A method, apparatus, equipment, medium, and procedure for inspecting equipment.

Abstract

This invention belongs to the technical field of equipment inspection, and specifically relates to a method, apparatus, equipment, medium, and program product for equipment inspection. This application provides an equipment inspection method comprising: acquiring inspection data of a target equipment; dividing the inspection data into multiple inspection sub-tasks according to a preset inspection task decomposition model; determining a processing terminal for each inspection sub-task based on the task characteristics of each inspection sub-task; and generating an inspection result for the target equipment based on the processing results of each processing terminal for each inspection sub-task. This solves the problems of manual inspection methods, which rely on human experience, are highly subjective, have the risk of missed or false inspections, and are difficult to digitize, retain, and analyze. It forms a complete intelligent inspection system, comprehensively improving the automation, intelligence, and decision-making efficiency of industrial equipment inspection.

Description

Technical Field

[0001] This invention belongs to the technical field of equipment inspection, and specifically relates to a method, apparatus, equipment, medium, and program product for equipment inspection. Background Technology

[0002] Industrial equipment inspection is a crucial step in ensuring production safety and efficiency. Currently, inspections mainly rely on methods such as regular manual inspections, fixed sensor monitoring, or robot / drone inspections. Manual inspections depend on human experience, are highly subjective, and carry the risk of missed or false inspections. Furthermore, the data is difficult to digitize, retain, and analyze. Summary of the Invention

[0003] This application addresses the problems of manual inspection, which relies on human experience, is highly subjective, and carries the risk of missed or false inspections, as well as the difficulty in digitizing and analyzing data, through a method, apparatus, equipment, medium, and program product for equipment inspection. It establishes a complete intelligent inspection system that comprehensively improves the automation, intelligence, and decision-making efficiency of industrial equipment inspection.

[0004] To address the aforementioned technical problems, this application proposes five aspects.

[0005] In a first aspect, this application proposes a method for inspecting equipment, comprising: acquiring inspection data of a target equipment; dividing the inspection data into multiple inspection sub-tasks according to a preset inspection task decomposition model; determining a processing terminal for each inspection sub-task based on the task characteristics of each inspection sub-task; and generating an inspection result for the target equipment based on the processing result of each processing terminal on each inspection sub-task.

[0006] In some embodiments, the task characteristics include: maximum allowable response latency, floating-point computational complexity, and task level; determining the processing end for each inspection subtask based on its task characteristics includes: determining the timeliness requirement of the inspection subtask based on the maximum allowable response latency and current network conditions; determining the computational complexity of the inspection subtask based on the floating-point computational complexity; determining the decision level of the inspection subtask based on the task level; and determining the processing end for the inspection subtask based on the timeliness requirement, the computational complexity, and the decision level.

[0007] In some embodiments, determining the timeliness requirement of the inspection subtask based on the maximum permissible response latency and the current network conditions includes: determining the expected network latency when processing the inspection subtask through a cloud processing terminal based on the current network conditions; and determining the timeliness requirement of the inspection subtask based on the expected network latency and the maximum permissible response latency.

[0008] In some embodiments, determining the computational complexity of the inspection subtask based on the floating-point operation volume includes: obtaining the maximum computing power of the GPU in the edge processing terminal and the current utilization rate of the GPU; determining the available computing power of the edge processing terminal based on the maximum computing power and the current utilization rate; and determining the computational complexity of the inspection subtask based on the floating-point operation volume and the available computing power.

[0009] In some embodiments, determining the processing end of the inspection sub-task based on the timeliness requirement, the computational complexity, and the decision level includes: establishing a comparison matrix of the timeliness requirement, the computational complexity, and the decision level; solving for the eigenvector corresponding to the largest eigenvalue of the comparison matrix; determining the comprehensive scheduling score of the inspection sub-task based on the eigenvector, the timeliness requirement, the computational complexity, and the decision level; obtaining the actual response delay and processing success rate of completed inspection sub-tasks; determining a current judgment threshold based on the actual response delay and the processing success rate; and determining the processing end of the inspection sub-task based on the current judgment threshold and the comprehensive scheduling score.

[0010] In some embodiments, acquiring inspection data of the target device includes: acquiring the current infrared image, local audio information, and current location and time information of the target device through the inspection device; acquiring multi-source sensor data of the target device during the inspection period through various types of sensors installed on the target device; and unifying the current infrared image, local audio information, and multi-source sensor data into the same spatiotemporal coordinate system according to the current location and time information to form the inspection data.

[0011] Secondly, this application provides an inspection device for equipment, comprising: a first acquisition module for acquiring inspection data of a target equipment; a first execution module for splitting the inspection data into multiple inspection sub-tasks according to a preset inspection task splitting model; a first determination module for determining a processing terminal for each of the inspection sub-tasks based on the task characteristics of each inspection sub-task; and a second execution module for generating an inspection result for the target equipment based on the processing result of each processing terminal on each of the inspection sub-tasks.

[0012] Thirdly, this application proposes an electronic device including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method described in any of the first aspects.

[0013] Fourthly, this application proposes a computer-readable storage medium having a computer program stored thereon, characterized in that the computer program, when executed by a processor, implements the steps of the method described in any one of the first aspects.

[0014] Fifthly, this application proposes a computer program product, including a computer program, characterized in that, when executed by a processor, the computer program implements the steps of the method described in any one of the first aspects.

[0015] This application provides a method for inspecting equipment, comprising: acquiring inspection data of a target equipment; dividing the inspection data into multiple inspection sub-tasks according to a preset inspection task decomposition model; determining a processing terminal for each inspection sub-task based on the task characteristics of each inspection sub-task; and generating an inspection result for the target equipment based on the processing results of each inspection sub-task by each processing terminal. This method solves the problems of manual inspection, which relies on human experience, is highly subjective, and carries the risk of missed or false inspections, as well as the difficulty in digitizing, storing, and analyzing data. It forms a complete intelligent inspection system, comprehensively improving the automation, intelligence, and decision-making efficiency of industrial equipment inspection. Attached Figure Description

[0016] One or more embodiments are illustrated by way of example with reference to the accompanying drawings, and these illustrative descriptions do not constitute a limitation on the embodiments.

[0017] Figure 1 A main flowchart of a device inspection method provided in an embodiment of this application; Figure 2 A main structural block diagram of an equipment inspection device provided in an embodiment of this application; Figure 3 This is a structural block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0018] Industrial equipment inspection is a crucial step in ensuring production safety and efficiency. Currently, inspections mainly rely on methods such as regular manual inspections, fixed sensor monitoring, or robot / drone inspections. Manual inspections depend on human experience, are highly subjective, and carry the risk of missed or false inspections. Furthermore, the data is difficult to digitize, retain, and analyze.

[0019] To address the aforementioned technical problems, this invention proposes a method for inspecting equipment. The implementation details of this method are described below. The following content is for illustrative purposes only and is not essential for implementing this solution.

[0020] Example 1: like Figure 1As shown, this application provides a method for inspecting equipment, applicable to electronic production equipment, which can be a server, mobile terminal, computer, cloud platform, etc. The data processing functionality of the production equipment provided in this application embodiment can be implemented by the processor of the electronic production equipment calling program code, wherein the program code can be stored in a computer storage medium. The equipment inspection method includes: Step S1: Obtain the inspection data of the target equipment.

[0021] In some embodiments, step S1, "acquiring inspection data of the target device", includes: Step S11: Obtain the current infrared image, local audio information, and current location and time information of the target device through the inspection equipment.

[0022] Step S12: Obtain multi-source sensor data of the target device during the inspection period through various types of sensors installed on the target device.

[0023] Step S13: Based on the current location information and the time information, unify the current infrared image, the local audio information, and the multi-source sensor data into the same spatiotemporal coordinate system to form the inspection data.

[0024] Equipment safety inspections require on-site inspections of the target equipment. Typical inspection methods include collecting sensor data from the sensors deployed on the target equipment, processing this data to determine the operational status of each sub-device within the target equipment, and simultaneously collecting external images or operational videos of the inspected equipment to assess its performance. The inspected equipment is typically large, containing different work areas, each with numerous sub-devices. Furthermore, a single sub-device may not be unique within the target equipment; multiple sub-devices of the same type may be used, and even within a single work area, multiple sub-devices of different types may exist. While existing technologies allow for the numbering of sensors during deployment and the determination of their actual locations within the target device, as well as the sub-devices they represent, inspections of the target device utilize not only sensor data but also image and audio data generated during the inspection process. Since these data are collected by inspection personnel, it becomes difficult to correlate the sub-devices corresponding to the data collected by the inspection personnel with those corresponding to the sensor data. This makes it challenging to unify and analyze the sensor data with the data collected by the inspection equipment for each sub-device and even the entire target device.

[0025] Therefore, in this application, when acquiring the inspection data of the target device, the data collected by the inspection device and the sensor data are first spatiotemporally aligned so that the data collected by the inspection device and the sensor data are unified under the same spatiotemporal coordinate system, so that the inspection of the target device can be achieved by combining the data collected by the inspection device and the sensor data.

[0026] The inspection equipment described in this application can be a head-mounted device worn by inspection personnel or a robot composed of inspection equipment. The inspection equipment integrates a BeiDou high-precision positioning module, a 5G communication module, an AR display unit, a microphone, and a camera (which can be a conventional optical camera or an infrared device). When inspecting the target area, the BeiDou module uses a BeiDou-3 dual-mode receiver (supporting BDS-3 / GLONASS) combined with RTK differential positioning technology to output the inspection equipment's own ±1cm-level three-dimensional coordinates in real time; it achieves nanosecond-level precise time synchronization through PTP (Precision Time Protocol) time synchronization protocol and sensor clock calibration. Simultaneously, the camera captures high-definition video streams from the equipment; an infrared thermal imager (temperature range -20℃~500℃, temperature accuracy ±0.5℃) captures infrared thermal images; a MEMS microphone captures local audio; and various IoT sensors pre-installed on the target equipment (temperature, vibration, pressure, and other multi-source sensor data) upload data in real time through a 5G NSA / SA dual-mode gateway (latency ≤20ms).

[0027] All collected data streams (video frames, sensor readings) are tagged with a unified "BeiDou spatiotemporal stamp," which contains the precise time of acquisition (format: UTC time + nanosecond offset) and the precise spatial location of the acquisition point (format: WGS84 coordinate system three-dimensional coordinates). Using the spatiotemporal stamp as an index, a Kalman filter algorithm is employed to eliminate redundancy and noise in multi-source data. Through data association rules (video frames are aligned at 100ms keyframe intervals, and sensor data is interpolated and matched at a 10Hz sampling rate), video streams, IoT data, and location data are unified under the same spatiotemporal coordinate system, forming a standardized data packet (data format: JSON-LD, containing device ID, spatiotemporal stamp, data type, value, and confidence field), which is the inspection data in this application.

[0028] This invention addresses the problem in existing technologies where, when multiple devices or multiple components of the same device malfunction simultaneously, timestamps alone cannot define the spatial range of the malfunction's source, making accurate fault location difficult. It also solves the problem that while time can guarantee data synchronization, it cannot establish a correspondence between data and the physical structure of the equipment. This makes it difficult for AI algorithms to learn the correlation between "location and fault" (e.g., bolts in specific locations are more prone to loosening due to stress characteristics, coastal equipment is more susceptible to salt spray corrosion, and inland equipment is affected by sand and dust abrasion), thus limiting the accuracy of fault prediction during inspections.

[0029] By utilizing BeiDou's ±1cm-level 3D coordinate system, visual defects (such as loose bolts) in video frames and operational parameters of IoT sensors (such as temperature and vibration) can be precisely linked to the physical structure of the equipment, forming a "position-state-time" 3D data model. This achieves a breakthrough from "equipment-level positioning" to "component-level positioning," improving positioning accuracy from meter-level to centimeter-level, providing precise navigation for fault location and maintenance. Furthermore, the validity of multi-source data can be cross-verified based on spatial location (e.g., if an infrared thermal image at a certain location shows an abnormal temperature, and the corresponding sensor data is normal, it will trigger data anomaly verification, eliminating sensor malfunctions or acquisition errors), thus enhancing data reliability.

[0030] Step S2: Divide the inspection data into multiple inspection sub-tasks according to the preset inspection task splitting model.

[0031] In existing technologies, inspection data generated by inspection tasks is typically processed uniformly across all inspection data for the target equipment. However, due to the sheer volume of inspection data, this unified processing requires significant time and computing power, hindering inspection personnel from obtaining timely inspection results. To address this issue and improve processing efficiency, this application further divides the inspection data into multiple sub-tasks based on a pre-defined task decomposition model. These sub-tasks include diagnostic tasks for individual devices, diagnostic tasks for work areas, diagnostic tasks for the correlation between upstream and downstream devices of a specific device, diagnostic tasks for a single process, and overall diagnostic tasks for the target equipment. While dividing the inspection data into different sub-tasks may increase the total computing power required to complete the inspection, the varying difficulty, computing power consumption, and time required for each sub-task allow for allocation to different processing endpoints based on their characteristics. This significantly reduces the overall time required and allows for the timely detection of any significant problems. In order to improve the efficiency of processing inspection data and to detect problems as quickly as possible, this application needs to define the processing terminals for each inspection sub-task. These processing terminals include cloud processing terminals and edge device processing terminals.

[0032] The task characteristics of the subtasks in this application include: maximum allowable response latency, floating-point computation cost, and task level. The maximum allowable response latency represents the maximum end-to-end response latency required by the inspection subtask, which is predefined by the task type to which the inspection subtask belongs. The floating-point computation cost is the amount of floating-point computation required to execute the task once. This value is pre-calculated by the structural parameters (number of network layers, number of parameters, input size) of each AI model during model registration and stored in the metadata of the model repository. The specific calculation method is as follows:

[0033] in This represents the total number of layers in the model. The area of ​​the convolutional kernel in layer I is... , These represent the number of input / output channels for layer I. , These represent the height and width of the output feature map of layer I, respectively.

[0034] The task level refers to the level to which the inspection sub-task belongs. In this application, the task level is divided into three levels: device level, area level, and global level. Device level tasks refer to sub-devices and tasks related to the upstream and downstream of the sub-devices. Area level tasks refer to tasks at the work area level, while global level tasks refer to tasks for the target device as a whole.

[0035] Therefore, in determining the processing end of each inspection subtask in this application, i.e. whether the inspection subtask is suitable to be processed on the cloud processing end or the edge processing end, it is necessary to refer to the task characteristics of the inspection subtask. Among them, the timeliness requirement of the subtask on the cloud processing end is determined by the maximum allowable response latency, the computational complexity of the inspection task on the edge processing end is calculated by the floating-point operation volume, and the decision level of the inspection subtask is calculated by the device level.

[0036] Step S3: Determine the processing terminal for each inspection subtask based on the task characteristics of each inspection subtask.

[0037] The task characteristics include, in some embodiments, step S3, "determining the processing end for each of the inspection sub-tasks based on the task characteristics of each inspection sub-task," including: Step S31: Determine the timeliness requirement of the inspection subtask based on the maximum allowable response delay and the current network conditions.

[0038] In some embodiments, step S31, "determining the timeliness requirement of the inspection sub-task based on the maximum allowable response delay and the current network conditions," includes: Step S311: Determine the expected network latency when processing the inspection subtask through the cloud processing terminal based on the current network conditions.

[0039] Step S312: Determine the timeliness requirement of the inspection subtask based on the expected network latency and the maximum allowable response latency.

[0040] The timeliness requirement U measures the sensitivity of the task to response latency, and is defined as the maximum end-to-end response latency that the task can tolerate. Compared to the expected latency of cloud processing under current network conditions The reciprocal mapping of the ratio:

[0041] Parameter description: (Unit: milliseconds) represents the maximum end-to-end response latency required for this task type, predefined by the task type. For example, a real-time video defect detection task. Sensor timing anomaly detection task Equipment life prediction task (1 hour).

[0042] (Unit: ms) represents the expected latency for uploading the task data to the cloud, for the cloud to complete the processing, and for the cloud to return the result under the current network conditions. The system calculates this in real time based on the moving average of the actual cloud round-trip time (RTT) within the past 5-minute sliding window:

[0043] in, This is the moving average of network RTT over the past 5 minutes. The historical average processing time for this task type in the cloud (obtained from cloud processing logs).

[0044] The range of values ​​is : The closer it is to 1, the more sensitive the task is to latency, and the more it should be prioritized for allocation to edge processing. The closer to 0, the less sensitive the task is to latency, making it suitable for assignment to cloud processing.

[0045] Step S32: Determine the computational complexity of the inspection subtask based on the floating-point operation quantity.

[0046] In some embodiments, step S32, "determining the computational complexity of the inspection subtask based on the floating-point operation quantity," includes: Step S321: Obtain the maximum computing power and current utilization rate of the GPU in the edge processing terminal.

[0047] Step S322: Determine the available computing power of the edge processing terminal based on the maximum computing power and the current utilization rate.

[0048] Step S323: Determine the computational complexity of the inspection subtask based on the floating-point operation volume and the available computing power.

[0049] Computational complexity C measures the feasibility of a task on an edge node and is defined as the ratio of the floating-point operations required by the task to the available computing power of the edge node.

[0050] Parameter description: (Unit: GFLOPS) is the number of floating-point operations required to perform the task once. This value is pre-calculated by the structural parameters (number of network layers, number of parameters, input size) of each AI model during model registration and stored in the metadata of the model repository.

[0051] (Unit: GFLOPS) represents the available computing power of the current edge node, collected in real time by the edge resource monitoring module.

[0052] in This represents the peak computing power of the GPU in the edge processing unit (determined by hardware specifications). This represents the current GPU utilization rate.

[0053] The range of values ​​for C is When C<1, it means that the edge processing end can complete the task within 1 second, which is suitable for execution on the edge processing end; when C≥1, it means that the edge processing end needs more than 1 second to complete the task, which indicates a computing power bottleneck, and it tends to be allocated to the cloud processing end.

[0054] Normalize C and map it to Interval:

[0055] The closer it is to 1, the less edge computing power is available to handle the task, and the more it should be allocated to cloud processing.

[0056] Step S33: Determine the decision level of the inspection sub-task based on the task level.

[0057] Decision level D measures the scope of decisions affected by the analytical conclusions of the task and the breadth of contextual information required. This invention divides industrial inspection decision-making into three levels:

[0058] Each analysis task is pre-labeled with its decision level during registration. The higher the level (D), the more extensive the data context and the stronger the computing power required, making it more suitable for execution on cloud processing platforms.

[0059] Step S34: Determine the processing end of the inspection sub-task based on the timeliness requirement, the computational complexity, and the decision level.

[0060] In some embodiments, step S34, "determining the processing endpoint of the inspection sub-task based on the timeliness requirement, the computational complexity, and the decision level," includes: Step S341: Establish a comparison matrix of the timeliness requirement, the computational complexity, and the decision level.

[0061] Step S342: Solve for the eigenvector corresponding to the largest eigenvalue of the comparison matrix.

[0062] Step S343: Determine the comprehensive scheduling score of the inspection sub-task based on the feature vector, the timeliness requirement, the computational complexity, and the decision level.

[0063] Step S344: Obtain the actual response delay and processing success rate of the completed inspection sub-tasks.

[0064] Step S345: Determine the current judgment threshold based on the actual response delay and the processing success rate.

[0065] Step S346: Determine the processing end of the inspection subtask based on the current judgment threshold and the comprehensive scheduling score.

[0066] For each inspection subtask j, the system calculates its edge processing execution priority score. :

[0067] Parameter description: , , These are the quantized values ​​of task j in three dimensions; , , The weight coefficients for each dimension satisfy...

[0068] The method for determining weight coefficients—the Analytic Hierarchy Process (AHP)—is as follows: Construct a three-dimensional pairwise comparison matrix A

[0069] in: This represents the ratio of the importance of timeliness requirements to computational complexity. This represents the ratio of the timeliness requirement to the importance of the decision-making level. This represents the ratio of computational complexity to the importance of the decision-making level.

[0070] Each ratio is determined based on the functional requirements of the industrial inspection scenario. In the context of safe production, real-time performance (timeliness) is the primary constraint, followed by computational feasibility (computational complexity), and then decision-making scope (decision level).

[0071] Find the largest eigenvalue of matrix A Corresponding feature vectors: After normalization, the weight coefficients for each dimension are obtained:

[0072]

[0073]

[0074] Rules for determining the processing end: like If so, it will be assigned to the edge processing end for execution; like Then it will be assigned to the cloud processing terminal for execution.

[0075] The current judgment threshold The following adaptive method is used to determine the response latency and processing success rate of each task under both edge processing and cloud processing conditions, with the objective function being to minimize the globally weighted latency:

[0076] Parameter description: in To count the total number of tasks within the window, The actual response delay for task j. This is an indicator function (it takes the value 1 if the timeout occurs). and These are the loss weights for the delayed and failed items, respectively. The system uses the previous day's task execution logs daily to search for the optimal [0,1] interval using a grid search (step size 0.01). And it will take effect the following day.

[0077] Step S4: Generate inspection results for the target device based on the processing results of each inspection sub-task from each processing terminal.

[0078] Each inspection subtask has a corresponding processing method at different processing terminals.

[0079] The processing flow at the edge processing end is as follows: Data transmission: Video streams, key vibration / temperature waveforms, and other latency-sensitive data (single frame video data size ≤ 10MB) are directly connected to the AI ​​server at the factory edge via a 5G URLLC (ultra-reliable low-latency communication, latency ≤ 10ms, reliability ≥ 99.999%) channel.

[0080] Core algorithms and functions: Computer vision model: A lightweight convolutional neural network (CNN) is used, specifically the MobileNetV3-Large variant (parameters ≤ 5M, inference speed ≥ 30fps). Through transfer learning, the defect features of industrial equipment are optimized to achieve real-time detection of appearance defects such as loose bolts (recognition accuracy ≥ 98%), paint peeling (minimum detection area ≥ 0.5cm²), and instrument panel reading recognition (error ≤ ±1%).

[0081] Temporal anomaly detection model: The isolated forest algorithm (100 decision trees, contamination parameter set to 0.01) or one-dimensional convolutional neural network (input dimension of sensor data with 100 time steps, convolution kernel size 3×1) is used to monitor real-time sensor streaming data and quickly detect anomalies such as excessive vibration (threshold: pump vibration acceleration ≥4.5m / s²) and sudden temperature rise (threshold: temperature rise ≥10℃ within 10s).

[0082] Lightweight decision-making: The edge server has a built-in rule engine, and the rule base is stored in JSON format (example: [{"Defect Type":"Loose Bolt","Defect Level":"Level 1","Trigger Condition":"Loose Angle ≥ 15°","Linked Action":"Audio-Visual Alarm + Push Repair Work Order"}]). The output of the above AI model is integrated, and once a fault or parameter in the predefined emergency defect library is identified as exceeding the safety threshold, a Level 1 alarm is immediately generated and simultaneously pushed to the AR terminal of the inspection personnel and the large screen of the monitoring center via the 5G network.

[0083] The detection methods at the edge processing end are mainly based on the multi-source data acquisition and edge AI analysis technologies of this solution, and are divided into four main categories: 1) Visual inspection: The device's appearance is captured by an AR terminal 4K camera, and features are identified by a lightweight CNN model. It is suitable for appearance defects such as loose bolts, leaking seals, and incorrect instrument readings. 2) IoT sensor detection: Real-time collection of operating parameters through pre-installed IoT sensors such as temperature and vibration sensors on the device body, suitable for abnormal operating conditions such as temperature and vibration; 3) Infrared thermal imager detection: The infrared thermal imager connected in the near field collects thermal imaging data of the equipment and captures temperature distribution differences. It is used to detect defects that are difficult to identify visually, such as leaks and hidden temperature anomalies. It can be compared with other detection methods to improve accuracy. 4) Audio-assisted recognition: Combined with audio feature extraction algorithms, it helps to judge defects such as abnormal vibration and component wear, and enhances the comprehensiveness of detection.

[0084] All detection methods carry a unified BeiDou time and space stamp to ensure the accuracy of defect location and data traceability.

[0085] The processing flow on the cloud processing end is as follows: Data transmission: The full amount of high-definition video recordings, complete sensor history sequences (storage period ≥ 90 days), and all metadata with time and space stamps are uploaded to the cloud AI analysis platform via a 5GeMBB (enhanced mobile broadband, downlink rate ≥ 1Gbps) channel.

[0086] Core algorithms and functions: Video Deep Analysis: The uploaded video is analyzed in a fine-grained manner using a deep learning model ResNet50 (with pre-trained weights based on ImageNet) or Vision Transformer (ViT-B / 16 architecture) to identify minute cracks (minimum detection width ≥ 0.1 mm) and early corrosion (corrosion area ≥ 3%) that may be missed on the edge side. Based on the defect type, location and severity, a weighted scoring method (weight: defect severity 0.6, location importance 0.3, and propagation risk 0.1) is used to generate a health score (0-100 points) for the appearance of the equipment.

[0087] Multi-source temporal prediction (deep model fusing spatial features): Existing temporal prediction models (such as LSTM and TCN) typically only input equipment operating parameters (such as temperature and vibration) and their timestamps, ignoring the significant spatial differences in the fault evolution process. For example, pumps of the same model located in coastal and inland environments, or at different positions upstream and downstream of the production line, exhibit systematic deviations in their degradation trajectories; similarly, the pre-fault precursor features of different components (bearings, impellers, seals) of the same equipment show heterogeneous spatial distribution. To address these issues, this invention uses BeiDou centimeter-level three-dimensional coordinates as explicit input features and deeply fuses them with sensor temporal data. The specific implementation method is as follows: For each group of devices, the multi-parameter historical sequence (time window length ≥ 72 hours) is synchronously stitched together with its corresponding BeiDou spatial coordinates. And spatial topological relationship encoding (such as the relative position vectors of upstream and downstream equipment on the same production line). Construct an LSTM + TCN fusion model, concatenating spatial coordinates and sensor parameters at the input to form a multidimensional feature vector. ,in Let be the sensor reading vector at time t. Encoding spatial topological relationships; The model automatically learns the three-dimensional coupling law of "location-parameter-fault" through the attention mechanism. For example, due to salt spray corrosion, the temperature parameter of coastal equipment has a significantly higher weight in the later stage of its life than that of inland equipment; the degree of influence of abnormal vibration of upstream equipment on the pressure of downstream equipment is determined by its spatial distance and connection relationship.

[0088] This fusion method reduces the remaining lifetime prediction error, enabling a leap in prediction capabilities from the "equipment level" to the "component level" and even the "space region level".

[0089] Unsupervised anomaly detection: Using an autoencoder (both encoder and decoder are 3-layer fully connected networks with hidden layer dimensions of 256-128-256) or a Gaussian mixture model (mixture component number K=5, trained by the EM algorithm), we learn from 100,000+ sets of normal operation data in the entire area to build a baseline of normal behavior and proactively discover rare potential anomaly patterns that are difficult to predefine (such as small data deviations caused by sensor drift) to achieve early warning.

[0090] Model Management and Deployment: The cloud platform is responsible for the continuous training and optimization of all AI models. The platform builds a model repository (supporting TensorFlow / PyTorch model formats) and uses a federated learning framework (FedAvg algorithm) to continuously train and optimize AI models, preventing data privacy leaks. Monthly or quarterly, optimized lightweight edge diagnostic models (model size compressed to ≤200MB, with INT8 quantization reducing computational power consumption) are deployed to edge servers via 5G network security as AES-256 encrypted firmware upgrade packages, enabling iterative evolution of the system algorithm.

[0091] Large Model + RAG Retrieval Enhancement Module: Addressing the pain points of insufficient knowledge accumulation, weak semantic understanding, and fragmented decision-making in traditional AI models, this solution constructs a three-in-one module of "dedicated knowledge base + RAG retrieval engine + lightweight large model" in the cloud AI analysis platform, which is deeply integrated with edge-cloud collaborative analysis capabilities to enhance the accuracy of auxiliary decision-making suggestions.

[0092] The AR smart terminal receives real-time analysis results, early warning information, and device history records from the edge and cloud. A dual-anchor registration mechanism utilizes the ±1cm absolute position provided by BeiDou as the initial anchor point, combined with terminal visual SLAM (based on the ORB-SLAM3 algorithm) and device feature point AI recognition (using the YOLOv8-nano model, with device component recognition speed ≥50fps). This achieves precise ±2cm overlay between virtual information and the real device, with an overlay latency ≤300ms, significantly suppressing information drift.

[0093] This application provides a method for inspecting equipment, comprising: acquiring inspection data of a target equipment; dividing the inspection data into multiple inspection sub-tasks according to a preset inspection task decomposition model; determining a processing terminal for each inspection sub-task based on the task characteristics of each inspection sub-task; and generating an inspection result for the target equipment based on the processing results of each inspection sub-task by each processing terminal. This method solves the problems of manual inspection, which relies on human experience, is highly subjective, and carries the risk of missed or false inspections, as well as the difficulty in digitizing, storing, and analyzing data. It forms a complete intelligent inspection system, comprehensively improving the automation, intelligence, and decision-making efficiency of industrial equipment inspection.

[0094] Example 2: Based on the foregoing embodiments, this application provides an inspection device for equipment. The various modules and units included in the device can be implemented by a processor in a computer device; of course, they can also be implemented by specific logic circuits. In the implementation process, the processor can be a central processing unit (CPU), a microprocessor unit (MPU), a digital signal processor (DSP), or a field programmable gate array (FPGA), etc.

[0095] like Figure 2 As shown, an inspection device for equipment includes: a first acquisition module 1, a first execution module 2, a first determination module 3, and a second execution module 4.

[0096] The first acquisition module 1 is used to acquire inspection data of the target device. The first execution module 2 is used to divide the inspection data into multiple inspection sub-tasks according to a preset inspection task splitting model. The first determination module 3 is used to determine the processing terminal for each inspection sub-task based on the task characteristics of each inspection sub-task. The second execution module 4 is used to generate inspection results for the target device based on the processing results of each processing terminal for each inspection sub-task.

[0097] The first determining module 3 is further configured to determine the timeliness requirement of the inspection subtask based on the maximum allowable response delay and the current network conditions; determine the computational complexity of the inspection subtask based on the floating-point operation volume; determine the decision level of the inspection subtask based on the task level; and determine the processing end of the inspection subtask based on the timeliness requirement, the computational complexity, and the decision level.

[0098] The first determining module 3 is further configured to determine the expected network latency when the inspection sub-task is processed through the cloud processing terminal based on the current network conditions; and to determine the timeliness requirement of the inspection sub-task based on the expected network latency and the maximum allowable response latency.

[0099] The first determining module 3 is further configured to obtain the maximum computing power of the GPU and the current utilization rate of the GPU in the edge processing terminal; determine the available computing power of the edge processing terminal based on the maximum computing power and the current utilization rate; and determine the computational complexity of the inspection subtask based on the floating-point operation volume and the available computing power.

[0100] The first determining module 3 is further configured to establish a comparison matrix of the timeliness requirement, the computational complexity, and the decision level; solve for the eigenvector corresponding to the largest eigenvalue of the comparison matrix; determine the comprehensive scheduling score of the inspection sub-task based on the eigenvector, the timeliness requirement, the computational complexity, and the decision level; obtain the actual response delay and processing success rate of the completed inspection sub-task; determine the current judgment threshold based on the actual response delay and the processing success rate; and determine the processing end of the inspection sub-task based on the current judgment threshold and the comprehensive scheduling score.

[0101] The first acquisition module 1 is further configured to acquire the current infrared image, local audio information, and current location and time information of the target device through the inspection device; acquire multi-source sensor data of the target device during the inspection period through various types of sensors installed on the target device; and unify the current infrared image, local audio information, and multi-source sensor data into the same spatiotemporal coordinate system according to the current location information and the time information to form the inspection data.

[0102] The various modules in the inspection device of the aforementioned equipment can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of the equipment in hardware form or independently of it, or stored in the memory of the processing device in software form, so that the processor can call and execute the operations corresponding to each module. It should be noted that the module division in this embodiment is illustrative and only represents a logical functional division; in actual implementation, there may be other division methods.

[0103] Example 3: Thirdly, this application provides an electronic device, such as Figure 3 As shown, it includes: at least one processor 100; and a memory 200 communicatively connected to the at least one processor 100; wherein the memory 200 stores instructions executable by the at least one processor 100, the instructions being executed by the at least one processor 100 to enable the at least one processor 100 to perform an inspection method for a device in the above embodiments.

[0104] The memory and processor are connected via a bus, which can include any number of interconnecting buses and bridges, connecting various circuits of one or more processors and memories. The bus can also connect various other circuits, such as peripheral devices, voltage regulators, and power management circuits, which are well known in the art and will not be described further herein. The bus interface provides an interface between the bus and the transceiver. The transceiver can be a single element or multiple elements, such as multiple receivers and transmitters, providing a unit for communicating with various other devices over a transmission medium. Data processed by the processor is transmitted over the wireless medium via an antenna, which further receives data and transmits it to the processor.

[0105] The processor manages the bus and general processing, and also provides various functions, including timing, peripheral interfaces, voltage regulation, power management, and other control functions. Memory is used to store data used by the processor during operation.

[0106] In addition, the computer device may include (but is not limited to) a data bus, an input / output (I / O) bus, a display, and input / output devices (e.g., keyboard, mouse, speakers, etc.).

[0107] The processor can communicate with external devices via the I / O bus through wired or wireless networks.

[0108] In one embodiment, the at least one computer-executable instruction may also be compiled into or comprise a software product / computer program product, wherein one or more computer-executable instructions are executed by a processor to perform the steps of the various functions and / or methods in the embodiments described herein.

[0109] Example 4: Fourthly, this application proposes a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in any of the first aspects.

[0110] Computer-readable storage media can be implemented by any type of volatile or non-volatile storage device or a combination thereof. Computer-readable storage media may include, but are not limited to, random access memory (RAM), read-only memory (ROM), flash memory, EPROM memory, EEPROM memory, registers, and computer storage media (e.g., hard disks, floppy disks, solid-state drives, removable disks, CD-ROMs, DVD-ROMs, Blu-ray discs, etc.).

[0111] Computer-readable storage media may also store at least one computer-executable program / instruction, such as computer-readable instructions. Computer-readable storage media include, but are not limited to, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and / or cache memory. Computer-readable storage media may include, for example, read-only memory (ROM), hard disk, flash memory, etc. For example, a non-transitory computer-readable storage medium may be connected to a computing device such as a computer, and then, when the computing device executes the computer-readable instructions stored on the computer-readable storage medium, the various methods described above can be performed.

[0112] Example 5: Fifthly, this application proposes a computer program product, including a computer program / instructions, characterized in that, when the computer program is executed by a processor, it implements the steps of the method described in any one of the first aspects.

[0113] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0114] It should be understood that the phrase "one embodiment" or "an embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this application. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. It should be understood that in the various embodiments of this application, the sequence numbers of the above-described processes do not imply a sequential order of execution; the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application. The sequence numbers of the above-described embodiments are merely descriptive and do not represent the superiority or inferiority of the embodiments.

[0115] It should be noted that, in this document, 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 a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0116] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.

[0117] In the embodiments provided in this disclosure, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus embodiments described above are merely illustrative; for example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0118] Alternatively, if the integrated units described above are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, or the parts that contribute to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a controller to execute all or part of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROMs, magnetic disks, or optical disks.

[0119] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein.

[0120] Those skilled in the art will understand that the above embodiments are specific embodiments for implementing this application, and in practical applications, various changes can be made to them in form and detail without departing from the spirit and scope of this application.

Claims

1. A method for inspecting equipment, characterized in that, include: Acquire inspection data of the target equipment; The inspection data is divided into multiple inspection sub-tasks according to the preset inspection task splitting model; The processing endpoint for each inspection subtask is determined based on its task characteristics. The inspection results for the target equipment are generated based on the processing results of each inspection sub-task by each processing terminal.

2. The method according to claim 1, characterized in that, The task characteristics include: maximum allowable response latency, floating-point computation load, and task level; determining the processing end for each inspection sub-task based on its task characteristics includes: The timeliness requirement of the inspection sub-task is determined based on the maximum allowable response delay and the current network conditions. The computational complexity of the inspection subtask is determined based on the floating-point operation volume. The decision level of the inspection sub-task is determined based on the task level. The processing endpoint for the inspection sub-task is determined based on the timeliness requirement, the computational complexity, and the decision-making level.

3. The method according to claim 2, characterized in that, The step of determining the timeliness requirement of the inspection sub-task based on the maximum allowable response delay and the current network conditions includes: Determine the expected network latency when processing the inspection subtask through the cloud processing terminal based on the current network conditions. The timeliness requirement of the inspection subtask is determined based on the expected network latency and the maximum allowable response latency.

4. The method according to claim 2, characterized in that, Determining the computational complexity of the inspection subtask based on the floating-point operation quantity includes: Obtain the maximum computing power and current utilization of the GPU in the edge processing terminal; The available computing power of the edge processing terminal is determined based on the maximum computing power and the current utilization rate; The computational complexity of the inspection subtask is determined based on the floating-point operation volume and the available computing power.

5. The method according to claim 2, characterized in that, The process of determining the processing endpoint for the inspection sub-task based on the timeliness requirement, the computational complexity, and the decision-making level includes: Establish a comparison matrix of the timeliness requirement, the computational complexity, and the decision level; Solve for the eigenvector corresponding to the largest eigenvalue of the comparison matrix; The comprehensive scheduling score of the inspection sub-task is determined based on the feature vector, the timeliness requirement, the computational complexity, and the decision level. Obtain the actual response latency and processing success rate of completed inspection sub-tasks; The current judgment threshold is determined based on the actual response delay and the processing success rate. The processing end of the inspection subtask is determined based on the current judgment threshold and the comprehensive scheduling score.

6. The method according to claim 1, characterized in that, The acquisition of inspection data for the target equipment includes: The current infrared image, local audio information, and current location and time information of the inspection equipment are obtained from the target equipment. The target device is equipped with various types of sensors to acquire multi-source sensor data during the inspection period. Based on the current location information and the time information, the current infrared image, the local audio information, and the multi-source sensor data are unified into the same spatiotemporal coordinate system to form the inspection data.

7. An inspection device for equipment, characterized in that, include: The first acquisition module is used to acquire inspection data of the target equipment; The first execution module is used to split the inspection data into multiple inspection sub-tasks according to a preset inspection task splitting model. The first determining module is used to determine the processing end for each of the inspection sub-tasks based on the task characteristics of each inspection sub-task. The second execution module is used to generate inspection results for the target equipment based on the processing results of each inspection sub-task by each processing terminal.

8. An electronic device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program performs the steps of the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program performs the steps of the method according to any one of claims 1 to 6.

Images