Intelligent manufacturing optimization method and system based on process mining and lean digital twinning

Abstract

The invention discloses an intelligent manufacturing optimization method and system based on process mining and lean digital twinning, and belongs to the technical field of computer application and industrial manufacturing. The method comprises the following steps: extracting real-time production data from a heterogeneous data source and converting the real-time production data into a process event log; analyzing the process event log through a process mining engine, and constructing a dynamic value flow digital twinborn body; performing lean state diagnosis on the process event log on the dynamic value flow digital twinborn body to generate a corresponding diagnosis result; based on the diagnosis result, taking the dynamic value flow digital twin as a simulation base, performing hypothesis analysis through a simulation engine, predicting the influence of different improvement measures on the key performance indicators, generating an optimization strategy, and pushing the optimization strategy to a production execution system; and on the basis of a newly generated process event log after the optimization strategy is executed, repeating automatic discovery and construction of the dynamic value flow digital twin and subsequent real-time diagnosis steps so as to carry out continuous verification and iterative optimization on an optimization effect.

Description

Technical Field

[0001] This application relates to the fields of computer applications and industrial manufacturing technology, specifically to an intelligent manufacturing optimization method and system based on process mining and lean digital twins. Background Technology

[0002] Lean Manufacturing is a widely recognized management system in industry for eliminating waste and improving efficiency. Its core tools include Value Stream Mapping (VSM), Overall Equipment Effectiveness (OEE) analysis, the Andon system, and Kaizen continuous improvement. However, traditional lean management practices are highly reliant on manual labor. For example, creating a VSM requires expert teams to spend weeks on-site conducting stopwatch measurements and interviews, and OEE statistics rely on manually filling out reports. This approach is not only inefficient and costly, but the data obtained is often outdated and subjective.

[0003] With the advancement of digitalization, attempts have emerged to "kanban" lean tools, such as displaying OEE calculation results on electronic screens. However, this "first-generation" digital lean system merely visualizes existing, static, and idealized standard operating procedures (SOPs), and it cannot reflect the actual (As-Is) production process.

[0004] Actual production processes are highly dynamic and subject to change. Existing lean manufacturing management systems, such as Manufacturing Execution Systems (MES) or Enterprise Resource Planning (ERP) systems, while recording results, cannot transparently display the process. This leads to several serious technical problems: 1. Static model is out of touch with dynamic reality: The pre-set SOP is inconsistent with the actual operation process of workers (such as non-standard operation, abnormal detour), which leads to management based on "what should be" rather than "what is".

[0005] 2. Hidden waste is difficult to detect: Among the traditional "seven wastes", process wastes such as "waiting", "unnecessary movement" or "rework cycle" are not visible in the aggregated reports of the MES system, and have become "dark data" of management.

[0006] 3. Unpredictable Improvement Measures: Any improvement (such as adjusting production line layout, replacing equipment, or modifying batches) carries risks. Without effective forecasting tools, companies can only resort to trial and error, which is costly and may disrupt production.

[0007] Therefore, traditional lean manufacturing management systems suffer from technical problems such as a disconnect between static models and reality, difficulty in identifying hidden waste, and unpredictable improvement effects. How to automatically discover the actual manufacturing process, how to diagnose hidden process waste, and how to predictively verify the effectiveness of improvement measures are the problems that this application urgently needs to solve. Summary of the Invention

[0008] The embodiments of this disclosure provide an intelligent manufacturing optimization method and system based on process mining and lean digital twins, which at least solves the technical problems existing in the prior art, such as the disconnect between static models and reality, difficulty in detecting hidden waste, and unpredictable improvement effects in lean manufacturing management systems.

[0009] According to one aspect of the present disclosure, a smart manufacturing optimization method based on process mining and lean digital twins is provided, comprising: extracting real-time production data from heterogeneous data sources and converting the real-time production data into standardized process event logs, the process event logs including at least case IDs, activities, and timestamps; analyzing the process event logs through a process mining engine to automatically discover and construct a dynamic value stream digital twin reflecting the actual production process, the dynamic value stream digital twin dynamically evolving as the process event logs are updated; performing real-time lean status diagnosis on the process event logs on the dynamic value stream digital twins to generate corresponding diagnostic results; based on the diagnostic results, using the dynamic value stream digital twins as a simulation base, performing hypothetical analysis through a simulation engine to predict the impact of different improvement measures on key performance indicators, generating corresponding optimization strategies and pushing them to the production execution system; and repeating the steps of automatically discovering and constructing the dynamic value stream digital twins and subsequent real-time diagnosis based on the newly generated process event logs after executing the optimization strategies, to continuously verify and iteratively optimize the optimization effect.

[0010] According to another aspect of the present disclosure, a storage medium is also provided, the storage medium including a stored program, wherein the methods described above are executed by a processor when the program is running.

[0011] According to another aspect of the present disclosure, a smart manufacturing optimization platform based on process mining and lean digital twins is also provided, comprising: an infrastructure layer, a process event and fusion layer, a smart lean twin layer, a decision and simulation optimization layer, and an application and improvement layer; the smart lean twin layer includes a process mining engine and a lean diagnostic module; the decision and simulation optimization layer includes a simulation engine and a decision execution module; the infrastructure layer is used to build an industrial network, edge computing, and cloud platform for running the smart manufacturing optimization platform; the process event and fusion layer is used to connect heterogeneous data sources, extract real-time production data from heterogeneous data sources, and convert the real-time production data into standardized process event logs, the process event logs including at least case ID, activity, and timestamp; the process mining engine is used to analyze the process event logs, automatically discover and build a system reflecting... A dynamic value stream digital twin of the actual production process is constructed, which evolves dynamically as process event logs are updated. A lean diagnostic module performs real-time lean status diagnostics on the process event logs within the dynamic value stream digital twin, generating corresponding diagnostic results. A simulation engine, based on the diagnostic results and using the dynamic value stream digital twin as a simulation base, performs hypothetical analysis to predict the impact of different improvement measures on key performance indicators. A decision execution module generates corresponding optimization strategies based on the simulation results and pushes them to the production execution system. The application and improvement layer repeats the automatic discovery and construction of the dynamic value stream digital twin and subsequent real-time diagnostic steps based on newly generated process event logs after executing the optimization strategy, to continuously verify and iteratively optimize the optimization effect.

[0012] According to another aspect of the present disclosure, a smart manufacturing optimization system based on process mining and lean digital twins is also provided, comprising: a production physical system and a smart manufacturing optimization platform; the production physical system includes an information system and physical assets; the smart manufacturing optimization platform interacts with the production physical system to execute the above-described smart manufacturing optimization method.

[0013] This application first extracts real-time production data from heterogeneous data sources and converts it into standardized process event logs, providing a foundation for building data-driven dynamic analysis. Then, it analyzes the process event logs using a process mining engine, automatically discovering and constructing a dynamic value stream digital twin reflecting the actual production process, achieving a technological leap from a "static ideal model" to a "dynamic reality mapping." Next, it performs real-time lean status diagnosis on the process event logs within the dynamic value stream digital twin, accurately locating bottlenecks and waste in the production process. Subsequently, using this digital twin as a simulation base, it conducts hypothetical analysis through a simulation engine to generate predictive optimization strategies, ensuring the reliability and predictability of the effectiveness evaluation of improvement measures. Afterward, the generated optimization strategies are pushed to the production execution system for execution, achieving a closed loop from virtual optimization to physical execution. Finally, based on newly generated event logs after execution, the construction and diagnosis steps are repeated, achieving continuous verification and iterative optimization of the improvement effect. Thus, this application, through the deep integration of process mining, dynamic digital twins, and simulation optimization, achieves the technical effects of making lean manufacturing processes transparent, making hidden waste explicit, and making improvement decisions more scientific. This solves the technical problems existing in the lean manufacturing management system, such as the disconnect between static models and reality, difficulty in detecting hidden waste, and unpredictable improvement effects. Attached Figure Description

[0014] The accompanying drawings, which are included to provide a further understanding of this disclosure and form part of this application, illustrate exemplary embodiments of this disclosure and are used to explain this disclosure, but do not constitute an undue limitation of this disclosure. In the drawings: Figure 1 This is an application scenario diagram of the intelligent manufacturing optimization platform based on process mining and lean digital twins described in the embodiments of this application; Figure 2 This is a flowchart of the intelligent manufacturing optimization method based on process mining and lean digital twins described in the embodiments of this application; Figure 3 This is a schematic diagram of the structure of the intelligent manufacturing optimization platform described in the embodiments of this application. Detailed Implementation

[0015] To enable those skilled in the art to better understand the technical solutions of this disclosure, the technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this disclosure, and not all embodiments. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this disclosure.

[0016] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0017] Key terms defined in this application: 1. Event Log: This is the data foundation of this platform. It is a data table with a specific format, where each row represents an event and must contain at least three columns of information: Case ID (identifying a process instance, such as an order), Activity (a step in the process, such as "start welding"), and Timestamp (the time the activity occurred).

[0018] 2. Process Mining: This is a set of techniques for analyzing processes from event logs. It mainly includes: 1. Discovery: Automatically drawing the actual process flowchart; 2. Conformance: Comparing the actual process with a standard process; 3. Performance Diagnosis: Overlaying time, cost, and other data onto the flowchart to analyze bottlenecks.

[0019] Example According to this embodiment, an intelligent manufacturing optimization platform based on process mining and lean digital twins is provided. Figure 1 The diagram illustrates application scenarios for the intelligent manufacturing optimization platform. (For reference) Figure 1As shown, this application scenario includes a terminal device 100 for technicians / managers 110, an intelligent manufacturing optimization platform 200, and a production physical system 300. The terminal device 100 provides a visual interactive interface for technicians / managers 110 to monitor production status, receive optimization decision suggestions, and perform manual intervention. The intelligent manufacturing optimization platform 200, as the core processing unit, deploys an intelligent analysis system including a process mining engine, a dynamic value stream digital twin, and a simulation engine. The production physical system includes information systems (such as MES, ERP, SCM (Supply Chain Management System)) and physical assets (such as PLCs (Programmable Logic Controllers), production lines, and sensors), responsible for providing production management data, receiving and executing optimization decisions, and collecting and feeding back real-time production data through physical assets.

[0020] Under the aforementioned operating environment, according to the first aspect of this embodiment, a smart manufacturing optimization method based on process mining and lean digital twins is provided, through... Figure 1 The intelligent manufacturing optimization platform 200 shown is implemented. Figure 2 A flowchart illustrating the method is shown below. (Refer to...) Figure 2 As shown, the method includes: S201: Extract real-time production data from heterogeneous data sources and convert the real-time production data into standardized process event logs, the process event logs including at least case ID, activity and timestamp; S202: The process event log is analyzed by the process mining engine to automatically discover and construct a dynamic value stream digital twin that reflects the actual production process. The dynamic value stream digital twin evolves dynamically as the process event log is updated. S203: On the dynamic value stream digital twin, perform real-time lean status diagnosis on the process event log and generate corresponding diagnostic results; S204: Based on the diagnostic results, using the dynamic value stream digital twin as the simulation base, perform hypothetical analysis through the simulation engine to predict the impact of different improvement measures on key performance indicators, generate corresponding optimization strategies, and push them to the production execution system. S205: Based on the newly generated process event logs after executing the optimization strategy, repeat the steps of automatically discovering and constructing a dynamic value stream digital twin and subsequent real-time diagnostics to continuously verify and iteratively optimize the optimization effect.

[0021] In this embodiment of the invention, real-time production data is first extracted from heterogeneous data sources and converted into standardized process event logs (corresponding to step S201). Specifically, by connecting to heterogeneous data sources such as manufacturing execution systems, programmable logic controllers, and sensors, real-time production data is extracted from these heterogeneous data sources, including order information, process records, equipment status signals, and process parameter readings. Then, data mapping rules are established to uniformly convert the heterogeneous data into standardized process event logs containing three elements: case ID, activity, and timestamp, providing a time-complete and formatted data foundation for subsequent process mining and twin construction.

[0022] Then, the process event logs are analyzed by a process mining engine to automatically discover and construct a dynamic value stream digital twin reflecting the actual production process (corresponding to step S202). Specifically, using process discovery algorithms in process mining technology (such as Alpha Miner and Heuristics Miner), the logical network of the actual business process is automatically derived from the event logs and visualized as a dynamic value stream digital twin. This twin is not a traditional 3D model twin of equipment, but a "value stream" twin. It is a process model discovered by the process mining engine, reflecting the "As-Is" process, and dynamically evolving as the event logs are updated. Thus, a precise mapping from a static preset process to a dynamic real process is achieved.

[0023] Next, on the dynamic value stream digital twin, real-time lean status diagnosis is performed on the process event logs to generate corresponding diagnostic results (corresponding to step S203). Specifically, based on the process topology and event sequence revealed by the dynamic value stream digital twin, abnormal patterns in the process are identified and quantified. For example, non-standard operation loops are discovered through path analysis, and time-wasting nodes are located through timestamp analysis. Thus, hidden waste that is difficult to detect in traditional management becomes measurable and analyzable.

[0024] Next, based on the diagnostic results, and using the dynamic value stream digital twin as a simulation platform, a hypothetical analysis is performed through a simulation engine to predict the impact of different improvement measures on key performance indicators (KPIs). Corresponding optimization strategies are then generated and pushed to the production execution system (corresponding to step S204). Specifically, the bottlenecks and waste points identified in the diagnosis are used as simulation inputs. Intervention measures such as resource reallocation and scheduling strategy adjustments are simulated on the twin, and their impact on key indicators such as overall equipment efficiency and delivery cycle is quantitatively predicted. Based on this, a data-driven optimization strategy is formed and pushed to the production execution system. Thus, the effectiveness of improvement measures can be scientifically evaluated before actual resources are invested, reducing trial-and-error costs.

[0025] Finally, based on the newly generated process event logs after executing the optimization strategy, the steps of automatically discovering and constructing a dynamic value stream digital twin and subsequent real-time diagnostics are repeated to continuously verify and iteratively optimize the optimization effect (corresponding to step S205). Specifically, the newly generated event logs after optimization are used as input, and steps S201 to S203 are re-executed. By comparing the process structure and performance indicators of the twin before and after optimization, the improvement effect is automatically verified, and the next round of optimization cycle is triggered based on the verification results. Thus, a continuous improvement closed loop of "monitoring-diagnosis-optimization-verification" is formed.

[0026] As described in the background section, existing lean manufacturing management systems, while recording results, suffer from problems such as a disconnect between static models and dynamic reality, difficulty in detecting hidden waste, and unpredictability of improvement measures.

[0027] In light of this, firstly, real-time production data is extracted from heterogeneous data sources and converted into standardized process event logs, providing a foundation for building data-driven dynamic analysis. Then, the process event logs are analyzed using a process mining engine to automatically discover and construct a dynamic value stream digital twin reflecting the actual production process, achieving a technological leap from a "static ideal model" to a "dynamic reality mapping." Secondly, real-time lean status diagnosis is performed on the process event logs within the dynamic value stream digital twin, accurately locating bottlenecks and waste in the production process. Subsequently, using this digital twin as a simulation base, hypothetical analysis is conducted through a simulation engine to generate predictive optimization strategies, ensuring the reliability and predictability of the effectiveness evaluation of improvement measures. Afterward, the generated optimization strategies are pushed to the production execution system for execution, achieving a closed loop from virtual optimization to physical execution. Finally, based on newly generated event logs after execution, the construction and diagnosis steps are repeated, achieving continuous verification and iterative optimization of the improvement effects. Thus, this application, through the deep integration of process mining, dynamic digital twins, and simulation optimization, achieves the technical effects of making lean manufacturing processes transparent, revealing hidden waste, and making improvement decisions more scientific. This solves the technical problems existing in the lean manufacturing management system, such as the disconnect between static models and reality, difficulty in detecting hidden waste, and unpredictable improvement effects.

[0028] Optionally, the operation of converting the real-time production data into a standardized process event log includes: binding the order number or batch number from the manufacturing execution system to a case ID of a complete production process instance; binding the equipment status, process reporting, and quality inspection results from the programmable logic controller, manufacturing execution system, or sensors to the activities of the production process instance; and aggregating and sorting all relevant activities based on the case ID to form the standardized process event log.

[0029] In this embodiment of the invention, a data acquisition interface is first established, connecting the database of the Manufacturing Execution System (MES), the status register of the Programmable Logic Controller (PLC), and the data stream of the sensor network. For each production process instance, a corresponding case ID is created using the work order number in the MES as a unique identifier. Then, by configuring data mapping rules, signals such as "equipment start / stop" and "processing status" transmitted by the PLC, nodes such as "process start" and "process completion" recorded in the MES, and events such as "temperature exceeding limit" and "vibration abnormality" detected by sensors are uniformly converted into activity names with clear semantic descriptions. Finally, by aligning timestamps and associating case IDs, activity records scattered across different data sources are arranged in ascending order according to their occurrence time, forming a process event log with a complete time sequence.

[0030] This breaks down the data barriers between the manufacturing execution system, equipment control system, and sensor network, reorganizing fragmented production information that was originally scattered across multiple heterogeneous systems into an event sequence with a unified standard format that fully reflects the entire production process, providing a high-quality and standardized input data foundation for subsequent process mining and analysis.

[0031] Optionally, the operation of analyzing the process event logs through a process mining engine to automatically discover and construct a dynamic value stream digital twin reflecting the actual production process includes: using a built-in process discovery algorithm to mine and analyze the standardized process event logs, identifying the various activities contained therein and the transition relationships between the activities; constructing a directed graph computation model based on the identified activities and transition relationships; wherein the nodes of the directed graph correspond to the activities, and the edges of the directed graph correspond to the transition relationships between the activities; and generating a dynamic value stream digital twin reflecting the actual production process with the directed graph computation model as its core.

[0032] In this embodiment of the invention, the process mining engine first uses its built-in process discovery algorithm to parse standardized process event logs and automatically identify all unique activity units by statistically analyzing activity names. Next, the process mining engine groups and sorts events by "case ID," reconstructing the complete activity sequence for each production order. Then, by statistically analyzing the direct follow-up relationships between activities and applying dependency measurement algorithms (such as calculating the ratio of the frequency of activity A following activity B to the frequency of activity B following activity A) to quantify the strength of causal relationships, noise is filtered out based on preset thresholds, thereby accurately identifying stable and meaningful process transition paths.

[0033] Then, the identified activities and transition relationships are formalized into a concrete directed graph G=(N,E) computational model. The node set N of this directed graph consists of all identified activities, with each node recording its name and total frequency of occurrence. The edge set E consists of all identified transition relationships. Each edge pointing from activity A to activity B serves as a data carrier, assigned key performance attributes such as transition frequency, transition probability, and average waiting time and processing time calculated based on timestamps, all calculated from the original logs.

[0034] Subsequently, a dynamic value stream digital twin, based on a directed graph computation model, is generated to reflect the actual production process. This dynamic value stream digital twin is technically implemented as a software service that encapsulates the aforementioned directed graph computation model. Its "dynamic evolution" characteristic is ensured through an incremental update mechanism: this mechanism continuously monitors newly arriving process event logs and triggers real-time updates to the directed graph nodes, edges, and their attributes. Simultaneously, the service has version management capabilities, saving model snapshots at different points in time to support comparative analysis of the process before and after optimization.

[0035] Thus, through the aforementioned progressively advanced technical implementation, a dynamic value stream digital twin is constructed, which is rooted in real production data, can evolve synchronously with actual production changes, and contains rich process performance information, providing a reliable process foundation for subsequent lean diagnosis and simulation optimization.

[0036] Optionally, the operation of performing real-time lean status diagnosis on the process event log and generating corresponding diagnostic results on the dynamic value stream digital twin includes: calculating the average waiting time for transfers between activities in the process event log based on the time-series data of the edges in the directed graph computation model; automatically identifying loop paths composed of activity nodes and transfer relationship edges based on the topology of the directed graph computation model, and quantifying the occurrence frequency and duration of the identified loop paths based on the process event log; and generating diagnostic results reflecting process bottlenecks and rework waste based on the average waiting time and the occurrence frequency and duration.

[0037] In this embodiment of the invention, firstly, for each edge in the directed graph, the timestamps of all corresponding transition instances in the process event log are extracted. By calculating the difference between the completion time and start time of adjacent activities, the average waiting time of each transition edge is calculated, and the activity connected to the edge with the longest waiting time is identified as the current process bottleneck. Simultaneously, a depth-first search algorithm is used to traverse the directed graph topology, detecting strongly connected components and identifying loop paths such as "quality inspection failure → rework → re-quality inspection," which consist of multiple nodes and edges. Next, based on the process event log, the number of times each identified loop path appears in all production cases is statistically analyzed using case IDs, and the total execution time for each loop is accumulated, thereby quantifying the scale of rework waste.

[0038] Finally, based on average waiting time and frequency of occurrence, diagnostic results are generated to reflect process bottlenecks and rework waste. For example, a diagnostic report containing specific quantitative data is generated: the bottleneck analysis section clearly points out an abnormal waiting time of 12 minutes on average in the "material loading to processing" stage, recommending priority optimization; the waste analysis section reveals that the "quality inspection failure → rework" cycle occurs 8.5% of the time, resulting in a cumulative monthly production capacity loss of 42 man-hours, and recommends carrying out special quality improvement projects. These diagnostic results are pushed to managers in real time through a visual dashboard, providing data support for precise improvement.

[0039] This transforms time waste (such as abnormal waiting) and quality waste (such as rework cycles) that are difficult to accurately capture in traditional lean management into quantifiable data indicators, making the hidden costs in the production process explicit and providing clear data support and decision-making basis for precise improvement.

[0040] Optionally, the method further includes: comparing the directed graph computation model with a preset standard operating procedure, the comparison including path consistency check and / or activity sequence integrity check; based on the comparison result, calculating the deviation between the actual operating procedure represented by the directed graph computation model and the standard operating procedure; when the deviation exceeds a preset threshold, automatically generating and issuing an alarm message containing specific deviation path information.

[0041] In this embodiment of the invention, a preset standard operating procedure is first modeled as a baseline directed graph, which defines standard process nodes and their legal transfer paths. During path consistency checks, all paths in the actual production directed graph calculation model are traversed to identify illegal transfer edges that do not exist in the baseline directed graph, such as "detour quality inspection" and other non-compliant paths. During activity sequence integrity checks, it is verified whether the actual process includes all key process nodes specified in the standard process, and whether their execution order conforms to preset constraints. Based on the check results, a comprehensive deviation is quantified by calculating the weighted sum of the frequency of illegal paths and the rate of missing key activities. When this deviation exceeds a preset threshold (e.g., 15%), a structured alarm is automatically generated, clearly identifying the specific non-compliant path, the involved case ID, and the severity level of the deviation.

[0042] This enables automated monitoring and quantitative evaluation of the compliance of actual production processes with standard specifications, allowing deviations from standard operating procedures to be detected, located accurately, and responded to quickly, effectively preventing quality inconsistencies and efficiency losses caused by long-term deviations from standard operating procedures.

[0043] Optionally, based on the diagnostic results, using the dynamic value stream digital twin as a simulation base, a hypothetical analysis is performed through a simulation engine to predict the impact of different improvement measures on key performance indicators (KPIs) and generate corresponding optimization strategies. This includes: simulating the execution of at least one predefined improvement scenario on the dynamic value stream digital twin; calculating the predicted values ​​of one or more KPIs based on the output of the simulation; comparing the predicted values ​​of KPIs under different improvement scenarios to generate an optimization strategy that includes recommended improvement measures and their expected effects.

[0044] In this embodiment of the invention, based on the specific problems identified by the diagnostic results, parameters for a "What-if" analysis scenario are configured on the dynamic value stream digital twin to simulate the execution of at least one predefined improvement scenario. For example, for the identified "rework cycle" problem, a simulation parameter of "adding one operator to the rework path" is set; as another example, for the discovered "bottleneck process A," a simulation parameter of "adjusting the batch size from 100 to 50" is set. The simulation engine uses discrete event simulation technology, based on the topology, transition probabilities, and temporal performance data of the directed graph computation model, to simulate the complete flow process of production orders under each improvement scenario.

[0045] By running "What-if" analyses, the impact of different improvement measures on key performance indicators (KPIs) can be predicted. For example, Simulation 1 predicts that adding one operator to the rework path will shorten the lead time by 18%; Simulation 2 predicts that adjusting the batch size of bottleneck process A to 50 will reduce work-in-process (WIP) by 25% and improve overall equipment efficiency (OEE) by 5%. Based on these predictions, a multi-objective decision matrix can be established, comprehensively considering the overall performance of each improvement scheme in terms of lead time, WIP inventory, and equipment efficiency, to generate an optimization strategy that includes specific recommended measures and their quantified expected effects.

[0046] Thus, a virtual testing environment based on a real process model was constructed, enabling managers to objectively evaluate the expected effects of various improvement plans before implementation through "What-if" analysis. This effectively avoids the limitations of relying on experience-based judgment in traditional improvement activities, realizes data-driven scientific decision-making, and significantly improves the success rate and return on investment of improvement activities.

[0047] Optionally, the method further includes: querying a pre-built lean knowledge graph based on the waste phenomena or performance indicators contained in the diagnostic results, obtaining one or more recommended countermeasures, and providing the recommended countermeasures to the manager; wherein the lean knowledge graph stores semantic relationship triples between waste phenomena, root causes, and recommended countermeasures.

[0048] In this embodiment of the invention, a lean knowledge graph incorporating knowledge from the manufacturing sector is first constructed. A graph database is used to store semantic triples representing "waste phenomenon - root cause - recommended countermeasures," such as establishing relationships like ("low OEE", "possible cause", "excessive equipment changeover time") and ("excessive equipment changeover time", "recommended countermeasures", "implement SMED rapid mold changeover"). When the diagnostic module identifies a specific waste phenomenon or abnormal performance indicator, the system semantically matches the key features in the diagnostic results with the waste phenomenon nodes in the knowledge graph. A graph traversal algorithm automatically discovers the associated root cause nodes and recommended countermeasure nodes. For example, when both "OEE below 85%" and "equipment changeover time exceeding 15%" are diagnosed simultaneously, the system can obtain an improvement suggestion of "implementing SMED rapid mold changeover" through graph reasoning. This suggestion and its reasoning basis are then pushed to the manager through the decision support module of the management interface.

[0049] This transforms the traditional improvement decision-making process, which relies on expert experience, into an intelligent reasoning process based on knowledge graphs. This significantly improves the depth of problem analysis and the accuracy of countermeasure recommendations, enabling managers to quickly obtain knowledge-verified improvement directions. It effectively shortens the decision-making cycle from problem diagnosis to countermeasure formulation and enhances the scientific and systematic nature of lean improvement.

[0050] In summary, the beneficial effects of the embodiments of this application are as follows: 1. Transform from "static" to "dynamic": Through the process mining engine, the real value stream map is automatically "discovered" from the production data, and a "dynamic lean digital twin" that is synchronously mapped with the physical world is constructed, solving the problem of the disconnect between the model and reality.

[0051] 2. Transform "implicit" into "explicit": By analyzing the process paths and timestamps of the twin, it is possible to accurately locate and quantify "implicit waste" such as rework cycles and abnormal waiting, making lean diagnosis more evidence-based.

[0052] 3. Shift from "trial and error" to "prediction": Through simulation engine, "What-if" analysis is performed on improvement measures on digital twins, which can predict their impact on OEE, WIP and Lead Time before actual resources are invested, greatly reducing the risk and cost of improvement activities.

[0053] 4. Transform from "passive" to "intelligent": By introducing lean knowledge graphs, the platform can proactively recommend improvement strategies after diagnosing problems, assisting managers in decision-making and realizing an intelligent closed loop of "diagnosis-recommendation-simulation-execution".

[0054] According to a second aspect of this embodiment, a smart manufacturing optimization platform 200 based on process mining and lean digital twins is provided. Figure 3 A schematic diagram of the platform's structure is shown for reference. Figure 3As shown, the intelligent manufacturing optimization platform 200 includes: an infrastructure layer 210, a process event and fusion layer 220, an intelligent lean twin layer 230, a decision and simulation optimization layer 240, and an application and improvement layer 250. The intelligent lean twin layer 230 includes a process mining engine and a lean diagnostic module. The decision and simulation optimization layer includes a simulation engine and a decision execution module. The infrastructure layer 210 is used to build an industrial network, edge computing, and cloud platform for running the intelligent manufacturing optimization platform. The process event and fusion layer 220 is used to connect heterogeneous data sources, extract real-time production data from heterogeneous data sources, and convert the real-time production data into standardized process event logs, which at least include case ID, activity, and timestamp. The process mining engine is used to analyze the process event logs, automatically discover and construct a system reflecting the actual production flow. The system comprises a dynamic value stream digital twin that evolves dynamically with updates to process event logs; a lean diagnostic module that performs real-time lean status diagnostics on the process event logs based on the dynamic value stream digital twin, generating corresponding diagnostic results; a simulation engine that, based on the diagnostic results and using the dynamic value stream digital twin as a simulation base, performs hypothetical analysis to predict the impact of different improvement measures on key performance indicators; a decision execution module that generates corresponding optimization strategies based on the simulation results and pushes them to the production execution system; and an application and improvement layer 250 that, based on newly generated process event logs after executing the optimization strategies, repeats the automatic discovery and construction of the dynamic value stream digital twin and subsequent real-time diagnostic steps to continuously verify and iteratively optimize the optimization effect.

[0055] In embodiments of the present invention, such as Figure 1 As shown, the application scenario of this application includes three main interactive objects: the terminal device 100 of the technician / manager 110, the intelligent manufacturing optimization platform 200, and the production physical system 300. The production physical system 300 includes information systems (such as MES and ERP) and physical assets (such as PLCs, machine tools, and sensors). The intelligent manufacturing optimization platform 200 collects data from the production physical system 300 through the process event and fusion layer, performs intelligent analysis and simulation, and feeds back optimization decisions to the physical system 300 for execution, or pushes insights to the terminal device 100 of the manager 110 through the application and improvement layer.

[0056] like Figure 3As shown, the intelligent manufacturing optimization platform 200 includes an infrastructure layer 210, a process event and fusion layer 220, an intelligent lean twin layer 230, a decision and simulation optimization layer 240, and an application and improvement layer 250. The infrastructure layer 210 provides the basic computing, storage, and network resources for the entire platform. Optionally, the infrastructure layer 210 includes an Industrial Internet of Things (IIoT) gateway, edge computing nodes, a 5G / TSN industrial network, and a cloud server. It is responsible for ensuring that data from the production physical system (300) can be reliably and with low latency collected.

[0057] The process event and fusion layer 220 is the platform's data entry point, responsible for addressing the issues of "dark data" and "data silos," and generating standard event logs required for process mining. This layer includes an event fusion module. In a specific embodiment, the event fusion module performs the following operations: 1. Connect to the MES system, extract work order data, and use the "work order number" or "product serial number" as the case ID.

[0058] 2. Connect to the PLC or device gateway, extract the device's "power on", "power off", and "processing" status signals, and bind the corresponding timestamps as activities.

[0059] 3. Connect to MES and extract nodes entered manually or by the system, such as "Process Report", "Quality Inspection Passed", "Quality Inspection NG", and "Rework Started", as activities.

[0060] 4. Aggregate and sort all the above data according to the case ID to form a standardized event log stream for use by the upper layer.

[0061] The Intelligent Lean Twin Layer 230 serves as the platform's "diagnostic brain," responsible for discovering knowledge and building models from event logs. It comprises a process mining engine, a lean diagnostic module, and a lean knowledge graph module. The process mining engine receives event logs and, using built-in process discovery algorithms (such as Alpha Miner, Heuristics Miner, or Inductive Miner), automatically calculates and draws a dynamic value stream digital twin. This twin is a directed graph where nodes represent "activities" and edges represent "transition relationships." The process mining engine also performs consistency checks, comparing this graph with the Standard Operating Procedure (SOP) and automatically highlighting "violation paths."

[0062] The Lean Diagnostics module performs performance analysis on the twin. It calculates the average wait time for each edge (transition relationship), thereby automatically identifying bottlenecks (edges with the longest wait times). It can also automatically identify cyclic structures in the graph (such as "quality inspection NG" -> "rework" -> "quality inspection") and quantify their frequency and duration, thus revealing "rework waste".

[0063] In an optional embodiment, the intelligent lean twin layer 230 also includes a lean knowledge graph module. This lean knowledge graph module pre-stores structured lean knowledge, such as triples (Subject: Low OEE, Relationship: Possible Cause, Object: Long Changeover Time) and (Subject: Long Changeover Time, Relationship: Recommended Countermeasure, Object: SMED Quick Changeover). When the lean diagnostic module detects "Low OEE" and "Changeover Time" as a bottleneck, the platform can automatically query the graph and recommend "SMED" as an improvement direction to the manager's terminal device.

[0064] The Decision and Simulation Optimization Layer 240 is the platform's "decision brain," responsible for moving from "diagnosis" to "action" and addressing the problem of "unpredictable improvement measures." It includes a simulation engine and a decision execution module. The core innovation of the simulation engine lies in its use of a non-idealized simulation model, instead directly employing the "As-Is" twin discovered by the process mining engine as its simulation foundation. Managers can perform "What-if" analysis on this realistic model. For example: Simulation 1: Increase the resources for the "rework" path by 1 person and predict the impact on the total lead time.

[0065] Simulation 2: Change the batch size of "bottleneck process A" from 100 to 50 and predict the impact on WIP (work in process) and OEE.

[0066] The decision execution module generates executable instructions based on the optimal strategy derived from simulation. For example, if simulation indicates that adjusting the scheduling rules is better, the decision execution module sends an update instruction to the MES system via the API interface; if a critical bottleneck is detected by diagnostics, an Andon alert is triggered on the dashboard of the physical assets.

[0067] The Application and Improvement Layer 250 is the platform's human-computer interaction interface. It includes the Application Kanban module and the Continuous Improvement module. The Application Kanban module visualizes the twin (i.e., a "live" VSM diagram), bottlenecks, waste diagnostic results, and KPIs such as OEE. The Continuous Improvement module provides lean teams with a digitalized PDCA tool, i.e., a continuous improvement (Kaizen) management tool.

[0068] The following will explain the meaning of the abbreviation "PDCA" in the PDCA tool: P (Plan): The team uses a simulation engine to plan and predict improvement measures.

[0069] D (Do): The team improves through decision-making and execution modules or offline execution.

[0070] C (Check): The core innovation lies in the fact that after the improvements are implemented, the intelligent manufacturing optimization platform continues to receive new event logs and automatically re-examines the process twins. Managers can immediately see the before-and-after process comparisons (Conformance Checking) and performance comparisons on the dashboard, achieving automatic closed-loop verification of data.

[0071] A (Act): Solidify the effective improvement measures into a new SOP.

[0072] The following section will describe the detailed steps of the intelligent manufacturing optimization platform 200 in executing the above method embodiments: I. The intelligent manufacturing optimization platform 200 extracts data from the production physical system 300 in real time and converts it into standard event logs through the event fusion module.

[0073] Second, the process mining engine analyzes event logs to automatically discover and build dynamic value stream digital twins.

[0074] Third, the Lean Diagnostics module analyzes the twin to diagnose waste, bottlenecks, and variations.

[0075] 4. (Optional) Query the Lean Knowledge Graph to match recommended countermeasures to the diagnostic results.

[0076] 5. The simulation engine loads the twin and the recommended strategy, runs a "What-if" simulation, and predicts the effects of different strategies.

[0077] VI. The manager selects the optimal simulation scheme, and the decision execution module generates and issues instructions to the physical system.

[0078] VII. The application and improvement layer continuously visualizes the results and manages the PDCA cycle. The process returns to step one, forming a closed loop of continuous improvement.

[0079] According to a third aspect of this embodiment, a smart manufacturing optimization system based on process mining and lean digital twins is provided, comprising: a production physical system (such as an automotive welding production line, including its MES and robot PLC) and the smart manufacturing optimization platform described above; the production physical system includes an information system and physical assets; the smart manufacturing optimization platform interacts with the production physical system to execute the smart manufacturing optimization method described above.

[0080] Optionally, the intelligent manufacturing optimization platform is deployed on edge or cloud servers and is tightly integrated with the production physical system through industrial networks, realizing all functions from data acquisition to intelligent diagnosis, simulation and closed-loop control.

[0081] It should be noted that the intelligent manufacturing optimization platform and system based on process mining and lean digital twin provided in this embodiment can realize all the functions and steps in the above method embodiments, solve the same technical problems, and achieve the same technical effects. The similarities will not be repeated.

[0082] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0083] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A smart manufacturing optimization method based on process mining and lean digital twins, characterized in that, include: Real-time production data is extracted from heterogeneous data sources and converted into standardized process event logs, which include at least case ID, activity and timestamp. By analyzing the process event logs through a process mining engine, a dynamic value stream digital twin reflecting the actual production process is automatically discovered and constructed. The dynamic value stream digital twin evolves dynamically as the process event logs are updated. On the dynamic value stream digital twin, the process event log is subjected to real-time lean status diagnosis, and corresponding diagnostic results are generated; Based on the diagnostic results, using the dynamic value stream digital twin as the simulation base, a hypothetical analysis is performed through the simulation engine to predict the impact of different improvement measures on key performance indicators, generate corresponding optimization strategies, and push them to the production execution system. Based on the newly generated process event logs after executing the optimization strategy, repeat the steps of automatically discovering and building a dynamic value stream digital twin and subsequent real-time diagnostics to continuously verify and iteratively optimize the optimization effect.

2. The method according to claim 1, characterized in that, The operation of converting the real-time production data into standardized process event logs includes: Bind an order number or batch number from the Manufacturing Execution System to a case ID for a complete production process instance; Activities that bind device status, process reports, and quality inspection results from programmable logic controllers, manufacturing execution systems, or sensors to the production process instance; The data from different data sources are then aggregated and sorted according to the case ID to form the standardized process event log.

3. The method according to claim 2, characterized in that, The process event logs are analyzed using a process mining engine to automatically discover and construct a dynamic value stream digital twin reflecting the actual production process, including: Using the built-in process discovery algorithm, the standardized process event log is mined and analyzed to identify the various activities contained therein and the transition relationships between the activities; Based on the identified activities and transition relationships, a directed graph computation model is constructed; wherein, the nodes of the directed graph correspond to the activities, and the edges of the directed graph correspond to the transition relationships between the activities; Generate a dynamic value stream digital twin based on the directed graph computation model to reflect the actual production process.

4. The method according to claim 3, characterized in that, On the dynamic value stream digital twin, the operation of performing real-time lean status diagnosis on the process event log and generating corresponding diagnostic results includes: Based on the temporal data of the edges in the directed graph computation model, the average waiting time for the transition between activities in the process event log is calculated. Based on the topology of the directed graph computation model, the loop path consisting of active nodes and transition relationship edges is automatically identified, and based on the process event log, the occurrence frequency and time of the identified loop path are quantified. Based on the average waiting time and the occurrence frequency and duration, diagnostic results are generated to reflect process bottlenecks and rework waste.

5. The method according to claim 3, characterized in that, Also includes: The directed graph computation model is compared with a preset standard operating procedure, and the comparison includes path consistency check and / or activity sequence integrity check. Based on the comparison results, the deviation between the actual work process represented by the directed graph computation model and the standard work process is calculated. When the deviation exceeds a preset threshold, an alarm message containing specific deviation path information is automatically generated and issued.

6. The method according to claim 1, characterized in that, Based on the diagnostic results, using the dynamic value stream digital twin as a simulation base, a hypothetical analysis is performed through a simulation engine to predict the impact of different improvement measures on key performance indicators and generate corresponding optimization strategies, including: On the dynamic value stream digital twin, at least one predefined improved scenario is simulated and executed; Based on the output of the simulation, calculate the predicted values ​​of one or more key performance indicators; By comparing the predicted values ​​of key performance indicators under different improvement scenarios, an optimization strategy is generated that includes recommended improvement measures and their expected effects.

7. The method according to claim 1, characterized in that, Also includes: Based on the waste phenomena or performance indicators contained in the diagnostic results, a pre-built lean knowledge graph is queried to obtain one or more recommended countermeasures, and the recommended countermeasures are provided to the manager; wherein, the lean knowledge graph stores semantic relationship triples between waste phenomena, root causes and recommended countermeasures.

8. A storage medium, characterized in that, The storage medium includes a stored program, wherein, when the program is executed, the method described in any one of claims 1 to 7 is performed by a processor.

9. A smart manufacturing optimization platform based on process mining and lean digital twins, characterized in that, include: Infrastructure layer, process event and fusion layer, intelligent lean twin layer, decision and simulation optimization layer, and application and improvement layer; The intelligent lean twin layer includes a process mining engine and a lean diagnostics module; The decision and simulation optimization layer includes a simulation engine and a decision execution module; The infrastructure layer is used to build industrial networks, edge computing, and cloud platforms that run the intelligent manufacturing optimization platform; The process event and fusion layer is used to connect heterogeneous data sources, extract real-time production data from heterogeneous data sources, and convert the real-time production data into standardized process event logs. The process event logs include at least case ID, activity, and timestamp. The process mining engine is used to analyze the process event logs, automatically discover and construct a dynamic value stream digital twin that reflects the actual production process, and the dynamic value stream digital twin evolves dynamically as the process event logs are updated. The lean diagnostic module is used to perform real-time lean status diagnosis on the process event log on the dynamic value stream digital twin and generate corresponding diagnostic results. The simulation engine is used to perform hypothetical analysis based on the diagnostic results, using the dynamic value stream digital twin as the simulation base, and to predict the impact of different improvement measures on key performance indicators. The decision execution module is used to generate corresponding optimization strategies based on the simulation results and push them to the production execution system; The application and improvement layer is used to repeat the automatic discovery and construction of a dynamic value stream digital twin and subsequent real-time diagnostic steps based on the process event logs newly generated after the optimization strategy is executed, so as to continuously verify and iteratively optimize the optimization effect.

10. A smart manufacturing optimization system based on process mining and lean digital twins, characterized in that, include: A production physical system and an intelligent manufacturing optimization platform; the production physical system includes an information system and physical assets; the intelligent manufacturing optimization platform interacts with the production physical system to execute the intelligent manufacturing optimization method according to any one of claims 1 to 7.

Images