Intelligent manufacturing digital monitoring method and system based on industrial Internet of Things

Abstract

The invention discloses an intelligent manufacturing digital monitoring method and system based on industrial Internet of Things, and relates to the technical field of industrial Internet of Things and intelligent manufacturing. Comprising a multi-source heterogeneous data acquisition module used for accessing a sensor, a controller and production equipment in a manufacturing field through an industrial Internet of Things protocol adapter to realize real-time data acquisition of an equipment state, a process parameter, an energy consumption level and a production progress; and the edge intelligent preprocessing module is used for performing protocol analysis on the collected multi-source heterogeneous data. According to the invention, the industrial protocol conversion unit of the multi-source heterogeneous data acquisition module supports multi-protocol parallel analysis and maps original data into standard data frames, so that the problems of protocol isomerism and interface sealing of a data acquisition layer are solved; the edge intelligent preprocessing module dynamically adjusts the window length through an adaptive sliding window algorithm, and improves the data quality in combination with local outlier factor anomaly filtering and time sequence autoregression missing interpolation, thereby laying a foundation for subsequent analysis.

Description

Technical Field

[0001] This invention relates to the field of industrial Internet of Things (IIoT) and intelligent manufacturing technology, specifically to a digital monitoring method and system for intelligent manufacturing based on IIoT. Background Technology

[0002] With the deepening of Industry 4.0 and smart manufacturing strategies, Industrial Internet of Things (IIoT) technology has become a core support for realizing the digitalization, networking, and intelligence of manufacturing processes. By deeply interconnecting sensors, controllers, production equipment, and information systems, it constructs a data perception and collaborative control system covering the entire production process, significantly improving the transparency and responsiveness of manufacturing systems. Against this backdrop, real-time, accurate, and comprehensive digital monitoring of the manufacturing process has become a key requirement for ensuring production stability, optimizing resource allocation, and improving product quality.

[0003] Among them, intelligent manufacturing digital monitoring based on the Industrial Internet of Things aims to achieve dynamic perception and visual management of equipment status, process parameters, energy consumption levels, and production progress through the collection, transmission, and fusion analysis of multi-source heterogeneous data. The core objective of this technology direction is to break down the information silos in traditional manufacturing systems and build an end-to-end closed-loop monitoring mechanism to support advanced applications such as predictive maintenance, intelligent scheduling, and adaptive control.

[0004] Existing technologies still face multiple challenges in achieving the above objectives: First, the data acquisition layer generally suffers from heterogeneous protocols, closed interfaces, and inconsistent sampling frequencies, making it difficult to efficiently access and uniformly model equipment data; second, monitoring systems often use static thresholds or simple rules for status judgment, lacking the ability to deeply mine the correlation of multi-dimensional data under complex operating conditions, making it difficult to accurately identify early anomalies or potential faults; third, existing monitoring architectures typically adopt a centralized processing mode, resulting in high data transmission latency and bandwidth pressure, which can easily lead to insufficient real-time performance and limited system scalability in large-scale production line deployments; finally, there is a lack of effective linkage between monitoring results and production decisions, making it impossible to dynamically adjust process parameters or scheduling strategies based on real-time status, causing the monitoring system to remain only at the "observable" level and failing to truly achieve "interventionist and optimizable" closed-loop control. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a digital monitoring method and system for intelligent manufacturing based on the Industrial Internet of Things (IIoT), which solves the problems of heterogeneous data acquisition protocols, weak anomaly identification capabilities, high latency in centralized processing, and disconnect between monitoring and decision-making in existing technologies.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a digital monitoring system for intelligent manufacturing based on the Industrial Internet of Things, comprising:

[0007] The multi-source heterogeneous data acquisition module is used to connect to sensors, controllers and production equipment in the manufacturing site through an industrial IoT protocol adapter to realize real-time data acquisition of equipment status, process parameters, energy consumption level and production progress.

[0008] The edge intelligent preprocessing module is used to perform protocol parsing, format unification and sampling frequency alignment on the collected multi-source heterogeneous data, and to perform abnormal data filtering and missing value imputation based on sliding window.

[0009] The dynamic modeling and feature extraction module is used to construct a multi-dimensional time series correlation analysis model and extract the deep coupling features between equipment operating status and process parameters;

[0010] The distributed collaborative decision-making module is used to generate equipment maintenance warnings, process parameter adjustment instructions, and production scheduling strategies based on feature extraction results.

[0011] The visualization and closed-loop control module is used to present the monitoring results in a graphical interface and drive the actuators to achieve adaptive adjustment of process parameters according to decision instructions.

[0012] Furthermore, the multi-source heterogeneous data acquisition module integrates an industrial protocol conversion unit, which supports parallel parsing of ModbusTCP, OPCUA, Profinet and EtherCAT protocols, and maps the raw data into standard data frames with timestamps, device identifiers, data values ​​and quality identifiers through a unified data model.

[0013] Furthermore, the edge intelligent preprocessing module is equipped with an adaptive sliding window algorithm. The window length is dynamically adjusted according to the data sampling frequency. Its abnormal data filtering adopts a detection mechanism based on local outlier factors, and missing value imputation uses a time series autoregressive model for prediction and filling.

[0014] Furthermore, the dynamic modeling and feature extraction module adopts a hybrid architecture of graph convolutional networks and long short-term memory networks. The graph convolutional network is used to model the topological dependencies between devices, while the long short-term memory network is used to capture long-term dynamic patterns of multi-dimensional time series. The dynamic modeling and feature extraction module calculates the correlation strength between device states and process parameters using the following formula:

[0015] ;

[0016] in, Indicates the first Each device at time Status indicators Indicates the first Each process parameter at time... The value, The length of the time series. This is the correlation coefficient.

[0017] Furthermore, the distributed collaborative decision-making module is composed of edge nodes and a cloud platform. The edge nodes are responsible for real-time anomaly detection and local decision-making, while the cloud platform is responsible for global optimization and long-term strategy generation. The edge nodes use a lightweight random forest classifier to classify the device status in real time, and the cloud platform dynamically adjusts the production scheduling strategy through a deep reinforcement learning algorithm.

[0018] Furthermore, the visualization and closed-loop control module integrates a digital twin engine, which uses 3D rendering technology to display the equipment operating status, process parameter curves, and abnormal alarm information in real time. Based on the output instructions of the decision module, the engine dynamically adjusts the control parameters of the actuator through the PID controller to achieve closed-loop optimization of process parameters.

[0019] This invention also provides a digital monitoring method for intelligent manufacturing based on the Industrial Internet of Things (IIoT), applicable to any of the aforementioned digital monitoring systems for intelligent manufacturing based on the IIoT, comprising the following steps:

[0020] S1. By connecting to sensors, controllers and production equipment on the manufacturing site through a multi-source heterogeneous data acquisition module, real-time data on equipment status, process parameters, energy consumption levels and production progress are collected.

[0021] S2. The edge intelligent preprocessing module is used to perform protocol parsing, format unification and sampling frequency alignment on the collected data, and to perform abnormal data filtering and missing value imputation based on sliding window.

[0022] S3. Construct a multi-dimensional time series correlation analysis model through dynamic modeling and feature extraction modules to extract the deep coupling features between equipment operating status and process parameters;

[0023] S4. Based on the feature extraction results, the distributed collaborative decision-making module generates equipment maintenance early warning, process parameter adjustment instructions and production scheduling strategies.

[0024] S5. The monitoring results are presented in a graphical interface through the visualization and closed-loop control module, and the actuator is driven to achieve adaptive adjustment of process parameters according to the decision instructions.

[0025] Furthermore, in step S2, the adaptive sliding window algorithm dynamically adjusts the window length according to the data sampling frequency, the abnormal data filtering adopts a detection mechanism based on local outlier factors, and the missing value imputation adopts a time series autoregressive model for prediction and filling.

[0026] Furthermore, in step S3, a hybrid architecture of graph convolutional networks and long short-term memory networks is used to construct a correlation analysis model. The graph convolutional network models the topological dependencies between devices, while the long short-term memory network captures long-term dynamic patterns of multidimensional time series. The correlation strength between device status and process parameters is calculated using the following formula:

[0027] ;

[0028] in, Indicates the first Each device at time Status indicators Indicates the first Each process parameter at time... The value, The length of the time series. This is the correlation coefficient.

[0029] Furthermore, in step S4, the distributed collaborative decision-making is jointly executed by the edge nodes and the cloud platform. The edge nodes use a lightweight random forest classifier for real-time anomaly detection and local decision-making, while the cloud platform uses a deep reinforcement learning algorithm for global optimization and long-term policy generation.

[0030] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0031] This invention addresses the issues of heterogeneous protocols and closed interfaces in the data acquisition layer by supporting parallel parsing of multiple protocols and mapping raw data into standard data frames through the industrial protocol conversion unit of the multi-source heterogeneous data acquisition module. The edge intelligent preprocessing module dynamically adjusts the window length using an adaptive sliding window algorithm, combined with local outlier anomaly filtering and time series autoregressive missing data imputation, improving data quality and laying the foundation for subsequent analysis. The dynamic modeling and feature extraction module employs a hybrid architecture of GCN and LSTM to mine equipment topology dependencies and multi-dimensional time series dynamic patterns, calculating equipment and process coupling characteristics through correlation strength formulas, thus solving the problem of static thresholds failing to identify potential anomalies. The distributed collaborative decision-making module combines real-time anomaly detection and local decision-making at edge nodes with global optimization on the cloud platform, solving the problems of high latency and limited scalability in centralized processing. The visualization and closed-loop control module achieves monitoring and execution linkage through digital twin display and PID closed-loop control, solving the problem of disconnect between monitoring and decision-making, comprehensively improving the transparency and responsiveness of the manufacturing system, ensuring stable production, and optimizing resource allocation. Attached Figure Description

[0032] Figure 1 This is a system structure diagram of the present invention;

[0033] Figure 2 This is a flowchart of the method of the present invention. Detailed Implementation

[0034] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0035] Please see Figure 1 This invention provides a smart manufacturing digital monitoring system based on the Industrial Internet of Things, comprising:

[0036] The multi-source heterogeneous data acquisition module is used to connect to sensors, controllers and production equipment in the manufacturing site through an industrial IoT protocol adapter to realize real-time data acquisition of equipment status, process parameters, energy consumption level and production progress.

[0037] The edge intelligent preprocessing module is used to perform protocol parsing, format unification and sampling frequency alignment on the collected multi-source heterogeneous data, and to perform abnormal data filtering and missing value imputation based on sliding window.

[0038] The dynamic modeling and feature extraction module is used to construct a multi-dimensional time series correlation analysis model and extract the deep coupling features between equipment operating status and process parameters;

[0039] The distributed collaborative decision-making module is used to generate equipment maintenance warnings, process parameter adjustment instructions, and production scheduling strategies based on feature extraction results.

[0040] The visualization and closed-loop control module is used to present the monitoring results in a graphical interface and drive the actuators to achieve adaptive adjustment of process parameters according to decision instructions.

[0041] Specifically, in this embodiment, the multi-source heterogeneous data acquisition module is deployed in the manufacturing site control cabinet via an industrial IoT protocol adapter. The adapter has a built-in multi-protocol compatible interface, allowing simultaneous access to devices such as temperature sensors, PLC controllers, and CNC machine tools. It collects real-time data on equipment status (e.g., vibration values), process parameters (e.g., injection pressure), energy consumption levels (e.g., workshop electricity meter data), and production progress (e.g., order completion rate in the MES system), solving the problem of multi-device data access. The edge intelligent preprocessing module interfaces with the acquisition module's output, first parsing the raw protocol data from different devices. It converts the Modbus TCP protocol sensor data and the OPCUA protocol controller data into a unified JSON format. Based on the sampling frequency of each device (e.g., 10Hz for sensors, 5Hz for controllers), it aligns the data to a unified 8Hz frequency. Then, it filters sudden abnormal vibration values ​​through a sliding window and uses interpolation to supplement missing data from temporarily offline sensors, improving data quality. The dynamic modeling and feature extraction module constructs a hybrid model. First, it represents the material transfer relationships between equipment using a graph structure, then uses a graph convolutional network to learn the topological dependencies of the equipment. Next, it uses an LSTM network to analyze the multi-dimensional time series over the past 24 hours, extracting the coupling features of "equipment vibration value - injection pressure," thus solving the problem that static thresholds cannot uncover correlations. In the distributed collaborative decision-making module, edge nodes are deployed on servers near the production line, analyzing equipment data in real time to generate local maintenance warnings, such as warning of bearing wear when vibration is abnormal. The cloud platform receives data from each edge node and adjusts the production scheduling of multiple production lines through a global algorithm, solving the problem of centralized latency. The visualization and closed-loop control module displays a digital twin interface in the monitoring center, showing the 3D status and parameter curves of the equipment in real time. When the decision module outputs an injection pressure adjustment command, the actuator drives the hydraulic valve through PID control to achieve adaptive adjustment of process parameters, forming a "collection-analysis-decision-control" closed loop. This breaks down the disconnect between monitoring and decision-making, improving the overall transparency and responsiveness of the manufacturing system.

[0042] In this embodiment, the multi-source heterogeneous data acquisition module integrates an industrial protocol conversion unit. This unit supports parallel parsing of ModbusTCP, OPCUA, Profinet and EtherCAT protocols, and maps the raw data into standard data frames with timestamps, device identifiers, data values ​​and quality identifiers through a unified data model.

[0043] Specifically, the industrial protocol conversion unit of the multi-source heterogeneous data acquisition module adopts a hardware-level protocol parsing chip. This chip integrates firmware for parsing ModbusTCP, OPCUA, Profinet, and EtherCAT protocols, enabling parallel processing of data streams from different devices. For example, when a Profinet-prone robotic arm and an EtherCAT-prone conveyor belt are simultaneously transmitting data in the workshop, the protocol conversion unit can simultaneously parse both types of data, avoiding delays caused by protocol queuing. A unified data model defines a standard data frame structure, including a 13-bit timestamp (accurate to milliseconds), an 8-bit device identifier (e.g., "ARM01" representing robotic arm number 1), a 32-bit data value (e.g., robotic arm joint angle), and a 2-bit quality flag (00 indicating normal data, 01 indicating suspicious data). During raw data mapping, the protocol conversion unit first extracts core information from each protocol data, such as converting register values ​​in ModbusTCP data into data values ​​and node IDs in OPCUA data into device identifiers, before encapsulating them according to the standard frame structure.

[0044] In this embodiment, the edge intelligent preprocessing module is equipped with an adaptive sliding window algorithm. The window length is dynamically adjusted according to the data sampling frequency. Its abnormal data filtering adopts a detection mechanism based on local outlier factors, and missing value imputation uses a time series autoregressive model for prediction and filling.

[0045] Specifically, the adaptive sliding window algorithm of the edge intelligent preprocessing module dynamically adjusts the window length by monitoring the data sampling frequency in real time. When the sampling frequency of the data output by the acquisition module is 10Hz (one data point every 0.1 seconds), the window length is set to 20 data points covering 2 seconds of data. When the sampling frequency drops to 5Hz, the window length is automatically adjusted to 10 data points, also covering 2 seconds of data, ensuring the stability of the window coverage time range and avoiding fluctuations in the amount of window data due to frequency changes. Abnormal data filtering uses a Local Outlier Factor (LOF) detection mechanism, calculating the local density of each data point within its k-neighborhood (k set to 5). If the local density of a data point is significantly lower than that of other points in the neighborhood, it is determined to be abnormal data and removed. For example, sudden over-range vibration values ​​of sensors can be accurately identified. Missing value imputation uses a time-series autoregressive model AR(p). First, the lag order p=3 is determined through the PACF plot, and then a regression equation is constructed using the normal data from the three time points before the missing value.

[0046] ;

[0047] in The predicted value for the missing value at time t. For regression coefficients, The error term is used to predict and fill in missing data.

[0048] In this embodiment, the dynamic modeling and feature extraction module adopts a hybrid architecture of graph convolutional networks and long short-term memory networks. The graph convolutional network is used to model the topological dependencies between devices, while the long short-term memory network is used to capture long-term dynamic patterns of multi-dimensional time series. The dynamic modeling and feature extraction module calculates the correlation strength between device status and process parameters using the following formula:

[0049] ;

[0050] in, Indicates the first Each device at time Status indicators Indicates the first Each process parameter at time... The value, The length of the time series. This is the correlation coefficient.

[0051] Specifically, the dynamic modeling and feature extraction module adopts a hybrid architecture of Graph Convolutional Network (GCN) and Long Short-Term Memory (LSTM). First, a device topology graph is constructed, with devices as nodes and material transfer or signal connections between devices as edges. GCN learns the topological dependencies between nodes through an adjacency matrix. For example, given the material transfer relationship between an injection molding machine and a conveyor belt, GCN can strengthen the correlation weights between their data. The LSTM network constructs a three-layer hidden layer structure, inputting normalized multidimensional time-series data. Device status and process parameters are normalized to the [0,1] interval to ensure dimensional consistency. Long-term dynamic patterns of the data are captured through gating units, such as the correlation between temperature changes and process parameters over 12 hours of equipment operation. The correlation strength between device status and process parameters is calculated using the following formula:

[0052] ;

[0053] in This refers to the normalized state index of the i-th device at time t, such as the normalized vibration value. This represents the normalized value of the j-th process parameter at time t, such as the normalized value of injection pressure. T is the length of the time series, such as 14400, which corresponds to 4 hours sampled at 1 second. Let be the time series mean of the i-th device status index. Let j be the time series mean of the j-th process parameter. The correlation coefficient has a range of [-1, 1], and the larger the absolute value, the stronger the correlation.

[0054] In this embodiment, the distributed collaborative decision-making module is composed of edge nodes and a cloud platform. The edge nodes are responsible for real-time anomaly detection and local decision-making, while the cloud platform is responsible for global optimization and long-term strategy generation. The edge nodes use a lightweight random forest classifier to classify the device status in real time, and the cloud platform dynamically adjusts the production scheduling strategy through a deep reinforcement learning algorithm.

[0055] Specifically, in the distributed collaborative decision-making module, edge nodes are deployed as embedded servers, equipped with a lightweight random forest classifier. During the classifier training phase, a dataset is constructed using historical equipment anomaly and normal data, with 100 decision trees used to balance accuracy and computational load. In real-time operation, edge nodes receive preprocessed equipment data, input it into the classifier, and output equipment status categories: normal, minor anomaly, and severe anomaly. When a minor anomaly is identified, a local maintenance alert is immediately generated and pushed to the production line team leader, achieving millisecond-level response and addressing real-time requirements. The cloud platform uses a cloud server cluster to deploy deep reinforcement learning algorithms, such as DQN. The algorithm uses maximizing overall workshop production efficiency as the reward function. The state space includes the status of equipment on each production line, order priorities, and energy consumption data, while the action space includes production task allocation and equipment start / stop strategies. The cloud platform receives aggregated data uploaded by each edge node every hour and iteratively optimizes the scheduling strategy through reinforcement learning. For example, when a piece of equipment on a production line experiences a minor anomaly, its tasks are dynamically allocated to other idle production lines, achieving global resource optimization.

[0056] In this embodiment, the visualization and closed-loop control module integrates a digital twin engine. This engine uses 3D rendering technology to display the equipment operating status, process parameter curves, and abnormal alarm information in real time. Based on the output instructions of the decision module, it dynamically adjusts the control parameters of the actuator through the PID controller to achieve closed-loop optimization of process parameters.

[0057] Specifically, the digital twin engine for the visualization and closed-loop control module is developed based on Unity3D. It updates the digital twin model in real time using data from industrial cameras and equipment sensors. The monitoring interface displays the equipment's appearance and operating posture in 3D, such as the rotation of robotic arm joints. Simultaneously, it generates process parameter curves, such as temperature-time and pressure-time curves. When an abnormality occurs, the corresponding part of the model flashes red and an alarm message pops up, such as "motor temperature too high," facilitating intuitive problem location for maintenance personnel. In the closed-loop control stage, the process parameter adjustment commands output by the decision module are transmitted to the PID controller. The controller adjusts the parameters based on the deviation between the command value and the actual collected value using the PID formula. Calculate the control quantity, where To control the output, This is the proportionality coefficient. For deviation, The integral coefficient is... The differential coefficient is used to drive actuators, such as hydraulic valves, to adjust their parameters. At the same time, the sensor collects the actual value after adjustment in real time and feeds it back to the controller, forming a closed-loop regulation.

[0058] Please see Figure 2 The present invention also provides a digital monitoring method for intelligent manufacturing based on the Industrial Internet of Things, comprising the following steps:

[0059] S1. By connecting to sensors, controllers and production equipment on the manufacturing site through a multi-source heterogeneous data acquisition module, real-time data on equipment status, process parameters, energy consumption levels and production progress are collected.

[0060] S2. The edge intelligent preprocessing module is used to perform protocol parsing, format unification and sampling frequency alignment on the collected data, and to perform abnormal data filtering and missing value imputation based on sliding window.

[0061] S3. Construct a multi-dimensional time series correlation analysis model through dynamic modeling and feature extraction modules to extract the deep coupling features between equipment operating status and process parameters;

[0062] S4. Based on the feature extraction results, the distributed collaborative decision-making module generates equipment maintenance early warning, process parameter adjustment instructions and production scheduling strategies.

[0063] S5. The monitoring results are presented in a graphical interface through the visualization and closed-loop control module, and the actuator is driven to achieve adaptive adjustment of process parameters according to the decision instructions.

[0064] Specifically, this method is applicable to the aforementioned system. Implementation begins by activating the multi-source heterogeneous data acquisition module, connecting its protocol adapter to equipment such as temperature sensors, PLC controllers, and CNC machine tools on the manufacturing site via Ethernet. The acquisition frequency is set to 5–10 Hz to collect real-time data on equipment status, process parameters, energy consumption levels, and production progress, ensuring data coverage of the entire production process and resolving the challenge of multi-source data access. Next, the edge intelligent preprocessing module is activated to parse the collected multi-protocol data, unifying different data formats into a standard format, aligning them to a unified frequency based on equipment sampling frequency differences, and then filtering out abnormal data through a sliding window and imputing missing values ​​using an autoregressive model to improve data quality and provide reliable input for subsequent modeling. Subsequently, the dynamic modeling and feature extraction module constructs a hybrid model, learning equipment topology dependencies and time-series dynamic patterns to extract deep coupling features between equipment and processes, accurately identifying their correlation patterns. After the distributed collaborative decision-making module is activated, edge nodes analyze data in real-time to generate local maintenance warnings, while the cloud platform aggregates data to optimize global scheduling strategies, balancing real-time performance and global optimization. Finally, the visualization and closed-loop control module displays the monitoring results and drives actuators to adjust process parameters according to decision commands, forming a complete closed loop.

[0065] In this embodiment, in step S2, the adaptive sliding window algorithm dynamically adjusts the window length according to the data sampling frequency, the abnormal data filtering adopts a detection mechanism based on local outlier factors, and the missing value imputation adopts a time series autoregressive model for prediction and filling.

[0066] Specifically, in step S2, the adaptive sliding window algorithm first obtains the sampling frequency of the collected data in real time through the frequency detection unit of the edge intelligent preprocessing module. When the sensor sampling frequency is detected to be 8Hz, i.e., 1 data point every 0.125 seconds, the window length is set to 16 data points to ensure that the window covers a data range of 2 seconds. When the sampling frequency drops to 4Hz, i.e., 1 data point every 0.25 seconds, the window length is automatically adjusted to 8 data points to maintain a coverage range of 2 seconds and avoid data processing deviation caused by frequency changes. The abnormal data filtering adopts the local outlier detection mechanism, calculates the local density of each data point in its 6 neighboring data points. If the local density of a data point is significantly lower than that of the neighboring data, it is judged as abnormal data and removed from the data sequence. For example, the instantaneous high value generated by sensor interference can be effectively filtered. The missing value imputation adopts the time series autoregressive model AR(4). By analyzing historical data, the lag order is determined to be 4. The regression equation is constructed using the normal data of the four moments before the missing value, the missing value is predicted and filled into the data sequence.

[0067] In this embodiment, in step S3, a hybrid architecture of graph convolutional networks and long short-term memory networks is used to construct a correlation analysis model. The graph convolutional network models the topological dependencies between devices, while the long short-term memory network captures long-term dynamic patterns of multidimensional time series. The correlation strength between device status and process parameters is calculated using the following formula:

[0068] ;

[0069] in, Indicates the first Each device at time Status indicators Indicates the first Each process parameter at time... The value, The length of the time series. This is the correlation coefficient.

[0070] Specifically, in step S3, a hybrid correlation analysis model combining a graph convolutional network and a long short-term memory network is first constructed. The graph convolutional network part constructs an undirected graph based on the physical connections and signal interactions of the equipment in the manufacturing site. Using equipment as nodes and connections between devices as edges, it learns the topological dependencies of the equipment through adjacency matrices and node feature matrices. For example, CNC machine tools and conveyor belts are considered adjacent nodes to strengthen the correlation weight between their data. The LSTM network part has two hidden layers. Normalized multidimensional time-series data is input, with equipment status and process parameters normalized to [0,1] to ensure dimensional consistency. Forget gates, input gates, and output gates are used to capture the long-term dynamic patterns of the data over 12 hours. The correlation strength between equipment status and process parameters is calculated using the following formula:

[0071] ;

[0072] in This refers to the normalized state index of the i-th device at time t, such as the normalized value of rotational speed. This represents the normalized value of the j-th process parameter at time t, such as the normalized depth of cut. T is the length of the time series, e.g., 7200, corresponding to 2 hours sampled at 1 second. Let be the mean of the status index of the i-th device. Let be the mean of the j-th process parameter. This is the correlation coefficient.

[0073] In this embodiment, in step S4, the distributed collaborative decision-making is jointly executed by the edge nodes and the cloud platform. The edge nodes use a lightweight random forest classifier for real-time anomaly detection and local decision-making, while the cloud platform uses a deep reinforcement learning algorithm for global optimization and long-term policy generation.

[0074] Specifically, during step S4, distributed collaborative decision-making is executed jointly by edge nodes and the cloud platform. Edge nodes are deployed on embedded servers near the production line, equipped with a lightweight random forest classifier. The classifier uses historical normal equipment data, slightly abnormal data, and severely abnormal data as training samples, and after training, it possesses real-time classification capabilities. During operation, the edge nodes receive preprocessed equipment data every 0.5 seconds, input it into the classifier, and quickly output the equipment status result. If a severely abnormality is detected, an equipment shutdown warning is immediately generated and sent to the production line control terminal, achieving local real-time decision-making and meeting millisecond-level response requirements. The cloud platform is deployed on a remote cloud server, equipped with a deep reinforcement learning algorithm. The algorithm uses the lowest overall energy consumption and shortest production cycle as the comprehensive reward function. The state space includes the equipment status, order information, and inventory data uploaded by each edge node, while the action space includes strategies such as production task allocation and equipment energy consumption threshold adjustment. The cloud platform aggregates global data from each edge node every 2 hours and iteratively updates the production scheduling strategy through the reinforcement learning algorithm. For example, when the energy consumption of equipment in a certain area is too high, the operating parameters and task allocation of the equipment in that area are dynamically adjusted to achieve global optimization.

[0075] In summary, this invention addresses the issues of heterogeneous protocols and closed interfaces in the data acquisition layer by supporting parallel parsing of multiple protocols and mapping raw data into standard data frames through the industrial protocol conversion unit of the multi-source heterogeneous data acquisition module. The edge intelligent preprocessing module dynamically adjusts the window length using an adaptive sliding window algorithm, combined with local outlier anomaly filtering and time-series autoregressive missing data imputation, improving data quality and laying the foundation for subsequent analysis. The dynamic modeling and feature extraction module employs a hybrid architecture of GCN and LSTM to mine equipment topology dependencies and multi-dimensional time-series dynamic patterns, calculating equipment and process coupling characteristics through correlation strength formulas, thus solving the problem of static thresholds failing to identify potential anomalies. The distributed collaborative decision-making module combines real-time anomaly detection and local decision-making at edge nodes with global optimization on the cloud platform, addressing the issues of high latency and limited scalability in centralized processing. The visualization and closed-loop control module achieves monitoring and execution linkage through digital twin display and PID closed-loop control, resolving the disconnect between monitoring and decision-making, and comprehensively improving the transparency and responsiveness of the manufacturing system, ensuring stable production and optimizing resource allocation.

[0076] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0077] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A smart manufacturing digital monitoring system based on the Industrial Internet of Things, characterized in that, include: The multi-source heterogeneous data acquisition module is used to connect to sensors, controllers and production equipment in the manufacturing site through an industrial IoT protocol adapter to realize real-time data acquisition of equipment status, process parameters, energy consumption level and production progress. The edge intelligent preprocessing module is used to perform protocol parsing, format unification and sampling frequency alignment on the collected multi-source heterogeneous data, and to perform abnormal data filtering and missing value imputation based on sliding window. The dynamic modeling and feature extraction module is used to construct a multi-dimensional time series correlation analysis model and extract the deep coupling features between equipment operating status and process parameters; The distributed collaborative decision-making module is used to generate equipment maintenance warnings, process parameter adjustment instructions, and production scheduling strategies based on feature extraction results. The visualization and closed-loop control module is used to present the monitoring results in a graphical interface and drive the actuators to achieve adaptive adjustment of process parameters according to decision instructions.

2. The intelligent manufacturing digital monitoring system based on the Industrial Internet of Things as described in claim 1, characterized in that, The multi-source heterogeneous data acquisition module integrates an industrial protocol conversion unit, which supports parallel parsing of ModbusTCP, OPCUA, Profinet and EtherCAT protocols, and maps the raw data into standard data frames with timestamps, device identifiers, data values ​​and quality identifiers through a unified data model.

3. The intelligent manufacturing digital monitoring system based on the Industrial Internet of Things as described in claim 1, characterized in that, The edge intelligent preprocessing module is equipped with an adaptive sliding window algorithm. The window length is dynamically adjusted according to the data sampling frequency. Its abnormal data filtering adopts a detection mechanism based on local outlier factors, and missing value imputation uses a time series autoregressive model for prediction and filling.

4. The intelligent manufacturing digital monitoring system based on the Industrial Internet of Things as described in claim 1, characterized in that, The dynamic modeling and feature extraction module adopts a hybrid architecture of graph convolutional networks and long short-term memory networks. The graph convolutional network is used to model the topological dependencies between devices, while the long short-term memory network is used to capture long-term dynamic patterns in multi-dimensional time series. The dynamic modeling and feature extraction module calculates the correlation strength between device status and process parameters using the following formula: ; in, Indicates the first Each device at time Status indicators Indicates the first Each process parameter at time... The value, The length of the time series. Let be the time series mean of the i-th device status index. Let j be the time series mean of the j-th process parameter. This is the correlation coefficient.

5. The intelligent manufacturing digital monitoring system based on the Industrial Internet of Things according to claim 1, characterized in that, The distributed collaborative decision-making module consists of edge nodes and a cloud platform. The edge nodes are responsible for real-time anomaly detection and local decision-making, while the cloud platform is responsible for global optimization and long-term strategy generation. The edge nodes use a lightweight random forest classifier to classify the device status in real time, and the cloud platform dynamically adjusts the production scheduling strategy through a deep reinforcement learning algorithm.

6. The intelligent manufacturing digital monitoring system based on the Industrial Internet of Things according to claim 1, characterized in that, The visualization and closed-loop control module integrates a digital twin engine, which uses 3D rendering technology to display the equipment operating status, process parameter curves and abnormal alarm information in real time. Based on the output instructions of the decision module, the engine dynamically adjusts the control parameters of the actuator through the PID controller to achieve closed-loop optimization of process parameters.

7. A digital monitoring method for intelligent manufacturing based on the Industrial Internet of Things (IIoT), applicable to any one of claims 1-6, characterized in that, Including the following steps: S1. By connecting to sensors, controllers and production equipment on the manufacturing site through a multi-source heterogeneous data acquisition module, real-time data on equipment status, process parameters, energy consumption levels and production progress are collected. S2. The edge intelligent preprocessing module is used to perform protocol parsing, format unification and sampling frequency alignment on the collected data, and to perform abnormal data filtering and missing value imputation based on sliding window. S3. Construct a multi-dimensional time series correlation analysis model through dynamic modeling and feature extraction modules to extract the deep coupling features between equipment operating status and process parameters; S4. Based on the feature extraction results, the distributed collaborative decision-making module generates equipment maintenance early warning, process parameter adjustment instructions and production scheduling strategies. S5. The monitoring results are presented in a graphical interface through the visualization and closed-loop control module, and the actuator is driven to achieve adaptive adjustment of process parameters according to the decision instructions.

8. The intelligent manufacturing digital monitoring method based on the Industrial Internet of Things according to claim 7, characterized in that, In step S2, the adaptive sliding window algorithm dynamically adjusts the window length according to the data sampling frequency, the abnormal data filtering adopts a detection mechanism based on local outlier factors, and the missing value imputation adopts a time series autoregressive model for prediction and filling.

9. The intelligent manufacturing digital monitoring method based on the Industrial Internet of Things according to claim 7, characterized in that, In step S3, a correlation analysis model is constructed using a hybrid architecture of graph convolutional network and long short-term memory network. The graph convolutional network models the topological dependencies between devices, while the long short-term memory network captures the long-term dynamic patterns of multidimensional time series. The correlation strength between equipment status and process parameters is calculated using the following formula: ; in, Indicates the first Each device at time Status indicators Indicates the first Each process parameter at time... The value, The length of the time series. Let be the time series mean of the i-th device status index. Let j be the time series mean of the j-th process parameter. This is the correlation coefficient.

10. The intelligent manufacturing digital monitoring method based on the Industrial Internet of Things according to claim 7, characterized in that, In step S4, the distributed collaborative decision-making is jointly executed by the edge nodes and the cloud platform. The edge nodes use a lightweight random forest classifier for real-time anomaly detection and local decision-making, while the cloud platform uses a deep reinforcement learning algorithm for global optimization and long-term policy generation.

Images