Intelligent management and control system, method, equipment and medium for flexible processing line of cylinder head

Abstract

This invention discloses an intelligent control system, method, equipment, and medium for a flexible cylinder head manufacturing production line, relating to the field of intelligent manufacturing technology. It includes: a dynamic topology network construction module, used to extract real-time operating status parameters of the production line from the digital twin production line model and construct a dynamic topology network for the production line; and a dynamic bottleneck diffusion identification module, used to calculate the load entropy of each workstation node in the dynamic topology network of the production line and determine the bottleneck diffusion domain based on the inter-node coupling diffusion strength matrix. This invention, through a production line control instruction generation and closed-loop execution module, transforms optimization schemes into process path adjustment instructions and scheduling task rearrangement instructions, and issues them to the production line execution layer. It collects execution feedback data in real time and dynamically triggers a new round of optimization based on the bottleneck diffusion domain coverage, forming a closed-loop adaptive control mechanism that ensures the continuous and efficient operation of the production line under dynamic disturbances.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing technology, and in particular to an intelligent control system, method, equipment and medium for cylinder head flexible processing production lines. Background Technology

[0002] As the manufacturing industry rapidly moves towards intelligence and flexibility, the cylinder head, as a key core component of engines, faces increasingly higher demands for precision, efficiency, and production flexibility in its processing. Traditional cylinder head manufacturing lines typically employ rigid assembly line structures with fixed process paths and simple scheduling methods, making them ill-suited for multi-variety, small-batch production models. In recent years, the gradual application of Flexible Manufacturing Systems (FMS) and intelligent manufacturing technologies has enabled cylinder head manufacturing lines to possess multiple process path selection capabilities, flexible equipment sharing, and dynamic scheduling capabilities. The operational status of these lines has become more complex and variable, posing new challenges to their intelligent management and control capabilities.

[0003] Currently, control technologies for flexible manufacturing production lines mainly include production scheduling based on Manufacturing Execution Systems (MES), equipment status monitoring based on the Industrial Internet of Things (IIoT), and production line simulation and optimization based on digital twins. In traditional technologies, bottleneck identification relies primarily on the analysis of static or semi-static indicators such as equipment utilization and buffer queue length. Scheduling optimization often employs rule-based heuristic algorithms or single-objective optimization models, and process path adjustments and task rearrangements are typically performed independently in stages. Some advanced control systems have introduced digital twin technology to achieve real-time mapping and visual monitoring of production line status; however, significant shortcomings remain in modeling bottleneck diffusion mechanisms, optimizing the coupling of process paths and scheduling, and implementing closed-loop adaptive control.

[0004] The main problems with existing technologies are as follows: First, bottleneck nodes in flexible manufacturing lines do not exist in isolation; their impact spreads downstream through material flow and process constraints. Traditional bottleneck identification methods lack the ability to quantitatively describe this dynamic diffusion effect, leading to inaccurate bottleneck domain location and insufficient targeting of optimization decisions. Second, there is a strong coupling relationship between process path adjustment and task rescheduling, but existing systems often handle these two separately, making it difficult to simultaneously optimize the completion time and operational stability of the production line from a globally optimal perspective. Furthermore, most control systems adopt open-loop or semi-open-loop control modes, lacking a closed-loop adaptive mechanism based on real-time feedback. When the production line status changes, the control strategy cannot be adjusted in a timely manner, causing the optimization effect to decay over time.

[0005] Therefore, it is essential to invent intelligent control systems, methods, equipment, and media for cylinder head flexible processing production lines to solve the above problems. Summary of the Invention

[0006] The purpose of this invention is to provide an intelligent control system, method, equipment, and medium for cylinder head flexible manufacturing production lines to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: an intelligent control system for a cylinder head flexible machining production line, comprising the following modules:

[0008] The digital twin production line mapping module is used to acquire twin data of the cylinder head flexible processing production line and build a digital twin production line model.

[0009] A dynamic topology network construction module is connected to the digital twin production line mapping module and is used to extract real-time operating status parameters of the production line from the digital twin production line model and construct a dynamic topology network for the production line.

[0010] The dynamic bottleneck diffusion identification module is connected to the dynamic topology network construction module and is used to calculate the load entropy of each workstation node in the production line dynamic topology network and determine the bottleneck diffusion domain based on the inter-node coupling diffusion strength matrix.

[0011] The process path and scheduling coupling optimization module is connected to the dynamic bottleneck diffusion identification module. It is used to construct a coupled optimization model with the bottleneck diffusion domain as the constraint boundary, with the dual objectives of minimizing the maximum completion time and minimizing the bottleneck diffusion domain coverage. The coupled optimization model is solved by a multi-objective optimization algorithm to generate an optimized solution set. The process path adjustment scheme and the scheduling task rearrangement scheme are output synchronously from the optimized solution set.

[0012] The production line control instruction generation and closed-loop execution module is connected to the process path and scheduling coupling optimization module. It is used to convert the process path adjustment scheme and scheduling task rearrangement scheme into production line control instructions and issue them to the production line execution layer. It collects execution feedback data in real time and inputs it into the digital twin production line mapping module to form a closed loop.

[0013] The technical effects and advantages of this invention are as follows:

[0014] 1. This invention extracts real-time operating status parameters of the production line from the digital twin production line model through a dynamic topology network construction module, constructs a dynamic topology network with workstations as nodes and material flow relationships and process sequence constraints as directed edges, and assigns a comprehensive load value to the nodes and a weight value to the edges, thereby realizing a structured representation of the complex coupling relationship of the production line.

[0015] 2. This invention uses a dynamic bottleneck diffusion identification module to calculate the load entropy of each workstation node to identify the main bottleneck node, and calculates the bottleneck propagation intensity level by level along the directed connection direction based on a pre-constructed coupling diffusion intensity matrix, thereby determining the bottleneck diffusion domain. This overcomes the shortcomings of traditional methods that ignore the dynamic diffusion effect of the bottleneck and improves the accuracy of bottleneck location.

[0016] 3. This invention constructs a coupled optimization model with the bottleneck diffusion domain as the constraint boundary through a process path and scheduling coupling optimization module. The model has two objectives: minimizing the maximum completion time and minimizing the bottleneck diffusion domain coverage. A multi-objective optimization algorithm is used to simultaneously output the process path adjustment scheme and the scheduling task reordering scheme, thereby realizing the global collaborative optimization of process path selection and scheduling task scheduling.

[0017] 4. This invention transforms the optimization scheme into process path adjustment instructions and scheduling task reordering instructions through the production line control instruction generation and closed-loop execution module, and sends them to the production line execution layer. It collects execution feedback data in real time and dynamically triggers a new round of optimization based on the bottleneck diffusion domain coverage, forming a closed-loop adaptive control mechanism to ensure the continuous and efficient operation of the production line under dynamic disturbances.

[0018] 5. This invention achieves intelligent control of the entire process of cylinder head flexible processing production line from state perception and bottleneck analysis to decision execution by deeply integrating digital twin mapping, dynamic topology network construction, bottleneck diffusion identification, and process path and scheduling coupling optimization. This significantly improves the production efficiency, operational stability, and flexible adaptability to multi-variety production tasks of the production line. Attached Figure Description

[0019] Figure 1 This is a diagram showing the overall modular structure of the system of the present invention;

[0020] Figure 2 Flowchart for determining the bottleneck diffusion domain of this invention;

[0021] Figure 3 This is a flowchart of the solution process for the coupled optimization model of the present invention;

[0022] Figure 4 This is a flowchart illustrating the closed-loop execution and control process of the present invention.

[0023] Figure 5 This is a flowchart of the overall method steps of the present invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] This invention provides, for example Figure 1 The intelligent control system for the cylinder head flexible manufacturing line shown includes the following modules:

[0026] The digital twin production line mapping module is used to acquire twin data of the cylinder head flexible processing production line and build a digital twin production line model.

[0027] Example 1: Detailed Implementation of the Digital Twin Production Line Mapping Module

[0028] This embodiment details how the digital twin production line mapping module acquires twin data and constructs a digital twin production line model.

[0029] 1. Acquisition of twin data

[0030] First, the digital twin production line mapping module acquires the twin data of the cylinder head flexible machining production line in real time through the following methods:

[0031] Data source access: The module connects to three types of data sources at the production line layer via industrial communication protocols such as OPCUA, MQTT, or ModbusTCP.

[0032] Equipment control system: Real-time acquisition of equipment status data (running, standby, fault, shutdown), axis position, spindle load, etc. from PLCs (programmable logic controllers) at each workstation (such as machining center, cleaning machine, leak testing machine).

[0033] Manufacturing Execution System (MES): Order information, process routes, workpiece models, work-in-process (WIP) quantities in each workstation buffer, and material handling instructions are obtained from the MES via REST API or database connection.

[0034] Sensor network: Vibration sensors, temperature sensors, and RFID readers for material identification are deployed at key workstations to collect equipment health status characteristics (such as spindle vibration spectrum characteristics) and workpiece processing progress information.

[0035] Data preprocessing: The acquired raw data undergoes timestamp alignment, outlier removal (e.g., using the 3σ principle to remove sensor noise), and unit normalization to form a standardized, time-synchronized twin data stream. This data stream includes, but is not limited to: the instantaneous processing cycle Tp(i,t) of each workstation i at time t, the number of workpieces in the buffer Bi(t), the overall equipment efficiency (OEE) index Hi(t), the current workpiece model Mi(t), and the transportation time Tr{ij}(t) between adjacent workstations.

[0036] 2. Construction of the digital twin production line model

[0037] The module uses a hybrid modeling approach based on discrete event simulation and state machine to construct a digital twin production line model. The specific steps are as follows:

[0038] Step A: Establish the basic model framework

[0039] In a simulation platform (such as Siemens Technomatix or AnyLogic), a simulation model containing all processing stations, buffer zones, and material handling equipment is created according to the physical layout and process sequence of the production line. Each station node is defined as a finite state machine with four states: "idle," "processing," "blocked," and "faulty."

[0040] Step B: Load static properties

[0041] Assign static attributes to each entity in the model, including: a unique identifier ID for the workstation (e.g., OP10, OP20), a rated processing capacity Ci (in pieces / hour), a list of workpiece models that the workstation can process Mlist(i), and the standard process time Ts(i, m) for each model.

[0042] Step C: Configure dynamic driver rules

[0043] The preprocessed real-time twin data stream is used as a driving signal and bound to the corresponding entity in the model:

[0044] Using the number of workpieces in the buffer Bi(t) transmitted from MES and the workpiece model Mi(t) identified by RFID, the queue length of each buffer and the attributes of the workpieces in the queue are dynamically updated in the model.

[0045] By utilizing the equipment status collected from the PLC, the current state of the corresponding workstation state machine in the model is forcibly updated. For example, when the PLC signal indicates a fault at a certain workstation, the corresponding workstation state in the model immediately switches from "processing" to "fault".

[0046] Using the measured instantaneous processing cycle Tp(i,t) and transport time Tr ij (t) replaces the static theoretical working time in the model, driving the progression of simulation time.

[0047] Step D: Establish a model synchronization update mechanism

[0048] To maintain consistency between the model and the physical production line, the module employs a dual mechanism of "event triggering + periodic synchronization":

[0049] Event Trigger: When a major event occurs on the physical production line (such as equipment failure or workpiece completion and exit from the workstation), a data acquisition is immediately triggered and the model status is refreshed.

[0050] Periodic synchronization: With a synchronization cycle of 100 milliseconds, parameters that change continuously but do not require event triggering (such as equipment temperature and vibration amplitude) are periodically updated.

[0051] Through the above steps, the digital twin production line mapping module constructs a state-driven digital twin production line model that is synchronized with the physical production line in real time. This model not only includes a three-dimensional visual geometric structure, but more importantly, it encapsulates the production line's behavioral logic and real-time operational data. It provides the required real-time operational status parameters of the production line to the dynamic topology network construction module in the form of standard data tables or API interfaces, such as: Bi(t) (the number of workpieces to be processed in the current buffer of each workstation), Tp(i,t) (the measured processing cycle time of each workstation), Hi(t) (the health status indicators of the equipment at each workstation), Mi(t) (the model of the workpiece currently being processed at each workstation), and Tr... ij (t) (material handling time between adjacent workstations).

[0052] A dynamic topology network construction module is connected to the digital twin production line mapping module and is used to extract real-time operating status parameters of the production line from the digital twin production line model and construct a dynamic topology network for the production line.

[0053] Furthermore, in the above technical solution, the dynamic topology network construction module constructs the production line dynamic topology network in the following manner:

[0054] The following real-time operating status parameters of the production line are extracted from the digital twin production line model: the number of workpieces to be processed in the current buffer of each workstation, the measured value of the processing cycle of each workstation, the health status index of the equipment at each workstation, the model of the workpiece currently being processed at each workstation, and the material handling time between adjacent workstations.

[0055] Each processing station is a network node. The first connection condition is the existence of material flow relationship between adjacent stations, and the second connection condition is the existence of process sequence constraint relationship between two stations. A directed connection between nodes is established when at least one of the first or second connection conditions is met.

[0056] Each node is assigned an attribute value, which includes at least the current comprehensive load value Li and the rated processing capacity Ci of the workstation. The comprehensive load value Li is calculated by weighting the number of workpieces to be processed in the current buffer of the workstation, the actual value of the processing cycle, and the standard working hours corresponding to the workpiece model.

[0057] Assign a weight value w to each directed edge. ij The weight value is calculated and determined based on the ratio of the material handling time between the two ends of the connecting station to the measured value of the processing cycle time of the downstream station;

[0058] The constructed set of nodes, set of directed edges, set of node attribute values, and set of edge weight values ​​are combined to form a dynamic production line topology network G(V, E, A, W) represented by an adjacency matrix, where V is the set of nodes, E is the set of directed edges, A is the node attribute matrix, and W is the edge weight matrix.

[0059] Example 2: Detailed Implementation of the Dynamic Topology Network Construction Module

[0060] This embodiment details how the dynamic topology network construction module extracts real-time operating status parameters of the production line from the digital twin production line model and constructs a dynamic topology network for the production line accordingly.

[0061] 1. Extraction of real-time operating status parameters of the production line

[0062] The dynamic topology network construction module periodically (sampling period ΔT = 1 second) reads the following five types of real-time status parameters from the digital twin production line model constructed by the digital twin production line mapping module through a standardized data interface (such as RESTAPI or shared memory database):

[0063] The number of workpieces waiting to be processed in the current buffer of each workstation, Bi(t): the total number of workpieces waiting to be processed in the upstream buffer of workstation i, which is directly taken from the queue length variable of the corresponding buffer in the digital twin model.

[0064] The measured value of the processing cycle time Tp(i,t) for each station: The actual time (in seconds) taken for station i to complete the processing of a single workpiece in the most recent time, which is recorded by the duration of the "processing" state of the station state machine in the digital twin model.

[0065] The health status index Hi(t) for each workstation's equipment is a continuous value ranging from [0, 1], where 1 represents complete health and 0 represents complete failure. This index is calculated in real-time by a digital twin model based on the fusion of vibration sensors, temperature sensors, and PLC fault codes. The specific calculation formula is as follows:

[0066] Hi(t)=a·(1-Vi(t) / Vth)+b·(1-Ti(t) / Tth)+c·Fi(t),

[0067] Where Vi(t) is the measured vibration amplitude, Vth is the vibration threshold, Ti(t) is the measured temperature, Tth is the temperature threshold, Fi(t) is the fault coefficient (1 when there is no fault, 0 when there is a fault), and a, b, and c are weighting coefficients, a+b+c=1. In a preferred embodiment, to balance the comprehensive impact of vibration, temperature, and fault status on the health of the equipment, the weighting coefficients a, b, and c can be calibrated based on field experience or historical fault data. For example, a=0.5, b=0.3, and c=0.2 can be set. Among them, the vibration weight a has the highest value because it can quickly reflect tool wear or spindle imbalance; the fault coefficient c is the second highest because it represents deterministic faults; and the temperature weight b is relatively low because its change is relatively gradual.

[0068] The model number Mi(t) of the workpiece currently being processed at each workstation: an integer variable representing the model code of the workpiece being processed at workstation i (e.g., 1 represents model A, 2 represents model B), which is obtained from the work order information synchronized from the MES system by the digital twin model.

[0069] Material handling time between adjacent workstations (Tr) ij (t): The time (in seconds) for the workpiece to be transported from the exit of station i to the entrance of station j (usually the downstream station). This value is reported in real time by the Automated Material Handling System (AMHS) or the manual handling record module integrated in the digital twin model.

[0070] 2. Construction of the dynamic topology network of the production line

[0071] After obtaining the above parameters, the module constructs a directed weighted dynamic topology network G(V, E, A, W) according to the following steps.

[0072] Step A: Determine the node set V

[0073] Each physical processing station in the production line is considered a node in the network, and a unique index i is assigned to each node (i = 1, 2, ..., N, where N is the total number of stations). The set of nodes is represented as V = {v1, v2, ..., vN}.

[0074] Step B: Determine the set of directed edges E

[0075] For any two nodes vi and vj (i ≠ j), determine whether to establish a directed edge e from vi to vj according to the following two conditions. ij :

[0076] First connection condition (material flow relationship): If there is a direct material flow from station i to station j in the physical production line (i.e., after the workpiece is processed at station i, it may be directly transported to station j in the next step), then a directed connection e is established. ij .

[0077] Second connection condition (process sequence constraint): If, in the process route of any cylinder head model, the processing step of station i is immediately adjacent to the processing step of station j (i.e., there is a process sequence constraint from i to j), a connection edge e is established regardless of whether there are other buffers or handling equipment between them. ij .

[0078] A directed edge from vi to vj is established if at least one of the first or second edge conditions is met. Finally, the set of all directed edges is denoted as E = {e...} ij}

[0079] Step C: Assign attribute values ​​to the nodes

[0080] Each node vi has the following attribute values: comprehensive load value Li and rated processing capacity Ci. Wherein:

[0081] Rated processing capacity Ci: This is a static attribute obtained from the process design parameters of the production line. It represents the theoretical maximum processing rate of station i under standard operating conditions (unit: pieces / hour).

[0082] The comprehensive load value Li(t) is a dynamic attribute, calculated at each sampling time t based on real-time parameters. The calculation formula is as follows:

[0083] Li(t)=α1·(Bi(t) / Bmax)+α2·(Tp(i,t) / Ts(i,Mi(t)))+α3·(1-Hi(t))

[0084] Where Bmax is the maximum capacity of the buffer at workstation i, which is a fixed constant; Ts(i, Mi(t)) is the standard process time (from the process database) for workstation i to process the current model Mi(t) workpiece; α1, α2, α3 are weighting coefficients that satisfy α1+α2+α3=1. In this embodiment, empirical values ​​α1=0.4, α2=0.4, and α3=0.2 are taken, which respectively reflect the contribution of buffer accumulation, cycle deviation, and health status to the load.

[0085] The higher the calculated Li(t) value, the greater the processing pressure that station i is currently under, and the more likely it is to become a bottleneck.

[0086] Step D: Assign weights to directed edges

[0087] For each directed edge e ij Its weight w ij (t) Calculate according to the following formula:

[0088] w ij (t) = Tr ij (t) / Tp(j,t)

[0089] Among them, Tr ij (t) represents the material handling time from workstation i to workstation j at the current moment; Tp ij This represents the measured processing cycle time of downstream workstation j at the current moment.

[0090] The physical meaning of this ratio is that if the material handling time is greater than the processing time of the downstream station, the material supply is more likely to become a constraint on the downstream station. A larger edge weight indicates a stronger coupling effect of station i on station j. When Tr ij When (t) or Tp(j,t) cannot be obtained (e.g., the connection exists only because of process sequence constraints and there is no actual material flow), the default value w is used. ij =1.0.

[0091] Step E: Form the network represented by the adjacency matrix.

[0092] Combining the node set V, directed edge set E, node attribute set, and edge weight set constructed above into an adjacency matrix, we obtain the production line dynamic topology network G(V, E, A, W):

[0093] Node attribute matrix A: A matrix of size N×2, where the first column of the i-th row is the rated processing capacity Ci of workstation i, and the second column is the comprehensive load value Li(t) at the current moment.

[0094] Edge weight matrix W: A sparse matrix of size N×N, where the element in the i-th row and j-th column W(i,j) = w ij (t) If there exists a directed edge e ij Otherwise, W(i,j) = 0.

[0095] The network structure G is dynamically updated every minute (or updated by an event) and is used as a topological description of the current production line status. It is then output to the dynamic bottleneck diffusion identification module for subsequent bottleneck analysis and diffusion calculation.

[0096] The dynamic bottleneck diffusion identification module is connected to the dynamic topology network construction module and is used to calculate the load entropy of each workstation node in the production line dynamic topology network and determine the bottleneck diffusion domain based on the inter-node coupling diffusion strength matrix.

[0097] Furthermore, in the above technical solutions, such as Figure 2 As shown, the dynamic bottleneck diffusion identification module determines the bottleneck diffusion domain in the following way:

[0098] For each workstation node i in the dynamic topology network of the production line, calculate its load entropy value Hi:

[0099] Arrange the calculated load entropy values ​​Hi of each workstation node in descending order, and take the node with the highest load entropy value as the main bottleneck node B0 at the current moment.

[0100] Starting from the main bottleneck node B0, based on the pre-constructed coupling diffusion strength matrix D, the propagation intensity of the bottleneck effect is calculated step by step along the directed connection directions of the production line dynamic topology network: for node j, which is d hops away from node B0, the bottleneck propagation intensity Sj = S B0 ·ΠD k，k+1 ·α^d, where S B0 The bottleneck strength value of the primary bottleneck node, D k，k+1 is the coupling diffusion coefficient between adjacent nodes on the propagation path, and α is the propagation attenuation factor and α∈(0,1);

[0101] The set of all workstation nodes whose propagation intensity Sj exceeds the preset diffusion threshold θ, along with the main bottleneck node B0, is defined as the bottleneck diffusion domain Ω at the current moment. B .

[0102] Furthermore, in the above technical solutions, such as Figure 3 As shown, the coupling diffusion strength matrix D is constructed as follows:

[0103] Obtain the time-series data of workstation status during each bottleneck event recorded in the historical operation data of the production line;

[0104] For each bottleneck event, extract the load change time series between the bottleneck workstation and other workstations, calculate the time-delay cross-correlation function between each pair of workstations, and take the maximum cross-correlation coefficient as the coupling diffusion coefficient between the workstation pairs.

[0105] For all bottleneck events, the average coupling diffusion coefficients of similar workstation pairs are used to construct a coupling diffusion strength matrix D=[d mn ], where d mn This represents the coupling diffusion strength of the bottleneck effect of workstation m to workstation n.

[0106] Example 3: Detailed Implementation of the Dynamic Bottleneck Diffusion Identification Module

[0107] This embodiment details how the dynamic bottleneck diffusion identification module calculates the load entropy of each workstation node in the dynamic topology network of the production line, and determines the bottleneck diffusion domain based on the pre-constructed coupling diffusion strength matrix.

[0108] 1. Preconstruction of the coupling diffusion strength matrix D

[0109] Before production line operation or during periodic offline phases, the dynamic bottleneck diffusion identification module first constructs a coupling diffusion strength matrix D using historical data. The specific steps are as follows:

[0110] Step A: Obtain historical bottleneck event data

[0111] Extract time-series data of workstation status during all bottleneck events recorded within the past 6 months (adjustable according to actual production line operating frequency) from the Manufacturing Execution System (MES) or historical database. A bottleneck event is defined as a period of time during which the buffer of a workstation is continuously overflowing or the processing cycle time continuously exceeds the standard working time by more than 20% and lasts for more than 5 minutes.

[0112] Step B: Calculate the coupling diffusion coefficient between each pair of workstations.

[0113] For each bottleneck event, execute the following sub-steps:

[0114] Determine the bottleneck location: Let m be the bottleneck location for this event.

[0115] Extracting the load change time series: Within a time window [t0-10min, t0+10min] from 10 minutes before the event occurrence time t0 to 10 minutes after the event, extract the comprehensive load value sequence Lm(t) of the bottleneck workstation m, and the comprehensive load value sequence Ln(t) of the candidate downstream (or upstream) workstation n. The sampling interval is 10 seconds, and approximately 120 sampling points are obtained.

[0116] Extract the load change time series: Within the time window [t0-10min, t0+10min] from 10 minutes before the event occurs to 10 minutes after it ends, extract the comprehensive load value sequence Lm(tk) of the bottleneck workstation m and the comprehensive load value sequence Ln(tk) of the candidate downstream workstation (or upstream workstation) n at 10-second sampling intervals, where k=1,2,...,N, and N is the total number of sampling points in the sequence (N≈120 in this example).

[0117] Calculate the cross-correlation coefficient with time delay and take the maximum value: Calculate the cross-correlation coefficient ρ{mn}(τ) for different time delay steps τ using the following formula:

[0118] ,

[0119] Where τ is the number of delay steps, and its value is an integer ranging from -N / 2 to N / 2 (corresponding to an actual delay of -300 seconds to +300 seconds); tk is the kth sampling time. , These are the mean values ​​of the two sequences during the current summation period (i.e., the overlapping part of the two sequences); the summation index k ranges from 1 to N-|τ|, ensuring that both Lm(tk) and Ln(tk+τ) are defined.

[0120] Find the maximum cross-correlation coefficient: search for ρ among all τ values. mn (τ) Maximum value, denoted as ρ mn This maximum value is the candidate value for the coupling diffusion coefficient of the bottleneck effect of workstation m to workstation n in this bottleneck event. When ρ mn When τ>0 reaches its maximum value, it indicates that the load change at station n lags behind that at station m, which is consistent with the physical meaning of downstream propagation of the bottleneck.

[0121] Step C: Construct matrix D

[0122] The candidate coupling diffusion coefficient ρ is calculated for the same workstation pair (m, n) in all historical bottleneck events. mn The arithmetic mean is taken to obtain the final coupling diffusion coefficient d. mn If a workstation has never experienced a bottleneck event, then d mnThe default value is 0.1 (indicating weak coupling).

[0123] Finally, a coupling diffusion strength matrix D of size N×N (where N is the total number of workstations) is constructed: D=[d mn ], where row index m represents the bottleneck workstation and column index n represents the affected workstations. Note d mn With d mn Generally asymmetrical, conforming to the directionality of bottleneck propagation.

[0124] 2. Load entropy calculation and main bottleneck node identification

[0125] During real-time production line operation, the dynamic bottleneck diffusion identification module periodically (e.g., every 30 seconds) reads the attributes of each workstation from the dynamic topology network G(V, E, A, W) and performs the following operations:

[0126] Step A: Calculate the load entropy Hi for each workstation node.

[0127] For workstation i, its load entropy value is defined as the degree of disorder of the comprehensive load of that workstation within its neighborhood, and the calculation formula is:

[0128] Hi=- p ij ·ln(p ij ),

[0129] Where Ni is the set of neighborhood nodes of workstation i, including workstation i itself and all nodes pointed to by its downstream directed edges (i.e., ∀j, e). ij ∈E).

[0130] p ij The normalized load transfer probability is calculated as follows:

[0131] p ij =Lj / Lk,

[0132] When j=i, Li is the total load value of workstation i itself; when j≠i, Lj is the total load value of downstream workstation j. This probability reflects the possibility of workstation i's load being distributed to downstream nodes (including itself).

[0133] The larger the load entropy Hi is, the more evenly the load of workstation i spreads downstream, and the more likely workstation i itself is a bottleneck source; conversely, if Hi is very small, it means that the load is mainly concentrated in a single workstation downstream, and workstation i may just be a forced node.

[0134] Step B: Determine the primary bottleneck node B0

[0135] The calculated load entropy values ​​Hi of each workstation node are sorted in descending order, and the node with the highest entropy value is selected as the main bottleneck node B0 at the current moment. If multiple nodes have the same entropy value (the difference is less than 0.01), the node with the highest comprehensive load value Li is selected as the main bottleneck node.

[0136] 3. Bottleneck diffusion domain Ω B The determination

[0137] Step A: Set propagation parameters

[0138] Propagation attenuation factor α: The value is 0.7 (an empirical value, which can be adjusted between 0.5 and 0.9 depending on the degree of coupling of the production line).

[0139] Diffusion threshold θ: The value is 0.3 (an empirical value, which can be adjusted between 0.2 and 0.5 depending on the production line's sensitivity to bottlenecks).

[0140] Main bottleneck node bottleneck strength S B0 : Take the product of the load entropy value H{B0} of the main bottleneck node itself and the normalized comprehensive load value, i.e., S B0 =H B0 · (L B0 / max(L)), where max(L) is the maximum value of the total load of all workstations at present.

[0141] Step B: Calculate the propagation intensity step by step along the directed edges.

[0142] Starting from the main bottleneck node B0, a breadth-first traversal is performed along the directed edges of the dynamic topology network (i.e., the directions indicated by the non-zero elements in matrix W). For a node j with a distance of d hops from B0 (d = 1, 2, 3, ..., not exceeding the network diameter), the bottleneck propagation intensity Sj is calculated using the following formula:

[0143] Sj=S B0 ·( D k，k+1 )·α^d,

[0144] in, D k，k+1 This represents the product of the coupling diffusion coefficients between adjacent nodes along the propagation path from B0 to j. If multiple paths exist, the maximum value of the product among all paths is taken (i.e., the strongest propagation path).

[0145] α^d is the decay factor raised to the power of d, which decays exponentially with the number of jumps.

[0146] For example: If the main bottleneck node is workstation 2, and the propagation occurs along the path 2→4→5, then the propagation intensity at node 5 is S5=S B0 ·d 24 ·d45 ·α^2.

[0147] Step C: Determine the bottleneck diffusion domain

[0148] Workstations that meet any of the following conditions are classified into the bottleneck diffusion domain Ω. B :

[0149] This node is the primary bottleneck node B0;

[0150] The propagation intensity Sj at this node is greater than or equal to θ (diffusion threshold).

[0151] The final set Ω B This represents all workstation nodes significantly affected by the main bottleneck at the current moment. This set will be output as the constraint boundary to the process path and scheduling coupling optimization module.

[0152] 4. Dynamic update mechanism

[0153] The dynamic bottleneck diffusion identification module repeats the above steps at the same interval as the network update (e.g., 30 seconds) to update Ω in real time. B If the Ω is calculated in two consecutive steps... B If the change is less than 20% (i.e., the ratio of the intersection to the union of the workstation sets is greater than 0.8), the current bottleneck domain remains unchanged to reduce system computational overhead; otherwise, the new bottleneck domain is pushed to the downstream optimization module in a timely manner.

[0154] The process path and scheduling coupling optimization module is connected to the dynamic bottleneck diffusion identification module. It is used to construct a coupled optimization model with the bottleneck diffusion domain as the constraint boundary, with the dual objectives of minimizing the maximum completion time and minimizing the bottleneck diffusion domain coverage. The coupled optimization model is solved by a multi-objective optimization algorithm to generate an optimized solution set. The process path adjustment scheme and the scheduling task rearrangement scheme are output synchronously from the optimized solution set.

[0155] Furthermore, in the above technical solution, the process path and scheduling coupling optimization module constructs the coupling optimization model in the following way:

[0156] With the bottleneck diffusion domain Ω B To constrain the boundary, the first optimization objective is defined as minimizing the maximum completion time: minf1 = (Ci), where Ci is the completion time of workstation i, and V is the set of workstation nodes;

[0157] The second optimization objective is defined as minimizing the bottleneck diffusion domain coverage: minf2 = |Ω B '| / |Ω B |, where |Ω B |To optimize the number of workstations within the bottleneck diffusion domain,|Ω B'| represents the number of workstations within the bottleneck diffusion domain after applying the current optimized solution;

[0158] Establish a constraint set consisting of the following constraints: processing capacity constraints of each workstation within the bottleneck diffusion domain, processing characteristic constraints of alternative process paths, changeover time constraints for switching between different types of cylinder heads, material handling resource constraints, and order delivery date constraints.

[0159] A multi-objective weighted function is constructed based on the first and second optimization objectives, and solved using a multi-objective optimization algorithm based on non-dominated sorting. The constraint set is set as the feasible region boundary.

[0160] Furthermore, in the above technical solution, the process path and scheduling coupling optimization module solves the coupling optimization model in the following way:

[0161] Based on the bottleneck diffusion domain Ω B Determine the set of affected workpieces J that require process path adjustments. aff ;

[0162] For the affected workpiece set J aff For each workpiece in the dataset, retrieve the set P of alternative process paths corresponding to that workpiece model. alt Each path in the set of alternative process paths satisfies the processing characteristic constraint of the alternative process path;

[0163] For each alternative process path, calculate the expected load change sequence for each station along that path, and then compare this expected load change sequence with the initial bottleneck diffusion domain Ω. B By comparing the results, we can predict the direction of change in the bottleneck diffusion domain coverage after adopting this alternative process path.

[0164] The decision to select alternative process routes is encoded as a decision variable, and the decision variables of scheduling tasks are rearranged, including work order sequence number and resource allocation number, and then incorporated into a unified coding scheme to form a set of decision variables for a multi-objective coupled optimization problem.

[0165] The NSGA-III optimization algorithm is used to perform a population iterative search within the feasible region defined by the constraint set, and outputs a non-dominated solution set.

[0166] Example 4: Detailed Implementation of the Process Path and Scheduling Coupling Optimization Module

[0167] This embodiment elaborates in detail how the process path and scheduling coupling optimization module utilizes the bottleneck diffusion domain Ω. B To constrain the boundaries, a dual-objective coupled optimization model is constructed and solved, simultaneously outputting process path adjustment schemes and scheduling task rearrangement schemes.

[0168] 1. Construction of the Coupled Optimization Model

[0169] Step A: Define decision variables

[0170] The optimization problem in this module involves two types of coupled decision variables:

[0171] Process path selection variable x j，p This is a binary variable; it takes the value 1 when workpiece j selects the alternative process path p, and 0 otherwise. Each workpiece has exactly one selected path. x j，p =1.

[0172] Scheduling rearrangement variables include the processing order vector πi at each workstation (representing the order of workpieces to be processed at workstation i) and the resource allocation variable y. j，p (Indicates whether workpiece j is assigned to workstation i for processing; valid for flexible workstations).

[0173] Step B: Define the optimization objective function

[0174] With the bottleneck diffusion domain Ω B To constrain the boundary, the first optimization objective is defined as minimizing the maximum completion time: minf1 = (Ci), where Ci is the completion time of workstation i, and V is the set of workstation nodes;

[0175] The second optimization objective is defined as minimizing the bottleneck diffusion domain coverage: minf2 = |Ω B '| / |Ω B |, where |Ω B |To optimize the number of workstations within the bottleneck diffusion domain,|Ω B '| represents the number of workstations within the bottleneck diffusion domain after applying the current optimized solution;

[0176] Step C: Establish constraint set

[0177] The constraint set defines the boundary of feasible solutions, specifically including the following five types of constraints:

[0178] Constraint 1 is a constraint on the processing capacity of each workstation within the bottleneck diffusion domain: for any workstation i∈Ω B The total processing time (including the remaining processing time of already assigned workpieces and the processing time of newly assigned workpieces) must not exceed the product of the rated processing capacity of the workstation and the remaining time window. That is:

[0179] ,

[0180] J i For the set of workpieces assigned to workstation i, Tp(i, m) j Workpiece j (model m) is processed at workstation i. j The cycle time, Ci is the rated capacity (pieces / unit time), Tavail This refers to the available time during the planning period.

[0181] Constraint 2 requires that the machining features of the alternative process path satisfy the following constraint: For any workpiece j, the selected alternative process path p must be able to machine all the key features of workpiece j (such as valve seat holes, spark plug holes, bolt holes, etc.). This constraint is verified using the "model-path" mapping table in the process knowledge base: if p∈P alt (j) is satisfied; otherwise, it cannot be selected.

[0182] Constraint 3 is the changeover time constraint for switching between different types of cylinder heads: Let the sequence of workpieces planned to be processed at station i be j1, j2, ..., j Ki The corresponding models are m i，1 m i，2 , ..., m i，Ki Therefore, the cumulative changeover time for this workstation must not exceed the upper limit.

[0183] ,

[0184] Among them, K i m is the total number of workpieces allocated to workstation i during the planning period; i，k S represents the model number of the k-th workpiece being processed at workstation i; i (a, b) represents the changeover time required for workstation i to switch from model a to model b (if a = b, then S...). i =0); This is the maximum cumulative production changeover time allowed within the planning period, usually taken as a fixed percentage of the total planning period time, such as 10%.

[0185] Constraint 4 is a material handling resource constraint: the material handling capacity of the equipment (AGV or conveyor belt) between any two workstations is limited. Let Q be the number of material handling tasks from workstation u to workstation v. uv The time for a single transport is Tr uv If so, the total handling time must not exceed the available time of the handling resources;

[0186] Constraint 5 is an order delivery date constraint: the completion time Cj of each workpiece j must not be later than the delivery date due specified in its order. j .

[0187] Step D: Construct a multi-objective optimization model

[0188] Taking the first objective f1 and the second objective f2 as parallel optimization objectives, and the constraint set as the boundary of the feasible region, we form a bi-objective optimization problem in the following standard form:

[0189] min(f1(x,π),f2(x,π)),

[0190] Constraints: Constraints 1 to 5.

[0191] 2. Coupled Optimization Solution Based on NSGA-III

[0192] This module uses a non-dominated sorting genetic algorithm (NSGA-III) based on reference points to solve the problem. The specific steps are as follows:

[0193] Step A: Determine the set of affected workpieces J aff

[0194] Based on the current bottleneck diffusion domain Ω B The following two types of workpieces were selected and included in J. aff :

[0195] Currently located at Ω B Workpieces (including those being processed) in any workstation buffer zone;

[0196] Not yet on the production line but planned to enter Ω in the next scheduling cycle. B Workpieces in the inner workstation.

[0197] Step B: Retrieve the set of alternative process paths P alt

[0198] For J aff For each workpiece j, retrieve all feasible alternative process paths P corresponding to its model from the process knowledge base. aff (j). Each path satisfies the "processing feature satisfies constraint". For example, the standard path for a certain cylinder head model is "OP10→OP20→OP30", and the alternative paths may be "OP10→OP25→OP30" or "OP10→OP20→OP35→OP40", depending on the equipment flexibility and process capability.

[0199] Step C: Predict the direction of change in bottleneck diffusion domain coverage

[0200] To accelerate convergence, for each alternative process path p∈P alt (j) Calculate its expected load change vector ΔLp = [ΔL1, ΔL2, ..., ΔLN], where ΔL1 represents the overall load change at workstation i after adopting path p (relative to the current path). Then compare ΔLp with the current bottleneck diffusion domain ΩB:

[0201] If ΔLi is negative at most stations within ΩB (i.e., the load is reduced), then it is estimated that this path can cause f2 to decrease.

[0202] Conversely, if ΔLi significantly increases the load on the external workstation of ΩB, then the estimated f2 may increase.

[0203] This estimate is used to guide the generation of the initial population, but is not a hard constraint.

[0204] Step D: Unified Coding Scheme

[0205] To simultaneously optimize process path selection and task scheduling, this module incorporates both types of decision variables into a unified chromosome coding scheme, using a hybrid coding method. Specifically, the size of the affected workpiece set Jaff is n, the number of workstations in the production line is N, and the number of flexible processes that can be allocated resources is R.

[0206] A chromosome is composed of three gene segments linked together:

[0207] Segment 1 (Process Path Selection) is J aff Each workpiece j in the process is assigned an integer gene g1. j ∈[0, |P alt (j)|-1], where |P alt (j) represents the number of feasible alternative process paths for workpiece j. For example, if workpiece A has 3 alternative paths, then g1 A The value can be 0, 1, or 2, corresponding to the path index. The length of this segment is equal to |J. aff |

[0208] Segment 2 (Processing sequence within a workstation): For each workstation i (i=1,...,N), let J be the set of workpieces currently to be processed. i (Including workpieces already existing in the buffer and workpieces newly allocated through path selection), the number of workpieces is denoted as K. i =|J i The processing sequence of the workpieces at this station is encoded into a sequence of length K. i The arrangement of π i , where π i (k) represents the workpiece ID at position k. All workstations are arranged sequentially by their workstation numbers, forming a sequence of length [length missing]. Segment 2. When there are no workpieces to be processed at a certain workstation, its arrangement is empty. This encoding satisfies the processing sequence variable at each workstation.

[0209] Segment 3 (Flexible Resource Allocation): For processes in a production line with flexible processing capabilities (i.e., one workpiece can be completed on multiple identical machines), there are R independent decision points of this type. Each decision point corresponds to a resource allocation gene g3. r The value ranges from 0 to M. r-1 integers, M r This represents the number of available machines at this decision point. For example, if a certain process can be performed on two identical machine tools, then g3... r ∈{0,1}. When the equipment allocation for all processes is fixed, the resource allocation fragment length is R.

[0210] In one embodiment, assuming J aff ={Workpiece 1, Workpiece 2}, where Workpiece 1 has 2 alternative paths and Workpiece 2 has 3 alternative paths; the production line has two workstations, A and B; and there is one flexible resource decision point (two machines can be selected). A chromosome can be represented as:

[0211] Segment 1: [1, 0] (Workpiece 1 uses the path with index 1, and workpiece 2 uses the path with index 0)

[0212] Segment 2: The arrangement at workstation A [workpiece 2, workpiece 1] and the arrangement at workstation B [workpiece 1, workpiece 2] are combined to form [2, 1, 1, 2].

[0213] Segment 3: [0] (Flexible resource decision point selection equipment 0)

[0214] Complete chromosome: [1, 0|2, 1, 1, 2|0]

[0215] Decoding process:

[0216] During the objective function evaluation, this module decodes the chromosome into a complete production plan according to the following steps:

[0217] Based on segment 1, determine the specific process path to be used for each workpiece, and obtain the processing time and process sequence of each workpiece at each workstation.

[0218] Based on segment 3, determine the equipment allocation for flexible resource decision points.

[0219] Based on segment 2, the processing sequence of the workpieces at each station is obtained. Combining the above path and resource allocation, the completion time Ci of each station is calculated using discrete event simulation, and then the two objective function values ​​are obtained.

[0220] Simultaneously, based on the optimized scheduling execution results, the bottleneck diffusion domain Ω is recalculated. B ', thus obtaining the second target value.

[0221] Through the unified coding scheme described above, this module can simultaneously search the process path space and the scheduling space, achieving coupled optimization of the two.

[0222] Step E: NSGA-III Population Iterative Search

[0223] Population initialization: Generate an initial population of size PopSize (200 in this example). 50% of the individuals are generated using the aforementioned predicted direction (i.e., choosing the path that reduces f2 as predicted), and 50% are generated randomly to ensure diversity.

[0224] Crossover operators: Single-point crossover is used for path selection segments, partial mapping crossover (PMX) is used for sorting segments, and uniform crossover is used for resource allocation segments.

[0225] Mutation operators: Randomly mutate the path selection segment to other paths with a probability of 0.1; use exchange mutation for the sorting segment; and use random reset for the resource allocation segment.

[0226] Non-dominated ranking: Stratify the population according to Pareto dominance and calculate the crowding distance for each individual.

[0227] Reference point generation: The Das-Dennis method is used to generate uniformly distributed reference points on the hyperplane, with the number set to 1 / 2 of the population size.

[0228] Selection and Evolution: Offspring populations are generated through tournament selection, crossover, and mutation. Parent and offspring generations are merged, and an elite retention strategy is adopted (retaining the top PopSize optimal individuals). The number of iterations, Gen, is set to 300 generations, or the process terminates early if the optimal frontier fails to improve for 50 consecutive generations.

[0229] Step F: Output the optimized solution set

[0230] After iteration, a set of non-dominated solutions (Pareto front) is obtained. Each non-dominated solution corresponds to a set of (f1, f2) values ​​and the complete set of decision variable values. This module passes the set of non-dominated solutions to the production line control instruction generation and closed-loop execution module, which selects the optimal solution based on actual preferences (such as prioritizing short delivery time or prioritizing stable production), or retains multiple candidate solutions for operators to choose from.

[0231] 3. Synchronously output process path adjustment plan and scheduling task rescheduling plan

[0232] The following two types of solutions are decoded from the selected optimal solution:

[0233] Process path adjustment plan: includes the set of affected workpieces J aff Each workpiece contains (workpiece ID, original process path number, new process path number, and the sequence of workstation numbers corresponding to each step).

[0234] Task rescheduling scheme: includes a list of processing sequences for all workpieces at all workstations. Each task entry includes (workpiece ID, target workstation number, estimated start time, and estimated completion time).

[0235] The above scheme outputs structured data (such as JSON or XML) to the production line control instruction generation and closed-loop execution module.

[0236] The production line control instruction generation and closed-loop execution module is connected to the process path and scheduling coupling optimization module. It is used to convert the process path adjustment scheme and scheduling task rearrangement scheme into production line control instructions and issue them to the production line execution layer. It collects execution feedback data in real time and inputs it into the digital twin production line mapping module to form a closed loop.

[0237] Furthermore, in the above technical solutions, such as Figure 4 As shown, the production line control instruction generation and closed-loop execution module completes closed-loop control in the following ways:

[0238] The optimal solution in the optimized solution set is decoded into a process path adjustment instruction and a scheduling task reordering instruction. The process path adjustment instruction includes the workpiece ID, the original process path number, the new process path number, and the workstation number corresponding to each step. The scheduling task reordering instruction includes the workpiece ID, the adjusted work order sequence number, the assigned target workstation number, and the expected start time.

[0239] The process path adjustment instructions and scheduling task rearrangement instructions are encapsulated into production line control instruction messages and sent to each workstation controller in the production line execution layer.

[0240] The digital twin production line mapping module collects twin data in real time after the production line execution, and calculates the bottleneck diffusion domain coverage |Ω at the current moment. B (t)| / |Ω B (t0)|, where t0 is the initial time when the process path adjustment scheme and the scheduling task rearrangement scheme begin to be executed. If the coverage exceeds the preset closed-loop feedback trigger threshold β for K consecutive collection cycles, a new round of bottleneck diffusion domain determination and coupling optimization solution will be triggered; otherwise, the current production line control instructions will remain unchanged.

[0241] Example 5: Detailed Implementation of the Production Line Control Instruction Generation and Closed-Loop Execution Module

[0242] This embodiment details how the production line control instruction generation and closed-loop execution module transforms the optimized solution into executable control instructions for the production line, sends them to the execution layer, and forms a closed-loop continuous optimization mechanism through real-time feedback data.

[0243] 1. Decoding the optimal solution and generating instructions

[0244] In the non-dominated solution set (Pareto front) output by the process path and scheduling coupling optimization module, the production line control instruction generation and closed-loop execution module selects the optimal solution based on the current production preferences. The selection strategy is as follows: if the current order delivery time is tight, the solution with the smallest maximum completion time f1 is selected first; if the current production line stability requirements are high, the solution with the smallest bottleneck diffusion domain coverage f2 is selected first; the operator can also manually select from the interface. After selecting the optimal solution, the module decodes and generates two types of instructions according to the following steps:

[0245] Step A: Generate process path adjustment instructions

[0246] For each workpiece whose "process path selection" decision variable in the optimal solution is not on the original path (i.e., the workpiece whose process route needs to be changed), a process path adjustment instruction is generated, and its data structure is shown in the table below:

[0247] Workpiece ID String unique identification code of the workpiece Original process path number Integer Path numbers currently in use or planned for use New process path numbering Integer Alternative path number selected after optimization Work Step Sequence array List the workstation numbers corresponding to each process step according to the processing sequence. Effective time Timestamp When should the switchover begin (usually the current time plus the shortest switchover preparation time)?

[0248] This instruction is not generated for workpieces for which the process path does not need to be adjusted.

[0249] Step B: Generate task rescheduling instructions

[0250] For each workstation i determined by the "processing sequence within the workstation" decision variable in the optimal solution, the updated workpiece processing sequence at that workstation is transformed into a scheduling task reordering instruction. Each instruction corresponds to a task at a workstation (i.e., the processing task of a workpiece at that workstation), and its data structure is shown in the table below:

[0251] Workpiece ID String unique identification code of the workpiece Target workstation number String Workstations that will perform this processing task Work order sequence number Integer The processing sequence number at this workstation (incrementing from 1). Expected start time Timestamp Based on the simulation evaluation of the start processing time Expected completion time Timestamp Based on the simulation evaluation of the completion processing time

[0252] For tasks that are canceled due to changes in the process path, an additional cancellation command needs to be generated to notify the corresponding workstation to delete the original task to be processed.

[0253] 2. Encapsulation and issuance of control commands

[0254] Step A: Command message encapsulation

[0255] The generated process path adjustment instructions and scheduling task rearrangement instructions are encapsulated in a unified message format, which is JSON in this embodiment.

[0256] Step B: Send to the execution layer

[0257] The module sends the encapsulated message to the production line execution layer through the following communication methods:

[0258] For process path adjustment instructions: These are sent to the Manufacturing Execution System (MES) or Process Management System on the production line. Upon receiving the instruction, the MES updates the corresponding workpiece's process route in the database and automatically calls the new machining program when the workpiece is in transit.

[0259] For task rescheduling instructions: they are sent to each workstation controller (PLC or workstation terminal) and the Automated Material Handling System (AMHS). The workstation controller updates its local task queue according to the sequence number and start time in the instruction; the AMHS adjusts the handling sequence according to the new scheduled completion time.

[0260] The command is sent using a "confirmation-retransmission" mechanism: if the workstation controller does not return a confirmation response within 1 second, the module will resend the command, with a maximum of 3 retries; if it still fails, the fault information will be reported to the human-machine interface (HMI) and the execution of the command at that workstation will be suspended, awaiting manual intervention.

[0261] 3. Closed-loop feedback mechanism

[0262] To achieve continuous optimization, this module executes closed-loop feedback control after executing instructions, as follows:

[0263] Step A: Real-time collection of execution feedback data

[0264] The digital twin production line mapping module collects twin data after the production line is executed in real time at a fixed sampling period Δt (30 seconds in this embodiment), including the status of each workstation load, buffer queue length, processing cycle time, bottleneck diffusion domain, etc.

[0265] Step B: Calculate the current bottleneck diffusion domain coverage.

[0266] At the initial time t0 (i.e., the moment when the process path adjustment scheme and the scheduling task reordering scheme begin execution), the initial bottleneck diffusion domain size is recorded as |Ω. B (t0)|。In each subsequent sampling period tk=t0+k·Δt, calculate the size of the bottleneck diffusion domain |Ω at the current moment. B (tk)|, and calculate the coverage using the following formula:

[0267] ,

[0268] This ratio reflects the degree of expansion or contraction of the bottleneck domain after optimization. If Coverage(tk) < 1, it indicates that the bottleneck domain has shrunk and the optimization effect is positive; if Coverage(tk) > 1, it indicates that the bottleneck domain has expanded and the production line stability has deteriorated.

[0269] Step C: Triggering condition judgment

[0270] This module sets the following parameters:

[0271] Closed-loop feedback trigger threshold β: The value is 1.2 (meaning that re-optimization is triggered when the bottleneck domain expands to 1.2 times the initial size).

[0272] Number of consecutive periods K: The value is 3 (meaning that the trigger is only triggered when the threshold is exceeded for K consecutive sampling periods, avoiding frequent re-optimization caused by instantaneous fluctuations).

[0273] If Coverage(tk) > β holds true for K consecutive sampling periods, it is determined that the current optimization scheme can no longer effectively suppress bottleneck diffusion, and the production line status has changed significantly. At this time, a new round of optimization is triggered. The trigger signal is sent to the dynamic bottleneck diffusion identification module and the process path and scheduling coupling optimization module to re-execute bottleneck identification and coupling optimization, and generate new control instructions.

[0274] Step D: Maintain or update instructions

[0275] If the coverage rate does not exceed the threshold β within K consecutive cycles, the current instruction is considered to be still valid, the existing production line control instructions remain unchanged, and monitoring continues.

[0276] To accommodate slow production line drift (such as a gradual increase in cycle time due to tool wear), this module also sets a time window T. window =2 hours: Even if the threshold condition is not triggered, every T window It also proactively triggers an optimization to eliminate accumulated bias.

[0277] Step E: Forming a closed loop

[0278] Through the above mechanism, the digital twin production line mapping module, dynamic topology network construction module, dynamic bottleneck diffusion identification module, process path and scheduling coupling optimization module, and this module form a continuously operating closed-loop control loop: executing instructions → collecting data → monitoring bottleneck coverage → re-optimizing when necessary → issuing new instructions, thereby realizing adaptive intelligent management and control of the cylinder head flexible processing production line.

[0279] Intelligent control methods for cylinder head flexible manufacturing production lines, such as Figure 5 As shown, it includes the following steps:

[0280] S1: Obtain twin data of the cylinder head flexible processing production line and construct a digital twin production line model;

[0281] S2: Extract real-time operating status parameters of the production line from the digital twin production line model, and construct a dynamic topology network for the production line;

[0282] S3: Calculate the load entropy of each workstation node in the dynamic topology network of the production line, and determine the bottleneck diffusion domain based on the coupling diffusion strength matrix between nodes;

[0283] S4: Using the bottleneck diffusion domain as the constraint boundary, construct a coupled optimization model with the dual objectives of minimizing the maximum completion time and minimizing the bottleneck diffusion domain coverage. Use a multi-objective optimization algorithm to solve the coupled optimization model to generate an optimized solution set. Simultaneously output the process path adjustment scheme and the scheduling task rearrangement scheme from the optimized solution set.

[0284] S5: Transform the process path adjustment scheme and scheduling task rearrangement scheme into production line control instructions and issue them to the production line execution layer. Collect execution feedback data in real time and return to step S1 to form a closed loop for continuous optimization.

[0285] Example 6: Computer Equipment

[0286] This embodiment provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the intelligent control method for the flexible cylinder head manufacturing line described above. This computer device can be an industrial control computer, an embedded industrial computer, or a cloud server, used to deploy all or part of the modules of the intelligent control system.

[0287] Example 7: Computer-readable storage medium

[0288] This embodiment provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, it implements the steps of the intelligent control method for the flexible cylinder head manufacturing line described above. The storage medium can be a non-volatile storage medium such as ROM, RAM, hard disk, or solid-state drive.

[0289] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An intelligent control system for a cylinder head flexible machining production line, characterized in that, Includes the following modules: The digital twin production line mapping module is used to acquire twin data of the cylinder head flexible processing production line and build a digital twin production line model. A dynamic topology network construction module is connected to the digital twin production line mapping module and is used to extract real-time operating status parameters of the production line from the digital twin production line model and construct a dynamic topology network for the production line. The dynamic bottleneck diffusion identification module is connected to the dynamic topology network construction module and is used to calculate the load entropy of each workstation node in the production line dynamic topology network and determine the bottleneck diffusion domain based on the inter-node coupling diffusion strength matrix. The process path and scheduling coupling optimization module is connected to the dynamic bottleneck diffusion identification module. It is used to construct a coupled optimization model with the bottleneck diffusion domain as the constraint boundary, with the dual objectives of minimizing the maximum completion time and minimizing the bottleneck diffusion domain coverage. The coupled optimization model is solved by a multi-objective optimization algorithm to generate an optimized solution set. The process path adjustment scheme and the scheduling task rearrangement scheme are output synchronously from the optimized solution set. The production line control instruction generation and closed-loop execution module is connected to the process path and scheduling coupling optimization module. It is used to convert the process path adjustment scheme and scheduling task rearrangement scheme into production line control instructions and issue them to the production line execution layer. It collects execution feedback data in real time and inputs it into the digital twin production line mapping module to form a closed loop.

2. The intelligent control system for the cylinder head flexible processing production line according to claim 1, characterized in that, The dynamic topology network construction module constructs the production line dynamic topology network in the following manner: The following real-time operating status parameters of the production line are extracted from the digital twin production line model: the number of workpieces to be processed in the current buffer of each workstation, the measured value of the processing cycle of each workstation, the health status index of the equipment at each workstation, the model of the workpiece currently being processed at each workstation, and the material handling time between adjacent workstations. Each processing station is a network node. The first connection condition is the existence of material flow relationship between adjacent stations, and the second connection condition is the existence of process sequence constraint relationship between two stations. A directed connection between nodes is established when at least one of the first or second connection conditions is met. Each node is assigned an attribute value, which includes at least the current comprehensive load value Li and the rated processing capacity Ci of the workstation. The comprehensive load value Li is calculated by weighting the number of workpieces to be processed in the current buffer of the workstation, the actual value of the processing cycle, and the standard working hours corresponding to the workpiece model. Assign a weight value w to each directed edge. ij The weight value is calculated and determined based on the ratio of the material handling time between the two ends of the connecting station to the measured value of the processing cycle time of the downstream station; The constructed set of nodes, set of directed edges, set of node attribute values, and set of edge weight values ​​are combined to form a dynamic production line topology network G(V, E, A, W) represented by an adjacency matrix, where V is the set of nodes, E is the set of directed edges, A is the node attribute matrix, and W is the edge weight matrix.

3. The intelligent control system for the cylinder head flexible processing production line according to claim 1, characterized in that, The dynamic bottleneck diffusion identification module determines the bottleneck diffusion domain in the following way: For each workstation node i in the dynamic topology network of the production line, calculate its load entropy value Hi: Arrange the calculated load entropy values ​​Hi of each workstation node in descending order, and take the node with the highest load entropy value as the main bottleneck node B0 at the current moment. Starting from the main bottleneck node B0, based on the pre-constructed coupling diffusion strength matrix D, the propagation intensity of the bottleneck effect is calculated step by step along the directed connection directions of the production line dynamic topology network: for node j, which is d hops away from node B0, the bottleneck propagation intensity Sj = S B0 ·ΠD k，k+1 ·α^d, where S B0 The bottleneck strength value of the primary bottleneck node, D k，k+1 is the coupling diffusion coefficient between adjacent nodes on the propagation path, and α is the propagation attenuation factor and α∈(0,1); The set of all workstation nodes whose propagation intensity Sj exceeds the preset diffusion threshold θ, along with the main bottleneck node B0, is defined as the bottleneck diffusion domain Ω at the current moment. B .

4. The intelligent control system for the cylinder head flexible processing production line according to claim 3, characterized in that, The coupling diffusion strength matrix D is constructed as follows: Obtain the time-series data of workstation status during each bottleneck event recorded in the historical operation data of the production line; For each bottleneck event, extract the load change time series between the bottleneck workstation and other workstations, calculate the time-delay cross-correlation function between each pair of workstations, and take the maximum cross-correlation coefficient as the coupling diffusion coefficient between the workstation pairs. For all bottleneck events, the average coupling diffusion coefficients of similar workstation pairs are used to construct a coupling diffusion strength matrix D=[d mn ], where d mn This represents the coupling diffusion strength of the bottleneck effect of workstation m to workstation n.

5. The intelligent control system for the cylinder head flexible processing production line according to claim 1, characterized in that, The process path and scheduling coupling optimization module constructs the coupling optimization model in the following way: With the bottleneck diffusion domain Ω B To constrain the boundary, the first optimization objective is defined as minimizing the maximum completion time: minf1 = (Ci), where Ci is the completion time of workstation i, and V is the set of workstation nodes; The second optimization objective is defined as minimizing the bottleneck diffusion domain coverage: minf2 = |Ω B '| / |Ω B |, where |Ω B |To optimize the number of workstations within the bottleneck diffusion domain,|Ω B '| represents the number of workstations within the bottleneck diffusion domain after applying the current optimized solution; Establish a constraint set consisting of the following constraints: processing capacity constraints of each workstation within the bottleneck diffusion domain, processing characteristic constraints of alternative process paths, changeover time constraints for switching between different types of cylinder heads, material handling resource constraints, and order delivery date constraints. A multi-objective weighted function is constructed based on the first and second optimization objectives, and solved using a multi-objective optimization algorithm based on non-dominated sorting. The constraint set is set as the feasible region boundary.

6. The intelligent control system for the cylinder head flexible processing production line according to claim 5, characterized in that, The process path and scheduling coupling optimization module solves the coupling optimization model in the following way: Based on the bottleneck diffusion domain Ω B Determine the set of affected workpieces J that require process path adjustments. aff ; For the affected workpiece set J aff For each workpiece in the dataset, retrieve the set P of alternative process paths corresponding to that workpiece model. alt Each path in the set of alternative process paths satisfies the processing characteristic constraint of the alternative process path; For each alternative process path, calculate the expected load change sequence for each station along that path, and then compare this expected load change sequence with the initial bottleneck diffusion domain Ω. B By comparing the results, we can predict the direction of change in the bottleneck diffusion domain coverage after adopting this alternative process path. The decision to select alternative process routes is encoded as a decision variable, and the decision variables of scheduling tasks are rearranged, including work order sequence number and resource allocation number, and then incorporated into a unified coding scheme to form a set of decision variables for a multi-objective coupled optimization problem. The NSGA-III optimization algorithm is used to perform a population iterative search within the feasible region defined by the constraint set, and outputs a non-dominated solution set.

7. The intelligent control system for the cylinder head flexible processing production line according to claim 1, characterized in that, The production line control instruction generation and closed-loop execution module completes closed-loop control in the following ways: The optimal solution in the optimized solution set is decoded into a process path adjustment instruction and a scheduling task reordering instruction. The process path adjustment instruction includes the workpiece ID, the original process path number, the new process path number, and the workstation number corresponding to each step. The scheduling task reordering instruction includes the workpiece ID, the adjusted work order sequence number, the assigned target workstation number, and the expected start time. The process path adjustment instructions and scheduling task rearrangement instructions are encapsulated into production line control instruction messages and sent to each workstation controller in the production line execution layer. The digital twin production line mapping module collects twin data in real time after the production line execution, and calculates the bottleneck diffusion domain coverage |Ω at the current moment. B (t)| / |Ω B (t0)|, where t0 is the initial time when the process path adjustment scheme and the scheduling task rearrangement scheme begin to be executed. If the coverage exceeds the preset closed-loop feedback trigger threshold β for K consecutive collection cycles, a new round of bottleneck diffusion domain determination and coupling optimization solution will be triggered; otherwise, the current production line control instructions will remain unchanged.

8. A method for intelligent control of a cylinder head flexible machining production line, applied to the intelligent control system for a cylinder head flexible machining production line as described in any one of claims 1 to 8, characterized in that, Includes the following steps: S1: Obtain twin data of the cylinder head flexible processing production line and construct a digital twin production line model; S2: Extract real-time operating status parameters of the production line from the digital twin production line model, and construct a dynamic topology network for the production line; S3: Calculate the load entropy of each workstation node in the dynamic topology network of the production line, and determine the bottleneck diffusion domain based on the coupling diffusion strength matrix between nodes; S4: Using the bottleneck diffusion domain as the constraint boundary, construct a coupled optimization model with the dual objectives of minimizing the maximum completion time and minimizing the bottleneck diffusion domain coverage. Use a multi-objective optimization algorithm to solve the coupled optimization model to generate an optimized solution set. Simultaneously output the process path adjustment scheme and the scheduling task rearrangement scheme from the optimized solution set. S5: Transform the process path adjustment scheme and scheduling task rearrangement scheme into production line control instructions and issue them to the production line execution layer. Collect execution feedback data in real time and return to step S1 to form a closed loop for continuous optimization.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the intelligent control method for the cylinder head flexible machining production line as described in claim 9.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the intelligent control method for the cylinder head flexible machining production line as described in claim 9.

Images