Iot-based production plant device state monitoring method and system

Abstract

This application provides a method and system for monitoring the condition of equipment in a production workshop based on the Internet of Things (IoT), relating to the field of equipment condition monitoring technology. The method includes: firstly, collecting multi-angle infrared data and control system switching data from the equipment via the IoT; then, processing the infrared data using a region growing algorithm to obtain three-dimensional temperature data, and calculating temperature gradient and divergence data accordingly; next, labeling the above data with the switching data to form thermal state data; then, constructing a structural relationship diagram of the equipment, and using a graph convolutional neural network in conjunction with the thermal state data to generate a feature vector reflecting the overall thermal state of the equipment; subsequently, inputting this feature vector and the switching data into a long short-term memory neural network to predict future temperature changes; finally, generating equipment health status monitoring results by comparing the prediction results with dynamic thresholds. This application achieves accurate predictive maintenance for early abnormal heating and potential failures of key equipment in a production workshop.

Description

Technical Field

[0001] This application relates to the field of equipment condition monitoring technology, and in particular to a method and system for monitoring the condition of production workshop equipment based on the Internet of Things. Background Technology

[0002] In modern industrial production systems, equipment condition monitoring in production workshops is a core component for ensuring continuous and stable operation of production lines and enabling predictive maintenance. Its accuracy and real-time performance directly impact production safety and economic efficiency. With the widespread adoption of Industrial Internet of Things (IIoT) technology, collecting equipment operation data through sensor networks and conducting remote monitoring has become a mainstream trend.

[0003] Currently, IoT-based device status monitoring methods mainly rely on single-type sensor data, such as deploying vibration sensors or single-point temperature sensors to collect time-series signals, and using traditional statistical analysis methods or shallow machine learning models to identify abnormal patterns. In addition, some solutions attempt to integrate discrete state signals from the device control system with sensor readings to establish a correlation between device operating conditions and operational status, thereby assessing device health.

[0004] However, existing methods focus on the independent analysis of local point-like physical quantities of equipment. The monitoring models constructed by these methods have limitations in characterizing the overall thermal state of equipment caused by the coupling effect of complex mechanical structures and heat conduction paths. Furthermore, the time-series prediction accuracy and early anomaly sensitivity of existing methods need improvement regarding the evolution of thermal behavior caused by dynamic changes in equipment operating conditions with varying load conditions. Therefore, existing technologies face technical challenges in achieving accurate and early predictive monitoring of the overall thermal state of equipment in production workshops. Summary of the Invention

[0005] This application provides a method and system for monitoring the status of equipment in a production workshop based on the Internet of Things, in order to solve the problem of insufficient accurate predictive maintenance for early abnormal heating and potential failure of key equipment in the production workshop in the prior art.

[0006] To address the aforementioned technical problems, in a first aspect, this application provides a method for monitoring the status of equipment in a production workshop based on the Internet of Things, comprising: The IoT platform in the production workshop collects multi-angle infrared data of the target device and switching data output by the control system of the target device. The multi-angle infrared data is processed using a region growing algorithm to generate three-dimensional temperature data of a specified region on the target device, and the corresponding temperature gradient data and divergence data are calculated based on the three-dimensional temperature data. Based on the state switching time points in the switch data, the three-dimensional temperature data, the temperature gradient data, and the divergence data are marked to form thermal state data; A structural relationship diagram of the target device is constructed, and based on the structural relationship diagram and the thermal state data, a feature vector is generated by combining a graph convolutional neural network. The feature vector is used to characterize the overall thermal state of the target device. Based on the feature vector and the switching data, the temperature change data of the target device in future time periods is predicted by a long short-term memory neural network. The temperature change data is compared with a dynamic threshold range to generate monitoring results that reflect the real-time health status of the target device.

[0007] Optionally, the step of processing the multi-angle infrared data using a region growing algorithm to generate three-dimensional temperature data of a specified region on the target device, and calculating the corresponding temperature gradient data and divergence data based on the three-dimensional temperature data, includes: From the multi-angle infrared data, pixels with a temperature value greater than a preset temperature threshold and located within a preset component area are selected as initial seed points. Based on the initial seed points, the region is expanded in the multi-angle infrared data according to the preset expansion rules to obtain the region image of the specified region; Based on the region image, temperature data of the specified region is extracted from the multi-angle infrared data. Combined with the internal and external parameters of the infrared thermal imager, the temperature data is converted into temperature point cloud data in three-dimensional space using a stereo vision three-dimensional reconstruction method. The temperature point cloud data is divided into grids to generate a surface grid model of the specified area. The temperature change of each node on the surface grid model in different directions is calculated to obtain temperature gradient data. Based on the surface mesh model, a three-dimensional mesh model of the specified region is constructed. According to the temperature gradient data and preset material parameters, the heat change of each unit in the three-dimensional mesh model is calculated, and divergence data is obtained based on the heat change.

[0008] Secondly, this application provides an Internet of Things-based equipment status monitoring system for production workshops, comprising: The acquisition module is used to acquire multi-angle infrared data of the target device and switch data output by the control system of the target device through the Internet of Things platform in the production workshop; The processing module is used to process the multi-angle infrared data using a region growing algorithm, generate three-dimensional temperature data of a specified region on the target device, and calculate the corresponding temperature gradient data and divergence data based on the three-dimensional temperature data. The marking module is used to mark the three-dimensional temperature data, the temperature gradient data, and the divergence data according to the state switching time points in the switch data to form thermal state data; A construction module is used to construct a structural relationship diagram of the target device, and based on the structural relationship diagram and the thermal state data, combined with a graph convolutional neural network, generate a feature vector, which is used to characterize the overall thermal state of the target device. The prediction module is used to predict the temperature change data of the target device in the future time period based on the feature vector and the switch data, using a long short-term memory neural network. The comparison module is used to compare the temperature change data with a dynamic threshold range to generate monitoring results that reflect the real-time health status of the target device.

[0009] Thirdly, this application provides an electronic device, comprising: Memory, used to store computer programs; A processor is used to execute the computer program to implement the steps of the IoT-based production workshop equipment status monitoring method as described in the first aspect above.

[0010] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the steps of the IoT-based production workshop equipment status monitoring method described in the first aspect above.

[0011] The technical solution provided in this application has the following beneficial effects: This application firstly collects infrared and switching signal data synchronously through an IoT platform, providing a foundation for multi-source data fusion analysis. Secondly, it uses a region growing algorithm to extract and reconstruct three-dimensional temperature data of a specified area of ​​the device from multi-angle infrared data, achieving accurate spatial characterization of the thermal field on the device surface. Then, it calculates temperature gradient and divergence data based on the three-dimensional temperature data, thereby physically characterizing the non-uniformity of heat distribution and the characteristics of internal heat sources. Next, it combines the thermal state data marked by switching signals to provide an accurate operating condition background for subsequent analysis. Then, by constructing a device structure relationship diagram and inputting it into a graph convolutional neural network, it achieves fusion learning of the device's physical topology and thermal state data, thereby extracting a feature vector that can characterize the overall thermal state of the device. This feature vector, along with the load history, is then input into a long short-term memory neural network for prediction, achieving accurate inference of future temperature change trends. Finally, by comparing the predicted trend with dynamic thresholds, it achieves early identification and warning of abnormal device states, thus completing precise state monitoring.

[0012] Furthermore, this application also achieves automatic and accurate segmentation of key components of the equipment through region growing, thereby locking in the target area for subsequent analysis; it transforms two-dimensional temperature information into three-dimensional spatial distribution through stereo vision reconstruction, improving the dimensionality and accuracy of thermal state characterization; and it transforms intuitive temperature data into physical quantities such as gradients and divergences, which are more revealing of internal heat conduction anomalies and heat source intensity, through gridded modeling and physical calculations, thus providing more discriminative and physically meaningful feature inputs for subsequent in-depth analysis.

[0013] These or other aspects of this application will become more apparent in the following description of the embodiments. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 A flowchart illustrating an IoT-based method for monitoring the status of equipment in a production workshop, as provided in this application embodiment; Figure 2 A schematic diagram illustrating a specific implementation of an IoT-based equipment status monitoring method for production workshops, provided in this application embodiment; Figure 3 This is a schematic diagram of the structure of an IoT-based equipment status monitoring system for a production workshop, provided as an embodiment of this application. Detailed Implementation

[0016] To address the problems existing in the prior art, this application proposes an IoT-based method for monitoring the condition of equipment in a production workshop. The core idea of ​​this method is as follows: First, multi-angle infrared data and switch data reflecting load changes are synchronously collected from the equipment through an IoT platform. Then, the infrared data is processed using a region growing algorithm to reconstruct the three-dimensional temperature field of a specified area of ​​the equipment, and gradient and divergence data characterizing heat distribution and flow are calculated accordingly. Next, based on the time of switch state changes, the corresponding operating conditions are labeled for the above thermophysical quantity data to form thermal state data. Then, by constructing a structural relationship diagram of the equipment and fusing the structural diagram with the thermal state data using a graph convolutional neural network, a feature vector that can characterize the overall thermal state of the equipment is generated. Afterward, this feature vector and the load history sequence are input into a long short-term memory neural network to predict the future temperature change trend of the equipment. Finally, the predicted trend is compared with a dynamic threshold to generate monitoring results.

[0017] Therefore, this solution enhances the ability to characterize complex thermal states from a holistic perspective by generating a three-dimensional thermophysical field from a two-dimensional infrared image and integrating it with the device topology for analysis. By combining the time-series prediction of load conditions, it achieves a more accurate grasp of the dynamic evolution of thermal states and earlier anomaly identification, thereby solving the problems of insufficient overall thermal state characterization and limited dynamic prediction accuracy in the background technology.

[0018] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some embodiments of the present application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] The core of this application is to provide a method for monitoring the status of equipment in a production workshop based on the Internet of Things (IoT). A flowchart of one specific implementation is shown below. Figure 1 As shown, the method includes: Step 101: Collect multi-angle infrared data of the target device and the switching data output by the control system of the target device through the Internet of Things platform in the production workshop.

[0020] In step 101, the target equipment refers to the key rotating or transmission equipment that needs to be monitored in the production workshop, such as machine tool spindles, gearboxes, or conveyor rollers. This embodiment does not limit this. Multi-angle infrared data refers to a sequence of images simultaneously acquired by multiple infrared thermal imagers placed at different locations around a target device. These images record the temperature distribution of the device surface at different observation angles at the same time. Switching data refers to a set of discrete status signals directly read from the device's own programmable logic controller or upper control system. These signals are used to indicate the device's real-time operating mode, such as the device's start-up, stop, overload alarm, and working status at different speeds.

[0021] In this embodiment, multiple fixed infrared thermal imagers arranged around the target equipment are synchronously triggered by IoT data acquisition nodes deployed on the production workshop site to acquire images, thereby obtaining a set of timestamp-aligned multi-angle infrared data. At the same time, switch data reflecting the current operating status of the equipment are read in real time from the network interface of the control system of the target equipment through a standard industrial communication protocol, and these two types of data are transmitted to the subsequent analysis and processing unit through the workshop IoT platform.

[0022] Step 102: Process the multi-angle infrared data using a region growing algorithm to generate three-dimensional temperature data of a specified region on the target device, and calculate the corresponding temperature gradient data and divergence data based on the three-dimensional temperature data.

[0023] The designated area refers to the key component area on the target device that needs to be monitored, such as the bearing housing or gearbox housing; the three-dimensional temperature data refers to a set of data with temperature attribute values ​​on the three-dimensional spatial coordinates generated by fusing two-dimensional infrared image information from different angles through stereo vision reconstruction methods. This data describes the complete temperature distribution pattern of the surface of the designated area of ​​the target device in three-dimensional space.

[0024] Temperature gradient data is calculated based on three-dimensional temperature data. It can characterize the rate of change of temperature along different directions in space at any point on or inside the surface of a device component, reflecting the direction and intensity of heat transfer in space. Divergence data is a physical quantity calculated based on temperature gradient data and combined with the thermal properties of materials. It quantifies the net inflow or outflow rate of heat per unit volume at a point in three-dimensional space, and is used to indicate whether there is an internal abnormal heat source or heat accumulation at that point.

[0025] This embodiment does not limit the specific process of the region growing algorithm. For example, in this embodiment, step 102 includes the following process, such as... Figure 2 As shown: Step 1021: Select pixels with a temperature value greater than a preset temperature threshold and located within a preset component area from the multi-angle infrared data as initial seed points.

[0026] In step 1021, the preset temperature threshold is a pre-set temperature value used to initially screen out pixels that may be in an abnormally high temperature state. In this embodiment of the application, the value of the preset temperature threshold is not specifically limited, and can be set according to the actual situation.

[0027] The preset component region is a region of interest predefined in the image based on the physical structure of the target device. Its function is to constrain the spatial position of the seed point and ensure that the segmentation process focuses on the target component. The preset component region provides spatial guidance and initial range for automatic and accurate segmentation of the specified region.

[0028] In this embodiment, by traversing the pixels in each frame of multi-angle infrared data, it is first determined whether the temperature value of the pixel is greater than a preset temperature threshold, and then it is further determined whether the coordinates of the pixel in the image fall within the boundary range of a preset component area. Pixels that meet both conditions are collected to form an initial seed point set for subsequent region growth.

[0029] In practical applications, such as monitoring a speed reducer, the preset temperature threshold can be set to 60 degrees Celsius, and the preset component area can be defined as a rectangular image range including the bearing housing of the speed reducer. In the infrared image acquired at a certain moment, all pixels in the image are traversed, and all pixels with a temperature value greater than 60 degrees and whose coordinates are within the rectangular range are selected as the initial seed points for this area growth.

[0030] Step 1022: Based on the initial seed point, perform region expansion in the multi-angle infrared data according to the preset expansion rules to obtain the region image of the specified region.

[0031] In step 1022, the preset expansion rule defines the specific conditions for region growth. The expansion rule may include temperature similarity threshold and spatial adjacency relationship, such as four-neighborhood, eight-neighborhood, etc. The region image is a binary image, in which pixels belonging to the specified region are marked as foreground and other pixels are marked as background.

[0032] In this embodiment, the region expansion begins with the initial seed point set obtained in step 1021. For each boundary pixel that already belongs to the region, its four-neighbor or eight-neighbor pixels on the image are checked. If a neighboring pixel has not yet been assigned to the region, and the absolute value of the difference between its temperature value and the average temperature value of the current region is less than a preset temperature similarity threshold, then the neighboring pixel is merged into the current region. This expansion process is iterated until no new pixel can meet the merging condition, and finally a clear binary region image that marks the specified region of the target device is generated.

[0033] In practical applications, a temperature similarity threshold of 5 degrees Celsius is set, and an eight-neighbor connection method is used for expansion. Starting from the initial seed point, neighboring pixels are continuously checked. If the temperature difference between the neighboring pixels and the average temperature of the current region is less than 5 degrees, the neighboring pixels are included. This process is repeated until the region no longer expands, and finally a binary image that only highlights the target bearing seat region is obtained.

[0034] Step 1023: Based on the region image, extract the temperature data of the specified region from the multi-angle infrared data, and combine the internal and external parameters of the infrared thermal imager to convert the temperature data into temperature point cloud data in three-dimensional space using a stereo vision three-dimensional reconstruction method.

[0035] Temperature data is a set of pixel coordinates and their corresponding temperature values; the internal parameters of the infrared thermal imager include focal length and imaging origin, which are used to describe the imaging geometry of the infrared thermal imager itself; the external parameters of the infrared thermal imager include the position and orientation of the thermal imager in the workshop world coordinate system, which are used to describe its observation angle; temperature point cloud data refers to a set of a large number of three-dimensional spatial points and their additional temperature attribute values.

[0036] This embodiment does not limit the specific implementation process of the stereo vision 3D reconstruction method. For example, step 1023 may specifically include the following steps: A1: Based on the pixel positions marked as the specified region in the regional image, extract the temperature values ​​of the pixel positions from the corresponding multi-angle infrared data to form temperature data.

[0037] In this embodiment of the application, the region image generated in step 1022 is first read, and the positions of all pixels marked as foreground are identified. Then, based on the coordinates of these pixels, the temperature values ​​recorded at the same coordinate positions are searched and extracted from the original multi-angle infrared data frame to form a set of temperature data. Each entry in the temperature data contains the two-dimensional image coordinates of the pixel and a temperature value.

[0038] In practical applications, if 3,000 pixels in the area image are marked as belonging to the bearing housing area, then based on the coordinates of these 3,000 pixels, 3,000 temperature readings are extracted from the corresponding infrared image to form temperature data containing 3,000 records.

[0039] A2: Using the internal parameters of the infrared thermal imager, the image coordinates of each pixel in the specified area are corrected to obtain the corrected pixel coordinates.

[0040] In this embodiment, pre-calibrated internal parameters of the infrared thermal imager are obtained, including radial distortion coefficient and tangential distortion coefficient; a distortion correction model is constructed using these parameters; then, the original image coordinates of each pixel in the temperature data are input into the correction model to calculate the corresponding ideal image coordinates that have eliminated lens distortion, i.e., the corrected pixel coordinates.

[0041] In practical applications, assuming the original coordinates of a pixel are [400, 300], after calculation and correction using known distortion coefficients, its corrected coordinates become [399.8, 300.1].

[0042] A3: Based on the external parameters of the infrared thermal imager, the corrected pixel coordinates are converted into spatial projection rays originating from the optical center of the thermal imager.

[0043] In this embodiment, the external parameters of the thermal imager corresponding to the frame of infrared data are obtained. These parameters describe the three-dimensional position and rotational attitude of the thermal imager in the workshop coordinate system. Using these external parameters and the internal parameters of the thermal imager, a projection relationship from two-dimensional image coordinates to a three-dimensional spatial ray is established. For each entry in the temperature data, its corrected pixel coordinates are substituted into the projection relationship to calculate a ray equation in three-dimensional space that originates from the optical center of the thermal imager and passes through the spatial direction corresponding to the pixel. This ray represents the three-dimensional spatial direction in which the object surface point observed by the pixel may be located.

[0044] In practical applications, for a pixel with corrected coordinates of [399.8, 300.1], combined with the location coordinates of the thermal imager [1.5, 2.0, 3.0] meters and its orientation, a spatial ray direction vector [0.2, -0.1, 0.975] is obtained through projection calculation. The projection calculation process can refer to relevant technologies, so this embodiment will not elaborate on the projection calculation process.

[0045] A4: Pair the spatial projection rays corresponding to the same pixel position in the specified area from infrared data at different angles to obtain ray pairs.

[0046] In step A4, the same pixel position refers to a pixel that actually corresponds to the same physical point on the surface of the device in three-dimensional space, and the same pixel position has different two-dimensional coordinates in multiple images from different angles.

[0047] In this embodiment of the application, based on the premise that the same physical component is marked in the area image obtained in step 1023 under different angle views, two spatial projection rays from two thermal imagers at different angles that observe the same physical point on the surface of the device are found through feature matching or epipolar geometric constraints, and these two spatial projection rays are paired to form a ray pair.

[0048] In practical applications, a ray obtained from a thermal imager at angle A and another ray obtained from a thermal imager at angle B are determined to be observing the same physical point by calculating their shortest distance or by using the pre-calibrated geometric relationship between the cameras, thus successfully pairing them.

[0049] A5: For each pair of rays, determine the intersection point of the two spatial projection rays in three-dimensional space, and assign the corresponding temperature value from the temperature data to the intersection point.

[0050] In the embodiments of this application, for each ray pair, the intersection point of the two rays in three-dimensional space is solved. Due to measurement errors, the two rays are usually not strictly intersecting. Therefore, the most likely three-dimensional intersection point coordinates are estimated by calculating the midpoint of the shortest line segment between the two rays or by using the least squares method. After determining the three-dimensional coordinates, a temperature value can be selected from the two temperature data records corresponding to the ray pair or its average value can be calculated, and the temperature value is assigned to the estimated three-dimensional space point.

[0051] In practical applications, if the coordinates of a ray pair of estimated three-dimensional spatial points are [0.5, 1.0, 0.2] meters and the corresponding average temperature is 65 degrees Celsius, then the three-dimensional point is recorded as a data point with coordinates [0.5, 1.0, 0.2] and a temperature of 65 degrees.

[0052] A6: Combine all the intersection points and their corresponding temperature values ​​to generate temperature point cloud data.

[0053] In this embodiment of the application, the coordinates of the three-dimensional spatial points and their associated temperature values ​​obtained from processing all ray pairs in step A5 are collected together to form a structured dataset as temperature point cloud data; each data point in this dataset contains three-dimensional coordinates and temperature values, which together describe the three-dimensional spatial morphology and temperature distribution of the surface of the specified area of ​​the target device.

[0054] In practical applications, after processing images from multiple angles, a temperature point cloud dataset containing tens of thousands of three-dimensional points may eventually be generated for subsequent analysis.

[0055] Step 1024: Divide the temperature point cloud data into a grid to generate a surface grid model of the specified area, and calculate the temperature change of each node on the surface grid model in different directions to obtain temperature gradient data.

[0056] In step 1024, the surface mesh model is a network of many interconnected triangular or quadrilateral patches that can approximately represent the three-dimensional surface shape of the specified region; a node is the vertex of these triangular or quadrilateral patches.

[0057] In this embodiment of the application, the temperature point cloud data generated in step 1023 is first processed into a triangular mesh to generate a continuous surface mesh model. Then, for each node on the mesh model, the temperature change rate of the node in the X, Y, and Z directions in the three-dimensional spatial coordinate system is calculated by the difference method based on the spatial coordinates and temperature values ​​of its neighboring nodes. The change rates in these three directions together constitute the temperature gradient data at the node.

[0058] In practical applications, for a node of a surface mesh model with coordinates [x, y, z] and temperature T, the rate of temperature change in the X direction dT / dx is obtained by calculating the difference between its coordinates and temperature in the X direction and its neighboring nodes. Similarly, the rate of temperature change in the Y direction dT / dy and the rate of temperature change in the Z direction dT / dz are obtained. These three values ​​constitute the temperature gradient vector of that point.

[0059] Step 1025: Based on the surface mesh model, construct a three-dimensional mesh model of the specified region. According to the temperature gradient data and preset material parameters, calculate the heat change of each unit in the three-dimensional mesh model, and obtain divergence data based on the heat change.

[0060] In step 1025, the three-dimensional mesh model is a mesh that is composed of many small cubes or other polyhedral units, filling the three-dimensional space inside and near the surface of the specified area, while the surface mesh model is a curved network composed of two-dimensional patches, which only describes the surface geometry of the object; the preset material parameters include at least the thermal conductivity of the material of the device component. A cell is a basic volume element that makes up a three-dimensional mesh model. It discretizes a continuous space into multiple small, regular geometric shapes, such as cubes or tetrahedrons. Each cell is used to independently calculate local physical properties. Cells are usually divided in three-dimensional space in a structured or unstructured array form according to a preset spatial resolution.

[0061] In this embodiment, the surface mesh model obtained in step 1024 is used as the boundary to extend into the interior and adjacent space, dividing it into a finer three-dimensional mesh. For each cell in the three-dimensional mesh, based on the temperature gradient data of the nodes located on the boundary of the cell calculated in step 1024 and the preset thermal conductivity of the component material, Fourier's law of heat conduction can be applied to calculate the heat flow through each surface of the cell. Then, the difference between the total heat flow into the cell and the total heat flow out of the cell is calculated as the net heat change of the cell per unit time. Finally, the ratio of the net heat change to the volume of the cell is used as divergence data characterizing whether the heat at that point is concentrated or diffused.

[0062] In practical applications, for a cubic element, given the temperature gradient along the normal direction on its six faces and the thermal conductivity k of the material, the heat flow through each face can be calculated using relevant formulas. The sum of the heat flows through all faces is then taken as the net heat flow. Then, by combining this with the volume V of the unit, the divergence value is obtained. .

[0063] This method reconstructs a three-dimensional temperature field from multidimensional infrared data and calculates its gradient and divergence to achieve a refined physical characterization of the internal heat distribution and abnormal heat source state of the equipment, thus providing a deeper and more sensitive state feature input for predictive maintenance.

[0064] Step 103: Based on the state switching time points in the switch data, mark the three-dimensional temperature data, the temperature gradient data, and the divergence data to form thermal state data.

[0065] In step 103, the state switching time point refers to the moment when the operating state of the equipment changes clearly, as identified from the switch data, such as when the equipment changes from stopped to started, or from normal operation to overload protection; thermal state data is a structured data set that integrates multiple physical quantities and has operating condition labels. This set associates three-dimensional temperature data, temperature gradient data, and divergence data with the operating conditions of the equipment at that time.

[0066] In this embodiment, the switching data stream from the equipment control system is first parsed to accurately identify the signal transition edges representing state changes and record their corresponding timestamps as state switching time points. Then, the three-dimensional temperature data, temperature gradient data, and divergence data generated in step 102 are acquired, and their respective acquisition timestamps are compared and aligned with the identified state switching time points. Next, based on the comparison results, each set of time-aligned physical quantity data is assigned a corresponding operating condition label, such as "steady-state operation after startup" or "overload condition". Finally, all data entries with operating condition labels are organized and encapsulated to form a complete thermal state data record, so that the subsequent analysis model can distinguish the equipment thermal behavior patterns under different operating loads.

[0067] Step 104: Construct a structural relationship diagram of the target device, and based on the structural relationship diagram and the thermal state data, generate a feature vector by combining a graph convolutional neural network. The feature vector is used to characterize the overall thermal state of the target device.

[0068] Among them, the structural relationship diagram is a graphical model used to express the physical connection relationship and heat conduction path between the components inside the target device; the feature vector is a fixed-dimensional numerical array that abstractly represents the overall thermal state of the device at a certain moment by integrating the structural information and thermal state information of the device.

[0069] In the structural design of a graph convolutional neural network, the first graph convolutional layer can be designed with 32 output channels, and its neighborhood aggregation weight matrix dimension matches the input feature dimension, using ReLU as the activation function of the first activation layer; the second graph convolutional layer can be designed with 64 output channels, and a skip connection is set from the output of the first activation layer to the output of this layer to achieve residual connection, also using ReLU as the activation function of the second activation layer; the graph pooling layer can use global average pooling operation to aggregate the feature vectors of multiple nodes into a single device-level feature vector; the fully connected layer can be designed as two layers, the first layer maps the dimension of the device-level feature vector from 64 to 32, and the second layer further maps it to the dimension of the final required feature vector, such as 16 dimensions, and uses the ReLU activation function between the two layers.

[0070] It should be noted that the above structure is exemplary. This application does not impose specific limitations on the layer and other structural designs used in the internal structure of the graph convolutional neural network, and corresponding settings can be made according to the actual situation.

[0071] In this embodiment, step 104 includes the following process: Step 1041: Based on the physical connection relationship and heat conduction path of the target device, establish a structural relationship diagram with components as nodes and connection or heat transfer relationship as edges.

[0072] In step 1041, physical connection relationship refers to the assembly relationship between various mechanical components inside the target equipment in the production workshop that are directly or indirectly fixed and coupled by physical means, such as mechanical connection achieved by bolt fastening, keyway mating, bearing nesting or gear meshing; The heat conduction path refers to the actual or equivalent channel through which heat is transferred from a high-temperature region to a low-temperature region during equipment operation due to component contact or the presence of a medium. This can be, for example, through direct contact conduction of metal components, indirect transfer through lubricating grease, or through heat exchange paths formed in the adjacent space by radiation and convection. Nodes represent independent physical components that make up the target device, such as bearings, gears, and shafts; edges represent physical connections between components, such as bolted connections, keyed connections, or known heat conduction paths.

[0073] In this embodiment, the key components that need to be monitored are first identified based on the mechanical assembly drawings or three-dimensional model of the target equipment, and each key component is defined as a node. Then, the actual physical connection between these components and the possible heat conduction paths are analyzed, such as direct contact conduction or indirect heat transfer through lubricating oil. Two components with connection or heat transfer relationship are connected by an edge. Finally, a network diagram consisting of multiple nodes and edges is formed as a structural relationship diagram, which accurately expresses the physical topology inside the equipment.

[0074] Step 1042: Use the multidimensional thermophysical quantities corresponding to each component in the thermal state data as the initial node features of the corresponding node.

[0075] In step 1042, the multidimensional thermophysical quantity refers to a set of parameter data with different physical meanings obtained from the monitoring of the target device and used to comprehensively describe its thermal state. This set includes at least three-dimensional temperature data characterizing the spatial distribution of temperature, temperature gradient data characterizing the rate of change of temperature space, and divergence data characterizing the intensity of local heat accumulation or diffusion. The initial node feature refers to a set of values ​​associated with each node in the structural relationship diagram, used to describe the thermal state of the component at the current moment.

[0076] In this embodiment of the application, for each component represented by a node in the structural relationship diagram, the three-dimensional temperature data, temperature gradient data, and divergence data corresponding to the component at the current moment are retrieved from the thermal state data; then these data are merged into a one-dimensional numerical array as the initial node feature of the node at the current moment.

[0077] Step 1043: Input the structural relationship graph and the initial node features of each node into the graph convolutional neural network. Through the first graph convolutional layer of the graph convolutional neural network, based on the edges in the structural relationship graph, aggregate the self-feature of each node and the features of adjacent nodes to generate initial component features.

[0078] In step 1043, the first graph convolutional layer is a computational layer in a graph convolutional neural network. Its function is to aggregate neighborhood information for the features of each node based on the edge connection relationship of the graph.

[0079] In this embodiment of the application, the structural relationship graph constructed in step 1041 and the initial node features assigned in step 1042 are fed into a pre-trained graph convolutional neural network as input; in the first graph convolutional layer of the neural network, the algorithm traverses each node in the structural relationship graph and reads its initial node features; at the same time, based on all the adjacent nodes directly connected to the node in the graph through edges, the initial node features of these adjacent nodes are read. Then, the features of the node itself are fused with the features of all its neighboring nodes through a preset weighted summation function, thereby generating an updated feature representation for each node that includes its own information and that of its direct neighbors. As the initial component features, the expression of the weighted summation function can be set with reference to relevant technologies. This embodiment will not elaborate on the expression of the weighted summation function.

[0080] Step 1044: The initial component features are processed through the first activation layer of the graph convolutional neural network, and neighborhood information aggregation is performed based on the processed component features through the second graph convolutional layer of the graph convolutional neural network. A residual connection mechanism is introduced to fuse the processed component features with the aggregation result to generate extended component features.

[0081] In step 1044, the first activation layer is a nonlinear transformation layer located after the first graph convolutional layer in the graph convolutional neural network. Its function is to apply a nonlinear mapping to the linear features output by the first graph convolutional layer in order to enhance the model's ability to fit and express complex thermal state patterns. The second graph convolutional layer is another graph convolutional computation layer located after the first activation layer in a graph convolutional neural network. Its function is to aggregate neighborhood information again based on the device structure relationship graph, aiming to capture the dependencies between nodes at greater distances. It is often used in combination with residual connection mechanisms to fuse shallow features.

[0082] Extended component features refer to the node feature representation generated after processing by the second graph convolutional layer and fusing residual information. It not only includes the thermal state information of the component itself and its direct neighbors, but also integrates a wider range of thermal influence information from indirectly connected components, thus forming a richer and deeper representation of the component in the overall thermal topology of the equipment.

[0083] In this embodiment, the initial component features output by the first graph convolutional layer are passed to a non-linear activation function layer for processing to enhance the expressive power of the features. Then, the processed features are fed into the second graph convolutional layer, which will again perform a new round of neighborhood information aggregation on the features of each node according to the structural relationship graph, so that each node can capture the influence of its neighbors outside its "two hops". Meanwhile, this step introduces a residual connection mechanism. Specifically, the processed component features output by the first activation layer are directly added to the result obtained by the aggregation of the second graph convolutional layer. Through this addition and fusion, the important original information of the node after one layer of processing is preserved, and new information from a wider range of neighborhoods is integrated, ultimately generating extended component features that can characterize more complex relationships between components.

[0084] Step 1045: The extended component features are processed through the second activation layer of the graph convolutional neural network, and the processed extended component features of all components in the structural relationship graph are integrated through the graph pooling layer of the graph convolutional neural network to output device-level features.

[0085] In this embodiment, the extended component features are further transformed nonlinearly through the second activation layer; then, the transformed extended component features of all nodes are input to the graph pooling layer; the graph pooling layer is used to globally integrate the features of all nodes in the entire graph. For example, this layer can use a symmetric aggregation function to take the maximum or average value of each element of the feature vector of all nodes, compressing multiple feature vectors that were originally equal in number to the number of nodes into a single, fixed-dimensional global feature vector as a device-level feature. This feature abstractly represents the thermal state of the entire device at the current moment.

[0086] Step 1046: Transform the device-level features through the fully connected layer of the graph convolutional neural network to generate a feature vector.

[0087] In this embodiment, the device-level features output by the graph pooling layer are input to one or more fully connected layers at the end of the graph convolutional neural network. The fully connected layers reduce the dimensionality, refine and format the device-level features through a series of linear transformations and nonlinear activations, mapping them into a more compact numerical vector that is more suitable for the input of the subsequent time-series prediction network, which serves as a feature vector to characterize the overall thermal state of the target device.

[0088] In step 104, this application utilizes a graph convolutional neural network to deeply fuse the physical structure topology of the device with multi-source thermal state data, thereby generating a feature vector that can comprehensively and accurately reflect the overall thermal coupling state of the device, laying a crucial and high-quality feature foundation for subsequent accurate time-series prediction.

[0089] Step 105: Based on the feature vector and the switching data, predict the temperature change data of the target device in the future time period using a long short-term memory neural network.

[0090] Among them, temperature change data refers to a sequence of temperature values ​​of the target device at a series of consecutive moments in the future, predicted by a neural network.

[0091] This application does not impose specific limitations on the structural design of layers and other components used in the internal structure of the Long Short-Term Memory Neural Network, and these can be set according to the actual situation.

[0092] In this embodiment, step 105 includes the following process: Step 1051: Arrange the switch data into a load sequence according to the time order, and combine the feature vectors at the same time with the load sequence to obtain a combined sequence.

[0093] In step 1051, the load sequence refers to a discrete signal sequence formed by arranging switch data in chronological order of their occurrence; the combined sequence refers to a new input sequence formed by splicing the feature vectors at the same timestamp with the corresponding data points in the load sequence, which integrates the overall thermal state characteristics of the equipment and the operating load state.

[0094] In this embodiment, firstly, switch data within the same time period as the feature vector generated in step 104 are extracted from a historical database or real-time data stream; then, these switch data are sorted in strictly ascending order of timestamps to form a load sequence, where each element represents a device load status code at a specific moment; next, the feature vector at each moment is concatenated with the load status code at the corresponding moment in the load sequence, that is, their ends are connected to form a longer vector, thereby obtaining a new combined sequence containing two parts of information at each time point. This combined sequence will be used as the input of a long short-term memory neural network.

[0095] Step 1052: Input the combined sequence into the memory unit of the long short-term memory neural network. In the memory unit, the information of the combined sequence at the current time and past time is processed through a gating mechanism to form a long-term time dependency.

[0096] In step 1052, the memory unit is the core computing unit of the long short-term memory neural network, which contains a gating structure for controlling the flow of information. The gating mechanism usually includes an input gate, a forget gate, and an output gate. These three structures work together to determine which historical information needs to be retained, which new information needs to be written, and what information should be output at the current moment. Long-term time dependency refers to the neural network's ability to learn and remember information associations that span a long historical time interval.

[0097] In this embodiment of the application, the combined sequence generated in step 1051 is input into the memory unit of the pre-trained long short-term memory neural network step by step in time sequence; for the input of the current time step, the memory unit first calculates a value between 0 and 1 through the forget gate, which determines how much information in the unit state of the previous time step needs to be discarded. Then, two values ​​are calculated through the input gate: one determines how much of the current input needs to be written, and the other generates a candidate new information based on the current input. Next, the memory unit adds the historical unit state decayed by the forget gate to the new candidate information filtered by the input gate, thereby updating the current unit state. This update process enables the network to selectively retain long-term historical information. Finally, output gates control how much information from the current cell state is output to the hidden state at the current time step; through this series of gating operations iterating at each time step, the memory cell gradually builds and maintains an internal state that can capture long-term temporal dependencies.

[0098] Step 1053: Based on the long-term time dependency, the memory unit outputs the predicted temperature value for the next moment.

[0099] In this embodiment of the application, after the information of the last historical time step of the combined sequence has been processed, a final hidden state that integrates the information of the entire historical sequence has been formed inside the memory unit. Based on this final hidden state, the neural network performs a linear transformation through its top fully connected output layer to map the high-dimensional hidden state into a specific value, which serves as the temperature prediction value of the target device at the first moment in the next future.

[0100] In practical applications, after processing all historical time steps, the final hidden state is assumed to be a 128-dimensional vector. This vector is input into a fully connected layer, which maps it to a single scalar value, such as the output value 70.5, which is interpreted as the predicted target device temperature at the next time step, in degrees Celsius.

[0101] Step 1054: Input the predicted temperature value at the next moment as new input information into the memory unit, repeat the gating mechanism process to obtain multiple predicted temperature values ​​at future moments, and combine the multiple predicted temperature values ​​at future moments to form temperature change data.

[0102] In this embodiment of the application, in order to predict a more distant future moment, an iterative prediction method is adopted; specifically, the temperature prediction value of the first future moment obtained in step 1053 is fed back to the memory unit of the long short-term memory neural network as part of a new "input"; since there is no real load data to predict the future moment, it can be assumed that the load state remains unchanged or its last known value can be used, and it can be combined with this predicted temperature value to form a new input vector for predicting the next moment. Then, starting from the updated memory cell state, the gating mechanism described in step 1052 is executed again on this new input vector to output the temperature prediction value for the second future moment. This process can be repeated multiple times as needed, with the latest prediction value fed back each time to predict the next moment, thereby obtaining the prediction values ​​for the third and fourth future moments in sequence, until the preset number of future moments is reached. Finally, all the predicted future temperature values ​​are arranged into a sequence in chronological order as the required temperature change data.

[0103] This application, through step 105, utilizes a long short-term memory neural network to model and perform multi-step iterative prediction on time-series data that integrates the overall thermal state and load history of the equipment, thereby achieving accurate prediction of the temperature change trend of the equipment over a future period and providing key and forward-looking data for early warning.

[0104] Step 106: Compare the temperature change data with the dynamic threshold range to generate a monitoring result that reflects the real-time health status of the target device.

[0105] The dynamic threshold range is a normal temperature fluctuation range that changes with time and equipment operating conditions, rather than a fixed value. It is obtained by analyzing historical normal data. The monitoring result is a structured output used to clearly indicate whether there is an abnormality in the equipment and the general situation of the abnormality.

[0106] In this embodiment, step 106 includes the following process: Step 1061: Obtain historical temperature data of the target device during normal operation.

[0107] In step 1061, historical temperature data refers to a series of temperature measurements recorded when the target device was previously confirmed to be in a fault-free and healthy operating state.

[0108] In this embodiment of the application, all relevant temperature data of the target device during normal operation and without alarm records within a preset historical time window are retrieved and extracted from the data warehouse of the storage device's operating history. These data can come from the prediction history in step 105 or from the direct measurement records of the basic temperature sensor, thereby forming a historical temperature dataset for subsequent statistical analysis.

[0109] Step 1062: Analyze the historical temperature data to obtain the dynamic threshold range that changes over time.

[0110] In this embodiment of the application, the historical temperature data obtained in step 1061 is first cleaned and organized; then, considering that the equipment temperature will change periodically with factors such as load, ambient temperature, and operating period, a time series analysis method is adopted, for example, the data is grouped by hour, day or load condition; for each specific time period or operating condition group, the statistical characteristics of the historical temperature data in the group are calculated, for example, the average value of the data in the group is calculated as the reference temperature of the time period, and the standard deviation of the data in the group is calculated. Next, based on the reference temperature and standard deviation, a reasonable offset is set to form the upper and lower temperature limits corresponding to that time period. Finally, the upper and lower limits corresponding to each time period or operating condition group are connected to form a normal temperature range that fluctuates dynamically with time or operating conditions, i.e., the dynamic threshold range.

[0111] Step 1063: When the temperature change data exceeds the dynamic threshold range within a preset time period, a monitoring result is generated.

[0112] In step 1063, the preset time period is a continuous time length used to determine persistent anomalies, such as three consecutive predicted time points.

[0113] In this embodiment, the temperature change data predicted in step 105 for multiple future moments are compared with the dynamic threshold range for the corresponding moment obtained in step 1062. Specifically, it is checked whether each predicted temperature value falls within the dynamic threshold range for the corresponding moment. If multiple consecutive predicted temperature values ​​are found to exceed the upper or lower limit of the dynamic threshold range for their corresponding moment, it is determined that the equipment may experience a continuous abnormality. Once the abnormality is determined to have occurred, a monitoring result is generated. This monitoring result typically includes key information such as the time when the abnormality started, the abnormal temperature value, and the extent to which the threshold was exceeded, thereby providing the operator with a clear status alarm.

[0114] In this embodiment, after step 106, the following process is also included: B1: Based on the abnormal start time and end time recorded in the monitoring results, extract the time sequence data of the target device from the preset database during the time from the abnormal start time to the end time.

[0115] In step B1, the preset database is a relational or time-series database used to store the device's comprehensive historical operating data.

[0116] In this embodiment of the application, after the monitoring result is generated in step 1063, the monitoring result will record the start and end times of the abnormality determined by the system; based on these two time points, a query is automatically initiated to the preset database to request the extraction of all relevant time series data recorded by the target device between these two time points; these time series data are a more original or richer set than thermal state data, and can usually include the original switch data stream, the time series of three-dimensional temperature data, the time series of temperature gradient data, and the time series of divergence data, etc., providing a complete data slice for subsequent root cause analysis.

[0117] B2: Input the time series data into the isolated forest model, calculate the anomaly value of each time point in the time series data through the isolated forest model, and merge all time point data with anomaly values ​​greater than a preset anomaly threshold into an anomaly data segment.

[0118] In step B2, the isolated forest model is an unsupervised anomaly detection algorithm. Its basic principle is that "abnormal data points are more likely to be isolated by randomly divided decision trees". The construction of this model is a conventional technique, and this embodiment does not limit it. The anomaly score is a quantitative score. The higher the score, the more likely the data point is to be an anomaly.

[0119] In this embodiment, the multidimensional time-series data extracted in step B1 is first organized into a set of data points according to time points, where each data point is a vector containing multiple features. Then, this set of data points is input into a pre-trained isolation forest model. The isolation forest model contains multiple randomly generated "isolation trees". The model evaluates the degree of anomalousness by calculating the average path length required for each data point to be isolated by these "isolation trees". The shorter the path, the higher the anomalous score, which is the anomalousness value. Next, a preset anomalousness threshold is set, and all time points with anomalousness values ​​exceeding the threshold are marked. Finally, the temporally continuous time point data marked as anomalous are merged together to form one or more continuous anomalous data segments, where each anomalous data segment represents the duration range of a suspected anomalous event.

[0120] B3: For each of the abnormal data segments, extract the corresponding feature data segments from the time-series data.

[0121] In step B3, the characteristic data segment refers to the portion of data extracted from the complete time-series data that is perfectly aligned in time with the abnormal data segment.

[0122] In this embodiment of the application, for each abnormal data segment identified in step B2, a data segment with the exact same time span is extracted from the original time-series data extracted in step B1 based on its start and end time points; this extraction operation is precise, ensuring that the data targeted by the subsequent analysis is completely consistent with the detected abnormal event in time.

[0123] B4: Input each of the feature data segments into a one-dimensional convolutional neural network classifier, and output a probability vector through the one-dimensional convolutional neural network classifier.

[0124] In step B4, for example, the structure of a one-dimensional convolutional neural network classifier is designed as follows: The input layer receives feature data segments; the first convolutional layer uses 32 convolutional kernels with a width of 3 for feature extraction and employs the ReLU activation function, followed by a max pooling layer with a width of 2; the second convolutional layer uses 64 convolutional kernels with a width of 3 and employs ReLU activation, followed by a global average pooling layer with a width of 2 to compress the feature map into a vector; finally, two fully connected layers are connected. The first fully connected layer has 128 neurons and is activated by ReLU, and the second fully connected layer has the same number of neurons as the predefined number of fault categories and uses the Softmax activation function to output a probability vector.

[0125] In this embodiment of the application, each feature data segment obtained in step B3 is taken as an independent input sample and input into a pre-trained one-dimensional convolutional neural network classifier. The classifier first slides its convolutional layer in the time dimension to extract local pattern features at different time scales in the feature data segment. Then, these features are downsampled through pooling layers to enhance robustness; finally, the extracted high-level features are mapped to the probability space of the fault category through a fully connected layer, and a probability vector is output; for example, if there are three predefined fault types, such as "bearing wear", "poor lubrication" and "misalignment", then the output probability vector may be [0.7, 0.2, 0.1], indicating that there is a 70% probability that the abnormal segment is caused by bearing wear.

[0126] B5: Based on the probability vector, determine the fault type with the highest probability, and combine the time information, anomaly degree value and confidence level of the fault type of the abnormal data segment to generate a diagnostic report, which is output through the IoT platform of the production workshop.

[0127] In this embodiment, the probability vector output in step B4 is parsed to find the probability value with the largest value and its corresponding fault type label. This type is taken as the main suspected fault of this anomaly, and its probability value is the confidence level. Then, by combining the occurrence time of the abnormal data segment, the overall or average anomaly value of the data segment, and the confidence level of the main suspected fault, a structured text or JSON format diagnostic report is generated. The content of this diagnostic report is more in-depth than the monitoring results generated in step 1063. It not only informs that "there is an anomaly," but also gives "what kind of anomaly it may be" and "how high the confidence level of the anomaly is." Finally, through the data interface of the workshop IoT platform, this diagnostic report is pushed to the workshop monitoring center or the terminal devices of relevant maintenance personnel to assist in making maintenance decisions.

[0128] This application, through step 106 and subsequent processes, not only achieves rapid judgment and alarm of abnormal states based on dynamic thresholds, but also conducts in-depth analysis of alarm events and preliminary identification of fault types by combining isolated forests and convolutional neural networks for secondary analysis. This upgrades simple state monitoring into intelligent analysis with certain diagnostic capabilities, improving the pertinence and efficiency of maintenance decisions.

[0129] Figure 3 A schematic diagram of an IoT-based equipment status monitoring system for a production workshop is provided as an embodiment of this application, as shown below. Figure 3 As shown, the system includes: The acquisition module 31 is used to acquire multi-angle infrared data of the target device and switch data output by the control system of the target device through the Internet of Things platform in the production workshop.

[0130] The processing module 32 is used to process the multi-angle infrared data using a region growing algorithm to generate three-dimensional temperature data of a specified region on the target device, and to calculate the corresponding temperature gradient data and divergence data based on the three-dimensional temperature data.

[0131] The marking module 33 is used to mark the three-dimensional temperature data, the temperature gradient data and the divergence data according to the state switching time points in the switch data to form thermal state data.

[0132] The construction module 34 is used to construct the structural relationship diagram of the target device, and based on the structural relationship diagram and the thermal state data, combined with a graph convolutional neural network, generate a feature vector, which is used to characterize the overall thermal state of the target device.

[0133] The prediction module 35 is used to predict the temperature change data of the target device in the future time period based on the feature vector and the switching data through a long short-term memory neural network.

[0134] The comparison module 36 is used to compare the temperature change data with a dynamic threshold range to generate a monitoring result that reflects the real-time health status of the target device.

[0135] The IoT-based production workshop equipment status monitoring system of this application is used to implement the aforementioned IoT-based production workshop equipment status monitoring method. Therefore, the specific implementation of the IoT-based production workshop equipment status monitoring system can be found in the embodiment section of the IoT-based production workshop equipment status monitoring method above. The specific implementation can be referred to the description of the corresponding embodiments, and will not be repeated here.

[0136] This application also provides an electronic device, including: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of any of the above-described IoT-based production workshop equipment status monitoring methods.

[0137] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of any of the above-described IoT-based production workshop equipment status monitoring methods.

[0138] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memory, random access memory, portable hard drives, magnetic disks, or optical disks.

[0139] The embodiments of this application also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the above embodiments of the IoT-based production workshop equipment status monitoring method.

[0140] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0141] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in one or more embodiments of this specification are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of related data must comply with relevant laws, regulations and standards, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0142] The foregoing has provided a detailed description of the IoT-based equipment status monitoring method and system for production workshops provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A method for monitoring the status of equipment in a production workshop based on the Internet of Things, characterized in that, include: The IoT platform in the production workshop collects multi-angle infrared data of the target device and switching data output by the control system of the target device. The multi-angle infrared data is processed using a region growing algorithm to generate three-dimensional temperature data of a specified region on the target device, and the corresponding temperature gradient data and divergence data are calculated based on the three-dimensional temperature data. Based on the state switching time points in the switch data, the three-dimensional temperature data, the temperature gradient data, and the divergence data are marked to form thermal state data; A structural relationship diagram of the target device is constructed, and based on the structural relationship diagram and the thermal state data, a feature vector is generated by combining a graph convolutional neural network. The feature vector is used to characterize the overall thermal state of the target device. Based on the feature vector and the switching data, the temperature change data of the target device in future time periods is predicted by a long short-term memory neural network. The temperature change data is compared with a dynamic threshold range to generate monitoring results that reflect the real-time health status of the target device.

2. The method according to claim 1, characterized in that, The process of using a region growing algorithm to process the multi-angle infrared data to generate three-dimensional temperature data for a specified region on the target device, and calculating the corresponding temperature gradient data and divergence data based on the three-dimensional temperature data, includes: From the multi-angle infrared data, pixels with a temperature value greater than a preset temperature threshold and located within a preset component area are selected as initial seed points. Based on the initial seed points, the region is expanded in the multi-angle infrared data according to the preset expansion rules to obtain the region image of the specified region; Based on the region image, temperature data of the specified region is extracted from the multi-angle infrared data. Combined with the internal and external parameters of the infrared thermal imager, the temperature data is converted into temperature point cloud data in three-dimensional space using a stereo vision three-dimensional reconstruction method. The temperature point cloud data is divided into grids to generate a surface grid model of the specified area. The temperature change of each node on the surface grid model in different directions is calculated to obtain temperature gradient data. Based on the surface mesh model, a three-dimensional mesh model of the specified region is constructed. According to the temperature gradient data and preset material parameters, the heat change of each unit in the three-dimensional mesh model is calculated, and divergence data is obtained based on the heat change.

3. The method according to claim 2, characterized in that, The step of extracting temperature data of the designated area from the multi-angle infrared data based on the regional image, and converting the temperature data into temperature point cloud data in three-dimensional space using a stereo vision 3D reconstruction method, in conjunction with the intrinsic and extrinsic parameters of the infrared thermal imager, includes: Based on the pixel positions marked as the specified region in the regional image, the temperature values ​​of the pixel positions are extracted from the corresponding multi-angle infrared data to form temperature data; Using the internal parameters of the infrared thermal imager, the image coordinates of each pixel in the specified area are corrected to obtain the corrected pixel coordinates; Based on the external parameters of the infrared thermal imager, the corrected pixel coordinates are converted into spatial projection rays originating from the optical center of the thermal imager; Pair spatial projection rays corresponding to the same pixel position in the specified region from infrared data at different angles to obtain ray pairs; For each pair of rays, determine the intersection point of the two spatial projection rays in three-dimensional space, and assign the corresponding temperature value from the temperature data to the intersection point. Combine all the intersection points and their corresponding temperature values ​​to generate temperature point cloud data.

4. The method according to claim 1, characterized in that, The process of constructing a structural relationship diagram of the target device, and generating feature vectors based on the structural relationship diagram and the thermal state data, combined with a graph convolutional neural network, includes: Based on the physical connection relationship and heat conduction path of the target device, a structural relationship diagram is established with components as nodes and connection or heat transfer relationship as edges. The multidimensional thermophysical quantities corresponding to each component in the thermal state data are used as the initial node features of the corresponding node. The structural relationship graph and the initial node features of each node are input into the graph convolutional neural network. Through the first graph convolutional layer of the graph convolutional neural network, the features of each node and the features of adjacent nodes are aggregated based on the edges in the structural relationship graph to generate initial component features. The initial component features are processed through the first activation layer of the graph convolutional neural network, and neighborhood information aggregation is performed based on the processed component features through the second graph convolutional layer of the graph convolutional neural network. A residual connection mechanism is introduced to fuse the processed component features with the aggregation result to generate extended component features. The extended component features are processed through the second activation layer of the graph convolutional neural network, and the processed extended component features of all components in the structural relationship graph are integrated through the graph pooling layer of the graph convolutional neural network to output device-level features. The device-level features are transformed through the fully connected layer of the graph convolutional neural network to generate feature vectors.

5. The method according to claim 1, characterized in that, The step of predicting the temperature change data of the target device in future time periods based on the feature vector and the switching data using a long short-term memory neural network includes: Arrange the switch data into a load sequence according to time order, and combine the feature vectors at the same time with the load sequence to obtain a combined sequence; The combined sequence is input into the memory unit of a long short-term memory neural network. In the memory unit, the information of the combined sequence at the current time and past time is processed through a gating mechanism to form a long-term time dependency. Based on the long-term time dependency, the memory unit outputs the predicted temperature value for the next moment; The predicted temperature value for the next moment is input into the memory unit as new input information. The gating mechanism process is repeated to obtain multiple predicted temperature values ​​for future moments. The multiple predicted temperature values ​​for future moments are combined to form temperature change data.

6. The method according to claim 1, characterized in that, The step of comparing the temperature change data with a dynamic threshold range to generate a monitoring result reflecting the real-time health status of the target device includes: Obtain historical temperature data of the target device during normal operation. By analyzing the historical temperature data, a dynamic threshold range that changes over time is obtained; When the temperature change data exceeds the dynamic threshold range within a preset time period, a monitoring result is generated.

7. The method according to claim 1, characterized in that, After generating monitoring results reflecting the real-time health status of the target device, the method further includes: Based on the abnormal start time and end time recorded in the monitoring results, extract the time-series data of the target device from the preset database during the time from the abnormal start time to the end time. The time series data is input into the isolated forest model, and the isolated forest model is used to calculate the anomaly degree value of each time point in the time series data. All time point data with an anomaly degree value greater than a preset anomaly degree threshold are merged into an anomaly data segment. For each of the abnormal data segments, a corresponding feature data segment is extracted from the time-series data; Each of the aforementioned feature data segments is input into a one-dimensional convolutional neural network classifier, and the one-dimensional convolutional neural network classifier outputs a probability vector. Based on the probability vector, the fault type with the highest probability is determined. Combining the time information, anomaly degree value, and confidence level of the fault type of the abnormal data segment, a diagnostic report is generated. The diagnostic report is output through the IoT platform of the production workshop.

8. A production workshop equipment status monitoring system based on the Internet of Things, characterized in that, include: The acquisition module is used to acquire multi-angle infrared data of the target device and switch data output by the control system of the target device through the Internet of Things platform in the production workshop; The processing module is used to process the multi-angle infrared data using a region growing algorithm, generate three-dimensional temperature data of a specified region on the target device, and calculate the corresponding temperature gradient data and divergence data based on the three-dimensional temperature data. The marking module is used to mark the three-dimensional temperature data, the temperature gradient data, and the divergence data according to the state switching time points in the switch data to form thermal state data; A construction module is used to construct a structural relationship diagram of the target device, and based on the structural relationship diagram and the thermal state data, combined with a graph convolutional neural network, generate a feature vector, which is used to characterize the overall thermal state of the target device. The prediction module is used to predict the temperature change data of the target device in the future time period based on the feature vector and the switch data, using a long short-term memory neural network. The comparison module is used to compare the temperature change data with a dynamic threshold range to generate monitoring results that reflect the real-time health status of the target device.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the steps of the IoT-based production workshop equipment status monitoring method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, enables the Internet of Things-based production workshop equipment status monitoring method as described in any one of claims 1 to 7.

Images