A Smart Analysis and Optimization System and Method for the Production of Traditional Chinese Medicine Oral Liquids

Abstract

This invention relates to an intelligent analysis and optimization system and method for the production of traditional Chinese medicine oral liquids, belonging to the field of industrial intelligent manufacturing and data analysis technology. The system adopts a three-layer architecture: the data layer collects multi-source heterogeneous data from SCADA systems, MES systems, and a digital twin platform, performs cleaning, storage, and feature engineering processing to construct production features for analysis; the algorithm layer deploys four core models: a production cycle time and bottleneck identification model, an equipment comprehensive efficiency root cause analysis model, a human efficiency and capacity matching optimization model, and a predictive maintenance model; the application layer displays the algorithm results through a visual dashboard and integrates them into the digital twin platform via an API interface, highlighting bottleneck equipment in a 3D scene to form a closed-loop optimization. This invention is mainly used in traditional Chinese medicine oral liquid workshops, enabling precise bottleneck location, quantitative OEE root cause analysis, dynamic matching of human efficiency and capacity, and predictive maintenance of equipment failures, significantly improving production efficiency and intelligence levels.

Description

Technical Field

[0001] This invention belongs to the field of industrial intelligent manufacturing and data analysis technology, and relates to an intelligent analysis and optimization system and method for the production of traditional Chinese medicine oral liquids in workshops. Background Technology

[0002] The production process of traditional Chinese medicine oral liquid preparation workshops is highly dependent on automated equipment such as filling machines, sterilizers, and cartoning machines. Its production management involves multi-source data, multiple processes, and numerous operators, making it a typical complex industrial system. Currently, workshop management mainly relies on traditional SCADA systems for data acquisition and monitoring, as well as equipment overall efficiency (OEE) calculations and bottleneck analysis based on human experience.

[0003] In existing technologies, there are several intelligent manufacturing systems for the pharmaceutical industry. For example, Chinese patent CN112163688A discloses a full-process intelligent manufacturing system for multi-dosage form Chinese and Western medicine products. This system achieves full lifecycle information integration through technologies such as the Internet of Things and big data. However, this system mainly focuses on macro-level factory layout and information integration, lacking refined and intelligent analysis and optimization of specific production processes in the workshop. Furthermore, Chinese patent CN115630839B proposes a production intelligent feedback control system based on data mining, using correlation analysis and cluster analysis to study the relationship between process parameters and quality indicators. However, its methods are relatively general and fail to deeply explore the root causes of production efficiency, nor does it address the dynamic matching and optimization of labor efficiency and production capacity.

[0004] In summary, the existing technology mainly has the following problems and shortcomings: Traditional OEE analysis can only provide an overall efficiency value and cannot accurately pinpoint the specific process, equipment, or root cause of inefficiency, such as specific equipment fault codes, material batch issues, or mismatches in cycle time between processes.

[0005] Production bottleneck identification and capacity matching optimization rely heavily on the subjective experience of managers, making it difficult to achieve data-driven scientific decision-making and resulting in suboptimal resource allocation.

[0006] Equipment operation data, production order data, and manually recorded data are scattered across different systems, lacking effective integration and real-time analysis. This makes it impossible to quickly warn and intelligently diagnose production anomalies, resulting in delayed problem handling.

[0007] Existing methods are mostly post-event statistics, which cannot predictively analyze potential equipment failures or efficiency decline trends based on historical and real-time data, making it difficult to achieve preventive maintenance.

[0008] Therefore, there is an urgent need in this field for a system that can integrate multi-source industrial data, apply targeted AI algorithms, and achieve automated, refined, and root-cause analysis of production cycle time and equipment OEE, and ultimately output executable optimization suggestions to comprehensively improve the overall production efficiency and intelligence level of traditional Chinese medicine oral liquid workshops. Summary of the Invention

[0009] The production process of traditional Chinese medicine oral liquid preparation workshops is highly dependent on automated equipment. Existing technologies mainly rely on traditional SCADA systems for data collection and monitoring, as well as OEE calculations and bottleneck analyses based on human experience. This approach suffers from problems such as limited analytical dimensions, reliance on experience without quantitative analysis, information silos leading to delayed responses, and a lack of predictability. Specifically, traditional methods cannot accurately pinpoint the specific processes, equipment, or root causes of inefficiency; production bottleneck identification and capacity matching optimization heavily rely on subjective experience; equipment operation data, production order data, and manually recorded data are scattered across different systems, lacking effective integration and real-time analysis; and existing methods are mostly post-event statistical analyses, unable to predictively analyze potential equipment failures or efficiency decline trends based on historical and real-time data.

[0010] In view of this, in order to solve the above-mentioned technical problems, the present invention provides an intelligent analysis and optimization system for the production of oral liquid traditional Chinese medicine, which adopts a three-layer architecture including a data layer, an algorithm layer and an application layer.

[0011] (1) Data layer, used to collect multi-source heterogeneous data from SCADA system, MES system and digital twin platform, and to clean, store and feature-engineer the data to construct production features for analysis.

[0012] Specifically, the data layer collects real-time equipment status timing data and downtime events from the SCADA system via the OPC UA protocol, extracts production work order information and material information from the MES system through interfaces, obtains 3D equipment status data from the digital twin platform, and supports manual entry of abnormal event data via web forms. After data collection, it is stored in a batch-stream combined manner and advanced production features are constructed based on business logic, such as the capacity difference between adjacent processes and the number of times a specific fault code occurs per unit time.

[0013] Furthermore, the feature engineering process includes: calculating the capacity difference between adjacent processes as a feature indicator for identifying blockages between processes; and counting the number of times a specific fault code occurs per unit time as a feature indicator for root cause analysis.

[0014] (2) The algorithm layer connects with the data layer and deploys four core models. These are: The production cycle time and bottleneck identification model, based on the aforementioned production characteristics, uses an association rule mining algorithm to identify bottleneck process combinations that cause production blockages.

[0015] Furthermore, the input features of the production cycle time and bottleneck identification model include: the real-time operating speed of each piece of equipment, the deviation rate between the speed of each piece of equipment and the design speed, the speed fluctuation coefficient of each piece of equipment, and the amount of material in the buffer area of ​​the preceding and following processes; the minimum support of the association rule mining is 0.1, and the minimum confidence is 0.7.

[0016] The root cause analysis model for overall equipment efficiency, based on the aforementioned production characteristics, employs a tree model to analyze the importance of features and quantitatively identify key factors affecting overall equipment efficiency. Furthermore, the input features of the root cause analysis model for overall equipment efficiency include: equipment characteristics, time characteristics, personnel characteristics, material characteristics, and environmental characteristics; the tree model is an XGBoost model, and its positive training data consists of equipment downtime events or events where overall equipment efficiency is below 70% for a duration exceeding 15 minutes.

[0017] The human efficiency and production capacity matching optimization model, based on the aforementioned production characteristics, uses discrete event simulation and genetic algorithms to generate personnel scheduling and production optimization schemes.

[0018] Furthermore, the fitness function for optimization using the genetic algorithm is:

[0019] in, For the first The completed output of each order. For the first The idle time of each process This is the penalty coefficient; The genetic algorithm uses preset population size, crossover probability, mutation probability and termination conditions for iterative optimization to generate personnel scheduling and production scheduling optimization schemes.

[0020] The predictive maintenance model, based on the aforementioned production characteristics, uses a long short-term memory network to learn from the time-series data of key equipment to predict the probability of future equipment failures.

[0021] Furthermore, the input layer of the predictive maintenance model uses a time window with a preset time step, and the dimension of the input features is determined based on the number of monitoring parameters for key equipment. The output layer activation function is Sigmoid, which outputs the probability of equipment failure within a preset future time period; its loss function is binary cross-entropy.

[0022] in, For real labels, This represents the probability predicted by the model.

[0023] (3) Application layer, connected to algorithm layer, is used to display the bottleneck process combination, the key factors, the optimization scheme and the failure probability through a visual dashboard, and to integrate it into the digital twin platform through API interface, so as to highlight the bottleneck equipment in the three-dimensional scene and form a closed loop optimization.

[0024] Furthermore, the application layer encapsulates the algorithm model as a RESTful API service and pushes intelligent early warning information through WeChat or an app; when the failure probability output by the predictive maintenance model is greater than a preset threshold, an early warning is triggered and a maintenance work order containing fault location and suggested measures is pushed.

[0025] This invention also provides an intelligent analysis and optimization method for the production of oral liquid traditional Chinese medicine in a workshop, which is applied to the above-mentioned system and includes steps such as data acquisition and processing, bottleneck identification, root cause analysis, optimization scheduling, predictive maintenance, and visualization closed loop.

[0026] Specifically, the data layer collects multi-source heterogeneous data and performs cleaning, storage, and feature engineering processing to construct production characteristics; the algorithm layer's production cycle time and bottleneck identification model, based on the production characteristics, uses association rule mining algorithms to identify bottleneck process combinations causing production blockages; the algorithm layer's equipment comprehensive efficiency root cause analysis model, based on the production characteristics, uses a tree model to analyze feature importance and quantitatively identify key factors affecting equipment comprehensive efficiency; the algorithm layer's human efficiency and capacity matching optimization model, based on the production characteristics, uses discrete event simulation and genetic algorithms to generate personnel scheduling and production scheduling optimization schemes; the algorithm layer's predictive maintenance model, based on the production characteristics, uses a long short-term memory network to learn from the time-series data of key equipment and predict the probability of future equipment failures; the application layer displays the bottleneck process combinations, key factors, optimization schemes, and failure probabilities through a visual dashboard and integrates them into a digital twin platform through an API interface to highlight bottleneck equipment in a 3D scene, forming a closed-loop optimization.

[0027] The present invention also provides a computer-readable storage medium.

[0028] Compared with the prior art, the beneficial effects of the present invention are as follows: First, it achieves refined production management and data-driven decision-making. This invention constructs a three-layer architecture to deeply integrate multi-source heterogeneous data from SCADA, MES, and digital twin platforms, and builds advanced production features, providing a data foundation for precise analysis. Based on this, the production cycle time and bottleneck identification model uses association rule mining algorithms to pinpoint capacity bottlenecks to specific bottleneck process combinations; the equipment OEE root cause analysis model uses a tree model for feature importance analysis, quantitatively attributing the causes of low OEE to specific key factors, replacing the traditional analysis mode that relies on manual experience. This achieves a shift from qualitative experience to quantitative data-driven approaches, significantly improving the scientific nature of production management.

[0029] Secondly, it achieves multi-dimensional collaborative optimization and predictive maintenance, significantly improving production efficiency and equipment reliability. This invention is not the application of a single model, but rather the organic integration of four core models to form a closed-loop system of collaborative optimization. The labor efficiency and capacity matching optimization model uses discrete event simulation and genetic algorithms to dynamically generate personnel scheduling and production plans based on real-time production status, effectively solving the problem of capacity mismatch between processes and achieving optimal resource allocation. The predictive maintenance model uses a long short-term memory network to learn from the time-series data of key equipment, enabling it to predict equipment failure probabilities in advance and achieve a shift from passive maintenance to proactive prevention. The collaborative work of these models significantly improves the overall production efficiency of the workshop and the reliability of equipment operation.

[0030] Third, it achieves a closed-loop visualization and highly integrated application of the analysis results. This invention encapsulates the algorithm model as a RESTful API service, displaying bottleneck processes, OEE root cause rankings, optimized scheduling schemes, and fault warning information in real time through a visual dashboard. The analysis results are seamlessly integrated into existing digital twin platforms via the API interface, highlighting bottleneck devices in a 3D scene, forming a closed-loop optimization process of perception-analysis-decision-feedback. This significantly reduces implementation costs and technical barriers, while providing managers with an intuitive and efficient decision support tool, significantly improving management efficiency and user experience.

[0031] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0032] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein: Figure 1This is a schematic diagram of the overall architecture of the intelligent analysis and optimization system for the production of oral liquid traditional Chinese medicine in a workshop, provided in an embodiment of the present invention. Figure 2 This is a schematic diagram illustrating the multi-source data acquisition, processing, storage, and feature engineering construction process of the data layer in this embodiment of the invention. Figure 3 This is a schematic diagram illustrating the relationship between the input features, algorithm logic, and output targets of the four core models of the algorithm layer in this embodiment of the invention. Figure 4 This is a schematic diagram illustrating the implementation process of service encapsulation, visual dashboard display, intelligent early warning push, and integration with the digital twin platform to form a closed-loop optimization in the application layer of this invention. Detailed Implementation

[0033] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0034] The accompanying drawings are for illustrative purposes only and are schematic diagrams, not actual pictures. They should not be construed as limiting the invention. To better illustrate the embodiments of the invention, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0035] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," "front," and "rear" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present invention. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0036] like Figure 1 As shown, the intelligent analysis and optimization system for the production of oral liquid traditional Chinese medicine in workshops provided by this invention adopts a three-layer architecture, including a data layer, an algorithm layer, and an application layer.

[0037] The data layer, located at the bottom of the architecture, is responsible for the acquisition, management, and feature engineering of multi-source data. Specifically, the system acquires real-time equipment status time-series data from the SCADA system via standardized protocols, including but not limited to the operating speed of the filling machine (unit: bottles / minute), the chamber temperature of the sterilizer (unit: °C), the counting pulses of the cartoning machine (unit: times / minute), and equipment shutdown event codes. Simultaneously, it extracts production work orders (including order number, product name, planned output, and delivery date), material batch information (including batch number, material name, supplier, and expiration date), and process parameter settings from the MES system via interfaces. Furthermore, the system obtains three-dimensional status data of the equipment (such as the current operating, shutdown, and fault status) and real-time location information from the existing digital twin platform. To supplement abnormal situations that automatic acquisition cannot cover, the system also provides a web-based form-based manual input interface for recording abnormal events such as material shortages and descriptions of sudden equipment failures.

[0038] After data acquisition, the system first performs data cleaning to remove obvious outliers and duplicate data. Next, it uses a batch-stream combined approach for storage: a MySQL relational database stores business relationship data such as production work orders, personnel information, and bills of materials, while a time-series database stores continuous time-series data collected by device sensors. Based on this, the system constructs model input features based on business logic. In one embodiment, the system calculates the capacity difference between adjacent processes, i.e., the difference between the output of the filling process and the output of the packaging process (in bottles / minute), as an important indicator for identifying inter-process blockages. In another embodiment, the system counts the number of times a specific fault code occurs per unit time for subsequent root cause analysis.

[0039] like Figure 2 As shown, in its specific implementation, the data layer of this invention further clarifies the acquisition path, storage medium, and feature construction logic for multi-source data. Real-time equipment status and downtime events from the SCADA system, production work orders and material information from the MES system, and 3D equipment status data from the digital twin platform are respectively imported. Business relationship data is stored in a MySQL database, while high-frequency time-series equipment data is stored in an InfluxDB time-series database. Based on this, feature engineering is performed according to the business construction logic to generate advanced features such as the capacity difference between adjacent processes and the planned equipment downtime rate, ultimately constructing a production analysis data warehouse and feature library for model analysis.

[0040] The algorithm layer, located in the middle of the architecture, deploys four core models that address optimization issues in four dimensions: production cycle time, equipment efficiency, human resource efficiency matching, and equipment health. The inputs to these models all originate from production characteristics processed by the data layer.

[0041] like Figure 3As shown, the input features, core algorithm logic, and output targets of each model within the algorithm layer have a clear correspondence. The production cycle time and bottleneck identification model is based on the actual capacity, speed deviation, and fluctuation of each process, and uses association rule mining to identify bottleneck process combinations that cause upstream and downstream blockages; the equipment OEE root cause analysis model takes equipment, personnel, materials, and environmental characteristics as inputs, and uses a tree model to output key factors affecting efficiency; the human efficiency and capacity matching optimization model generates personnel scheduling and production scheduling suggestions through discrete event simulation and genetic algorithm optimization; and the predictive maintenance model learns from the time-series data of key equipment such as temperature, current, and vibration based on a long short-term memory network, outputs the failure probability, and provides data support for planned maintenance.

[0042] (1) Production cycle time and bottleneck identification model In one specific embodiment, this model employs a combination of statistical analysis and association rule mining. The system is based on features constructed from the data layer, including the real-time operating speed of each device, the deviation rate between the speed of each device and the design speed, the speed fluctuation coefficient of each device, and the amount of material in the buffer areas of the preceding and following processes.

[0043] An event where "the speed of a certain process is lower than 80% of the design speed for more than 5 minutes" is defined as a transaction. The minimum support is set to 0.1, and the minimum confidence to 0.7. The system uses the Apriori algorithm to mine frequent itemsets and identify bottleneck process combinations. For example, the frequently mined itemset {filling line 3, packaging line 5} indicates that this combination frequently experiences insufficient speed simultaneously, which is the main cause of production blockage.

[0044] In another embodiment, to reduce computational load, the FP-Growth algorithm can be used instead of the Apriori algorithm for frequent pattern mining.

[0045] In another embodiment, the minimum support and minimum confidence levels can be dynamically adjusted. For example, the support threshold can be appropriately lowered during peak order periods to capture more potential bottleneck combinations. The system outputs a list of current bottleneck processes and associated rules in real time, which is then pushed to administrators through the application layer.

[0046] (2) Equipment OEE Root Cause Analysis Model In one specific embodiment, this model uses a tree model as the core algorithm for feature importance analysis. The input features provided by the data layer mainly include equipment features (equipment type, equipment ID, equipment model), time features (shift, time), personnel features (operator ID), material features (material batch, material type), and environmental features (ambient temperature). Among them, the category features are vectorized using one-hot encoding.

[0047] Labels are defined as follows: events of "equipment downtime" or "OEE below 70% for more than 15 minutes" are considered positive samples with a label of 1; other events are considered negative samples with a label of 0. The formula for calculating Overall Equipment Effectiveness (OEE) is: OEE = Availability × Performance × Quality. Where Availability is the equipment utilization rate, Performance is the performance utilization rate, and Quality is the first-pass yield.

[0048] The model was trained using historical data from the past two months, containing approximately 20,000 samples, of which about 5,000 were positive samples. The training and validation sets were split in an 8:2 ratio, and 5-fold cross-validation was used to fine-tune the tree model's parameters, such as adjusting the maximum depth, learning rate, and number of weak learners. After training, the model outputs a ranking of feature importance.

[0049] In one embodiment, if the number of times the "hoist fault code F001 occurs" ranks first, then the fault can be determined to be a key factor causing the decline in OEE.

[0050] In another embodiment, if the feature importance of "Operator ID003" is high, a suggestion is made that additional training or skills assessment is needed for that operator. In yet another embodiment, a random forest or LightGBM model can be used instead of XGBoost.

[0051] In yet another embodiment, to improve the interpretability of the model, SHAP (SHapley Additive ex Planations) values ​​can be used to visualize the contribution of each feature to a single prediction result, thereby more accurately pinpointing the specific reasons for the decline in OEE.

[0052] (3) Optimization model for matching human efficiency with production capacity In one specific embodiment, this model employs an optimization strategy that combines discrete event simulation with genetic algorithms.

[0053] First, a discrete event simulation model of the workshop production process is constructed, including each step of preparation, filling, sterilization, and packaging. Parameters such as the capacity limit of each piece of equipment, standard working hours, buffer capacity, and personnel skill matrix are set. For example, the capacity limit of the filling machine is set to 200 bottles / minute, and the buffer capacity is set to 5000 bottles.

[0054] Then, a genetic algorithm is used for optimization. The population size is set. =100; Crossover probability =0.8; mutation probability pm=0.1. Termination condition is reaching the maximum number of generations. =200 generations, or 30 consecutive generations with no improvement in fitness. The fitness function is set to maximize overall output P or minimize total production line idle time while meeting order delivery deadlines. The fitness function that maximizes overall output is:

[0055] in, For the first The completed output of each order. For the first The idle time of each process This is the penalty coefficient.

[0056] After iterative optimization using a genetic algorithm, the system outputs a staffing schedule and production plan for the next 24 hours. For example, the output might be: "Add two skilled packaging workers to the morning shift and prioritize assigning orders for product A to lines 1-5."

[0057] In another embodiment, particle swarm optimization (PSO) can be used instead of genetic algorithm.

[0058] In another embodiment, the optimization objective can be set as minimizing total production cost, taking into account factors such as energy consumption and material loss.

[0059] In another embodiment, when the urgency of orders differs, different weighting coefficients can be set for different orders, so that the optimization scheme prioritizes the delivery of urgent orders.

[0060] (4) Predictive maintenance model In one specific embodiment, this model uses a Long Short-Term Memory (LSTM) network to learn from the time-series data of critical equipment to predict failure probabilities. The data layer provides historical operating data of the critical equipment over the past six months, containing at least 100 failure events. The network structure is designed as follows: Input layer time step T=60, using data from the past 60 minutes, feature dimension F=3 (temperature, current, vibration amplitude). First LSTM layer: number of neurons=64, return sequence=True. Second LSTM layer: number of neurons=32, return sequence=False. A Dropout layer is then added with a ratio of 0.2 to prevent overfitting. Fully connected layer: number of neurons=16, activation function=ReLU. Output layer: number of neurons=1, activation function=Sigmoid, used to output the failure probability value (0-1).

[0061] During training data preparation, historical data is divided into 60-minute windows. If a failure occurs within 30 minutes after the window ends, the window is labeled 1; otherwise, it is labeled 0. The loss function used is binary crossentropy, with the following formula:

[0062] in, For real labels, This represents the probability predicted by the model. The optimizer is Adam, and the initial learning rate is set to 0.001.

[0063] The model outputs the probability of equipment failure within the next 30 minutes in real time. When the failure probability is greater than a preset threshold (e.g., 0.7), an early warning is triggered, and maintenance personnel are notified in advance through the application layer.

[0064] In another embodiment, a gated recurrent unit (GRU) can be used instead of an LSTM to reduce model parameters and training time.

[0065] In another embodiment, different LSTM models can be constructed according to different equipment types. For example, vibration features can be added to rotating equipment, and temperature gradient features can be added to thermal equipment.

[0066] In another embodiment, in order to obtain a longer prediction lead time, the prediction target of the time window can be set to a fault event within the next 1 or 2 hours, and the network structure can be adjusted accordingly to accommodate the longer temporal dependencies.

[0067] The application layer, located at the top of the architecture, is responsible for service-oriented encapsulation, visualization, and business integration of the algorithm model's output. Algorithms are encapsulated as RESTful API services, displaying analysis results through visual dashboards and generating intelligent alerts. Ultimately, all analytical capabilities are seamlessly integrated into the existing digital twin platform via APIs and highlighting functionality, forming a closed loop.

[0068] like Figure 4 As shown, the application layer encapsulates the algorithm model into an HTTP interface via a RESTful API service, allowing the visualization dashboard and third-party systems to call it. The visualization dashboard displays real-time production line status, OEE dynamics, bottleneck process alarms, and key factor conclusions, and can be embedded in the analysis tab of the digital twin platform. When the algorithm identifies a bottleneck or predicts a failure risk, the system automatically generates an early warning event containing key information and suggested measures, which is pushed via WeChat or the app. Simultaneously, after the digital twin platform calls the algorithm API service, it highlights the bottleneck equipment in the 3D scene, achieving closed-loop optimization from data perception to feedback control.

[0069] In one specific embodiment, the application layer encapsulates the four models mentioned above as independent RESTful API services for front-end web applications and mobile apps to call. The front-end builds a visual analysis dashboard to display in real time the association rule graph of bottleneck processes, the feature importance ranking of OEE root causes, the Gantt chart of human efficiency and capacity optimization schemes, and the failure probability trend curve of predictive maintenance.

[0070] The system establishes an intelligent early warning and push mechanism, pushing key early warning information (such as "Packaging line 5, failure probability 0.85, recommended to check guide rail sensors") to relevant personnel via WeChat or DingTalk robots. Simultaneously, all analytical capabilities and results are seamlessly integrated into the existing digital twin platform via API and highlighting functions. For example, after the digital twin platform obtains the "bottleneck process combination" results via API, the equipment marked as a bottleneck is highlighted in red and flashes as a warning in the 3D scene, forming a closed loop of perception-analysis-decision-feedback.

[0071] In another embodiment, the application layer can provide what-if analysis capabilities, allowing managers to manually adjust personnel or equipment configurations in a simulation model and view the impact on production line capacity in real time.

[0072] In another embodiment, the system can automatically generate periodic analysis reports that summarize the bottleneck distribution, OEE change trends, and the effectiveness of optimization suggestions over a period of time, and send them to management via email.

[0073] Taking a traditional Chinese medicine oral liquid production line as an example, the system operation process is explained as follows: Data aggregation: The system collects real-time data on equipment speed, temperature, and counts for processes such as filling, sterilization, and packaging, as well as production batch information from the MES, through the data layer.

[0074] Bottleneck identification: The production cycle model of the algorithm layer, through correlation analysis, found that when producing product A, the speed of the packaging process is consistently lower than that of the filling process, and this combination occurs frequently, and is marked as a bottleneck by the system.

[0075] Root cause diagnosis: The system triggers the OEE root cause analysis model to analyze the data during the bottleneck period. The feature importance ranking results show that "cartoning machine material guide rail jamming alarm" is the most important feature factor causing the OEE decline during this period.

[0076] Intelligent Early Warning and Optimization: The application layer immediately pushes an early warning to the maintenance supervisor's APP, including fault location (packaging line 5) and a suggestion to "check the guide rail sensors." Simultaneously, the efficiency and capacity matching optimization model begins calculating production scheduling adjustments for subsequent batches, suggesting that subsequent batches of product A orders be prioritized for allocation to other packaging lines in good maintenance condition.

[0077] Results: Maintenance personnel can quickly address issues based on early warning information, avoiding prolonged downtime. The system continuously learns, and long-term operation can achieve quantifiable goals such as improving overall OEE and reducing troubleshooting time.

[0078] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A smart analysis and optimization system for the production of oral liquid traditional Chinese medicine in a workshop, characterized in that, include: The data layer is used to collect multi-source heterogeneous data from SCADA systems, MES systems, and digital twin platforms, and to clean, store, and feature-engineer the data to construct production features for analysis. The algorithm layer, connected to the data layer, includes: A production cycle time and bottleneck identification model is used to identify bottleneck process combinations that cause production blockages based on the aforementioned production characteristics and an association rule mining algorithm. The equipment overall efficiency root cause analysis model is used to analyze the importance of features based on the aforementioned production characteristics using a tree model, and to quantitatively identify the key factors affecting the overall efficiency of the equipment. A human efficiency and production capacity matching optimization model is used to generate personnel scheduling and production optimization schemes based on the aforementioned production characteristics, employing discrete event simulation and genetic algorithms; and, A predictive maintenance model is used to learn from the time-series data of key equipment using a long short-term memory network based on the aforementioned production characteristics, and to predict the probability of future equipment failures. The application layer, connected to the algorithm layer, is used to display the bottleneck process combination, the key factors, the optimization scheme, and the failure probability through a visual dashboard, and is integrated into the digital twin platform through an API interface to highlight the bottleneck equipment in a three-dimensional scene, forming a closed-loop optimization.

2. The system according to claim 1, characterized in that, The multi-source heterogeneous data collected by the data layer includes: equipment status time-series data and shutdown events collected from the SCADA system via the OPC UA protocol; production work order information and material information extracted from the MES system via the interface; equipment three-dimensional status data obtained from the digital twin platform; and abnormal event data manually entered via Web forms.

3. The system according to claim 1 or 2, characterized in that, The feature engineering process includes: calculating the capacity difference between adjacent processes as a feature indicator for identifying blockages between processes; and counting the number of times a specific fault code occurs per unit time as a feature indicator for root cause analysis.

4. The system according to claim 1, characterized in that, The input features of the production cycle time and bottleneck identification model include: real-time operating speed of each device, deviation rate of each device speed from the design speed, speed fluctuation coefficient of each device, and material quantity in the buffer area of ​​the preceding and following processes; the minimum support of the association rule mining is 0.1, and the minimum confidence is 0.

7.

5. The system according to claim 1, characterized in that, The input features of the equipment comprehensive efficiency root cause analysis model include: equipment features, time features, personnel features, material features, and environmental features; the tree model is an XGBoost model, and the positive samples of its training data are equipment downtime events or events where the equipment comprehensive efficiency is less than 70% and lasts for more than 15 minutes.

6. The system according to claim 1, characterized in that, The optimization model for matching human efficiency with production capacity uses a genetic algorithm for optimization, and its fitness function is: in, For the first The completed output of each order. For the first The idle time of each process This is the penalty coefficient; The genetic algorithm uses preset population size, crossover probability, mutation probability and termination conditions for iterative optimization to generate personnel scheduling and production scheduling optimization schemes.

7. The system according to claim 1, characterized in that, The predictive maintenance model adopts a multi-layer long short-term memory network structure. The input layer uses a time window with a preset time step. The input feature dimension is determined based on the number of monitoring parameters of key equipment. The output layer activation function is Sigmoid, which is used to output the probability of equipment failure within a preset future time period. Its loss function is the binary cross-entropy: in, For real labels, This represents the probability predicted by the model.

8. The system according to claim 1, characterized in that, The application layer encapsulates the algorithm model as a RESTful API service and pushes intelligent early warning information through WeChat or an app. When the failure probability output by the predictive maintenance model is greater than a preset threshold, an early warning is triggered and a maintenance work order containing fault location and suggested measures is pushed.

9. A method for intelligent analysis and optimization of traditional Chinese medicine oral liquid production workshop, applied to the system described in any one of claims 1 to 8, characterized in that, Includes the following steps: The data layer collects heterogeneous data from multiple sources and performs cleaning, storage, and feature engineering processing to construct production features; Based on the aforementioned production characteristics, the production cycle time and bottleneck identification model at the algorithm layer uses an association rule mining algorithm to identify the bottleneck process combinations that cause production blockages. The root cause analysis model of equipment overall efficiency at the algorithm layer is based on the aforementioned production characteristics. It uses a tree model to analyze the importance of characteristics and quantitatively identify the key factors affecting equipment overall efficiency. The algorithm layer's human efficiency and production capacity matching optimization model, based on the aforementioned production characteristics, employs discrete event simulation and genetic algorithms to generate personnel scheduling and production optimization schemes. The predictive maintenance model at the algorithm layer, based on the aforementioned production characteristics, uses a long short-term memory network to learn from the time-series data of key equipment and predict the probability of future equipment failures. The application layer displays the bottleneck process combination, the key factors, the optimization scheme, and the failure probability through a visual dashboard, and integrates them into the digital twin platform through an API interface to highlight the bottleneck equipment in a 3D scene, forming a closed-loop optimization.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in claim 9.

Images