Digital Twin-Driven Fault Prediction Method and System for Packaging Production Line

Abstract

This invention provides a digital twin-driven method and system for predicting packaging production line faults, belonging to the field of fault prediction technology. The method includes: analyzing equipment operating parameters and packaging quality constraints of process nodes in the packaging production line to establish a parameter-quality mapping relationship; segmenting packaging operations to obtain monitoring production line blocks; establishing small digital twin models for production status detection and analysis, and fault prediction for each production line block; connecting the production status of the blocks and the fault prediction results of the production line blocks through a production line-level twin platform to perform production line status fusion analysis and obtain fused production line fault results; and providing production line fault feedback based on the fault prediction results of the production line blocks and the fused production line fault results. This invention solves the technical problem that existing technologies typically use a single digital twin model to represent the state of the entire production line, which is insufficient to handle the differences and dynamic changes in various parts of the production line, leading to insufficient fault prediction accuracy.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of fault prediction, in particular to a digital-twin-driven packaging production line fault prediction method and system. BACKGROUND

[0002] A packaging production line is usually composed of multiple mechanical devices, sensors, control systems, etc. These devices and processes are interconnected, forming a complex production system. Due to the high automation, complexity and diversity of the packaging production line, it is often difficult to diagnose faults in time, and it may lead to large-scale downtime or quality problems. The prior art usually uses a single digital twin model to represent the state of the entire production line, relying on a comprehensive large model to predict device faults and quality problems. Since the entire production line involves multiple devices, process flows and parameters, a single model cannot effectively handle complex and heterogeneous production data. Such a single large model often cannot cope with the differences and dynamic changes of each block in the production line, making it difficult to achieve accurate fault prediction and optimization, thereby increasing the downtime of the production line, reducing production efficiency, and affecting product quality and production costs. SUMMARY

[0003] The application provides a digital-twin-driven packaging production line fault prediction method and system, aiming to solve the technical problem that the prior art usually uses a single digital twin model to represent the state of the entire production line, which is difficult to cope with the differences and dynamic changes of each part of the production line, resulting in insufficient fault prediction accuracy.

[0004] The first aspect of the application provides a digital-twin-driven packaging production line fault prediction method, which comprises: analyzing the device operation parameters of the packaging production line and the packaging quality constraints of the process nodes, establishing a parameter-quality mapping relationship; based on the device operation parameters and the packaging quality constraints of the process nodes, the packaging operation is segmented to obtain a monitoring production line block; based on the monitoring production line block, a small digital twin model is established respectively, and the parameter-quality mapping relationship is combined to analyze the production state of each monitoring production line block and predict the fault of the production line block; the output of the small digital twin model is connected through a production line level twin platform to obtain the production state of the block and the fault prediction result of the production line block, and the production line state fusion analysis is performed to obtain the production line fusion fault result; according to the production line block fault prediction result and the production line fusion fault result, the production line fault feedback is performed.

[0005] In a second aspect, the application discloses a digital-twin-driven packaging production line fault prediction system, which is used for the above-mentioned digital-twin-driven packaging production line fault prediction method, and comprises: a mapping relationship establishing module, which is used for analyzing equipment operation parameters of a packaging production line and packaging quality constraints of process nodes, and establishing a parameter-quality mapping relationship; a packaging operation segmentation module, which is used for packaging operation segmentation based on the equipment operation parameters and the packaging quality constraints of the process nodes, and obtaining monitoring line blocks; a fault prediction module, which is used for establishing small digital twin models based on the monitoring line blocks respectively, and combining the parameter-quality mapping relationship to perform production state detection analysis and line block fault prediction on each monitoring line block; a fusion analysis module, which is used for connecting block production states and line block fault prediction results of outputs of the small digital twin models through a line-level twin platform, performing line state fusion analysis, and obtaining line fusion fault results; and a fault feedback module, which is used for performing line fault feedback according to the line block fault prediction results and the line fusion fault results.

[0006] The one or more technical solutions provided in the application have at least the following beneficial effects:

[0007] By analyzing equipment operation parameters of a packaging production line and packaging quality constraints of process nodes, a parameter-quality mapping relationship is established. Through the mapping relationship, the dependency between equipment operation parameters and final packaging quality can be accurately understood, thereby providing accurate data basis for subsequent fault prediction and quality management. By packaging operation segmentation based on equipment operation parameters and packaging quality constraints, different monitoring line blocks are obtained, and a complex production line is disassembled into multiple blocks, each of which is independently monitored, which can improve the accuracy of data acquisition and processing, and avoid excessive generalization of the influence of single block fault on the whole line. By establishing small digital twin models for each monitoring line block respectively, and combining the parameter-quality mapping relationship, production state detection and fault prediction are performed on each block. Each small digital twin model focuses on a specific monitoring line block, can accurately model the equipment state, quality and potential fault of the block, and makes the prediction more targeted and accurate. The output results of each small digital twin model are connected through a line-level twin platform, line state fusion analysis is performed, and line fusion fault results are obtained. Through line-level analysis, potential fault problems across blocks and equipment can be identified, and the globality and collaboration of prediction can be improved. According to the line block fault prediction results and the line fusion fault results, line fault feedback is performed, decision support is provided for production scheduling and equipment maintenance, helps management optimize production plan and resource allocation, and reduces the maintenance cost and fault occurrence frequency of the production line.

[0008] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0009] Figure 1 A schematic diagram of the digital twin-driven packaging production line fault prediction method provided in this application embodiment.

[0010] Figure 2 A schematic diagram of the structure of a digital twin-driven packaging production line fault prediction system provided in an embodiment of this application.

[0011] Figure labeling: Module 10 for establishing mapping relationship, Module 20 for packaging operation segmentation, Module 30 for fault prediction, Module 40 for fusion analysis, and Module 50 for fault feedback. Detailed Implementation

[0012] This application provides a digital twin-driven method and system for predicting faults in packaging production lines, which solves the technical problem that the existing technology usually uses a single digital twin model to represent the state of the entire production line, which is difficult to cope with the differences and dynamic changes of various parts of the production line, resulting in insufficient fault prediction accuracy.

[0013] After introducing the basic principles of this application, various non-limiting embodiments of this application will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit this application.

[0014] Example 1, as Figure 1 As shown in the embodiment of this application, a method for predicting failures in a packaging production line driven by a digital twin is provided. The method includes:

[0015] Analyze the equipment operating parameters and packaging quality constraints of the packaging production line and process nodes, and establish parameter-quality mapping relationships.

[0016] Equipment operating parameters include the operational parameters of each piece of equipment in the packaging production line (such as carton sealing machines, filling machines, labeling machines, etc.), such as speed, temperature, pressure, time, and vibration. Each piece of equipment has certain operating standards during operation, and how these parameters are adjusted directly affects the quality of the final packaged product. A packaging production line consists of multiple processes, each with specific tasks, and the equipment and its parameters involved will have different impacts on packaging quality. At each process node, there are certain standards for packaging quality, such as the sealing of the packaging bag, the clarity of the label, and the strength of the seal. Based on actual process requirements, reasonable packaging quality constraints are set for the operating parameters of each piece of equipment. For example, the pressure of a sealing machine cannot be too high, otherwise it may cause the packaging bag to break; the temperature of the filling machine must be within a certain range to prevent food spoilage, etc. The mapping relationship between equipment operating parameters and packaging quality constraints is established through experiments, historical data, or simulation models; in other words, it clarifies how each parameter affects the quality of the final packaged product.

[0017] The packaging operation is segmented based on the equipment operating parameters and packaging quality constraints of the process nodes to obtain monitoring production line blocks.

[0018] The equipment and process nodes on the packaging production line are interconnected, with a certain sequence and dependency. By analyzing these relationships, it is clear how each piece of equipment and process node affects the overall production process. The packaging quality constraints of each process node will have different dependencies on the operating parameters of upstream equipment. The degree of dependence of each process node on the parameters of upstream equipment can be assessed using sensitivity analysis or coupling analysis. For example, if the speed of a certain piece of equipment has a significant impact on the downstream quality, it indicates that the coupling between the equipment and that node is strong. Equipment and process nodes with coupling exceeding a certain threshold and physically adjacent are clustered to form monitoring production line blocks. Through this segmentation, the entire packaging production line is divided into multiple monitoring production line blocks, and each monitoring production line block can independently perform status monitoring and fault prediction.

[0019] Based on the monitored production line blocks, small digital twin models are established for each block. Combined with the parameter-quality mapping relationship, production status detection and analysis and fault prediction are performed on each monitored production line block.

[0020] A small digital twin model is a digital copy of each monitored production line block. It simulates the interaction between physical equipment, processes, and parameters, including the geometric model, physical mechanism model, and data-driven predictive model of the equipment within each monitored production line block. When the small digital twin model is running, it utilizes the established parameter-quality mapping relationship to specifically perform production status detection and analysis, collect equipment operating parameters in real time, drive the physical mechanism model to perform simulation, and detect whether the actual operating status of the production line is normal by comparing it with the actual equipment's operating indicators. It also performs fault prediction for production line blocks, using a trained data-driven predictive model to predict the remaining effective lifespan and quality deviation trends of the equipment within the block. Based on this, it can identify in advance when equipment is about to fail or when quality deviations may occur.

[0021] By connecting the block production status and production line block fault prediction results output by the small digital twin model through the production line-level twin platform, a production line status fusion analysis is performed to obtain the production line fusion fault results.

[0022] Using a production line-level digital twin platform, the output data from each small digital twin model is first aggregated, including production status data and fault prediction results for each monitoring block. During this process, data from different blocks are aligned and standardized to ensure unified processing. A production line status fusion analysis is then performed. This includes identifying bottleneck blocks on the production line—those critical blocks that limit overall production line efficiency and output—based on production data from each block, such as cycle time and throughput. The impact of bottleneck blocks on the entire production line is analyzed, particularly regarding quality and equipment failures. Bottleneck blocks may affect the production efficiency and quality of upstream and downstream blocks. Based on the prediction results for bottleneck blocks, the overall production line status is assessed. By comprehensively evaluating production data, equipment health status, and quality deviations from different blocks, a fusion fault analysis of the production line is generated.

[0023] Production line fault feedback is provided based on the production line block fault prediction results and the production line fusion fault results.

[0024] Based on the fault prediction results of production line blocks and the fault results of production line fusion, the results are compared with the preset multi-level alarm rule base. When the prediction result meets a certain preset threshold, an early warning message is generated, including the location of the warning device or block, the predicted fault event, the expected occurrence time, and the warning level. The early warning message is transmitted to the operators through the display screen, alarm system, or system automation feedback mechanism to remind them to take measures, such as checking the equipment and adjusting the production parameters.

[0025] Furthermore, based on the equipment operating parameters and packaging quality constraints of the process nodes, the packaging operation is segmented to obtain monitoring production line blocks, including:

[0026] Based on the physical layout and process flow diagram of the production line, the sequence and connection relationship of each piece of equipment and process node are determined; the sensitivity and dependence of the packaging quality constraints of downstream process nodes on the operating parameters of upstream equipment are analyzed, and their coupling strength is quantified; a group of equipment and their associated process nodes whose coupling strength exceeds the preset threshold and are physically adjacent are clustered into a monitoring production line block.

[0027] Physical layout refers to the spatial location and arrangement of equipment in a production line. Through physical layout, the actual position of each piece of equipment on the production line and the physical connections between them can be determined. A process flow diagram describes the production flow of each process, showing each production step from raw materials to the final packaged product, including the equipment and tasks for each process. Each node in the process flow diagram represents a process or piece of equipment, and the arrows between nodes indicate the sequence of material or product flow. Combining the physical layout and the process flow diagram, the order of each piece of equipment is determined; for example, the filling machine follows the sealing machine, and the labeling machine follows the sealing machine. In some cases, a process depends on the result of the previous process. For example, the accuracy of the filling machine directly affects the operation of the subsequent sealing machine; if the filling is substandard, the sealing machine cannot function properly. The process flow diagram clarifies the dependencies between processes and equipment, helping to divide monitoring areas later.

[0028] The packaging quality constraints at each process node refer to the quality standards that the process needs to achieve. The packaging quality constraints of downstream processes directly affect the operational requirements of upstream equipment. For example, if the downstream packaging process has extremely high requirements for sealing, then the accuracy and filling volume stability of the upstream filling machine are crucial. The coupling degree describes the degree of influence of the packaging quality constraints of a process node on the operating parameters of upstream equipment. The higher the coupling strength, the greater the influence of upstream equipment on the quality of downstream processes. Specifically, sensitivity analysis is conducted to evaluate the impact of different equipment operating parameters (such as temperature, pressure, speed, etc.) on downstream quality constraints (such as packaging accuracy, sealing, etc.). Based on the sensitivity analysis results, the influence intensity of each equipment operating parameter on the quality of downstream processes is quantified. Statistical analysis, regression analysis, and other methods can be used for quantification.

[0029] Based on the impact of equipment operating parameters on packaging quality, a preset threshold is set. When the coupling strength exceeds this preset threshold, it indicates that the influence between the equipment and downstream processes is significant. The preset threshold can be set based on historical data, process standards, or empirical rules. Based on the analysis results of coupling strength, equipment and process nodes with coupling strength exceeding the preset threshold are selected. In addition to coupling strength, it is also necessary to ensure that the clustered equipment and process nodes are physically adjacent. The adjacency of equipment means that they transfer products through physical connections. Therefore, these equipment and process nodes have a significant mutual influence. For each pair of equipment and process nodes, if their coupling strength exceeds the preset threshold and they are physically adjacent, they are assigned to the same monitoring production line block.

[0030] During the clustering process, the goal is to ensure that the equipment and process nodes within the same monitoring production line block have a strong coupling relationship. This ensures that the fault prediction within the monitoring block is more accurate and effective. At the same time, it is also necessary to ensure that the coupling between different monitoring production line blocks is as small as possible. The smaller the coupling between blocks, the more independent and accurate the monitoring and fault prediction of each block will be. By analyzing the interaction between equipment and process nodes, it is possible to determine which blocks can be ignored in terms of coupling, thus ensuring their relative independence.

[0031] Furthermore, small digital twin models are established based on the monitored production line blocks, including:

[0032] Based on the geometric model, physical mechanism model and data-driven prediction model of the equipment in each monitoring production line block, a small digital twin model is constructed. The data-driven prediction model is obtained by training historical equipment parameter data, historical quality data and historical fault data through machine learning algorithms, and is used to predict the remaining effective life or quality deviation trend of the equipment in the block.

[0033] A geometric model is a mathematical description of the structure, size, spatial arrangement, and interconnection of equipment. It is used to represent the physical form and spatial position of the equipment. By collecting the physical parameters of the equipment, such as size, weight, and shape, as well as layout information, a geometric model of the equipment is constructed. These geometric models help to understand the spatial configuration of the equipment and the relative positions between the equipment, such as whether the relative positions of the conveyor belt and the packaging machine are prone to friction or other mechanical interference.

[0034] Physical mechanism models are mathematical models built based on the physical and engineering principles of equipment. They are used to describe the behavior of equipment under normal or fault conditions, including thermodynamic models, mechanical models, electrical models, etc. They can be combined with real-time data to perform physical mechanism simulation calculations, simulate the state of equipment in actual operation, and predict the failure modes of equipment under specific working conditions, such as overload, overheating, and component wear.

[0035] Data-driven predictive models, based on historical data and trained using machine learning algorithms, predict equipment status and quality. They are used to predict the remaining effective lifespan of equipment and quality deviation trends. Specifically, these models rely on historical equipment parameter data, historical quality data, and historical failure data for training. Historical data is input into the machine learning model to train the equipment's behavior patterns under different operating conditions. During training, methods such as cross-validation are used for evaluation to ensure the model has sufficient generalization ability. Data-driven predictive models can predict the remaining effective lifespan of equipment—how long it can operate before reaching failure—indicating early maintenance or component replacement. Based on process parameters and the output of the equipment status prediction model, quality deviation trends, such as product dimensional errors and sealing differences, can also be predicted.

[0036] In small digital twin models, geometric models, physical mechanism models, and data-driven prediction models are complementary. Their combination can provide comprehensive support for production line fault prediction and health management. Based on the prediction results of small digital twin models, production line operation and maintenance personnel can make corresponding adjustment decisions, such as adjusting equipment operating parameters and carrying out preventive maintenance.

[0037] Furthermore, the data-driven prediction model includes two parallel output sub-models: an equipment health prediction model and a quality risk prediction model. The equipment health prediction model takes the equipment status parameter sequence as input and outputs the remaining effective life of the equipment to predict performance degradation failures of mechanical or electrical components. The quality risk prediction model takes the real-time values ​​and trends of process execution parameters as input and, combined with the parameter-quality mapping relationship, outputs the quality deviation trend to predict quality failures caused by process drift.

[0038] Equipment health prediction models are used to predict the remaining usable life of equipment, i.e., how long the equipment can continue to operate normally before failure. They primarily focus on the performance degradation of mechanical and electrical components, helping to determine whether equipment is about to fail, thus providing a timeline for maintenance and replacement. Equipment health prediction models take a sequence of equipment state parameters as input, which is real-time monitoring data of the equipment, including mechanical equipment parameters, electrical equipment parameters, and operating environment parameters. The output is the remaining usable life, typically expressed in time (e.g., hours, days, cycles) or number of operations (e.g., production units). Based on the input equipment state data and learning from historical data, the model predicts how long the equipment can safely operate under current operating conditions. Equipment health prediction models can employ deep learning, such as LSTM neural networks, to extract the evolution patterns of equipment health from time-series data, making them particularly suitable for predicting remaining usable life.

[0039] Quality risk prediction models focus on quality problems caused by process drift during production. The goal is to predict potential quality deviations by monitoring key process parameters. These models take real-time values ​​and trends of process execution parameters as input. These parameters are real-time collected production parameters, such as temperature, pressure, speed, time, and environmental conditions. Based on these input parameters and trends, the model outputs quality deviation trends, including deviation values, predicted pass rates, and probability distributions of quality deviations. Support vector regression models can be used to predict the relationship between process data and quality indicators.

[0040] Furthermore, combining the aforementioned parameter-quality mapping relationship to perform production status detection and analysis for each monitored production line block, as well as fault prediction for that production line block, includes:

[0041] The real-time collected equipment operation parameter stream is input into the block-level digital twin model, driving the operation of each block-level digital twin model. The physical mechanism model simulates operation based on real-time parameters and compares it with the actual equipment operation indicators to achieve status detection and anomaly identification. The real-time parameter sequence is input into the equipment health prediction model and the quality risk prediction model respectively. The equipment health prediction model outputs a time-based remaining effective life prediction result, and the quality risk prediction model outputs a probability-based quality deviation prediction result. The remaining effective life prediction result and the quality deviation prediction result are jointly encapsulated into a production line block failure prediction result.

[0042] The system collects real-time data on equipment operating parameters from the production line, including equipment status such as speed, temperature, pressure, and vibration. The block-based digital twin model is a digital simulation model built based on each monitoring block in the production line. Each monitoring block contains multiple equipment or process nodes and is simulated based on the physical, chemical, and mechanistic models of that block. After inputting the equipment operating parameter stream, the block-based digital twin model simulates the operation of the equipment and calculates its simulated operating status. By comparing the simulated operating status with the ideal state predicted by the model, it identifies whether the equipment is within the normal operating range. If the simulated operating status deviates from the normal operating value of the model, and this deviation reaches a certain threshold, it is automatically identified as an anomaly.

[0043] Real-time parameter sequences are input into an equipment health prediction model, which outputs a time-based prediction of remaining effective life, i.e., how long the equipment can continue to operate safely, predicting the degradation of mechanical components and the time of possible failure. Real-time parameter sequences are also input into a quality risk prediction model, which outputs a probability-based prediction of quality deviations, i.e., predicting future quality deviations based on real-time parameters and historical data, identifying potential quality problems such as dimensional errors and product defect rates.

[0044] The remaining effective life of the equipment and the predicted quality deviation are combined to form an overall production line block failure prediction result. This result can be directly used to identify potential failures or quality problems in the production line.

[0045] Furthermore, a production line status fusion analysis is performed to obtain production line fusion fault results, including:

[0046] The production monitoring data, remaining effective life prediction results, and quality deviation prediction results of each block are aggregated. Data alignment and connection are performed based on the temporal and physical connection relationships between blocks to establish a production line-level data chain. Based on the production line-level data chain, bottleneck blocks of the production line are identified by comparing the cycle time and throughput of each block. Based on the bottleneck blocks of the production line, the production line-level data chain is used to predict and evaluate product packaging quality and evaluate the production line impact of equipment operation failures to obtain the production line fusion failure results.

[0047] Time alignment involves synchronizing data from different blocks using a unified clock standard (such as a global clock). For example, if monitoring data for one block is collected hourly while data for another block is collected minutely, they are aligned by time windows to ensure data consistency and accuracy. In addition to time alignment, data is also correlated based on the physical connections between blocks, such as the connection order and dependencies between devices. For instance, a failure in one block can affect the quality of subsequent blocks; therefore, data is sorted and correlated based on these connections when establishing the data chain. Through the combination of data alignment and physical connections, a unified production line-level data chain is ultimately generated. This chain connects the status data, fault prediction data, and quality prediction data of each block, forming a global view that accurately reflects the overall operating status of the production line.

[0048] Cycle time is the time required to complete one unit of product on a production line. During analysis, cycle time data helps identify which sections or equipment are slow, leading to reduced overall production efficiency. Comparing the cycle time of each section identifies bottlenecks in production. Generally, the section with the longest cycle time is the bottleneck section of the production line because it limits the speed of the entire line. Throughput refers to the number of products produced per unit of time. By comparing and analyzing the throughput of each section, bottleneck sections with low efficiency in the production process can be identified. These bottleneck sections may be caused by equipment problems, improper process settings, or human error, leading to a decrease in the overall throughput of the production line.

[0049] Based on data from bottleneck areas in the production line, the overall product packaging quality of the production line can be predicted. For example, if there is a quality risk in the bottleneck area, the packaging quality of the entire production line may be affected. Quality assessment can be based on historical quality data and real-time quality deviation prediction results, combined with the health status of the bottleneck area, to predict the overall production process quality. Besides quality issues, equipment failures in bottleneck areas can cause the entire production line to stop or become unstable. Based on the remaining effective life prediction results of each area, the timing of potential equipment failures can be predicted, and the impact of such failures on the entire production line can be assessed. Combining bottleneck area failure prediction with quality risk analysis yields a fused failure result for the production line, providing a comprehensive evaluation of the failure status of the entire production line.

[0050] Furthermore, by aggregating production monitoring data, remaining usable life prediction results, and quality deviation prediction results from each block, and aligning and connecting the data based on the temporal and physical connectivity between blocks, a production line-level data chain is established, including:

[0051] The system receives production monitoring data streams, remaining usable life prediction results, and quality deviation prediction results from small digital twin models of each block, and converts heterogeneous data into standardized data objects, marking them with time-series identifiers based on the production line global clock. Using the production line global clock as a reference, a fixed time window is defined, and the production monitoring data arriving from each block within each time window is aggregated to generate a unified production line status within the window period. For predictive results, the original prediction timestamp and release timestamp are retained, and the relative time difference with the current time is calculated, aligning the data according to time sequence relationships. Based on the production line topology structure obtained from the time sequence and physical connection relationships between blocks, the time-aligned data is associated with the production line topology structure. The production line data is stored in a time-series database in primary chronological order, forming a production line-level data chain. This production line-level data chain supports query and traceability analysis by time dimension, spatial dimension, and process logic dimension.

[0052] Since the data from different blocks may have different formats, units, and collection frequencies, they need to be converted into standardized data objects. This means that data such as equipment parameters and prediction results from different sources need to be unified into a standard format, such as unified units, unified data types, and unified data ranges, to ensure that subsequent calculations and analyses can be processed correctly. Each data object is timestamped, and the data collection time is identified based on the production line's global clock. The global clock ensures that the data from all blocks is sorted and synchronized according to a unified time standard, avoiding data distortion or processing errors due to time asynchrony.

[0053] Based on actual needs, a fixed time window is defined, such as every minute, every hour, or every production cycle. This fixed time window is used to aggregate and integrate data from various blocks. For example, if the production data of a certain monitoring block is updated every second, but for full-line status analysis, the data is aggregated into a data stream updated every minute. All block data are synchronized according to a global clock to ensure that the data from each block is aggregated under the same time reference. Within each time window, production monitoring data, remaining effective life prediction results, quality deviation prediction results, etc., from different blocks are aggregated. Aggregation can involve averaging, summing, finding maximum or minimum values ​​from multiple blocks, depending on the analysis requirements. For example, equipment status can be aggregated into the average status of the entire production line, and prediction results can be aggregated into the total failure probability of the entire production line. These aggregated data generate a unified production line status, representing the overall health and production status of the entire production line within that time window. This is a global perspective that can provide real-time basis for production scheduling, fault prediction, and resource allocation.

[0054] For predictive results, such as remaining usable life predictions and quality deviation predictions, the original prediction timestamps and release timestamps are retained because these predictions are based on historical data and future trends and are time-sensitive. The relative time difference between these timestamps and the current time is calculated and aligned in chronological order. For example, if the equipment health prediction result for a certain block indicates that the equipment will fail within the next 3 hours, then the time difference between the timestamp of the result and the current time is calculated so that its warning significance can be correctly understood in the overall production line status analysis.

[0055] To ensure data accuracy and comparability, the data is time-aligned according to a global clock. Even if the data update times of some blocks are different, adjustments can be made based on the time difference, so that all data can be compared and analyzed using a unified time standard. The time-aligned data can be sorted by time, ensuring that data from different blocks can be accurately compared and analyzed on the same time dimension, helping to more accurately identify potential fault risks and quality issues.

[0056] Production line topology refers to the physical or logical layout of various equipment, process nodes, and their connections in a production line, including the order, dependencies, signal transmission paths, and data flow relationships between equipment. The time-aligned data is associated with the production line topology. For example, if a failure in block A leads to quality problems in block B or production delays in block C, these impacts need to be reflected in the production line topology. In this step, based on the production line topology, it is determined how changes in time-series data affect the performance of different equipment or process nodes. Each data point is marked as corresponding to a specific topology node or edge and dynamically updated according to the time sequence and physical connections.

[0057] Production line data is stored in a time-series database in chronological order. Time-series databases are used to process high-frequency, time-series data, supporting efficient querying and large-scale data processing. Chronological order means sorting and storing data based on timestamps, allowing subsequent queries and analyses to be performed efficiently on a time-based basis. A production line-level data chain can provide a comprehensive view of the entire production line's status, recording all information from equipment status and process parameters to production quality.

[0058] The production line-level data chain supports multi-dimensional queries. The time dimension refers to the ability to query the production line status and related data within a specific point in time or time interval; the spatial dimension refers to the ability to query based on the spatial layout of the production line; and the process logic dimension refers to the ability to query the impact between different processes based on the sequence and dependencies of the processes.

[0059] Furthermore, based on the bottleneck blocks in the production line, the production line-level data chain is used to predict and evaluate product packaging quality and assess the production line impact of equipment malfunctions, thereby obtaining integrated production line fault results, including:

[0060] Based on the prediction results of the bottleneck block in the production line, the upstream and downstream block impact analysis is performed along the spatial dimension of the production line-level data chain to simulate the block failure state of the upstream and downstream blocks at the time point of the bottleneck block prediction result. The production line impact analysis of product packaging quality prediction and equipment failure prediction state is performed along the time and logical dimensions of the production line-level data chain to obtain the failure impact prediction results of the upstream and downstream blocks. Based on the production line-level data chain, the upstream and downstream blocks undergo iterative impact analysis of adjacent blocks until the correlation analysis of all blocks is completed. The failure impact prediction results of all blocks are then fused into the production line process to obtain the fused failure result of the production line.

[0061] Based on the failure prediction results of the bottleneck block in the production line, an impact analysis of upstream and downstream blocks is performed along the spatial dimension. The occurrence of a bottleneck block failure will not only affect that block, but also have a chain reaction on the upstream and downstream blocks of the entire production line. Therefore, it is necessary to simulate the state changes of upstream and downstream blocks when a failure occurs. Specifically, the spatial dimension within the production line-level data chain refers to the physical layout of the production line and the connection relationships between various blocks. Analyzing the impact of bottleneck block failures on directly connected upstream and downstream blocks includes: conducting reverse tracing analysis (upstream) to simulate how bottleneck block failures lead to material blockages, analyzing the backlog of output from upstream blocks at the time of bottleneck failure, and assessing the potential cascading shutdowns or speed reductions based on the backlog situation. For example, if a bottleneck block failure prevents timely processing of upstream materials, it may lead to shutdowns or speed reductions in upstream equipment. Conducting forward propagation analysis (downstream) to simulate how bottleneck block failures lead to downstream material shortages, analyzing the degree of interruption in downstream input, and assessing the potential equipment idleness or waiting states based on the shortage situation. For example, a bottleneck block failure may cause downstream production lines to halt due to delayed material arrivals, potentially leading to idle downstream equipment.

[0062] Time-dimensional analysis is conducted to examine the status changes of various blocks on the production line at different points in time. The occurrence of failures alters the production rhythm, affecting the production line's throughput and capacity. By analyzing the time dimension, the specific time of failure occurrence, the duration of its impact, and the propagation process of these impacts are identified. Based on the failure prediction results at each time point, the potential impact on product quality at that moment and in subsequent moments is assessed. Furthermore, by analyzing the equipment operating status over time, the potential for equipment to stop or slow down during a failure is evaluated, and the possible duration of the equipment failure and the recovery process are predicted.

[0063] Logical dimension analysis is conducted, which is an impact analysis based on the material flow path of the packaging production line and the process relationships in the production process. Based on the dependency relationship between material flow and production steps, the analysis examines the impact of various aspects such as material flow direction and equipment coordination when a failure occurs. According to the working logic of the production line, the analysis examines the impact of various processes caused by bottleneck block failures. For example, if the process of producing a certain product is affected by a bottleneck block failure, it will lead to a slowdown in the production speed of that product or even quality problems.

[0064] A recursive iterative analysis is performed on all impacts. Specifically, the impact of bottleneck block failures on other blocks is analyzed. First, an initial impact assessment is conducted. Then, upstream and downstream blocks are considered as new sources of impact, and the impact analysis is repeated until the state changes of all blocks tend to stabilize or their impact becomes negligible. Bottleneck block failures may lead to backlogs or delays in output from upstream blocks, creating a chain reaction. Bottleneck block failures can also affect material supply to downstream blocks, causing equipment idleness or stagnation, thus impacting overall production line efficiency. The impact prediction results of each block are integrated by considering process sequence and time relationships. Combined with the production processes and timestamps of each block, a global production line failure fusion result is formed. Global indicators include total capacity loss, estimated final product quality pass rate, and expected order delivery delay time, all of which reflect the impact of bottleneck block failures on the entire production line.

[0065] Furthermore, based on the production line block failure prediction results and the production line fusion failure results, production line failure feedback is provided, including:

[0066] Based on the production line block fault prediction results and the production line fusion fault results, a real-time comparison is performed with the preset multi-level alarm rule library. When the predicted data meets the trigger threshold of the warning level, a fault warning information is generated, including the location of the warning device or block, the predicted fault event, the expected occurrence time, and the warning level.

[0067] The multi-level alarm rule base consists of a series of preset rules used to trigger different alarm levels based on varying fault severity and prediction results. These rules are designed according to the characteristics of different equipment, the importance of the production process, and potential risks, including different fault prediction rules for different equipment, trigger thresholds for different data items, and fault severity levels. It receives fault prediction results from each monitoring block and the fused fault results of the production line, and compares them with preset thresholds in the multi-level alarm rule base. If the predicted data exceeds the set trigger threshold, the fault alarm condition for that monitoring block is triggered. For example, if the remaining effective lifespan of a certain device is predicted to be less than 20 hours, and this data meets the high-level alarm threshold, a high-priority alarm will be triggered.

[0068] When the comparison results meet the triggering conditions of a certain warning level, corresponding fault warning information is generated. Among them, the location of the warning device or block indicates the monitoring block where the problem occurs; the predicted fault event describes the predicted fault type or event; the estimated occurrence time is based on the fault occurrence time or trend given by the prediction model, providing the estimated time of fault occurrence; the warning level is the urgency level of the warning specified according to the severity of the fault. For example, a high-level warning requires immediate shutdown and inspection, while a low-level warning only suggests monitoring.

[0069] Example 2, based on the same inventive concept as the digital twin-driven packaging production line fault prediction method in the foregoing examples, such as... Figure 2 As shown in the figure, this application provides a digital twin-driven packaging production line fault prediction system, the system comprising:

[0070] The mapping relationship establishment module 10 is used to parse the equipment operation parameters and packaging quality constraints of the process nodes of the packaging production line and establish a parameter-quality mapping relationship; the packaging operation segmentation module 20 is used to segment the packaging operation based on the equipment operation parameters and packaging quality constraints of the process nodes to obtain monitoring production line blocks; the fault prediction module 30 is used to establish small digital twin models based on the monitoring production line blocks respectively, and combine the parameter-quality mapping relationship to perform production status detection analysis and fault prediction of each monitoring production line block; the fusion analysis module 40 is used to connect the block production status and production line block fault prediction results output by the small digital twin model through the production line-level twin platform to perform production line status fusion analysis and obtain production line fusion fault results; the fault feedback module 50 is used to provide production line fault feedback based on the production line block fault prediction results and production line fusion fault results.

[0071] Furthermore, the packaging operation segmentation module 20 is used to perform the following operation steps:

[0072] Based on the physical layout and process flow diagram of the production line, the sequence and connection relationship of each piece of equipment and process node are determined; the sensitivity and dependence of the packaging quality constraints of downstream process nodes on the operating parameters of upstream equipment are analyzed, and their coupling strength is quantified; a group of equipment and their associated process nodes whose coupling strength exceeds the preset threshold and are physically adjacent are clustered into a monitoring production line block.

[0073] Furthermore, the fault prediction module 30 is used to perform the following operation steps:

[0074] Based on the geometric model, physical mechanism model and data-driven prediction model of the equipment in each monitoring production line block, a small digital twin model is constructed. The data-driven prediction model is obtained by training historical equipment parameter data, historical quality data and historical fault data through machine learning algorithms, and is used to predict the remaining effective life or quality deviation trend of the equipment in the block.

[0075] Furthermore, the data-driven prediction model includes two parallel output sub-models: an equipment health prediction model and a quality risk prediction model. The equipment health prediction model takes the equipment status parameter sequence as input and outputs the remaining effective life of the equipment to predict performance degradation failures of mechanical or electrical components. The quality risk prediction model takes the real-time values ​​and trends of process execution parameters as input and, combined with the parameter-quality mapping relationship, outputs the quality deviation trend to predict quality failures caused by process drift.

[0076] Furthermore, the fault prediction module 30 is used to perform the following operation steps:

[0077] The real-time collected equipment operation parameter stream is input into the block-level digital twin model, driving the operation of each block-level digital twin model. The physical mechanism model simulates operation based on real-time parameters and compares it with the actual equipment operation indicators to achieve status detection and anomaly identification. The real-time parameter sequence is input into the equipment health prediction model and the quality risk prediction model respectively. The equipment health prediction model outputs a time-based remaining effective life prediction result, and the quality risk prediction model outputs a probability-based quality deviation prediction result. The remaining effective life prediction result and the quality deviation prediction result are jointly encapsulated into a production line block failure prediction result.

[0078] Furthermore, the fusion analysis module 40 is used to perform the following operation steps:

[0079] The production monitoring data, remaining effective life prediction results, and quality deviation prediction results of each block are aggregated. Data alignment and connection are performed based on the temporal and physical connection relationships between blocks to establish a production line-level data chain. Based on the production line-level data chain, bottleneck blocks of the production line are identified by comparing the cycle time and throughput of each block. Based on the bottleneck blocks of the production line, the production line-level data chain is used to predict and evaluate product packaging quality and evaluate the production line impact of equipment operation failures to obtain the production line fusion failure results.

[0080] Furthermore, the fusion analysis module 40 is used to perform the following operation steps:

[0081] The system receives production monitoring data streams, remaining usable life prediction results, and quality deviation prediction results from small digital twin models of each block, and converts heterogeneous data into standardized data objects, marking them with time-series identifiers based on the production line global clock. Using the production line global clock as a reference, a fixed time window is defined, and the production monitoring data arriving from each block within each time window is aggregated to generate a unified production line status within the window period. For predictive results, the original prediction timestamp and release timestamp are retained, and the relative time difference with the current time is calculated, aligning the data according to time sequence relationships. Based on the production line topology structure obtained from the time sequence and physical connection relationships between blocks, the time-aligned data is associated with the production line topology structure. The production line data is stored in a time-series database in primary chronological order, forming a production line-level data chain. This production line-level data chain supports query and traceability analysis by time dimension, spatial dimension, and process logic dimension.

[0082] Furthermore, the fusion analysis module 40 is used to perform the following operation steps:

[0083] Based on the prediction results of the bottleneck block in the production line, the upstream and downstream block impact analysis is performed along the spatial dimension of the production line-level data chain to simulate the block failure state of the upstream and downstream blocks at the time point of the bottleneck block prediction result. The production line impact analysis of product packaging quality prediction and equipment failure prediction state is performed along the time and logical dimensions of the production line-level data chain to obtain the failure impact prediction results of the upstream and downstream blocks. Based on the production line-level data chain, the upstream and downstream blocks undergo iterative impact analysis of adjacent blocks until the correlation analysis of all blocks is completed. The failure impact prediction results of all blocks are then fused into the production line process to obtain the fused failure result of the production line.

[0084] Furthermore, the fault feedback module 50 is used to perform the following operation steps:

[0085] Based on the production line block fault prediction results and the production line fusion fault results, a real-time comparison is performed with the preset multi-level alarm rule library. When the predicted data meets the trigger threshold of the warning level, a fault warning information is generated, including the location of the warning device or block, the predicted fault event, the expected occurrence time, and the warning level.

[0086] Through the foregoing detailed description of the digital twin-driven packaging production line fault prediction method, those skilled in the art can clearly understand the digital twin-driven packaging production line fault prediction system in this embodiment. Since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and relevant parts can be referred to the method section.

[0087] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A digital twin driven method for failure prediction of a packaging line, characterized in that, The method comprises: Resolving the equipment operation parameters and packaging quality constraints of the process nodes of the packaging production line to establish a parameter-quality mapping relationship; Based on the equipment operation parameters and packaging quality constraints of the process nodes, the packaging operation is segmented to obtain a monitoring production line block; Based on the monitoring production line block, a small digital twin model is established respectively, and the production state detection analysis and fault prediction of the production line block are carried out in combination with the parameter-quality mapping relationship; The output block production state and production line block fault prediction result of the small digital twin model are connected through the production line level twin platform to carry out production line state fusion analysis and obtain the production line fusion fault result; According to the production line block fault prediction result and the production line fusion fault result, the production line fault feedback is carried out; Based on the equipment operation parameters and packaging quality constraints of the process nodes, the packaging operation is segmented to obtain a monitoring production line block, comprising: Based on the physical layout and process flow diagram of the production line, the sequence and connection relationship of each device and process node are determined; The sensitivity and dependence of the packaging quality constraints of the downstream process nodes on the upstream equipment operation parameters are analyzed, and the coupling strength is quantified; A group of devices and their associated process nodes with coupling strength exceeding a preset threshold and being physically adjacent are clustered and divided into a monitoring production line block; The data-driven prediction model includes two parallel output sub-models, namely the equipment health prediction model and the quality risk prediction model, wherein the equipment health prediction model takes the sequence of equipment state parameters as input and outputs the remaining useful life of the equipment, which is used to predict the performance degradation fault of mechanical or electrical components; the quality risk prediction model takes the real-time value and trend of process execution parameters as input, combines the parameter-quality mapping relationship, and outputs the quality deviation trend, which is used to predict the quality fault caused by process drift; The production line state fusion analysis is carried out to obtain the production line fusion fault result, comprising: The production monitoring data, remaining useful life prediction result and quality deviation prediction result of each block are summarized, the data are aligned and connected based on the time sequence and physical connection relationship between the blocks, and the production line level data chain is established; According to the production line level data chain, the production line bottleneck block is identified by comparing the beat and throughput of each block; According to the production line bottleneck block, the production line level data chain is evaluated in terms of product packaging quality prediction and equipment operation fault impact, and the production line fusion fault result is obtained.

2. The digital twin driven packaging line failure prediction method of claim 1, wherein, Based on the monitoring production line block, a small digital twin model is established respectively, comprising: According to the geometric model, physical mechanism model and data-driven prediction model of the equipment in each monitoring production line block, a small digital twin model is constructed, wherein the data-driven prediction model is obtained by training historical equipment parameter data, historical quality data and historical fault data through a machine learning algorithm, and is used to predict the remaining useful life or quality deviation trend of the equipment in the block.

3. The digital twin driven packaging line failure prediction method of claim 1, wherein, In combination with the parameter-quality mapping relationship, the production state detection analysis and fault prediction of each monitoring production line block are carried out, comprising: The real-time collected equipment operation parameter stream is input into the block small digital twin model, each block small digital twin model is driven to run, the physical mechanism model is simulated according to the real-time parameters, and the actual operation index of the equipment is compared to realize state detection and abnormality identification; The real-time parameter sequence is input into the equipment health prediction model and the quality risk prediction model respectively, the time-based remaining effective life prediction result is output through the equipment health prediction model, and the quality deviation prediction result based on probability is output through the quality risk prediction model; The remaining effective life prediction result and the quality deviation prediction result are jointly packaged as a line block fault prediction result.

4. The digital twin driven packaging line failure prediction method of claim 1, wherein, The production monitoring data, the remaining effective life prediction result and the quality deviation prediction result of each block are summarized, the data are aligned and connected based on the time sequence and the physical connection relationship between the blocks, and a line level data chain is established, including: The production monitoring data stream, the remaining effective life prediction result and the quality deviation prediction result from each block small digital twin model are received, the heterogeneous data are converted into standardized data objects, and the time sequence identifiers based on the line global clock are stamped; Each block production monitoring data arriving in each time window is aggregated based on the line global clock as a reference to define a fixed time window, a unified state of the line in the window period is generated, for the predictive result, the original prediction timestamp and the publishing timestamp are retained, and the relative time difference with the current time is calculated, the data are aligned according to the time sequence relationship; The time sequence aligned data are associated with the line topology structure based on the line topology structure obtained based on the time sequence and the physical connection relationship between the blocks; The line data are stored in a time sequence database in time order to form a line level data chain, and the line level data chain supports query and trace analysis in the time dimension, the space dimension and the process logic dimension.

5. The digital twin driven packaging line failure prediction method of claim 4, wherein, The line level data chain is subjected to product packaging quality prediction evaluation and line influence evaluation of equipment operation failure according to the line bottleneck block to obtain a line fusion failure result, including: The upstream and downstream block influence analysis is performed along the space dimension of the line level data chain based on the prediction result of the line bottleneck block to simulate the block failure state of the upstream and downstream blocks at the prediction result time point of the line bottleneck block; The product packaging quality prediction evaluation and the line influence analysis of the equipment failure prediction state are performed along the time dimension and the logic dimension of the line level data chain to obtain the failure influence prediction result of the upstream and downstream blocks; Based on the line level data chain, the adjacent block iterative influence analysis is performed on the upstream and downstream blocks until the correlation analysis of all blocks is completed, the failure influence prediction results of all blocks are fused in the line process to obtain the line fusion failure result.

6. The digital twin driven packaging line failure prediction method of claim 1, wherein, The line fault feedback is performed according to the line block fault prediction result and the line fusion failure result, including: The line block fault prediction result and the line fusion failure result are compared with a preset multi-level alarm rule library in real time, when the prediction data meet the triggering threshold of the warning level, the fault warning information is generated, including the warning equipment or block position, the predicted failure event, the predicted occurrence time and the warning level.

7. A digital twin driven packaging line failure prediction system, characterized in that, The system comprises: a mapping relationship establishing module configured to analyze equipment operation parameters of a packaging production line and packaging quality constraints of process nodes, and establish a parameter-quality mapping relationship; a packaging operation segmentation module configured to segment packaging operations based on the equipment operation parameters and the packaging quality constraints of the process nodes, and obtain monitored line blocks; a fault prediction module configured to establish small digital twin models based on the monitored line blocks, respectively, and combine the parameter-quality mapping relationship to perform production state detection analysis and line block fault prediction for each monitored line block; a fusion analysis module configured to connect block production states and line block fault prediction results of outputs of the small digital twin models through a line-level twin platform, perform line state fusion analysis, and obtain a line fusion fault result; a fault feedback module configured to perform line fault feedback according to the line block fault prediction result and the line fusion fault result.

Images