Distributed manufacturing data acquisition system based on NB-IoT

Abstract

The invention relates to the technical field of manufacturing data acquisition, and discloses a distributed manufacturing data acquisition system based on NB-IoT. The system comprises a heterogeneous node sensing layer, a narrowband signaling scheduling layer, an edge fusion processing layer, a cloud collaborative decision-making layer and a closed-loop optimization feedback layer. The heterogeneous node sensing layer obtains multi-source data through a sensor array, generates an equipment operation state thermodynamic diagram and divides region priorities; the narrowband signaling scheduling layer deploys NB-IoT nodes, transmits time division multiplexing signaling and generates a node load distribution diagram; the edge fusion processing layer synchronizes the clock, fuses the multi-source data and the load distribution map, and generates fusion confidence; the cloud collaborative decision-making layer converts the fusion confidence into an executable instruction and issues the executable instruction; and the closed-loop optimization feedback layer monitors the load change of the system, generates an optimization index and dynamically optimizes an acquisition strategy. The system can penetrate through network congestion, improves the real-time performance and accuracy of data acquisition, dynamically adapts to the change of a manufacturing environment, and is suitable for a distributed manufacturing scene.

Description

Technical Field

[0001] This invention relates to the field of manufacturing data acquisition technology, specifically to a distributed manufacturing data acquisition system based on NB-IoT. Background Technology

[0002] In the process of modern manufacturing's transformation towards intelligent manufacturing, data acquisition in distributed manufacturing scenarios faces numerous challenges. Traditional manufacturing data acquisition methods largely rely on wired networks or short-range wireless communication technologies, which reveal significant limitations in complex industrial environments. Wired networks are costly to install and lack flexibility when equipment is moved or its layout adjusted, making them difficult to adapt to the dynamic changes in distributed production units. Short-range wireless communication technologies, limited by transmission distance and interference resistance, often experience signal instability and data transmission interruptions in large manufacturing workshops or scenarios involving multiple devices working together, significantly compromising the integrity and timeliness of the acquired data. With the intelligent upgrading of manufacturing equipment, the data streams generated during the production process are experiencing explosive growth, making the fusion and processing of multi-source heterogeneous data a major challenge. Different types of sensors output data with significantly different formats and sampling frequencies. Traditional data acquisition systems lack effective fusion mechanisms and often can only perform simple processing on single types of data, failing to achieve deep correlation analysis of multi-source data. This makes it difficult for management to fully grasp the production status, thus affecting the accuracy and timeliness of decision-making. In terms of network communication, distributed manufacturing environments are characterized by dense equipment, complex signals, and frequent network congestion. Traditional communication scheduling mechanisms, which use fixed time slot allocation, cannot dynamically adjust according to real-time network load, resulting in significant fluctuations in signaling response delays. Furthermore, data from some nodes may fail to upload due to timeouts, leading to data loss. Simultaneously, the lack of an effective clock synchronization mechanism causes discrepancies in the time dimension of data collected by different nodes, reducing spatiotemporal consistency and posing challenges to subsequent data analysis and applications. There is a significant disconnect between the decision-making and execution phases of existing data acquisition systems. Command transmission between the cloud platform and edge devices largely relies on traditional protocols, which lack real-time performance and effective feedback mechanisms. When system load changes, the acquisition strategy cannot be adjusted in a timely manner, leading to decreased acquisition efficiency and even impacting the stability of the production process. In terms of data processing, traditional systems largely rely on centralized cloud processing. Uploading large amounts of raw data not only consumes valuable network bandwidth but also increases the computing pressure on the cloud, leading to increased data processing latency. Furthermore, cloud processing demands extremely high network stability; a network failure can paralyze the entire data acquisition system. While edge computing technology has alleviated this problem to some extent, achieving efficient collaboration between the edge and the cloud, as well as data fusion and sharing among edge nodes, remains a significant technical challenge. As the manufacturing industry increasingly demands real-time, accurate, and reliable data acquisition, existing data acquisition systems are struggling to meet practical needs. Therefore, developing a system that can adapt to distributed manufacturing environments and possesses efficient data acquisition, fusion processing, and dynamic optimization capabilities has become a crucial issue to be addressed in the current intelligent development of the manufacturing industry. Summary of the Invention

[0003] The purpose of this invention is to provide a distributed manufacturing data acquisition system based on NB-IoT to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides a distributed manufacturing data acquisition system based on NB-IoT, the system comprising: The heterogeneous node perception layer acquires multi-source perception data through a sensor array, applies an edge filtering network to the multi-source perception data, generates a heat map of the device's operating status, and divides the heat map of the device's operating status into regions based on priority. The narrowband signaling scheduling layer deploys NB-IoT communication nodes according to the priority of the divided areas, transmits time-division multiplexed signaling to penetrate network congestion, calculates signaling response delay through time slot allocation algorithm, and generates node load distribution map; The edge fusion processing layer establishes a spatiotemporal mapping between sensing and signaling, uses the NTP protocol to synchronize clocks, and fuses multi-source sensing data and node load distribution maps through the edge computing network to generate fusion confidence. The cloud-based collaborative decision-making layer uses a rule engine to transform fused confidence into executable collection commands, and then sends these executable collection commands to the edge gateway in real time via the MQTT protocol. The closed-loop optimization feedback layer monitors the system load changes after the edge gateway executes the collection command in real time, calculates the deviation between the system load change and the preset system load change threshold, generates the collection strategy optimization index, and dynamically optimizes the manufacturing data collection strategy until the collection strategy optimization index reaches stability.

[0005] Preferably, the method for acquiring multi-source sensing data through a sensor array includes: The distributed manufacturing data acquisition system deploys three types of sensing nodes: vibration sensors, temperature sensors, and pressure sensors, which collect multi-source sensing data from the equipment. The multi-source sensing data includes vibration frequency data of the equipment, temperature distribution data of the equipment surface, and pressure fluctuation data of key parts of the equipment. A non-uniform sampling mechanism simulating industrial scenarios is used to dynamically adjust the sensor sampling frequency and transmission cycle. Based on the physical layout of the equipment, the bearing joints, gear meshing areas, and motor heat dissipation areas are designated as high-frequency sampling areas, while the remaining areas on the equipment surface are designated as low-frequency sampling areas. A sampling function is dynamically defined to divide the equipment surface into high-frequency and low-frequency sampling areas. Dense time-series sampling is used in the high-frequency sampling areas, while sparse time-series sampling is used in the low-frequency sampling areas. Sensor parameters are dynamically adjusted through control commands to obtain the final multi-source sensing data.

[0006] Preferably, the method for generating a heatmap of device operating status by applying an edge filtering network to multi-source sensing data includes: The multi-source sensing data is normalized by mean, and the normalized multi-source sensing data is stacked into a four-dimensional matrix according to the dimensions. The dimensions of the four-dimensional matrix include time, space, type and value. An edge filtering network structure is constructed, which includes forward processing path, backward correction path and lateral correlation. Multi-source sensing data is input into an edge filtering network structure. In the forward processing path, a convolutional neural network is used to extract features from the input multi-source data, generating feature matrices of different granularities. More specific state information is extracted layer by layer. The feature matrices of different granularities include basic feature matrices, intermediate feature matrices, and high-level feature matrices. In the backward correction path, the high-level feature matrices are downsampled to gradually reduce the data dimensionality. A horizontal correlation is established between the forward processing path and the backward correction path to fuse feature matrices of the same granularity. 3×3 convolutions are applied to each output layer of the edge filtering network structure to adjust the feature dimensions. The feature matrices of different granularities output by the edge filtering network structure are fused to generate a heatmap of the device's operating status.

[0007] Preferably, the method for prioritizing the regions in the heat map of equipment operating status includes: The equipment status assessment score for each region in the equipment operation status heat map is calculated by weighting and summing the vibration frequency data of equipment operation, the temperature distribution data of equipment surface, and the pressure fluctuation data of key parts of equipment included in the multi-source sensing data. The system sets a first threshold and a second threshold for device status evaluation scores. The device status evaluation score is then compared to each of these thresholds. If the score is less than the first threshold, the corresponding area is marked green; if the score is greater than the first threshold but less than the second threshold, the corresponding area is marked yellow; and if the score is greater than the second threshold, the corresponding area is marked red. Different colors are used to divide the heatmap of device operating status into regions with different priorities. The green region is defined as the low priority region, the yellow region as the medium priority region, and the red region as the high priority region.

[0008] Preferably, the method for generating the node load distribution map includes: Based on the assigned regional priorities, communication nodes of different densities are deployed on the device surface. The communication nodes are NB-IoT dual-mode communication nodes, and differentiated signaling strategies are adopted for different priority areas. Time-division multiplexing signaling is transmitted to the network side through the communication nodes to achieve initial full network coverage. The signaling transmission angle is dynamically adjusted according to the regional priority of the device operation status heatmap. The signaling transmission angle interval of high priority areas is smaller than that of medium priority areas, and the signaling transmission angle interval of medium priority areas is smaller than that of low priority areas. An angle optimization algorithm is applied to dynamically adjust the signaling transmission angle. A network traffic load model is introduced, and the delay parameter in the load model is continuously corrected through an iterative time slot allocation algorithm. The algorithm stops when the maximum number of iterations is reached. A signaling propagation path matrix is ​​established based on a path planning algorithm, and the load distribution is solved by a linear regression method. Finally, a node load distribution map with priority marking is generated.

[0009] Preferably, the method for establishing the spatiotemporal mapping between sensing and signaling includes: Using any point on the surface of the manufacturing equipment as the origin, with the X and Y axes parallel to the surface and the Z axis perpendicular to the surface, a unified global coordinate system is defined. A calibration board is used to calibrate the sensing nodes, obtaining their intrinsic and extrinsic parameters. The pixel coordinates of the equipment operating status heatmap are transformed to node coordinates using the intrinsic parameters of the sensing nodes, and then the node coordinates are transformed to global coordinates using the extrinsic parameters of the sensing nodes, thus obtaining the equipment operating status heatmap in the global coordinate system. The communication nodes are calibrated using the origin of the global coordinate system as a reference point, obtaining their extrinsic parameters. The node load distribution map is transformed into the global coordinate system by using the extrinsic parameters of the communication nodes, thereby obtaining the node load distribution map in the global coordinate system. In the global coordinate system, the device operation status heat map and the node load distribution map are spatially aligned to establish a spatial mapping between sensing and signaling. The time source connected to the sensing node is set as the master clock, and the time source connected to the communication node is set as the slave clock. Time synchronization is performed through the NTP protocol to establish a time mapping between sensing and signaling.

[0010] Preferably, the method for fusing multi-source sensing data and node load distribution maps through an edge computing network includes: Based on the device operation status heatmap and node load distribution map in the global coordinate system, a directed graph is constructed. An edge computing network is built using a multi-layer RNN structure based on a bidirectional gating mechanism to fuse the device operation status heatmap and node load distribution map in the global coordinate system. The edge computing network includes an input layer, a feature extraction layer, a temporal interaction layer, a cross-modal fusion layer, and an output layer. The directed graph is used as the input to the input layer of the edge computing network, and the fusion confidence is generated through the output layer.

[0011] Preferably, the method for constructing a directed graph includes: Each priority region in the device operation status heatmap under the global coordinate system is taken as a sensing node, and the feature vector of each priority region is extracted as the feature of the sensing node; each load concentration region in the node load distribution map under the global coordinate system is taken as a signaling node, and the time sequence feature vector of each load concentration region is extracted as the feature of the signaling node; all sensing nodes and signaling nodes are collected to obtain the node set; Traverse all sensing nodes and calculate the Manhattan distance between any two sensing nodes in the global coordinate system. Set a preset sensing distance threshold. If the Manhattan distance between any two sensing nodes in the global coordinate system is less than the preset sensing distance threshold, add a one-way edge between the two sensing nodes. If the Manhattan distance between any two sensing nodes in the global coordinate system is greater than or equal to the preset sensing distance threshold, do not add an edge. Traverse all signaling nodes and calculate the Manhattan distance between any two signaling nodes in the global coordinate system. A preset signaling distance threshold is used; if the Manhattan distance between any two signaling nodes is less than the threshold, a one-way edge is added between them. If the Manhattan distance is greater than or equal to the threshold, no edge is added. Through sequential search, find the next signaling node for each sensing node and add a one-way edge between it and the next sensing node. Collect all one-way edges to obtain an edge set. Construct a directed graph based on the obtained node set and edge set.

[0012] Preferably, the method for converting fused confidence scores into executable collection instructions based on a rule engine includes: The manufacturing equipment space is divided into small cubic units, each of which records the current load value and fusion confidence. At the same time, all adjustable acquisition parameters are listed, including the adjustment range of each parameter, energy consumption, and the correlation between historical load impact and other parameters. Three control objectives are established, including a primary stability objective, a secondary stability objective, and an efficiency objective; two types of constraints are set, including fixed constraints and variable constraints. Using a rule engine algorithm, n sets of parameter optimization schemes are randomly generated. The completion status of each scheme is evaluated for three control objectives, and the completion status scores of the three control objectives are obtained. The completion status scores of the three control objectives are added together to obtain a comprehensive score. The scheme with the highest comprehensive score is selected from the n sets of schemes as the final optimization scheme. The selected final optimization scheme is converted into actual executable acquisition instructions. The executable acquisition instructions include a first instruction, a second instruction, and a monitoring instruction.

[0013] Preferably, the methods for obtaining the completion scores of the three control objectives include: The percentage reduction in the maximum load value of the observation system is used as the score for the completion of the first stability objective; the variance of the fusion confidence in different regions is used as the score for the completion of the second stability objective; and the total energy consumption resulting from the adjustment of all acquisition parameters is used as the score for the completion of the efficiency objective.

[0014] Compared with the prior art, the beneficial effects of the present invention are: Through a multi-layered collaborative architecture, it demonstrates numerous advantages in the field of distributed manufacturing data acquisition. The heterogeneous node perception layer uses sensor arrays to acquire multi-source sensing data and applies an edge filtering network to generate a heatmap of equipment operating status and perform regional priority division. This approach can accurately capture the real-time operating status of equipment, making data acquisition more targeted, avoiding the waste of resources caused by indiscriminate collection of all areas, and improving the efficiency of data acquisition. The narrowband signaling scheduling layer deploys NB-IoT communication nodes based on regional priorities. It leverages time-division multiplexing signaling to penetrate network congestion and then uses a time-slot allocation algorithm to calculate signaling response delays and generate a node load distribution map. This effectively solves the data transmission delay problem caused by network congestion in distributed manufacturing environments. Time-division multiplexing signaling fully utilizes communication resources, while the time-slot allocation algorithm dynamically adjusts node load, making the entire communication network operate more smoothly, reducing data loss or delays caused by network problems, and ensuring the timeliness and stability of data transmission. The edge fusion processing layer establishes a spatiotemporal mapping between sensing and signaling, uses the NTP protocol to synchronize clocks, and generates fusion confidence scores by fusing multi-source sensing data and node load distribution maps through the edge computing network. This solves the problem of inconsistencies between multi-source data in time and space. The application of the NTP protocol ensures the synchronization of clocks across nodes, providing a unified time reference for data fusion. The edge computing network can then perform rapid data fusion processing locally, avoiding the bandwidth pressure and processing latency caused by uploading large amounts of raw data to the cloud. This improves the efficiency and accuracy of data fusion, making the generated fusion confidence scores more reflective of actual production conditions. The cloud-based collaborative decision-making layer, based on a rule engine, transforms fused confidence levels into executable data collection commands and distributes them to the edge gateway in real time via the MQTT protocol, achieving efficient collaboration between the cloud and edge devices. The rule engine can quickly transform abstract fused confidence levels into specific operational commands, while the MQTT protocol ensures the real-time and reliable delivery of commands, enabling the edge gateway to respond promptly to cloud decisions and ensuring that data collection is conducted according to the optimal strategy. The closed-loop optimization feedback layer monitors system load changes in real time, calculates deviations, and generates an optimization index for the acquisition strategy. It dynamically optimizes the acquisition strategy until the index stabilizes, enabling the system to self-adjust and optimize. During manufacturing, system load fluctuates with changes in production conditions. The closed-loop optimization mechanism can promptly detect these changes and adjust the acquisition strategy, ensuring the system always operates at high efficiency, adapting to different production scenarios and load conditions, and improving the flexibility and adaptability of the entire data acquisition system. Attached Figure Description

[0015] Figure 1 This is a schematic diagram illustrating the working principle of the NB-IoT-based distributed manufacturing data acquisition system described in this invention. Figure 2 A flowchart of a multi-source sensing data acquisition method; Figure 3 A flowchart generated from a heatmap of equipment operating status; Figure 4 A flowchart generated for the node load distribution diagram; Figure 5 A flowchart for establishing the spatiotemporal mapping of sensing and signaling. Detailed Implementation

[0016] 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.

[0017] Please see Figures 1-5 This invention provides a distributed manufacturing data acquisition system based on NB-IoT, the system comprising: The heterogeneous node perception layer acquires multi-source perception data through a sensor array, applies an edge filtering network to the multi-source perception data, generates a heat map of the device's operating status, and divides the heat map of the device's operating status into regions based on priority. The narrowband signaling scheduling layer deploys NB-IoT communication nodes according to the priority of the divided areas, transmits time-division multiplexed signaling to penetrate network congestion, calculates signaling response delay through time slot allocation algorithm, and generates node load distribution map; The edge fusion processing layer establishes a spatiotemporal mapping between sensing and signaling, uses the NTP protocol to synchronize clocks, and fuses multi-source sensing data and node load distribution maps through the edge computing network to generate fusion confidence. The cloud-based collaborative decision-making layer uses a rule engine to transform fused confidence into executable collection commands, and then sends these executable collection commands to the edge gateway in real time via the MQTT protocol. The closed-loop optimization feedback layer monitors the system load changes after the edge gateway executes the collection command in real time, calculates the deviation between the system load change and the preset system load change threshold, generates the collection strategy optimization index, and dynamically optimizes the manufacturing data collection strategy until the collection strategy optimization index reaches stability.

[0018] Example 1: In a distributed manufacturing data acquisition system, when acquiring multi-source sensing data through a sensor array, three types of sensing nodes need to be deployed: vibration sensors, temperature sensors, and pressure sensors. Vibration sensors are used to collect vibration frequency data during equipment operation, capturing periodic or non-periodic vibration signals generated by internal components due to rotation, friction, etc. Temperature sensors are responsible for collecting temperature distribution data on the equipment surface, reflecting the heat dissipation in different areas of the equipment during operation. Pressure sensors, targeting key parts of the equipment such as connection interfaces and load-bearing structures, collect pressure fluctuation data to monitor changes in components under stress.

[0019] The non-uniform sampling mechanism simulating industrial scenarios aims to better match the monitoring needs of different areas of equipment in actual production, thereby dynamically adjusting the sampling frequency and transmission cycle of the sensors. Based on the physical layout diagram of the equipment, the criteria for dividing the high-frequency and low-frequency sampling areas are first clarified: the bearing joints, gear meshing areas, and motor heat dissipation vents of the equipment are designated as high-frequency sampling areas. These areas are critical parts of equipment operation, and their state changes have a significant impact on the overall performance of the equipment, requiring more intensive monitoring; while the remaining parts of the equipment surface are designated as low-frequency sampling areas. The state of these areas is relatively stable and does not require excessively high-frequency sampling.

[0020] A dynamically defined sampling function precisely divides the device surface into high-frequency and low-frequency sampling zones. Dense temporal sampling is employed in the high-frequency sampling zone, meaning multiple data acquisitions are performed per unit time to obtain more detailed information on state changes. Sparse temporal sampling is used in the low-frequency sampling zone, with fewer samples per unit time to reduce unnecessary data redundancy. By sending control commands to the sensors, parameters such as the sensor sampling frequency and data transmission interval can be dynamically adjusted. Ultimately, data acquired by the three sensors under different sampling strategies are integrated to form complete multi-source sensing data.

[0021] When using an edge filtering network to generate a heatmap of equipment operating status from multi-source sensing data, the first step is to perform mean normalization on the multi-source sensing data. This step transforms sensing data of different types and magnitudes into a unified data range, eliminating the impact of differences in data magnitude, and enabling data such as vibration frequency, temperature distribution, and pressure fluctuations to be processed on the same dimension.

[0022] The normalized multi-source sensing data is stacked into a four-dimensional matrix according to its dimensions, which include time, space, type, and value. The time dimension records the time sequence of data acquisition, the space dimension corresponds to the coordinates of different positions on the device surface, the type dimension distinguishes three data types: vibration, temperature, and pressure, and the value dimension is the specific data value after normalization.

[0023] The constructed edge filtering network structure includes a forward processing path, a backward correction path, and lateral correlation. After inputting multi-source sensing data in four-dimensional matrix form into this network structure, a convolutional neural network is used to extract features from the input data in the forward processing path. The convolution operation captures local features of the data through a sliding window, generating feature matrices of different granularities layer by layer. The basic feature matrix contains the original state features of the device surface, the intermediate feature matrices further refine the correlation features within the region, and the high-level feature matrix covers more abstract and core device operating state features. Furthermore, as the number of network layers increases, the extracted state information becomes increasingly specific.

[0024] In the backward correction path, the data dimensionality of the high-level feature matrix is ​​gradually reduced through downsampling operations. The downsampling process selects the feature extrema or mean in a local region, which reduces the amount of data while retaining key feature information, so that the dimensionality of the high-level feature matrix matches the feature matrices at different levels in the forward processing path.

[0025] A horizontal correlation is established between the forward processing path and the backward correction path, fusing feature matrices of the same granularity. This fusion, achieved through element-wise addition or concatenation, allows the detailed features extracted by the forward path to complement the high-level features fed back by the backward path, enhancing the feature matrix's ability to represent the device state.

[0026] 3×3 convolutions are applied to each output layer of the edge filtering network structure to adjust the feature dimensions, ensuring that the feature matrices output from different layers maintain consistency in dimensions such as the number of channels and size, facilitating subsequent fusion operations. The adjusted feature matrices of different granularities are then integrated to generate a heatmap of the device's operating status. This heatmap visually displays the differences in operating status across different areas of the device's surface using color gradient changes.

[0027] When prioritizing regions in the heat map of equipment operating status, it is necessary to calculate the equipment status assessment score for each region. The calculation process involves weighted summation of vibration frequency data, temperature distribution data, and pressure fluctuation data from multi-source sensing data within that region. The weight allocation is determined based on the degree of influence of different data on the equipment status assessment. For example, temperature anomalies have a greater impact on the motor region, so its weight may be relatively higher.

[0028] A first threshold and a second threshold are preset for device status assessment scores, with the first threshold being less than the second threshold. The device status assessment score for each region is compared to these two thresholds: if the score is less than the first threshold, the region is marked green; if the score is greater than the first threshold but less than the second threshold, it is marked yellow; and if the score is greater than the second threshold, it is marked red. Green represents low-priority regions, indicating stable device operation and requiring no special monitoring; yellow represents medium-priority regions, suggesting slight abnormal tendencies and requiring continued attention; and red represents high-priority regions, indicating potentially significant device status anomalies requiring priority monitoring and handling. This process completes the regional priority division of the device operation status heatmap.

[0029] Example 2: The process of generating a node load distribution map begins with deploying NB-IoT dual-mode communication nodes at varying densities on the device surface based on the completed regional priority division results. These communication nodes possess two communication modes and can automatically switch according to the network environment and data transmission requirements to adapt to complex industrial manufacturing scenarios. The signaling strategies employed differ significantly for regions with different priorities. High-priority regions, due to their higher requirements for real-time and reliability data transmission, have a relatively higher deployment density of communication nodes and a correspondingly higher signaling transmission frequency; medium-priority regions have a moderate density of communication nodes and a moderate signaling transmission frequency; low-priority regions deploy fewer communication nodes and have a lower signaling transmission frequency. This differentiated deployment and strategy setting can meet the data collection needs of different regions while avoiding resource waste.

[0030] Communication nodes transmit time-division multiplexing signaling to the network side. This signaling method divides time into multiple non-overlapping time slots. Each communication node transmits signaling within its assigned specific time slot, thus avoiding signal interference between different nodes and achieving initial network-wide coverage. During signaling transmission, the signaling transmission angle is dynamically adjusted based on the priority of each area in the device operation status heatmap. Specifically, high-priority areas have smaller signaling transmission angle intervals, resulting in denser signaling coverage and ensuring accurate and timely data transmission. Medium-priority areas have larger signaling transmission angle intervals than high-priority areas, with moderate signaling coverage density. Low-priority areas have the largest signaling transmission angle intervals, resulting in relatively sparse signaling coverage. This angle adjustment mechanism allows signaling resources to be more concentrated on important areas.

[0031] An angle optimization algorithm is used to dynamically adjust the signaling transmission angle. This algorithm continuously monitors the signaling reception quality and network status in each area and corrects the transmission angle in real time based on feedback information. When the signaling reception strength in a certain area is insufficient or interference occurs, the algorithm automatically reduces the signaling transmission angle interval in that area to enhance signaling coverage; conversely, if the signaling reception in a certain area is good and the network load is low, the transmission angle interval is appropriately increased to reduce unnecessary signaling transmission.

[0032] During the adjustment process, a network traffic load model is introduced. This model comprehensively considers factors such as the signaling transmission volume, data transmission rate, and network congestion level of each communication node to assess the network load status. An iterative time slot allocation algorithm continuously corrects the latency parameters in the load model. Each iteration reallocates time slot resources based on the current network load, optimizing the timing and duration of signaling transmission. When the number of iterations reaches a preset maximum, the iteration process stops, at which point the latency parameters can better reflect the actual latency characteristics of the network.

[0033] A signaling propagation path matrix is ​​established based on a path planning algorithm. This algorithm analyzes the physical structure of the device surface, the distribution of obstacles, and the location information of communication nodes to plan the optimal propagation path of signaling from the sending node to the receiving node. The path matrix records the path connection relationships and path quality parameters between each node, such as signal attenuation and transmission time.

[0034] Load distribution was solved using linear regression. A linear relationship model between load and signaling transmission parameters was established using data from the path matrix and corrected delay parameters. Through analysis and calculation of extensive historical and real-time monitoring data, the coefficients in the model were determined, thereby identifying the load values ​​for each region. Finally, combined with regional priority information, a node load distribution map with priority markers was generated. This map clearly shows the load status of each communication node in different priority regions, providing an intuitive basis for subsequent network optimization and data acquisition strategy adjustments.

[0035] Example 3: When establishing the spatiotemporal mapping of sensing and signaling, an arbitrary point on the surface of the manufacturing equipment is used as the origin. The X and Y axes are parallel to the surface of the manufacturing equipment, and the Z axis is perpendicular to the surface of the manufacturing equipment, defining a unified global coordinate system. This coordinate system needs to cover all areas of the equipment to be monitored, ensuring that each position on the equipment surface can be identified by a unique coordinate value. The unit of the coordinate value can be selected as millimeters or centimeters according to the equipment size to meet the accuracy requirements.

[0036] The sensing node is calibrated using a calibration board with multiple markers having known precise coordinates. The calibration board is fixed at different positions on the device surface, and images of the calibration board are captured by the sensing node. Image recognition technology is used to extract the pixel coordinates of the markers in the sensing node's coordinate system. Based on the actual physical coordinates and pixel coordinates of the markers, the intrinsic and extrinsic parameters of the sensing node are calculated. The intrinsic parameters include the sensing node's focal length, principal point coordinates, and other internal optical parameters, while the extrinsic parameters include the sensing node's position and attitude parameters in the global coordinate system.

[0037] The pixel coordinates of the device operating status heatmap are transformed to node coordinates using the intrinsic parameters of the sensing nodes. This process is based on the principle of perspective projection, mapping the two-dimensional pixel coordinates to the sensing node's own three-dimensional coordinate system. Then, the node coordinates are transformed to global coordinates using the extrinsic parameters of the sensing nodes. This transformation involves rotation and translation operations to align the coordinate values ​​in the sensing node's coordinate system with the global coordinate system, thereby obtaining the device operating status heatmap in the global coordinate system. At this point, each pixel in the heatmap corresponds to a specific position of the device surface in the global coordinate system.

[0038] The communication node is calibrated using the origin of the global coordinate system as a reference point. A high-precision positioning device is placed at the communication node to measure its actual position in the global coordinate system. Combined with the node's installation angle, the extrinsic parameters of the communication node are determined. These extrinsic parameters also include rotation and translation parameters, used to describe the spatial attitude of the communication node relative to the global coordinate system.

[0039] The node load distribution map is transformed to a global coordinate system using extrinsic parameters of the communication nodes. The load data in the node load distribution map, originally based on the local coordinate system of the communication nodes, is unified to the global coordinate system after the extrinsic parameter transformation, thus obtaining the node load distribution map in the global coordinate system. At this point, each load data point in the load distribution map has a corresponding global coordinate position.

[0040] In the global coordinate system, the device operation status heatmap and node load distribution map are spatially aligned to establish a spatial mapping between sensing and signaling. During the alignment process, the scaling ratio and positional offset of the images are adjusted by comparing the coordinate information of the same physical location in the two maps to ensure that the heatmap and load distribution map completely overlap in space. For areas with slight deviations, interpolation algorithms are used for correction to ensure that sensing data and signaling data at the same location can accurately correspond.

[0041] The time source connected to the sensing nodes is set as the master clock, and the time source connected to the communication nodes is set as the slave clock. Time synchronization is achieved through the NTP protocol, establishing a time mapping between sensing and signaling. The master clock uses a high-precision atomic clock or GPS clock to provide a unified time reference for the entire system. The slave clock obtains the current time of the master clock by periodically sending time synchronization requests to it, and calculates the difference between its local time and the master clock time based on network transmission latency, thereby adjusting its local clock accordingly.

[0042] The specific process of time synchronization is as follows: The slave clock sends a synchronization message containing its current timestamp to the master clock. Upon receiving the message, the master clock records the received timestamp and immediately returns a response message containing both its own timestamp and the received timestamp. After receiving the response message, the slave clock records the received timestamp and calculates the time deviation using the following formula:

[0043] in, This indicates the time deviation between the slave clock and the master clock. For the timestamp of the synchronization message sent from the clock, The timestamp for receiving synchronization messages from the master clock. The timestamp of the master clock sending the response message. This is the timestamp for receiving the response message from the clock. Based on the calculated time deviation, the clock corrects its own time to ensure consistency between the sensing node and the communication node. Through multiple synchronization operations, the time deviation can be controlled to the millisecond or even microsecond level, ensuring the accuracy of sensing data and signaling data in the time dimension.

[0044] By establishing the aforementioned spatial and temporal mappings, the sensing data and signaling data are precisely correlated in the spatiotemporal dimensions, providing a unified spatiotemporal benchmark for subsequent multi-source data fusion and system decision-making. This spatiotemporal mapping mechanism can effectively eliminate data errors caused by spatial location differences and temporal asynchrony, enabling the system to more accurately analyze the relationship between equipment operating status and network load.

[0045] Example 4: When fusing multi-source sensing data and node load distribution map through edge computing network, it is necessary to first construct a directed graph based on the device operation status heat map and node load distribution map in the global coordinate system, and then build an edge computing network using a multi-layer RNN structure based on bidirectional gating mechanism to fuse these two types of graph data.

[0046] The heatmap of equipment operation status in the global coordinate system clearly defines the priority of each region and its corresponding status data, while the node load distribution map marks the load situation at different locations. When constructing the directed graph, each priority region in the equipment operation status heatmap is first treated as a sensing node, and the feature vector of each region is extracted as the sensing node feature. These feature vectors contain fused information from multi-source sensing data such as vibration, temperature, and pressure for that region. Simultaneously, each load concentration region in the node load distribution map is treated as a signaling node, and the temporal feature vector of each load concentration region is extracted as the signaling node feature. The temporal feature vector covers the load change data of that region at different time periods. All sensing nodes and signaling nodes are collected to form a node set.

[0047] Iterate through all sensing nodes and calculate the Manhattan distance between any two sensing nodes in the global coordinate system. For example, if one sensing node is located at coordinates (10, 20, 5) and another is located at (15, 25, 5), then their Manhattan distance is |10-15|+|20-25|+|5-5|=10. A preset sensing distance threshold, say 15, is used. When the Manhattan distance between two sensing nodes is less than this threshold (e.g., 10 is less than 15 in the example above), a one-way edge is added between the two sensing nodes; if the distance is greater than or equal to the threshold, no edge is added.

[0048] Next, iterate through all signaling nodes and calculate the Manhattan distance between any two signaling nodes in the global coordinate system. Assume one signaling node has coordinates (8, 18, 5) and another has coordinates (20, 22, 5). Their Manhattan distance is |8-20|+|18-22|+|5-5|=16. The preset signaling distance threshold is 20. Since 16 is less than 20, a one-way edge is added between these two signaling nodes; if the distance exceeds the threshold, no edge is added.

[0049] By sequentially searching, the next signaling node is found for each sensing node. For example, starting from sensing node A, the nearest signaling node is searched in a clockwise direction. When signaling node B is found, a one-way edge is added between sensing node A and signaling node B. All one-way edges obtained through the above steps are collected to form an edge set, and then a directed graph is constructed based on the node set and the edge set.

[0050] The constructed edge computing network consists of an input layer, a feature extraction layer, a temporal interaction layer, a cross-modal fusion layer, and an output layer. The constructed directed graph is used as the input to the edge computing network's input layer, which converts the node features and edge information of the directed graph into a tensor form that the network can process.

[0051] The feature extraction layer uses convolution operations to process the input tensor and extract local correlation information from the node features. For example, convolution on the features of sensing nodes can capture the potential relationships between vibration, temperature, and pressure data in the same region; convolution on the features of signaling nodes can extract the local trends of load changes.

[0052] The temporal interaction layer, based on a multi-layer RNN structure with a bidirectional gating mechanism, performs temporal modeling on the extracted feature sequence data. The forward RNN processes temporal information from past to present, while the backward RNN processes it from present to past. Combining the two allows for the capture of long-term dependencies in the data. For example, for temporal data on temperature changes in a specific region, the forward RNN can analyze the gradual increase in temperature, while the backward RNN can trace back to the state before the temperature increase, thus providing a comprehensive understanding of the temporal characteristics of temperature changes.

[0053] The cross-modal fusion layer fuses features related to sensing nodes and features related to signaling nodes, using an attention mechanism to allow the network to automatically focus on important feature information. For example, when the temperature data of a certain area is abnormal and the corresponding load data also fluctuates, the attention mechanism will increase the weight of these two parts of features, making the fused features more reflective of the actual state of the area.

[0054] The output layer processes the fused features through a fully connected network to generate a fusion confidence score. This score integrates information from multi-source sensing data and node load distribution, characterizing the reliability of the operational status assessment across different areas of the device. The entire edge computing network processing is completed at the edge nodes, reducing data transmission latency to the cloud and improving the real-time performance of the fusion processing.

[0055] Example 5: When converting fusion confidence into executable acquisition commands based on a rule engine, the manufacturing equipment space needs to be divided into small cubic units. The division process must consider the actual size and structural characteristics of the equipment. The side length of each unit can be determined according to the precision of the equipment to ensure accurate reflection of the state of each fine area of ​​the equipment. Each unit records its current load value and fusion confidence. The load value reflects the pressure exerted on the system by the corresponding equipment area during data acquisition, while the fusion confidence integrates multi-source sensing data and node load distribution information, reflecting the reliability of the state assessment of that area. Simultaneously, all adjustable acquisition parameters are listed, including sensor sampling frequency, data transmission period, signaling transmission power, and time slot allocation duration. For each parameter, its adjustment range needs to be clearly defined; for example, the sampling frequency can be adjusted between 10Hz and 100Hz. The energy consumption during parameter adjustment is recorded; for example, increasing the sampling frequency increases sensor energy consumption. The historical correlation of this parameter's impact on the load is statistically analyzed; for example, adjusting the time slot allocation duration usually has a significant impact on system load.

[0056] Three control objectives are established: the first stability objective focuses on the overall stability of the system load; the second stability objective emphasizes the consistency of fused confidence across different regions; and the efficiency objective focuses on energy consumption control during data acquisition. Two types of constraints are set: fixed constraints include insurmountable physical limitations such as the maximum sampling frequency and minimum transmission period supported by the device hardware; and variable constraints are dynamically adjusted according to real-time network conditions and device operation, such as appropriately relaxing restrictions on certain non-critical parameters when network congestion occurs.

[0057] When using the rule engine algorithm, n sets of optimization schemes for the acquisition parameters are first randomly generated. The number of n can be set according to actual needs to ensure that there are enough schemes to choose from. For each scheme, the achievement of the three control objectives is evaluated. When evaluating the first stability objective, the percentage reduction in the maximum load value of the observation system under the scheme is used. That is, the proportion of the maximum load value of the system after implementation compared with the maximum load value before implementation. This proportion is used as the achievement score of the first stability objective. When evaluating the second stability objective, the variance of the fusion confidence in different regions is calculated. The smaller the variance, the more uniform the distribution of fusion confidence in each region. This variance value is used as the achievement score of the second stability objective. When evaluating the efficiency objective, the total energy consumption caused by the adjustment of all acquisition parameters is calculated. That is, the sum of the additional energy consumed after the adjustment of each parameter. This total amount is used as the achievement score of the efficiency objective.

[0058] The scores for the completion of the three control objectives are summed to obtain a comprehensive score for each scheme. The scheme with the highest comprehensive score from the n schemes is selected as the final optimized scheme, which achieves optimal control of the data acquisition process while balancing system stability and energy consumption. The selected final optimized scheme is converted into actual executable acquisition commands. The first command may involve specific adjustments to the sensor sampling frequency and transmission period, such as adjusting the sensor sampling frequency in a high-frequency sampling area from 50Hz to 80Hz and shortening the transmission period from 2 seconds to 1 second. The second command may include specific parameters for signaling transmission power and time slot allocation, such as increasing the signaling transmission power in high-priority areas by 10% and increasing the time slot allocation duration by 0.5 seconds. The monitoring command specifies the areas and parameters that need to be monitored, such as requiring the reporting of temperature distribution data and corresponding load changes in a gear meshing area every 5 minutes. These executable acquisition commands are distributed to each sensor and communication node through the edge gateway to guide them in data acquisition and transmission operations.

[0059] 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.

[0060] 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 distributed manufacturing data acquisition system based on NB-IoT, characterized in that, include: The heterogeneous node perception layer acquires multi-source perception data through a sensor array, applies an edge filtering network to the multi-source perception data, generates a heat map of the device's operating status, and divides the heat map of the device's operating status into regions based on priority. The narrowband signaling scheduling layer deploys NB-IoT communication nodes according to the priority of the divided areas, transmits time-division multiplexed signaling to penetrate network congestion, calculates signaling response delay through time slot allocation algorithm, and generates node load distribution map; The edge fusion processing layer establishes a spatiotemporal mapping between sensing and signaling, uses the NTP protocol to synchronize clocks, and fuses multi-source sensing data and node load distribution maps through the edge computing network to generate fusion confidence. The cloud-based collaborative decision-making layer uses a rule engine to transform fused confidence into executable collection commands, and then sends these executable collection commands to the edge gateway in real time via the MQTT protocol. The closed-loop optimization feedback layer monitors the system load changes after the edge gateway executes the collection command in real time, calculates the deviation between the system load change and the preset system load change threshold, generates the collection strategy optimization index, and dynamically optimizes the manufacturing data collection strategy until the collection strategy optimization index reaches stability.

2. The distributed manufacturing data acquisition system based on NB-IoT according to claim 1, characterized in that, The method for acquiring multi-source sensing data through a sensor array includes: The distributed manufacturing data acquisition system deploys three types of sensing nodes: vibration sensors, temperature sensors, and pressure sensors, which collect multi-source sensing data from the equipment. The multi-source sensing data includes vibration frequency data of the equipment, temperature distribution data of the equipment surface, and pressure fluctuation data of key parts of the equipment. A non-uniform sampling mechanism simulating industrial scenarios is used to dynamically adjust the sensor sampling frequency and transmission cycle. Based on the physical layout of the equipment, the bearing joints, gear meshing areas, and motor heat dissipation areas are designated as high-frequency sampling areas, while the remaining areas on the equipment surface are designated as low-frequency sampling areas. A sampling function is dynamically defined to divide the equipment surface into high-frequency and low-frequency sampling areas. Dense time-series sampling is used in the high-frequency sampling areas, while sparse time-series sampling is used in the low-frequency sampling areas. Sensor parameters are dynamically adjusted through control commands to obtain the final multi-source sensing data.

3. The distributed manufacturing data acquisition system based on NB-IoT according to claim 2, characterized in that, The method for generating a heatmap of device operating status by applying an edge filtering network to multi-source sensing data includes: The multi-source sensing data is normalized by mean, and the normalized multi-source sensing data is stacked into a four-dimensional matrix according to the dimensions. The dimensions of the four-dimensional matrix include time, space, type and value. An edge filtering network structure is constructed, which includes forward processing path, backward correction path and lateral correlation. Multi-source sensing data is input into an edge filtering network structure. In the forward processing path, a convolutional neural network is used to extract features from the input multi-source data, generating feature matrices of different granularities. More specific state information is extracted layer by layer. The feature matrices of different granularities include basic feature matrices, intermediate feature matrices, and high-level feature matrices. In the backward correction path, the high-level feature matrices are downsampled to gradually reduce the data dimensionality. A horizontal correlation is established between the forward processing path and the backward correction path to fuse feature matrices of the same granularity. 3×3 convolutions are applied to each output layer of the edge filtering network structure to adjust the feature dimensions. The feature matrices of different granularities output by the edge filtering network structure are fused to generate a heatmap of the device's operating status.

4. The distributed manufacturing data acquisition system based on NB-IoT according to claim 3, characterized in that, The method for prioritizing regions in the heat map of equipment operating status includes: The equipment status assessment score for each region in the equipment operation status heat map is calculated by weighting and summing the vibration frequency data of equipment operation, the temperature distribution data of equipment surface, and the pressure fluctuation data of key parts of equipment included in the multi-source sensing data. The system sets a first threshold and a second threshold for device status evaluation scores. The device status evaluation score is then compared to each of these thresholds. If the score is less than the first threshold, the corresponding area is marked green; if the score is greater than the first threshold but less than the second threshold, the corresponding area is marked yellow; and if the score is greater than the second threshold, the corresponding area is marked red. Different colors are used to divide the heatmap of device operating status into regions with different priorities. The green region is defined as the low priority region, the yellow region as the medium priority region, and the red region as the high priority region.

5. The distributed manufacturing data acquisition system based on NB-IoT according to claim 4, characterized in that, The method for generating the node load distribution map includes: Based on the assigned regional priorities, communication nodes of different densities are deployed on the device surface. The communication nodes are NB-IoT dual-mode communication nodes, and differentiated signaling strategies are adopted for different priority areas. Time-division multiplexing signaling is transmitted to the network side through the communication nodes to achieve initial full network coverage. The signaling transmission angle is dynamically adjusted according to the regional priority of the device operation status heatmap. The signaling transmission angle interval of high priority areas is smaller than that of medium priority areas, and the signaling transmission angle interval of medium priority areas is smaller than that of low priority areas. An angle optimization algorithm is applied to dynamically adjust the signaling transmission angle. A network traffic load model is introduced, and the delay parameter in the load model is continuously corrected through an iterative time slot allocation algorithm. The algorithm stops when the maximum number of iterations is reached. A signaling propagation path matrix is ​​established based on a path planning algorithm, and the load distribution is solved by a linear regression method. Finally, a node load distribution map with priority marking is generated.

6. The distributed manufacturing data acquisition system based on NB-IoT according to claim 5, characterized in that, The method for establishing the spatiotemporal mapping between sensing and signaling includes: Using any point on the surface of the manufacturing equipment as the origin, with the X and Y axes parallel to the surface and the Z axis perpendicular to the surface, a unified global coordinate system is defined. A calibration board is used to calibrate the sensing nodes, obtaining their intrinsic and extrinsic parameters. The pixel coordinates of the equipment operating status heatmap are transformed to node coordinates using the intrinsic parameters of the sensing nodes, and then the node coordinates are transformed to global coordinates using the extrinsic parameters of the sensing nodes, thus obtaining the equipment operating status heatmap in the global coordinate system. The communication nodes are calibrated using the origin of the global coordinate system as a reference point, obtaining their extrinsic parameters. The node load distribution map is transformed into the global coordinate system by using the extrinsic parameters of the communication nodes, thereby obtaining the node load distribution map in the global coordinate system. In the global coordinate system, the device operation status heat map and the node load distribution map are spatially aligned to establish a spatial mapping between sensing and signaling. The time source connected to the sensing node is set as the master clock, and the time source connected to the communication node is set as the slave clock. Time synchronization is performed through the NTP protocol to establish a time mapping between sensing and signaling.

7. The distributed manufacturing data acquisition system based on NB-IoT according to claim 6, characterized in that, The method for fusing multi-source sensing data and node load distribution maps through an edge computing network includes: Based on the device operation status heatmap and node load distribution map in the global coordinate system, a directed graph is constructed. An edge computing network is built using a multi-layer RNN structure based on a bidirectional gating mechanism to fuse the device operation status heatmap and node load distribution map in the global coordinate system. The edge computing network includes an input layer, a feature extraction layer, a temporal interaction layer, a cross-modal fusion layer, and an output layer. The directed graph is used as the input to the input layer of the edge computing network, and the fusion confidence is generated through the output layer.

8. The distributed manufacturing data acquisition system based on NB-IoT according to claim 7, characterized in that, The method for constructing a directed graph includes: Each priority region in the device operation status heatmap under the global coordinate system is taken as a sensing node, and the feature vector of each priority region is extracted as the feature of the sensing node; each load concentration region in the node load distribution map under the global coordinate system is taken as a signaling node, and the time sequence feature vector of each load concentration region is extracted as the feature of the signaling node; all sensing nodes and signaling nodes are collected to obtain the node set; Traverse all sensing nodes and calculate the Manhattan distance between any two sensing nodes in the global coordinate system. Set a preset sensing distance threshold. If the Manhattan distance between any two sensing nodes in the global coordinate system is less than the preset sensing distance threshold, add a one-way edge between the two sensing nodes. If the Manhattan distance between any two sensing nodes in the global coordinate system is greater than or equal to the preset sensing distance threshold, do not add an edge. Traverse all signaling nodes and calculate the Manhattan distance between any two signaling nodes in the global coordinate system. A preset signaling distance threshold is used; if the Manhattan distance between any two signaling nodes is less than the threshold, a one-way edge is added between them. If the Manhattan distance is greater than or equal to the threshold, no edge is added. Through sequential search, find the next signaling node for each sensing node and add a one-way edge between it and the next sensing node. Collect all one-way edges to obtain an edge set. Construct a directed graph based on the obtained node set and edge set.

9. The distributed manufacturing data acquisition system based on NB-IoT according to claim 8, characterized in that, The method for converting fused confidence scores into executable data collection instructions based on a rule engine includes: The manufacturing equipment space is divided into small cubic units, each of which records the current load value and fusion confidence. At the same time, all adjustable acquisition parameters are listed, including the adjustment range of each parameter, energy consumption, and the correlation between historical load impact and other parameters. Three control objectives are established, including a primary stability objective, a secondary stability objective, and an efficiency objective; two types of constraints are set, including fixed constraints and variable constraints. Using a rule engine algorithm, n sets of parameter optimization schemes are randomly generated. The completion status of each scheme is evaluated for three control objectives, and the completion status scores of the three control objectives are obtained. The completion status scores of the three control objectives are added together to obtain a comprehensive score. The scheme with the highest comprehensive score is selected from the n sets of schemes as the final optimization scheme. The selected final optimization scheme is converted into actual executable acquisition instructions. The executable acquisition instructions include a first instruction, a second instruction, and a monitoring instruction.

10. The distributed manufacturing data acquisition system based on NB-IoT according to claim 9, characterized in that, The methods for obtaining scores on the completion of the three control objectives include: The percentage reduction in the maximum load value of the observation system is used as the score for the completion of the first stability objective; the variance of the fusion confidence in different regions is used as the score for the completion of the second stability objective; and the total energy consumption resulting from the adjustment of all acquisition parameters is used as the score for the completion of the efficiency objective.

Images