to, will continuously listen to and receive data packets broadcast from these three downstream candidate nodes, and update its internally maintained data in real time based on the received information. A downstream node status table; when a new workpiece arrives at the distribution node waiting for assignment, it does not need to request instructions from any superior system, but directly assigns the workpiece to the downstream candidate node with the smallest back pressure value recorded in its status table. In this way, a self-organizing control loop that is completely driven by real-time data and achieves dynamic balance of production flow through autonomous collaboration of each node is formed locally in the production line. In other words, in order to suppress the load oscillation that may be caused by the inherent small decision delay or information noise, that is, to avoid the workpiece being continuously sent to a node whose back pressure value is temporarily the lowest but whose load is accumulating rapidly, this invention adds a predictive stabilization mechanism to the communication protocol.When broadcasting its backpressure value, each node also encapsulates a trend code generated based on the changing trend of its own backpressure value within a preset time window; the algorithm for generating this trend code is embedded in the microcontroller, and its procedure is as follows: the microcontroller cyclically caches its own most recent value; Back pressure value per broadcast cycle For configurable parameters, such as By calculating the first difference or linear regression slope of the time series, the result is compared with two asymmetric preset thresholds. and For comparison, if the slope is consistently higher than The trend code is set to upward; if it remains below... If the back pressure value is low, it is set to decrease; otherwise, it is considered stable. When an upstream node makes a diversion decision and detects that the back pressure values ​​of multiple downstream candidate nodes are the same as the current minimum, its internal diversion logic will automatically switch to the second priority, that is, prioritize guiding the workpiece to the downstream candidate node indicated by the trend code as stable or decreasing. This design injects effective predictability and damping effect into the entire distributed system, which can actively avoid the congestion points that are about to be formed and significantly improve the smoothness of the overall operation of the production line.

[0029] In rare cases, when the back pressure values ​​and trend codes of multiple downstream nodes remain identical, the system will automatically trigger a final arbitration mechanism based on network health detection to make the ultimate and undisputed optimal decision. In this situation, the upstream node will temporarily suspend workpiece allocation and instead broadcast a lightweight heartbeat probe packet in parallel to all downstream candidate nodes with identical conditions. It will also activate its internal high-precision timer to measure the round-trip time required from sending the probe packet to receiving acknowledgment signals from each node. ; As a comprehensive indicator, its value not only reflects the real-time communication latency of the wireless network, but also indirectly reveals the current computing load of the downstream node microcontroller. This is because a node on the verge of lag due to performing complex internal calculations will inevitably respond to external communication requests at an abnormally slow speed. Therefore, the upstream node will direct the actual work to the measured value. The shortest downstream candidate node is selected, thus deepening the decision-making basis from the macro-level state of production load to the micro-level state of node equipment operational health. It should be noted that, to prevent unnecessary decision-making fluctuations due to normal network jitter, this arbitration logic also integrates an activation condition for a polling strategy. The procedure is as follows: before assigning a workpiece, first check all candidate nodes... The value is determined only if the absolute value of the difference between any two satisfies the condition. Only then will the above be executed. The strategy of prioritizing the shortest option, or conversely, if all The absolute values ​​of the differences are all less than or equal to the preset round-trip time difference threshold. If the system determines that there is no significant difference in their network and device health, the upstream node will adopt a simple polling strategy to allocate tasks in order to achieve a more balanced long-term wear and task allocation.

[0030] Finally, to enable the production line to adapt to unknown workpieces or sudden process changes, this invention integrates a seamless human-machine collaborative rule-based self-evolving workflow. When a node cannot process an unknown workpiece due to a missing rule base, it automatically enters a preset abnormal state, manifested by illuminating a bright red indicator light physically bound to its workstation and continuously broadcasting an abnormal state message containing the workpiece's identification identifier on the network. At this time, any production line operator authorized by the system can intervene using a handheld teach pendant integrated with a wireless communication module and RFID read / write function. The operating procedure is designed to be extremely intuitive: the operator first uses the teach pendant to perform a physical scan of the stuck abnormal workpiece to read the unique identification identifier stored in its RFID tag. Then, based on their professional knowledge and experience, the operator moves to the target node they deem most suitable for processing the workpiece and performs a second physical scan of the RFID identification identifier fixed at the target node's workstation. After capturing these two identification identifiers, the handheld teach pendant automatically matches the workpiece identification identifier with the target node's identification identifier locally. The new rule association data packet is encapsulated into a structured data packet and immediately broadcast to the entire network. All nodes in the network, especially the target node designated as the executor of the new rule, will automatically parse the data packet content upon receiving the broadcast and add a permanent or temporary rule to their local rule base, thus formally gaining the authorization and capability to process this type of workpiece. This mechanism instantly and error-free transforms the valuable on-site experience and tacit knowledge of operators into structured data that the machine can strictly execute, enabling the production line to dynamically learn and adapt to new production tasks without interruption or the need for any software engineers to program. To ensure data consistency and final state convergence of the distributed system rule base when multiple authorized operators use handheld teach pendants for concurrent intervention, this invention integrates a conflict resolution procedure based on event timestamps. This procedure stipulates that each handheld teach pendant, when generating and broadcasting an association data packet containing workpiece and target node identifiers, must encapsulate a unique event timestamp generated by a local high-precision clock within the data packet. Its physical meaning is the absolute moment the rule is created; any node in the network, after receiving this rule broadcast, must execute the deterministic logic of subsequent overriding, that is, the node queries its locally stored timestamps of existing rules for that workpiece type. Only when the incoming Strictly greater than Only then will the node accept and atomically update its local rule base. Otherwise, the data packet will be silently discarded; furthermore, the target node that successfully updates the rules will immediately broadcast a message containing the artifact's identity and new information to the entire network. The rule confirmation message provides a traceable final consistency proof for the upper-level monitoring system and forms the technical foundation for achieving highly reliable and serialized execution of industrial field control commands in complex electromagnetic environments and multi-entity concurrent operations.

[0031] Simultaneously, in the extended implementation, a backend data monitoring and management system can be added. This system utilizes the existing wireless LAN in the workshop to asynchronously collect key data generated by each autonomous node. This includes the node's real-time load (backpressure value and trend code), the actual processing time of the workpiece, the equipment network health (RTT), and system event logs such as unknown workpiece handling and new rule creation. Specifically, this backend system mainly includes four functional modules: a real-time production line monitoring center: transforming the dispersed production line status into a centralized, visualized heatmap through a graphical dashboard, facilitating managers to quickly identify operational bottlenecks; a production performance analysis module: analyzing historical data... The system automatically generates reports on key performance indicators (KPIs) such as system throughput, order lead time, and load balancing, providing quantitative basis for continuous improvement. The equipment health and early warning module tracks the changing trends of RTT values ​​at each node over a long period, enabling predictive maintenance and automatic early warning for potential computing power or network problems. The process adaptation and knowledge base management module automatically compares standard working hours with actual working hours and prompts for calibration, forming a data-driven process verification closed loop. Simultaneously, it centrally manages production rules created by operators through handheld devices, gradually building a traceable and evolving enterprise production knowledge base. These are all extended implementation methods known to those skilled in the art.

[0032] Example 1: This example demonstrates the specific operation of the described technical solution in a particular industrial application scenario. It aims to reveal the synergistic effect and underlying logic among the various technical features of the solution when dealing with extreme dynamic disturbances. In a continuously operating flexible manufacturing unit for automotive electronic controllers, the unit needs to process orders for multiple conventional models simultaneously. Its production line has achieved stable high-load operation by deploying the distributed production planning method of this invention. At a specific moment, the unit receives an urgent order with the highest priority, requiring the delivery of a batch of brand-new ECU products within a limited time. The process complexity of this product is significantly higher than all conventional models. While the workpieces can flow into the same physical production line, the standard time values ​​for multiple processes have increased significantly. For any traditional production system relying on static planning, this event usually means suspending existing production or performing a global recalculation and production line rebalancing, easily leading to widespread production stagnation and instruction confusion. When the first batch of urgent order workpieces is put into the production line, the system's adaptive adjustment mechanism is immediately triggered. At the first diversion point, three parallel surface mount processing nodes A, B, C, and D are connected downstream. Before the arrival of the emergency task, the back pressure values ​​of nodes B and C remained at similarly low levels. When node A received and began processing the first emergency task, its back pressure value generation procedure, based on standard time-weighted calculations, immediately included the longer standard time value corresponding to this high-complexity task in its total load. This caused a significant jump in its back pressure value reading compared to processing regular tasks. The accuracy of this signal is a prerequisite for the effective execution of all subsequent decisions. Since nodes B and C were still processing regular tasks, their back pressure values ​​remained stable. Therefore, the upstream diversion nodes, based on their assignments... The deterministic procedure of the node with the minimum back pressure value guides the subsequently arriving workpieces sequentially to nodes B and C, thereby facilitating the use of parallel production capacity to absorb the impact instantaneously at the first moment of disturbance. Here, the back pressure value calculation mechanism based on standard working hours and the distributed decision-making mechanism based on back pressure value broadcasting produce a key synergistic effect. The former ensures the authenticity and comparability of the load signal, while the latter ensures the response speed based on this high-fidelity signal, enabling the system to complete an effective load balancing within microseconds without waiting for the intervention of any central system.

[0033] Furthermore, in this scenario, the system architecture of this invention reconciles the seemingly contradictory goals of global static optimization and local real-time response in the field of industrial production scheduling through its inherent design. Traditional advanced planning and scheduling systems, in pursuit of theoretical global optimization, must freeze all states within a time slice during their calculation process, thus losing the ability to respond to high-frequency disturbances within the calculation cycle. This invention, however, completely delegates decision-making power to each local node. The system no longer pursues a time-consuming and static theoretical optimal solution, but instead achieves a dynamic and continuously approaching equilibrium state. When a node D on the production line completes its current task ahead of schedule due to operator skill or accidental exceptional performance of the equipment, its work-in-process queue is quickly cleared, and its state-aware device... The system immediately detects this change and updates its back pressure value to an extremely low value. This back pressure value, containing information about the new available capacity, is broadcast, and the upstream distribution node begins directing new workpieces to node D. In this way, the system directly transforms microscopic random fluctuations, typically considered uncontrollable noise in traditional production management, into optimization opportunities that can be captured and utilized by the entire system in real time. It doesn't attempt to plan or predict these fluctuations, but rather, through its underlying architecture design, allows it to naturally and seamlessly benefit from them when they occur. The essence of this operating mode is a restructuring of the core issue in traditional production management—how to make the production line conform to the plan—instead focusing on how to make the plan continuously learn from and adapt to the production line. As urgent orders arrive on the production line… As the internal flow continues, a deeper stabilization mechanism begins to emerge. At a certain moment, the backpressure values ​​of downstream nodes E and F become identical due to processing similar tasks. At this point, the upstream branch node activates its trend code judgment during decision-making. If the backpressure value of node E has shown a decreasing trend over the past few cycles, while the backpressure value of node F has shown a stable trend, the task will be assigned to node E according to the arbitration rule that prioritizes nodes with stable or decreasing trend codes. This design, with its added predictive stabilization mechanism, upgrades the system's decision-making basis from a snapshot of the current load to dynamic information about the current load and its change vector. This allows the system to not only avoid existing congestion but also proactively avoid impending congestion, thereby effectively suppressing congestion caused by task-related factors. Uneven task allocation could cause production line load fluctuations, but this ensured the smooth operation of the entire production line even under extreme conditions of handling tasks of varying complexity. Ultimately, the emergency order for this batch was integrated into the existing production flow without causing a command vacuum or large-scale shutdown on the production line. Its delivery cycle was completed within the specified time, and the production of regular models was not interrupted unplanned. The entire process of this embodiment reveals a core architectural principle of the present invention: instead of attempting to forcibly plan and control a complex system through an omniscient central unit, it empowers each basic unit in the system with interaction rules based on local information and determinism, thereby fostering a system behavior that can calmly cope with external uncertainties and operate efficiently.

[0034] Example 2: To quantitatively verify the actual effectiveness of the technical solution of this invention in dealing with dynamic disturbances and to compare its performance with that of the traditional centralized scheduling method, a comparative verification experiment described in this example was designed and executed. The core purpose of this experiment is to evaluate the specific effectiveness of the distributed production planning method proposed in this invention in improving production line throughput efficiency and enhancing system responsiveness, especially under conditions of sudden high load shocks. To this end, a discrete event simulation test platform was built. This platform reproduces a flexible manufacturing cell in a software environment. Its topology includes a workpiece inlet, a main conveyor belt, and a... The platform includes a standard branching node that splits the main path into three parallel processing nodes, and a confluence point that merges three branch paths into a single exit. Two workpiece types with different process complexities are set up in the platform: Type A workpieces with shorter standard time values ​​and Type B workpieces with longer standard time values. A disturbance generator is also included to simulate emergency order insertion scenarios in real production. As a control group, the platform runs in parallel a control model that uses a traditional Advanced Planning and Scheduling (APS) system for centralized scheduling. This APS system is set to collect full production line data and perform a global plan recalculation every sixty seconds; this cycle represents the inherent response latency of such systems.

[0035] Regarding the setting of experimental parameters, the core parameter of the present invention, namely the period of the broadcast back pressure value of each node, was calibrated. The setting of this parameter requires a technical trade-off between the real-time nature of decision information and the processing load of the system network and nodes. An excessively long broadcast period will lead to outdated information in the diversion decision, which may cause workpiece assignment errors, while an excessively short period will unnecessarily increase the computational burden on the communication bus and node microcontrollers. Therefore, the decision rule adopted is that the broadcast period should be significantly shorter than the shortest standard working time value of any workpiece on the production line, so as to ensure that the change in the load of downstream nodes caused by a workpiece can be perceived by the upstream at least once within the processing cycle of a workpiece. Based on this rule, and considering that the shortest standard working time value of type A workpiece in this experiment is five seconds, we set the broadcast period to a non-limiting example value of 500 milliseconds to focus on examining the real-time response capability of the scheme in this experiment.

[0036] The experiment ran for a total of eight simulation hours, divided into three phases: the initial two hours were a stable operation period, during which the production line processed type A and type B workpieces according to a preset mixing ratio; at 0:00 on the second hour, the disturbance generator instantaneously injected one hundred type B workpieces into the production line entrance, creating a disturbance impact; thereafter, the system continued to run for six hours as a recovery and continuous observation period; during the experiment, the behavior of the two models showed essential differences. In the control group, when a batch of type B workpieces arrived at the diversion node, the physical queue in front of the diversion node accumulated rapidly because the APS system was in its sixty-second calculation and decision-making cycle, and the congestion only began to ease after the next scheduling instruction was issued; in the experimental group of the present invention, the load pressure brought by the emergency workpieces was rapidly transmitted upward through high-frequency back pressure value broadcasting, and the diversion node dynamically allocated the workpieces to the parallel node with the lowest back pressure value at that time in almost real time, without forming an observable continuous backlog on the main conveyor belt; Table 1 shows the comparative data of some performance indicators of the two methods one hour after the disturbance occurred (T=3h) and at the end of the experiment (T=8h), see Table 1.

[0037] Table 1: Performance Index Comparison and Analysis Table.

[0038]

[0039] Given that the simulation topology, workpiece combinations, and disturbance events of the two models were identical in the experiment, the performance differences shown in Table 1 can necessarily be attributed to the inherent differences between the two scheduling architectures. The total throughput of the proposed solution reached 352 pieces per hour one hour after the disturbance, significantly higher than the 210 pieces per hour of the control group, and remained at a higher level at the end of the experiment. This indicates that its continuous fine-tuning capability avoids production line stoppages caused by global recalculation. The average lead time for emergency orders was shortened by more than 60%, and the maximum queue length of the diversion nodes remained at an extremely low level. The underlying mechanism is that the distributed backpressure mechanism of the proposed solution decentralizes the decision-making power of load distribution to the local level, making each workpiece allocation a near-instantaneous decision based on the latest on-site information, thereby avoiding the formation of upstream bottlenecks. In addition, the lower standard deviation of the parallel node load also indicates the effectiveness of the method in dynamically maintaining production line balance.

[0040] Example 3: This example combines Figures 1 to 3 This paper describes the implementation of a production planning method based on industrial big data, such as... Figure 1 As shown, Figure 1 This demonstrates how a distribution node located in the decision center, when faced with multiple downstream nodes A, B, and C, assigns workpieces using a four-level decision logic. In the first-level back pressure value comparison, the distribution node obtains the back pressure values ​​of each downstream node. And its changing trend, as shown in the figure, the relationship between node A and node C The values ​​are all 45, which is less than 128 for node B, so there are multiple minimum values. Therefore, the comparison proceeds to the second-level trend code comparison. In the second-level arbitration, only nodes A and C, whose back pressure values ​​are both the smallest, are compared. Since the trend of node A is downward (i.e.,...),... Figure 1 "in The trend of node C is stable (i.e., Figure 1 "in Both are considered priority paths, and the decision is not unique, so a third-level RTT probe is initiated; in the third-level probe, the splitter node sends probe packets to node A and node C and measures their round-trip time. The measured value of node A for Node C for The absolute value of the difference between the two This value is greater than the preset threshold. Therefore, the system selects The shorter node A is selected as the target node for the workpiece; the figure also shows a fourth-level alternative strategy, namely the polling strategy, which considers all candidate nodes in the third-level probing. The difference is less than or equal to When enabled, the system will perform cyclic allocation according to the preset node list order (such as A→B→C→A…).

[0041] like Figure 2 As shown, the graph is based on time. The graph plots the performance curves of the two methods over eight hours, with the horizontal axis representing the present invention and the vertical axis representing the traditional APS system. At the second hour, a disturbance event was applied, causing the throughput of the traditional APS system to plummet from nearly 300 units / hour to below 200 units / hour. Although it recovered slowly, it did not reach its initial level by the end of the test. In contrast, the present invention's throughput only fluctuated slightly during the disturbance and quickly climbed to over 350 units / hour, maintaining a stable high level thereafter. This graph clearly demonstrates the significant superiority of the present invention in terms of response speed and maintaining stable production efficiency when dealing with dynamic disturbances.

[0042] like Figure 3As shown in the diagram, this illustrates the complete processing flow starting from the arrival of an unknown type of workpiece at node 1: When the unknown workpiece arrives, node 1 enters an abnormal state due to the lack of results in the rule base query, illuminates its physical indicator light, and broadcasts an abnormal message; after an operator discovers the abnormality, they use a handheld teach pendant to sequentially scan the RFID tag of the unknown workpiece and the RFID tag of the target node 2, which they determine can handle the workpiece. The handheld teach pendant generates a new association rule based on this and broadcasts the rule data packet to all nodes in the network; upon receiving the broadcast, node 2, as the target node, gains the new permission to handle this type of workpiece and updates its local rule base. All other nodes in the network, including node 1, also synchronously update this new rule, thus enabling the entire production system to dynamically adapt to new production tasks without downtime or manual programming.

[0043] Example 4: In a specific system deployment and debugging application, to ensure the reliable operation of the production planning method of the present invention on a brand-new automotive dashboard assembly line, it is necessary to accurately set the multi-level arbitration logic within its shunting nodes, especially the key judgment thresholds in trend code judgment and round-trip time comparison. This requires a systematic and data-driven parameter calibration procedure to ensure the system's performance after actual production, used to determine the relevant parameters for trend code generation, including the number of historical data points. The threshold for determining whether a trend is upward or downward. and The procedure is as follows: First, the production line is operated in a baseline mode. In this mode, the autonomous decision-making logic of all distribution nodes is temporarily disabled, and workpieces are allocated in a fixed or simple cyclic manner. Simultaneously, each production node records and stores its own back pressure value time-series data at a high frequency. After data acquisition, the back pressure value time-series data recorded by each node is analyzed, and its first-order difference sequence is calculated. This difference sequence reflects the instantaneous rate of change of the back pressure value. Subsequently, the statistical standard deviation of this difference sequence is calculated. The rules for setting the threshold are as follows: and Any absolute value of the rate of change of back pressure exceeding this range is considered a statistically significant trend caused by actual changes in production load; number of historical data points. This is determined by analyzing the autocorrelation function of the back pressure signal, selecting a value that causes the autocorrelation function to first drop to a specific low value (e.g., The number of data points corresponding to the time delay is used to ensure that the analysis window can cover a complete typical load fluctuation cycle, and is used to set the round-trip time difference threshold. The polling strategy implementation mechanism is defined as follows: Each branch node continuously broadcasts a specific number of heartbeat probe packets to all its downstream candidate nodes within a specified time, and accurately records each successful response. Value; after the detection is completed, the values ​​collected for each upstream-downstream link are... Analyze the sample values ​​and calculate their statistical standard deviation. ; The value selection rule is to take all downstream links. The largest result of multiplying the standard deviation by a coefficient greater than three (such as five times), i.e. This ensures that only when two nodes... The system only activates the shortest time when the difference significantly exceeds the range of high-probability random network fluctuations. Priority assignment logic: For the implementation of the polling strategy, each distribution node broadcasts an identity query request to its downstream nodes during initialization. Based on the received response, it internally constructs a static list containing unique identifiers of all downstream candidate nodes in a fixed order. At the same time, it initializes an internal pointer pointing to the first element of the list. Whenever the polling strategy is triggered, the node assigns the workpiece to the downstream node currently pointed to by the pointer. After the assignment is completed, the pointer is moved one position to the right. If the pointer has reached the end of the list, it wraps back to the beginning of the list, forming a deterministic and unambiguous circular assignment.

[0044] During the deployment and debugging phase of this invention, procedural guidelines are provided to ensure the determinism of all values, including the back pressure value mapping rules and round-trip time difference thresholds in the decision-making logic of the diversion node. The following procedures are followed for calibration: First, the system is based on the maximum design capacity of the node ingress physical buffer. Compared with the longest standard working time value of a single workpiece stored in the query table Calculate the theoretical maximum time load of the node. , it is and The product of these, and then any node broadcasts its instantaneous production load. Previously, it was unambiguously converted into an 8-bit integer backpressure value using a linear function that rounds down. The specific form of this function is: , and when Exceed hour, The value is constant at 255; secondly, the round-trip time difference threshold. The value of is derived from the statistical analysis of the RTT sample set collected in no less than one hundred heartbeat detection cycles for each upstream-downstream link, by calculating the statistical standard deviation of the RTT sample set for each link. A physical quantity characterizing the discreteness of normal network communication delays, which will ultimately... Set to five times the maximum standard deviation across all downstream links, i.e. By executing the complete offline calibration procedure described above, several core algorithm parameters that previously required experience or trial and error to determine were all set using a deterministic process. After all parameters were calibrated, the production planning system for the automotive dashboard assembly line was placed in an operating state that was locally calibrated and highly adapted to the dynamic characteristics of the current production line, enabling it to perform dynamic load balancing based on real-time back pressure values ​​and handle various complex boundary and congestion scenarios based on precise arbitration logic.

[0045] Example 5: To ensure the long-term stability and data accuracy of the method of the present invention, its deployment and maintenance include procedures for the initial construction of the core data table and subsequent adaptive calibration. When the system is first deployed on a production line, its internal storage and lookup tables for different workpiece types and their corresponding standard working hours, which serve as the basis for back pressure value calculation, are populated with data through the following standardized procedure: For each type of workpiece to be processed on the production line, a statistically representative sample batch is taken and continuously processed at a designated standard station. The timing system automatically records the actual time consumed by each sample in the batch from entering the station to completion of processing. After all sample data are collected, statistical outliers are removed, and the average processing time of the remaining valid samples is calculated. This average value is determined as the initial standard working hour value for this type of workpiece and is fixed in the lookup tables of all relevant nodes in the entire production line. This procedure ensures that the basic data on which the system operates is traceable and its values ​​are obtained through objective measurement and statistics.

[0046] To address the potential long-term, gradual deviation between actual processing time and initial standard time values ​​caused by equipment wear or fine-tuning of process parameters, the system also includes a periodic data validity verification and fine-tuning mechanism. Each production node, while performing its main function, is configured to record the actual processing time for each workpiece it processes in the background, and store this data after associating it with the workpiece type identifier. The system automatically executes a data comparison program at a user-defined cycle. This program calculates the average actual processing time for each workpiece type over the past cycle and compares it with the standard time value stored in the lookup table. If the relative deviation exceeds a preset allowable fluctuation threshold, the system automatically generates a maintenance alert, prompting engineering technicians to review and update the standard time value for that workpiece.

[0047] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0048] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A production planning method based on industrial big data, characterized in that, Includes the following steps: Step a: Configure multiple workstations on the production line as independently operating nodes. Each node is equipped with a status sensing device to obtain the physical stacking degree of the work-in-process queue at its own inlet in real time. Step b: Each node generates a cumulative time value as a back pressure value, which represents the instantaneous production load of the node, based on the physical stacking degree of the work-in-process queue and the time weighting calculation rule determined by the standard working time value of different workpiece types stored in its internal memory. Step c: Each node periodically broadcasts a data packet to all its directly connected upstream nodes, the data packet containing the node's unique identifier and its currently generated backpressure value; Step d: An upstream node located at the production line diversion point continuously listens to and receives data packets broadcast from all its downstream candidate nodes, and makes a workpiece diversion decision based on the back pressure value carried by each downstream candidate node in the data packet; the upstream node assigns the workpiece to be diverted to the downstream node with the smallest back pressure value among all downstream candidate nodes. In step c, each node, when broadcasting its back pressure value, also attaches a trend code generated based on the changing trend of its own back pressure value within a preset time window; and in step d, when multiple downstream candidate nodes have the same minimum back pressure value, the upstream node prioritizes guiding the workpiece to the downstream candidate node indicated by the trend code as stable or declining. When an upstream node detects that multiple downstream candidate nodes have the same minimum back pressure value and that the trend codes of the downstream candidate nodes all indicate that the trend is stable or declining, the upstream node performs the following steps before assigning a workpiece: broadcast a heartbeat detection packet to all downstream candidate nodes with the same minimum back pressure value and trend codes indicating that the trend is stable or declining; measure and compare the round-trip time (RTT) of the signals confirming receipt of the detection packet from each downstream candidate node. And guide the actual workpiece to the downstream candidate node with the shortest round-trip time (RTT).

2. A production planning method based on industrial big data according to claim 1, characterized in that, In step a, the state sensing device is a photoelectric sensor or a basic vision sensor.

3. A production planning method based on industrial big data according to claim 1, characterized in that, In step b, each node's microcontroller stores a lookup table of different workpiece types and their corresponding standard working hours; and the back pressure value of the node is calculated in real time as the sum of the standard working hours of all workpieces to be processed in its work-in-process queue. The back pressure value is mapped to an integer value from zero to two hundred and fifty-five.

4. A production planning method based on industrial big data according to claim 1, characterized in that, In step a, the microcontroller configured on each node communicates with the upstream node via the wireless local area network inside the workshop.

5. A production planning method based on industrial big data according to claim 1, characterized in that, When a node is unable to process a workpiece according to its configured preset rules, the node issues an abnormality indication signal and indicates the abnormality by illuminating the indicator light associated with its workstation and broadcasting the workpiece's identification identifier. In response to an authorized operator using a handheld teach pendant, the operator successively reads the workpiece's identification identifier and the identification identifier of the target node that can process the workpiece. The handheld teach pendant packages the association between the workpiece's identification identifier and the target node's identification identifier and broadcasts it. Upon receiving the broadcast, the target node automatically adds a new rule to its configured preset rules that allows it to process this type of workpiece.

6. A production planning method based on industrial big data according to claim 5, characterized in that, The handheld teach pendant integrates a wireless communication module and RFID reading and writing functions, and completes the reading of the workpiece and target node identification and broadcasting of association information through two physical scanning actions.

7. A production planning method based on industrial big data according to claim 1, characterized in that, When an upstream node makes a workpiece allocation decision, if multiple downstream candidate nodes have the same minimum back pressure value and their trend codes all indicate stability or decline, then the upstream node decides on the workpiece allocation according to the following rule: If the absolute value of the difference between any two of the round-trip times (RTT) of all these downstream candidate nodes is less than or equal to a preset round-trip time difference threshold, that is, the following condition is met: in, Indicates the first Round-trip time of each downstream candidate node Indicates the first Round-trip time of each downstream candidate node If the preset round-trip time difference threshold is used, the upstream node will use a polling strategy to allocate workpieces.

Images