Data visualization processing method of smart factory

Abstract

The invention discloses a data visualization processing method for a smart factory, and relates to the field of data visualization, and the method comprises the steps: obtaining goods spatio-temporal data, a production plan change instruction, process flow data and equipment operation and maintenance data, and constructing a dynamic association strength matrix; constructing a three-dimensional thermodynamic diagram according to the dynamic association strength matrix, defining cargo nodes, equipment nodes and process nodes, creating a heterogeneous graph, and calculating implicit weights of equipment faults and cargos by adopting a graph neural network model; carrying out heterogeneous data fusion on the dynamic association intensity matrix, the process flow data and the equipment operation and maintenance data to construct a three-dimensional association map; through federated learning driven parameter optimization and pulse warning area marking, the three-dimensional relevance map can automatically reconstruct the relevance network along with the change of a production plan, an emergency scheme (such as dispersed high-risk goods) can be quickly generated after abnormal triggering, and the problem that a static visual architecture is difficult to adapt to dynamic adjustment of a production process is solved.

Description

Technical Field

[0001] The present invention relates to the field of data visualization, and in particular to a data visualization processing method for a smart factory. Background Art

[0002] With the deepening development of Industry 4.0 and intelligent manufacturing concepts, smart factories, as the core of modern manufacturing, have undergone an evolution in data visualization management from basic monitoring to intelligent decision-making. Early factory visualization relied primarily on static charts (such as bar charts and line graphs) and pre-set rule-driven alarm mechanisms. These methods collected equipment status data through discrete sensor nodes and displayed production progress and fault information in a two-dimensional format. While these approaches achieved basic data presentation capabilities, they struggled to meet the real-time integration requirements of heterogeneous multi-source data in complex manufacturing scenarios.

[0003] In recent years, with breakthroughs in the Internet of Things (IoT), artificial intelligence (AI), and 3D visualization technologies, smart factories have gradually introduced dynamic modeling methods based on digital twins. These methods use high-precision 3D rendering technology to map physical spaces into virtual environments and construct multidimensional visualization interfaces in conjunction with real-time data streams. However, visualization still primarily focuses on static topological relationships and lacks in-depth exploration of dynamic correlations within the production process. This leads to delayed identification of key anomalies and rigid resource scheduling strategies. Furthermore, traditional methods typically employ independent modular designs, failing to fully consider the implicit coupling between equipment failures, process changes, and material flow, making it difficult to form a decision support system based on a global optimization perspective. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a data visualization processing method for a smart factory to solve the problems of insufficient dynamic correlation modeling of multi-source heterogeneous data and difficulty in adapting static visualization architecture to dynamic adjustment of production processes.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides a data visualization processing method for a smart factory, which comprises:

[0008] Obtain cargo spatiotemporal data production plan change instructions, process flow data, and equipment operation and maintenance data to build a dynamic correlation strength matrix;

[0009] A three-dimensional heat map is constructed based on the dynamic correlation strength matrix. Goods nodes, equipment nodes, and process nodes are defined, a heterogeneous graph is created, and a graph neural network model is used to calculate the implicit weights of equipment failures and goods.

[0010] The dynamic correlation strength matrix, process flow data and equipment operation and maintenance data are integrated into heterogeneous data to construct a three-dimensional correlation map;

[0011] Mark the pulse warning area in the three-dimensional correlation map and generate a distributed storage path planning map;

[0012] Create a line chart of inventory turnover and a bar chart of production disruptions, and update the parameters of a graph neural network model through federated learning.

[0013] As a preferred solution of the data visualization processing method of the smart factory of the present invention, wherein:

[0014] The graph neural network model is used to calculate the implicit weights of equipment failures and goods. The specific steps are as follows:

[0015] Normalize the cargo nodes, equipment nodes, and process nodes to obtain their normalized features;

[0016] Filter out the neighboring nodes of the cargo node, and concatenate the normalized features of the cargo node with the normalized features of the neighboring nodes to form a merged feature vector;

[0017] The neighbor nodes of the cargo node refer to the equipment nodes and process nodes connected to the cargo node;

[0018] Perform linear transformation on the merged feature vector through a single-layer neural network to obtain the hidden layer feature vector;

[0019] Calculate the similarity score between the cargo node and the neighbor node based on the hidden layer feature vector;

[0020] Normalize the similarity scores between cargo nodes and neighbor nodes into probability distribution to generate attention coefficients;

[0021] The implicit weights of equipment failure and goods are calculated based on the normalized features and attention coefficients of neighboring nodes.

[0022] As a preferred solution of the data visualization processing method for the smart factory described in the present invention, the construction of a three-dimensional correlation map refers to normalizing the dynamic correlation strength matrix, mapping the process steps in the process flow data to the X-axis, mapping the equipment status to the Y-axis, and mapping the normalized values in the dynamic correlation strength matrix to the Z-axis.

[0023] As a preferred solution of the data visualization processing method of the smart factory described in the present invention, the construction of the dynamic association strength matrix refers to calculating the dynamic association strength between goods, and constructing the dynamic association strength matrix based on the dynamic association strength between the goods ID and the goods.

[0024] As a preferred solution of the data visualization processing method for the smart factory described in the present invention, the marking of pulse warning areas in the three-dimensional correlation map refers to defining a risk threshold based on the implicit weights of historical equipment failures and goods, and marking the cargo storage areas where the implicit weights of equipment failures and goods are not lower than the risk threshold as pulse warning areas.

[0025] As a preferred solution of the data visualization processing method for the smart factory of the present invention, the method of updating the graph neural network model parameters by the federated learning algorithm comprises the following specific steps:

[0026] Collect equipment failure data and calculate the false alarm rate of equipment failures, and trigger the update of graph neural network model parameters based on the false alarm rate of equipment failures;

[0027] Based on the equipment failure data, a federated feature extraction layer is used to generate a failure feature vector;

[0028] Add the fault feature vector to the node attribute of the device node and update the heterogeneous graph;

[0029] Perform global average pooling and maximum pooling operations on the hidden layer feature vector to generate channel weights, and use the channel weights to enhance the hidden layer feature vector;

[0030] Perform global average pooling and global maximum pooling on the merged feature vector, and use spatial convolution kernel to enhance the merged feature vector;

[0031] Based on the enhanced hidden layer feature vector and the merged feature vector, backpropagation is used to update the parameters of the fully connected layer and attention mechanism of the graph neural network model;

[0032] Replace the symmetric convolution of the graph neural network model with an asymmetric convolution combination;

[0033] The weight matrix and bias vector of the graph neural network model are updated by calculating their average values.

[0034] As a preferred solution of the data visualization processing method of the smart factory of the present invention, wherein:

[0035] The specific steps of creating a heterogeneous graph are as follows:

[0036] Based on the cargo nodes, equipment nodes and process nodes, establish directed edges from cargo to equipment, cargo to process and process to equipment;

[0037] The edge weights of the goods-equipment edge, the edge weights of the goods-process edge, and the edge weights of the process-equipment edge are calculated to form a heterogeneous graph.

[0038] As a preferred solution of the data visualization processing method of the smart factory described in the present invention, the generation of a distributed storage path planning map includes distributing the goods in the pulse warning area, calculating the shortest path of the AGV, and generating a distributed storage path planning map through augmented reality operations.

[0039] In a second aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the data visualization processing method for a smart factory as described in the first aspect of the present invention is implemented.

[0040] In a third aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the data visualization processing method for a smart factory as described in the first aspect of the present invention.

[0041] The beneficial effects of the present invention are as follows: by constructing a dynamic correlation strength matrix and a three-dimensional heat map, the implicit coupling relationship between cargo flow and equipment status (such as the impact of equipment failure on inventory turnover) can be captured, which solves the problem of insufficient dynamic correlation modeling of multi-source heterogeneous data, makes the implicit correlation between equipment failure and cargo flow explicit, and provides a decision-making basis for resource scheduling; through federated learning-driven parameter optimization and pulse warning area marking, the three-dimensional correlation map can automatically reconstruct the correlation network as the production plan changes, breaking through the limitations of the static architecture, and can quickly generate emergency plans (such as dispersing high-risk goods) after an anomaly is triggered, shortening the response time, improving the efficiency of resource scheduling, and solving the problem that the static visualization architecture is difficult to adapt to the dynamic adjustment of the production process. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0043] Figure 1 Flowchart of the data visualization processing method for smart factories.

[0044] Figure 2 Schematic diagram of the three-dimensional correlation map structure.

[0045] Figure 3 Schematic diagram of heterogeneous graph structure.

[0046] Figure 4 Generate a schematic diagram for the dynamic association strength matrix between cargoes. DETAILED DESCRIPTION

[0047] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0048] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0049] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0050] Reference Figures 1 to 4 , is an embodiment of the present invention, which provides a data visualization processing method for a smart factory, comprising the following steps:

[0051] S1: Obtain cargo spatiotemporal data, production plan change instructions, process flow data, and equipment operation and maintenance data to build a dynamic correlation strength matrix.

[0052] The specific steps are as follows:

[0053] RFID reader arrays are evenly arranged in the warehouse to obtain the storage time of goods. Temperature and humidity sensors are deployed along the AGV path to obtain ambient temperature and humidity values. UWB positioning tags are installed in the cargo storage area to obtain the storage coordinates of the goods.

[0054] It should also be noted that by collaboratively collecting spatiotemporal data of goods through multi-source heterogeneous sensors, we can achieve full-dimensional perception of the dynamic state of the warehouse, solve the problem of incomplete spatial correlation characterization caused by relying on a single data source, and provide high-precision basic data support for the subsequent dynamic correlation intensity matrix.

[0055] Obtain production plan change instructions (order ID, process priority, estimated delivery time) through the enterprise resource planning (ERP), and obtain equipment operation and maintenance data (fault code, maintenance start timestamp, maintenance end timestamp and equipment status) and process flow data (process steps, process sequence, process time and production batch records) from the manufacturing execution unit (MES).

[0056] It should also be noted that the integration of ERP and MES order priorities, process flows, and equipment failure data enables the dynamic correlation strength matrix to reflect the coupling relationship between production plan changes and equipment status, breaking through the limitation of static label classification that cannot capture cross-domain dynamic features and improving the comprehensiveness of risk prediction.

[0057] Calculate the time difference between the goods' warehousing time and the expected delivery time, use linear regression to analyze the correlation between the number of days of historical order delays and the degree of production loss, obtain the decay rate, and use the exponential decay function to simulate the decay process and calculate the goods' time-space factor coefficient. The longer the time (the larger the time difference), the smaller the goods' time-space factor coefficient.

[0058] Measure the Euclidean distance (straight-line distance) between all cargo storage locations in the warehouse, divide the actual distance by the total length of the warehouse diagonal, and convert it into a proportional value between 0 and 1 to obtain the spatial correlation weight. The closer the distance between the goods, the higher the spatial correlation weight.

[0059] It should also be noted that the exponential decay function is used to quantify the time value (the time difference between entering and leaving the warehouse) and the spatial proximity (normalized value of the Euclidean distance) of the goods, which solves the problem of ignoring the impact of time dynamics and spatial density on risk prediction and provides a quantitative basis for dynamic correlation strength modeling.

[0060] Extract the process execution sequence and collaborative frequency (the frequency of coordination, synchronization or resource interaction between multiple processes or departments to achieve common production goals) from historical production batch records, and use association rule mining (such as prior algorithms) to quantify the process dependency strength (the closeness or importance of the mutual dependence relationship between processes). The greater the process dependency strength, the closer the process dependency relationship.

[0061] The difference in the arrangement position of the current goods and the target goods in the production process and the process dependence intensity are multiplied as the denominator, and the spatiotemporal factor coefficient of the goods is multiplied by the spatial correlation weight and then added to 1 as the denominator to calculate the dynamic correlation intensity between the goods.

[0062] It should also be noted that: based on association rule mining (Apriori algorithm), the frequency of collaborative occurrence between processes is quantified. Combined with the difference in the arrangement position of the current goods and the target goods in the production process and the spatiotemporal factor coefficient, the implicit dependency relationship of the processes is explicitly modeled, providing risk association analysis capabilities across process dimensions for three-dimensional heat maps and heterogeneous maps.

[0063] Construct a dynamic association strength matrix: rows and columns are arranged in order of cargo ID to form an N×N matrix (N is the total number of cargoes). The matrix element value is the dynamic association strength between cargoes in rows and columns. The dynamic association strength of the same cargo is set to 0 (meaningless).

[0064] It should also be noted that by integrating RFID, UWB positioning, and environmental sensor data, a comprehensive depiction of the spatial distribution and temporal flow of goods is achieved, providing a dynamic correlation foundation for subsequent three-dimensional heat maps. The dynamic correlation strength matrix quantifies the temporal decay factors and spatial distance weights between goods, addressing cross-domain coupling relationships that traditional tag classification cannot capture (such as the impact of equipment failures on inventory turnover), providing a basis for multi-dimensional dynamic mapping for visualization.

[0065] S2: Construct a three-dimensional heat map based on the dynamic correlation strength matrix, define cargo nodes, equipment nodes, and process nodes, create a heterogeneous graph, and use a graph neural network model to calculate the implicit weights of equipment failures and cargo.

[0066] The specific steps are as follows:

[0067] The dynamic correlation strength matrix is normalized using the minimum-maximum normalization method to map the dynamic correlation strength between goods to the [0,1] interval.

[0068] The normalized dynamic correlation intensity matrix is mapped onto the surface of the 3D warehouse model. The normalized values of the dynamic correlation intensity matrix are mapped to a red-blue gradient, with 0 corresponding to pure blue (RGB: 0, 0, 255) and 1 corresponding to pure red (RGB: 255, 0, 0). Intermediate values are linearly interpolated to create a gradient transition by blending the red and blue channels. A dynamic light spot is generated at the cargo coordinates, with its movement speed proportional to the square of the normalized value of the dynamic correlation intensity matrix. For example, a normalized value of 0.6 for the dynamic light spot results in a speed of 0.5 m / s, while a normalized value of 0.9 for the dynamic light spot results in a speed of 1.125 m / s. The 3D heat map updates the dynamic light spot position in real time.

[0069] It should also be noted that by linking the color gradient (red-blue) with the light spot speed, the correlation intensity is converted into a dynamic visual signal perceptible to the human eye, which improves the recognition efficiency of highly correlated areas (such as areas with a high incidence of equipment failures) and enhances the operator's intuitive judgment ability.

[0070] The equipment unique identifier, equipment type, historical failure times, mean time between failures, and the time of the most recent maintenance (the difference between the maintenance end timestamp and the maintenance start timestamp of the most recent maintenance) are defined as equipment nodes; the cargo unique identifier, storage coordinates, storage time, expected usage time (the difference between the expected outbound time and the storage time), and current dynamic association strength are defined as cargo nodes; the process number, process time, process priority, and associated equipment list are defined as process nodes.

[0071] Connect the device node and the cargo node, establish a directed edge from cargo to device, record the total number of times the target device is used (the cumulative number of times all cargo use the target device) and the number of times cargo uses the target device, calculate the device usage frequency weight (the number of times cargo uses the target device divided by the total number of times the target device is used) as the edge weight of the cargo-device edge. If the total number of times the target device is used or the number of times cargo uses the target device is zero, force the edge weight of the cargo-device edge to be zero.

[0072] Connect the goods node and the process node, establish a directed edge from the goods to the process, count the total number of runs of the target process (the cumulative number of times all goods pass through the target process) and the number of times the goods pass through the target process, and calculate the process usage frequency weight (the number of times the goods pass through the target process divided by the total number of runs of the target process) as the edge weight of the goods-process edge. If the total number of runs of the target process or the number of times the goods pass through the target process is zero, force the edge weight of the goods-process edge to be zero.

[0073] Connect the process nodes and equipment nodes, establish a directed edge from the process to the equipment, count the number of times the equipment is used in the target process and the total number of times the equipment is used, calculate the process equipment usage frequency weight (the number of times the equipment is used in the target process divided by the total number of times the equipment is used) as the edge weight of the process-equipment edge. If the number of times the equipment is used in the target process or the total number of times the equipment is used is zero, then force the edge weight of the process-equipment edge to be zero.

[0074] Create a device node for each device, a cargo node for each cargo, and a process node for each process; connect the device node and cargo node, cargo node and process node, process node and device node, establish directed edges from cargo to equipment, cargo to process, and process to equipment, and calculate the edge weights of the cargo-equipment edge, the cargo-process edge, and the process-equipment edge to form a heterogeneous graph.

[0075] It should also be noted that by dynamically calculating edge weights based on usage frequency statistics (equipment / process / goods), heterogeneous graphs can adapt to changes in production processes, solving the problem that static topology graphs cannot respond to real-time task changes and improving the adaptability of graph neural networks to dynamic production scenarios.

[0076] Divide the historical number of failures of the device by the highest historical number of failures among all devices to obtain the normalized value of the historical number of failures of the device; divide the average failure interval of the device by the maximum failure interval among all devices to obtain the normalized value of the average failure interval of the device; assign a unique binary code of fixed length according to the device type, for example, the injection molding machine is coded as 001, the welding machine is coded as 010, and the cutting machine is coded as 100; divide the value of the storage coordinate by the maximum value of the corresponding dimension of the warehouse to obtain the normalized storage coordinate; divide the time difference between the goods' entry time and the expected exit time by the historical maximum storage time to obtain the normalized value of the time difference between the goods' entry time and the expected exit time; divide the process time by the historical maximum process time to obtain the normalized value of the process time.

[0077] For each cargo node, according to the connection relationship between cargo-equipment edges and cargo-process edges, all equipment nodes and process nodes connected to the cargo node are screened as neighbor nodes of the cargo node.

[0078] The normalized storage coordinates, the normalized value of the time difference between the goods' entry time and the expected exit time, and the dynamic association strength between goods are defined as the normalized features of the goods node. The normalized value of the equipment's historical failure count, the normalized value of the equipment's mean time between failures, and the binary code are defined as the normalized features of the equipment node. The normalized value of the process time and the process priority are defined as the normalized features of the process node. The normalized features of the goods node and the normalized features of the neighboring nodes are spliced by dimension to form a merged feature vector. For example, the spliced feature vector of goods ID-205 and equipment ID-07 contains 7 dimensions (3 dimensions for goods + 3 dimensions for equipment + 1 dimension for edge weight).

[0079] The merged feature vector is input into a single-layer neural network, and a linear transformation is performed on the merged feature vector (first multiplied by the weight matrix and then added with the bias vector) to generate a hidden layer feature vector. The LeakyReLU activation function is applied to the hidden layer feature vector (the negative slope is usually fixed at 0.2) to obtain the similarity score between the cargo node and the neighboring node, which is used to measure the importance of the neighboring node to the current cargo node; the Softmax function is used to normalize the similarity score between the cargo node and the neighboring node into a probability distribution to generate the attention coefficient.

[0080] It should also be noted that by generating the attention coefficient through neighbor node feature concatenation and Softmax normalization, the hidden layer feature vector focuses on key related factors (such as the impact of high-failure equipment on cargo risks), improving the pertinence and accuracy of the implicit weight calculation.

[0081] The normalized features of each neighbor node are multiplied by the corresponding attention coefficient, and then added together to obtain the hidden features of the current cargo node. The Sigmoid function is used to compress the value of the hidden features of the current cargo node to the range of 0 to 1 to obtain the implicit weight of equipment failure and cargo.

[0082] It should also be noted that the three-dimensional heat map, by linking red-blue gradients with dynamic spot speed, intuitively displays the spatiotemporal evolution of the dynamic correlation strength between goods. Compared to traditional two-dimensional heat maps, this enhances the sensitivity of abnormal areas. The normalized feature concatenation and attention mechanism of heterogeneous graph nodes (goods / equipment / process) enable visualization of the implicit weights between equipment failures and goods, overcoming the limitations of static topological models in identifying implicit correlations and providing quantitative support for early warning of high-risk areas.

[0083] S3: Heterogeneous data fusion of dynamic correlation intensity matrix, process flow data and equipment operation and maintenance data to construct a three-dimensional correlation map.

[0084] The specific steps are as follows:

[0085] In a three-dimensional Cartesian coordinate system, the X-axis is based on the topological structure of the process chain. The process nodes are arranged from left to right according to the execution order in the process flow data. Each process node corresponds to a process number. The entity of the process node is a sphere, and the radius of the process node is proportional to the process duration (for example, the radius of the process node whose process duration exceeds 150% of the average value is enlarged to 1.5 times);

[0086] The Y-axis is based on the device status. Device nodes are arranged from front to back based on their physical location. The entity of the device node is a red cube with a fixed side length and a red-yellow-green gradient fill. The device failure frequency is calculated by dividing the number of device failures in a specific time period by the total device operating time. The failure threshold is set based on the statistical distribution of historical failure frequencies (for example, the 95% quantile or the mean of the historical failure frequencies plus 2 times the standard deviation). When the device failure frequency exceeds the failure threshold, a red flashing alarm is triggered.

[0087] The Z-axis displays the dynamic correlation strength between goods in the form of a three-dimensional bar chart. The cargo nodes are arranged from bottom to top according to the storage coordinate height. The entity of the cargo node is a three-dimensional bar chart. The column height of the three-dimensional bar chart is directly related to the normalized value in the dynamic correlation strength matrix. For example, a normalized value of 0.9 corresponds to a column height of 1.8 meters.

[0088] When the process flow data conflicts with the equipment status (for example, the process requires the use of faulty equipment), the process node is displayed in gray and a warning icon (such as an exclamation mark) is superimposed, indicating that the process node cannot be used; when the cargo node and the equipment node are associated with a conflict (for example, the cargo is associated with faulty equipment), a red flashing border is superimposed on the surface of the column of the three-dimensional bar chart, indicating a "high-risk association".

[0089] It's also worth noting that the three-dimensional correlation map provides a three-dimensional perspective on the entire production process lifecycle through the collaborative modeling of the process chain on the X-axis, the equipment state gradient on the Y-axis, and the dynamic correlation strength column on the Z-axis. When process-equipment conflicts or high-risk goods-equipment correlations occur, a gray warning icon with a flashing red border overlay prompts operators more efficiently respond to potential bottlenecks, resolving the difficulty of traditional static topology maps in dynamically adapting to production changes.

[0090] S4: Mark the pulse warning area in the three-dimensional correlation map and generate a distributed storage path planning map.

[0091] The specific steps are as follows:

[0092] The normalized value of the dynamic correlation intensity matrix is multiplied by the implicit weight of equipment failure and cargo, and then multiplied by (1 plus the product of the equipment failure impact coefficient and the equipment failure frequency) as the numerator. The sum of the spatial correlation weight and the edge weight of the cargo-equipment edge is used as the denominator to calculate the cargo risk index. The equipment failure impact coefficient is usually 1.5, which is used to amplify the impact of the equipment failure frequency on the cargo risk index.

[0093] Colors are superimposed on the surface of the columns in the 3D histogram of the 3D correlation map, and the column height corresponds to the dynamic correlation strength between the goods. The risk threshold and pulse frequency threshold are divided by the statistical distribution of the implicit weights of historical equipment failures and goods. For example, the pulse frequency threshold of 90% corresponds to 0.8, indicating that the implicit weight of 90% of equipment failures and goods is ≤0.8. The risk threshold of 75% corresponds to 0.6, indicating that the implicit weight of 75% of equipment failures and goods is ≤0.6.

[0094] In the three-dimensional correlation map, the cargo storage area where the implicit weight of equipment failure and cargo is greater than or equal to the risk threshold is marked as a pulse warning area, and the pulse warning area is projected as a translucent cube. When the implicit weight of equipment failure and cargo is greater than or equal to the risk threshold, a red pulse ripple is triggered. When the implicit weight of equipment failure and cargo is greater than or equal to the risk threshold and less than or equal to the pulse frequency threshold, a red pulse ripple is triggered once at a fixed time (such as 1 second). When the implicit weight of equipment failure and cargo is greater than or equal to the pulse frequency threshold, three red pulse ripples are triggered at a fixed time (such as 1 second).

[0095] It should also be noted that through pulse frequency grading (fixed time triggering 1 time / 3 times) combined with green cube marking of target storage areas, differentiated visualization of risk levels is achieved, guiding AGV to quickly complete the decentralized transfer of high-risk goods, reducing the risks brought by centralized storage.

[0096] The goods in the pulse warning area are stored in a dispersed manner, and the area used for dispersed storage of goods is used as the target dispersed storage area. The target dispersed storage area is marked as a green cube. The Dijkstra algorithm is used to calculate the shortest path from the current position of the AGV to the target dispersed storage area, avoiding the coordinates of the faulty equipment and the pulse warning area; the task queue is sorted according to the implicit weight of equipment failure and goods, and tasks are executed according to the task queue. Obstacles (such as people and piles) are detected in real time through lidar, and the path is dynamically adjusted (for example, detouring 2 meters).

[0097] Generate a distributed storage path planning map through AR (augmented reality): map the real-time position of the AGV to a light ball, and the moving speed of the light ball is synchronized with the actual speed of the AGV. For example, when the AGV speed is 1 meter / second, the light ball moves 0.016 meters per frame (60 frames / second); the path is displayed as a guide line, and when there is a path conflict (an obstacle is detected and a detour is required), the color of the guide line changes; map tasks to blocks, and sort tasks by the implicit weight of equipment failure and goods. Unfinished tasks are displayed in red, yellow and green blocks on the right panel of the AR interface, and completed tasks are displayed in gray blocks. For example, a task with an implicit weight of 0.8 for equipment failure and goods is displayed in red, yellow and green blocks, a task with an implicit weight of 0.5 for equipment failure and goods is displayed in yellow, and a task with an implicit weight of 0.3 for equipment failure and goods is displayed in green.

[0098] It's also worth noting that the interplay of translucent cubes and red ripples in the pulse warning area, combined with the AGV shortest path algorithm and AR guidance lines, enables closed-loop visual control from risk warning to strategy execution. The AR interface simultaneously displays task priorities (red / yellow / green blocks) and real-time obstacle avoidance dynamics, reducing the need for manual intervention and making distributed storage path planning adaptive, meeting the demands for rapid decision-making in dynamic production scenarios.

[0099] S5: Create a line chart of inventory turnover and a bar chart of production disruptions, and update the parameters of the graph neural network model through federated learning.

[0100] The specific steps are as follows:

[0101] Calculate the improvement in inventory turnover: Calculate the inventory turnover rate before decentralized storage (goods shipped divided by average inventory) and the inventory turnover rate after decentralized storage. Subtract the inventory turnover rate before decentralized storage from the inventory turnover rate after decentralized storage, and then divide the result by the inventory turnover rate before decentralized storage to obtain the improvement in inventory turnover.

[0102] Create an inventory turnover line chart with time as the X-axis and inventory turnover rate improvement as the Y-axis. Data points are marked as blue dots and the line color is green. When the inventory turnover rate improvement is lower than the average of the historical inventory turnover rate improvement, the corresponding data point is displayed in red and a warning icon (such as a triangle) is superimposed.

[0103] Extract the timestamp of the fault event from the equipment operation and maintenance data, and count the number of production interruptions caused by equipment failure within a unit time (for example, per month) before and after distributed storage.

[0104] A production interruption bar chart is created with unit time as the X-axis and the number of production interruptions due to equipment failure as the Y-axis. The data column before decentralized storage is red, and the data column after decentralized storage is green. The top of the data column shows the number of production interruptions due to equipment failure.

[0105] It should also be noted that the red / green column comparison (data before and after dispersion) intuitively demonstrates the optimization effect of the decentralized storage strategy on inventory turnover and production interruption rate due to equipment failure, providing management with a visual basis for operational improvement and enhancing the scientific nature of decision-making.

[0106] Collect equipment failure data, including equipment vibration frequency abnormal waveform data, temperature sensor timing data, downtime series data, and the number of goods where the implicit weight of equipment failure and goods is greater than or equal to the risk threshold but no equipment failure has occurred. Divide the number of misjudged goods by the number of goods where the implicit weight of equipment failure and goods is greater than or equal to the risk threshold, calculate the false alarm rate of equipment failure, set the false alarm threshold based on historical false alarm data, and trigger the update of the graph neural network model parameters if the false alarm rate of equipment failure exceeds the false alarm threshold.

[0107] Based on equipment fault data (e.g., ), a federated feature extraction layer is used to generate fault feature vectors. Fault feature vectors of low-frequency fault types (e.g., gearbox tooth breakage, shafting offset) are oversampled to ensure that the weight of tail category features is increased. The fault feature vectors are added to the node attributes of the equipment node and the heterogeneous graph is updated.

[0108] Perform global average pooling and global maximum pooling operations on the hidden layer feature vector of the ResNet network, input the results of global average pooling and global maximum pooling into the fully connected layer, use an activation function (such as ReLU) for nonlinear transformation, add the results of global average pooling and global maximum pooling after nonlinear transformation, use Sigmoid activation function to generate channel weights, multiply the channel weights by the hidden layer feature vector element by element, and enhance the feature response of the hidden layer feature vector;

[0109] Perform global average pooling and global maximum pooling on the merged feature vector to generate a global average pooling feature map and a global maximum pooling feature map. After concatenating the global average pooling feature map and the global maximum pooling feature map, perform feature fusion through a spatial convolution kernel to generate a two-dimensional spatial mask. Use the Sigmoid activation function to normalize the two-dimensional spatial mask. Multiply the normalized two-dimensional spatial mask by the merged feature vector element-by-element to enhance the feature response of the spatial region of the merged feature vector (such as the mutation point of the device vibration waveform).

[0110] The enhanced hidden layer feature vector and the merged feature vector are input into the fully connected layer and attention mechanism of the graph neural network model to generate the predicted output; the loss gradient is calculated based on the loss function, and the loss gradient is transferred to the weights and biases of the fully connected layer and the parameters of the attention mechanism using backpropagation. The parameters are adjusted according to the direction of the loss gradient using the optimizer;

[0111] Symmetric convolution is replaced with an asymmetric convolution combination. For example, vertical convolution uses a 5×1 convolution kernel sliding in the vertical direction to capture the timing trend of the equipment fault waveform. Horizontal convolution uses a 1×5 convolution kernel sliding in the horizontal direction to capture spatial distribution anomalies. The results of the vertical and horizontal convolutions are added to form an asymmetric convolution feature map. The asymmetric convolution kernel parameters are constrained by L2 regularization to prevent overfitting caused by excessive elements in the weight matrix.

[0112] The classification loss function is minimized using the margin calibration algorithm, and the classification margin is optimized using gradient descent. The weight matrix and bias vector of the graph neural network model are converted into ciphertext format using an encryption algorithm and uploaded to the cloud server. After the cloud server decrypts the encrypted parameters of each factory, it calculates the arithmetic mean of the weight matrices of all factories to generate a global weight matrix, and calculates the arithmetic mean of the bias vectors to generate a global bias vector.

[0113] The global weight matrix and global bias vector are re-encrypted and sent to each factory. The updated weight matrix and updated bias vector are obtained by weighted averaging the global weight matrix and global bias vector. The expression is:

[0114] U=m·V+(1-m)·W;

[0115] d=m·c+(1-m)·b;

[0116] Among them, U is the updated weight matrix, m is the global weight coefficient, V is the global weight matrix, W is the weight matrix, d is the updated bias vector, c is the global bias vector, and b is the bias vector.

[0117] Replace the weight matrix and bias vector of the graph neural network with the updated weight matrix and updated bias vector.

[0118] It's also worth noting that the comparison of the inventory turnover line chart and the production interruption bar chart (red and green bars) clearly demonstrates the effectiveness of the decentralized storage strategy in improving operational efficiency. The false alarm rate-driven federated learning parameter update mechanism ensures that the graph neural network model can continuously adapt to new failure modes. The combination of asymmetric convolutions and encrypted federated parameter updates enhances the model's spatial-temporal feature extraction capabilities while ensuring data privacy in multi-factory collaboration, providing a sustainable technical framework for long-term dynamic optimization.

[0119] This embodiment also provides a computer device suitable for the data visualization processing method of a smart factory, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the data visualization processing method of a smart factory proposed in the above embodiment.

[0120] The computer device may be a terminal, comprising a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner may be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a button, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse.

[0121] This embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the data visualization processing method for a smart factory proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, disk or optical disk.

[0122] In summary, the present invention solves the problem of insufficient dynamic correlation modeling of multi-source heterogeneous data by: constructing a dynamic correlation strength matrix and a three-dimensional heat map, which can capture the implicit coupling relationship between cargo flow and equipment status (such as the impact of equipment failure on inventory turnover), making the implicit correlation between equipment failure and cargo flow explicit, and providing a decision-making basis for resource scheduling; through federated learning-driven parameter optimization and pulse warning area marking, the three-dimensional correlation map can automatically reconstruct the correlation network as the production plan changes, breaking through the limitations of the static architecture, and can quickly generate emergency plans (such as dispersing high-risk goods) after an anomaly is triggered, shortening the response time, improving the efficiency of resource scheduling, and solving the problem that the static visualization architecture is difficult to adapt to the dynamic adjustment of the production process.

[0123] 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 the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A data visualization processing method for a smart factory, characterized by: include, Obtain cargo spatiotemporal data, production plan change instructions, process flow data, and equipment operation and maintenance data to construct a dynamic correlation strength matrix; A three-dimensional heat map is constructed based on the dynamic correlation strength matrix. Goods nodes, equipment nodes, and process nodes are defined, a heterogeneous graph is created, and a graph neural network model is used to calculate the implicit weights of equipment failures and goods. The dynamic correlation strength matrix, process flow data and equipment operation and maintenance data are integrated into heterogeneous data to construct a three-dimensional correlation map; Mark the pulse warning area in the three-dimensional correlation map and generate a distributed storage path planning map; Create a line chart of inventory turnover and a bar chart of production disruptions, and update the parameters of a graph neural network model through federated learning.

2. The data visualization processing method for a smart factory according to claim 1, characterized in that: The graph neural network model is used to calculate the implicit weights of equipment failures and goods. The specific steps are as follows: Normalize the cargo nodes, equipment nodes, and process nodes to obtain their normalized features; Filter out the neighboring nodes of the cargo node, and concatenate the normalized features of the cargo node with the normalized features of the neighboring nodes to form a merged feature vector; The neighbor nodes of the cargo node refer to the equipment nodes and process nodes connected to the cargo node; Perform linear transformation on the merged feature vector through a single-layer neural network to obtain the hidden layer feature vector; Calculate the similarity score between the cargo node and the neighbor node based on the hidden layer feature vector; Normalize the similarity scores between cargo nodes and neighbor nodes into probability distribution to generate attention coefficients; The implicit weights of equipment failure and goods are calculated based on the normalized features and attention coefficients of neighboring nodes.

3. The data visualization processing method for a smart factory according to claim 1, characterized in that: The construction of the three-dimensional correlation map refers to normalizing the dynamic correlation strength matrix, mapping the process steps in the process flow data to the X-axis, mapping the equipment status to the Y-axis, and mapping the normalized values in the dynamic correlation strength matrix to the Z-axis.

4. The data visualization processing method for a smart factory according to claim 1, characterized in that: The constructing of the dynamic association strength matrix refers to calculating the dynamic association strength between goods, and constructing the dynamic association strength matrix according to the goods ID and the dynamic association strength between the goods.

5. The data visualization processing method for a smart factory according to claim 1, characterized in that: Marking the pulse warning area in the three-dimensional correlation map means defining a risk threshold based on the implicit weight of historical equipment failures and goods, and marking the cargo storage area where the implicit weight of equipment failures and goods is not less than the risk threshold as a pulse warning area.

6. The data visualization processing method for a smart factory according to claims 1 and 2, characterized in that: The specific steps of updating the graph neural network model parameters through the federated learning algorithm are as follows: Collect equipment failure data and calculate the false alarm rate of equipment failures, and trigger the update of graph neural network model parameters based on the false alarm rate of equipment failures; Based on the equipment failure data, a federated feature extraction layer is used to generate a failure feature vector; Add the fault feature vector to the node attribute of the device node and update the heterogeneous graph; Perform global average pooling and maximum pooling operations on the hidden layer feature vector to generate channel weights, and use the channel weights to enhance the hidden layer feature vector; Perform global average pooling and global maximum pooling on the merged feature vector, and use spatial convolution kernel to enhance the merged feature vector; Based on the enhanced hidden layer feature vector and the merged feature vector, backpropagation is used to update the parameters of the fully connected layer and attention mechanism of the graph neural network model; Replace the symmetric convolution of the graph neural network model with an asymmetric convolution combination; The weight matrix and bias vector of the graph neural network model are updated by calculating their average values.

7. The data visualization processing method for a smart factory according to claim 2, characterized in that: The specific steps of creating a heterogeneous graph are as follows: Based on the cargo nodes, equipment nodes and process nodes, establish directed edges from cargo to equipment, cargo to process and process to equipment; The edge weights of the goods-equipment edge, the edge weights of the goods-process edge, and the edge weights of the process-equipment edge are calculated to form a heterogeneous graph.

8. The data visualization processing method for a smart factory according to claim 5, characterized in that: The generating of the distributed storage path planning map includes distributing the storage of goods in the pulse warning area, calculating the shortest path of the AGV, and generating the distributed storage path planning map through augmented reality operation.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the data visualization processing method for a smart factory described in any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the data visualization processing method for a smart factory described in any one of claims 1 to 8 are implemented.

Images