Equipment maintenance operation management system and method based on industrial vision intelligence

Abstract

The invention relates to the technical field of industrial equipment maintenance and repair, and discloses an intelligent equipment maintenance and repair operation management system and method based on industrial vision, and the system comprises an industrial vision collection module which is responsible for obtaining the working state and appearance data of equipment in real time through an industrial-grade vision sensor; the multi-sensor fusion module is used for integrating the working state and appearance data of the equipment acquired by the industrial vision acquisition module in real time, and providing multi-dimensional perception capability for the state of the equipment; and the edge calculation processing module is responsible for carrying out localization processing and real-time analysis on the working state and appearance data of the equipment acquired by the industrial visual acquisition module in real time. Through an RGB camera, an infrared camera and a 3D camera in the industrial vision acquisition module, the working state and appearance data of the equipment can be comprehensively acquired from multiple angles, the multi-dimensional perception capability of the equipment state is provided, and more comprehensive and accurate equipment state monitoring is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial equipment maintenance and repair, and specifically to an equipment maintenance and repair operation management system and method based on industrial visual intelligence. Background Art

[0002] In modern industrial production, the stable operation of equipment is of vital importance. With the development of industrial technology, the complexity and intelligence of equipment are increasing, and traditional equipment maintenance and repair methods are facing many challenges.

[0003] On the one hand, traditional equipment monitoring mainly relies on manual inspection and regular maintenance, which is inefficient and difficult to accurately grasp the operating status of the equipment in real time. At the same time, manual inspection is easily affected by subjective factors and may miss potential fault hazards, while regular maintenance may lead to excessive maintenance or untimely maintenance, increasing maintenance costs or reducing equipment reliability.

[0004] On the other hand, with the increasing intelligence and automation of equipment, the amount of data generated by the equipment has increased explosively. Traditional data processing methods are unable to cope with the huge amount of data and are unable to extract valuable information. At the same time, different types of sensor data are often processed independently and lack effective fusion methods, making it difficult to fully and accurately reflect the status of the equipment.

[0005] In addition, in terms of equipment maintenance decisions, traditional methods mainly rely on experience and fixed maintenance plans, lack flexibility and adaptability, and cannot be dynamically adjusted according to the actual operating status and fault conditions of the equipment, making it difficult to achieve the optimal maintenance effect. Summary of the invention

[0006] In view of the deficiencies in the prior art, the present invention provides an equipment maintenance and repair operation management system and method based on industrial visual intelligence to solve the problems raised in the above background technology.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: an equipment maintenance and repair operation management system based on industrial visual intelligence, characterized in that it includes: an industrial visual acquisition module, which is responsible for obtaining the working status and appearance data of the equipment in real time through industrial-grade visual sensors; The multi-sensor fusion module is used to integrate the working status and appearance data of the equipment obtained in real time from the industrial vision acquisition module, providing multi-dimensional perception capabilities of the equipment status; The edge computing processing module is responsible for local processing and real-time analysis of the working status and appearance data of the equipment acquired by the industrial vision acquisition module in real time; The deep learning analysis module is responsible for in-depth analysis of the working status and appearance data of the equipment acquired in real time through deep learning algorithms, extracting the complex characteristics of equipment failures and identifying potential problems; The fault detection module is responsible for comprehensively evaluating the analysis results of the deep learning analysis module and identifying the fault status of the current equipment; Intelligent maintenance decision module, which generates adaptive maintenance plans based on the reinforcement learning algorithm of the deep learning analysis module; Augmented reality and user interaction module, which is used by maintenance personnel to obtain equipment status in real time, visualize fault information and provide remote collaboration functions; The system integration and security module is responsible for system integration and data transmission security.

[0008] Preferably, the industrial vision acquisition module includes: RGB camera unit, used to capture images of the device surface; Infrared camera unit to detect thermal radiation anomalies of the equipment; A 3D camera unit, used to obtain three-dimensional deformation information of the device surface; The multi-sensor fusion module fuses the sensor data through a weighted fusion algorithm, and the fusion formula is: , in, is the data of the i-th sensor, is the corresponding weight, n is the number of sensors, Represents the fused features or results.

[0009] Preferably, the edge computing processing module includes: The lightweight deep learning model is used for real-time fault detection. The lightweight deep learning model adopts deep separable convolution, and its convolution operation formula is: , in, is the input feature map, is the depth convolution kernel, is the convolution kernel, Represents the output feature map or result matrix.

[0010] Preferably, the edge computing processing module further includes a Sobel edge detector for image feature extraction, and its calculation formula is: , in, and Represents the gradient of the image in the horizontal and vertical directions. represents the image gradient magnitude, Represents the gray value distribution function of the image, item.

[0011] Preferably, the deep learning analysis module performs fault classification based on a convolutional neural network model, and the output of the convolutional layer is calculated by the following formula: , in, is the convolution kernel of the lth layer, is the input feature map, is the bias, is the nonlinear activation function, represents the hidden state of the i+1th layer or time step; The deep learning analysis module includes a Softmax layer for fault classification, and the fault probability of its output is calculated by the following formula: , in, Indicates that given an input feature vector The probability of category j under the condition of ; is the input feature vector, For input features The score corresponding to category j; K represents the total number of categories, The denominator represents the sum of the index scores of all categories, which is used for normalization so that the sum of the probabilities of all categories is 1.

[0012] Preferably, the intelligent maintenance decision module generates a maintenance plan based on a reinforcement learning algorithm and uses a Q learning algorithm to update the state-action value function , and its update formula is: , in, represents the Q value of taking action a in state s, represents the learning rate, Indicates immediate reward, represents the discount factor, Indicates the next state Determine the optimal action The maximum Q value, It represents the next state reached after taking action a in the current state s. represents the optimal action.

[0013] Preferably, the intelligent maintenance decision module optimizes maintenance task scheduling by genetic algorithm, and the fitness function is: , in, is the cost of the ith maintenance, Maintenance is time-consuming. is the balance coefficient, represents the total cost, and N represents the total number of cost items or time items.

[0014] A method for equipment maintenance, repair and operation management based on industrial visual intelligence, comprising the following steps: S1. First, use the RGB camera in the industrial vision acquisition module to collect the surface image of the equipment, then use the infrared camera to detect the abnormal thermal radiation of the equipment and detect the overheating parts in time. Finally, use the 3D camera to obtain the three-dimensional deformation information of the equipment surface; S2, using a multi-sensor fusion module, using a weighted fusion algorithm to integrate data from different types of sensors to obtain multi-dimensional perception data of the device status; S3, deploy edge computing processing modules at the equipment site. First, a lightweight deep learning model is used for real-time fault detection to detect potential faults. Then, a Sobel edge detector is used to extract image features and key features of the image. S4, conducts in-depth analysis of equipment data based on the convolutional neural network model, outputs the fault probability through the Softmax layer after identifying potential problems, and determines the probability of different types of equipment faults according to the calculation formula; S5. Comprehensively evaluate the analysis results of the deep learning module to accurately identify the fault status of the current equipment; S6. Generate an adaptive maintenance plan based on a reinforcement learning algorithm, use a Q learning algorithm to update the state-action value function, optimize the maintenance strategy, and optimize the maintenance task scheduling through a genetic algorithm to determine the optimal maintenance task schedule; S7. Using augmented reality and user interaction modules, maintenance personnel can obtain equipment status and visualize fault information in real time, providing remote collaboration functions; S8. Responsible for system integration through system integration and security module to ensure the collaborative work among modules.

[0015] Preferably, in step S4, an attention mechanism is introduced to the output of the convolutional neural network model to highlight important features and fault types; Assume that the original output of the convolutional neural network is , the attention weight is , the improved output Calculated by the following formula: , in, Represents element-wise multiplication; Attention Weight Calculated through a separate network or algorithm, it automatically adjusts the attention paid to different features and fault types based on different input data.

[0016] Preferably, in step S6, a more comprehensive maintenance decision is made based on the results of the reinforcement learning algorithm and the genetic algorithm; Assume that the maintenance plan value obtained by the reinforcement learning algorithm is , the maintenance task scheduling fitness obtained by genetic algorithm is The maintenance cost is , the equipment importance coefficient is , the comprehensive evaluation value is calculated by the following formula:

[0017] in, , , , is the corresponding weight coefficient, which is adjusted according to actual conditions.

[0018] The present invention provides an equipment maintenance and repair operation management system and method based on industrial visual intelligence. It has the following beneficial effects: 1. The present invention can comprehensively acquire the working status and appearance data of the equipment from multiple angles through the RGB camera, infrared camera and 3D camera in the industrial vision acquisition module. The RGB camera provides a clear image of the equipment surface, which is convenient for observing details. The infrared camera detects abnormal thermal radiation and timely detects potential overheating problems. The 3D camera obtains three-dimensional deformation information on the equipment surface, which helps to more accurately evaluate the physical state changes of the equipment. The multi-sensor fusion module further integrates the data of different types of sensors, provides multi-dimensional perception capabilities of the equipment status, and realizes more comprehensive and accurate equipment status monitoring.

[0019] 2. The present invention deploys the edge computing processing module at the equipment site, which can perform local processing and real-time analysis on the data collected by the sensor. The lightweight deep learning model can perform real-time fault detection, quickly detect potential faults, and improve the efficiency of fault detection. At the same time, the Sobel edge detector performs image feature extraction and can extract the key features of the image to provide more valuable information for subsequent fault analysis. Its edge computing processing method reduces the delay and bandwidth requirements of data transmission, and improves the response speed and reliability of the system.

[0020] 3. The present invention uses a deep learning analysis module to conduct in-depth analysis of equipment data based on a convolutional neural network model, which can extract complex features of equipment failures and identify potential problems. The Softmax layer outputs the failure probability, and can determine the probability of different types of equipment failures according to the calculation formula, providing a more accurate basis for fault detection. The fault detection module conducts a comprehensive evaluation of the analysis results of the deep learning module, which can accurately identify the fault status of the current equipment and provide reliable guarantees for timely maintenance measures.

[0021] 4. The present invention generates an adaptive maintenance plan based on a reinforcement learning algorithm through an intelligent maintenance decision module, uses a Q learning algorithm to update the state-action value function, and optimizes the maintenance strategy. At the same time, a genetic algorithm optimizes maintenance task scheduling and can determine the optimal maintenance task arrangement based on the fitness function. Its intelligent maintenance decision-making method can improve maintenance efficiency, reduce maintenance costs, and extend the service life of equipment.

[0022] 5. The present invention uses augmented reality and user interaction modules to enable maintenance personnel to obtain equipment status and visualize fault information in real time, and provide remote collaboration functions. Maintenance personnel can understand the fault conditions of equipment more intuitively through augmented reality technology, thereby improving the accuracy and efficiency of fault diagnosis. The remote collaboration function allows experts to remotely guide maintenance work, solve the problem of insufficient technical skills of on-site maintenance personnel, and improve the quality and efficiency of maintenance work.

[0023] 6. The present invention is responsible for system integration through system integration and security modules to ensure the collaborative work between modules. At the same time, it guarantees the security of data transmission, prevents data leakage and tampering, provides reliable guarantee for the stable operation of the equipment maintenance and operation management system, and ensures the smooth progress of equipment maintenance work. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 is a system architecture diagram of the present invention; Figure 2 It is a schematic diagram of the industrial vision acquisition module of the present invention. DETAILED DESCRIPTION

[0025] In order to make the technical personnel in the technical field understand the scheme of the present invention, the technical scheme in the embodiment of the present invention will be clearly and completely described below in combination with the drawings in the embodiment of the present invention. Obviously, the described embodiment is a partial embodiment of the present invention, not a complete embodiment. Based on the embodiment of the present invention, other embodiments obtained by ordinary technicians in the field without creative work should fall within the scope of protection of the present invention.

[0026] The present invention is described in detail below in conjunction with the accompanying drawings: Example: Please see attached Figure 1and attached Figure 2 ,The embodiment of the present invention provides an equipment maintenance and repair operation management system based on industrial visual intelligence, including: an industrial visual acquisition module, responsible for acquiring the working status and appearance data of the equipment in real time through industrial-grade visual sensors; The multi-sensor fusion module is used to integrate the working status and appearance data of the equipment obtained in real time from the industrial vision acquisition module, providing multi-dimensional perception capabilities of the equipment status; The edge computing processing module is responsible for local processing and real-time analysis of the working status and appearance data of the equipment acquired by the industrial vision acquisition module in real time; The deep learning analysis module is responsible for in-depth analysis of the working status and appearance data of the equipment acquired in real time through deep learning algorithms, extracting the complex characteristics of equipment failures and identifying potential problems; The fault detection module is responsible for comprehensively evaluating the analysis results of the deep learning analysis module and identifying the fault status of the current equipment; Intelligent maintenance decision module, which generates adaptive maintenance plans based on the reinforcement learning algorithm of the deep learning analysis module; Augmented reality and user interaction module, which is used by maintenance personnel to obtain equipment status in real time, visualize fault information and provide remote collaboration functions; The system integration and security module is responsible for system integration and data transmission security.

[0027] The benefits of industrial vision acquisition modules help detect appearance damage, promptly warn of potential overheating failures, and more accurately assess changes in the physical state of equipment; Benefits of multi-sensor fusion module: It realizes multi-dimensional perception of equipment status, avoids the limitations of a single sensor, improves the accuracy and reliability of equipment status judgment, and more accurately reflects the actual situation of the equipment; Benefits of edge computing processing modules Reduce data transmission delays and bandwidth requirements, detect potential failures in a timely manner, and provide timely maintenance for equipment; Benefits of the deep learning analysis module Automatically learn patterns and rules in data without manually designing complex feature extraction methods, providing a quantitative basis for fault detection and maintenance decisions; The benefits of the fault detection module are to avoid misjudgment of a single analysis method, improve the accuracy of fault detection, provide a reliable basis for taking effective maintenance measures in a timely manner, and reduce the losses caused by equipment failures; The benefits of the intelligent maintenance decision module are that it can make dynamic adjustments based on the actual operating status and fault conditions of the equipment, improve the flexibility and adaptability of maintenance decisions, reduce maintenance costs, and ensure efficient maintenance work; Benefits of augmented reality and user interaction module Improve the accuracy and efficiency of fault diagnosis, quickly locate problems, provide remote collaboration functions, and improve the quality of maintenance work; The benefits of system integration and security modules improve the overall performance and stability of the system, ensure the security of data transmission, prevent data leakage and tampering, protect the security and reliability of the equipment operation management system, and provide stable technical support for equipment maintenance and repair work.

[0028] Industrial vision acquisition modules include: RGB camera unit, used to capture images of the device surface; Infrared camera unit to detect thermal radiation anomalies of the equipment; A 3D camera unit, used to obtain three-dimensional deformation information of the device surface; The multi-sensor fusion module fuses the sensor data through a weighted fusion algorithm. The fusion formula is: , in, is the data of the i-th sensor, is the corresponding weight, n is the number of sensors, Represents the fused features or results.

[0029] The RGB camera unit helps maintenance personnel observe whether there are scratches, wear, corrosion, etc. on the surface of the equipment, or whether the equipment casing has cracks due to long-term vibration. Accurately identifying and locating equipment parts helps to quickly find the parts that need to be repaired or replaced during maintenance. The surface images provided are important basic data for subsequent analysis of the working status of the equipment. The infrared camera unit can detect overheated electrical components, such as transformers and motors, in a timely manner. It can detect the temperature distribution on their surfaces, help determine whether energy is being used effectively, whether there is excessive heat loss, and optimize the operating parameters of the equipment to reduce energy loss. The 3D camera unit monitors the three-dimensional deformation of key components of the equipment in real time, detects component deformation caused by long-term stress, fatigue and other factors in advance, measures the gap and relative position between components such as the engine cylinder and piston, ensures that the assembly accuracy meets the requirements, and improves the performance and reliability of the equipment; The weighted fusion algorithm can integrate device information of multiple different dimensions to assign weights based on factors such as the importance, accuracy, and reliability of different sensor data, thereby increasing the weight of infrared camera data in a temperature-sensitive device fault detection scenario.

[0030] The edge computing processing module includes: The lightweight deep learning model is used for real-time fault detection. The lightweight deep learning model uses deep separable convolution, and its convolution operation formula is: , in, is the input feature map, is the depth convolution kernel, is the convolution kernel, Represents the output feature map or result matrix.

[0031] The role of depthwise separable convolution in lightweight deep learning models is to reduce the computational complexity of convolution operations, which cannot meet the real-time requirements. However, depthwise separable convolution significantly reduces the computational complexity while ensuring a certain accuracy, allowing the model to run quickly on edge devices. Depth-wise separable convolution improves the efficiency of the production process by optimizing the production steps. In edge computing scenarios, it can extract useful information from the input feature map for fault detection more quickly, thereby improving the efficiency of real-time fault detection. The lightweight model using depthwise separable convolution can realize real-time processing of device data without taking up too much storage space, providing timely support for subsequent fault detection. At the same time, the low computational complexity enables the model to run quickly on the low-power processor of the edge device, reducing energy consumption.

[0032] The edge computing processing module further includes a Sobel edge detector for image feature extraction, and its calculation formula is: , in, and Represents the gradient of the image in the horizontal and vertical directions. represents the image gradient magnitude, Represents the gray value distribution function of the image, item.

[0033] The role of horizontal and vertical gradient calculations is to produce values ​​at the crack location and accurately locate the horizontal position of the crack. Similarly, locating the vertical position of the crack helps to identify minor damage on the surface of the equipment, boundaries between components, etc. By determining the main direction of texture change, a basis is provided for subsequent fault type judgment. Scratches on the surface of the equipment will cause weaker edge strength, while deeper cracks will cause stronger edge strength. Through calculation, edge strength information is extracted to provide quantitative indicators for fault severity assessment. Maintenance personnel can set thresholds. If the threshold is exceeded, it is judged that the equipment has a more serious fault and needs timely maintenance. Through calculation, the areas with obvious brightness changes, namely the edge areas, are highlighted, while the areas with smaller brightness changes are suppressed. In the subsequent fault analysis, more attention is paid to the equipment features represented by the edge areas, such as the outline of the components, the boundaries of the deformation areas, etc., thereby improving the accuracy and efficiency of fault detection.

[0034] The deep learning analysis module performs fault classification based on a convolutional neural network model. The output of the convolutional layer is calculated by the following formula: , in, is the convolution kernel of the lth layer, is the input feature map, is the bias, is the nonlinear activation function, represents the hidden state of the i+1th layer or time step; The deep learning analysis module includes a Softmax layer for fault classification, and its output fault probability is calculated by the following formula: , in, Indicates that given an input feature vector The probability of category j under the condition of ; is the input feature vector, For input features The score corresponding to category j; K represents the total number of categories, The denominator represents the sum of the index scores of all categories, which is used for normalization so that the sum of the probabilities of all categories is 1.

[0035] The output formula of the convolution layer can automatically extract local features from the input feature map. For the image data of industrial equipment, the convolution layer extracts the shape, texture and other features of the equipment parts. Through multiple layers of convolution, the model can learn abstract and advanced feature representations, which are very critical for the identification of equipment failures. When the model identifies overheating failures of equipment, by adjusting the convolution kernel and bias, the convolution layer can pay more attention to the texture and shape features related to the overheating area in the infrared image; Through nonlinear activation functions, the convolutional layer can better fit complex relationships. The characteristics of fatigue damage to equipment will be obvious after a certain threshold. Nonlinear activation functions can help the model capture nonlinear changes and enhance the model's ability to express fault characteristics. Provides a quantitative basis for fault classification. For various types of equipment faults, the formula can calculate the probability of the equipment belonging to various fault types, allowing maintenance personnel to determine the fault type of the equipment based on the probability; In actual applications, it helps maintenance personnel assess risks. At the same time, by observing the changing trend of probability, it monitors the development process of equipment failures and provides early warning of failures.

[0036] The intelligent maintenance decision module generates maintenance plans based on the reinforcement learning algorithm and uses the Q learning algorithm to update the state-action value function. , and its update formula is: , in, represents the Q value of taking action a in state s, represents the learning rate, Indicates immediate reward, represents the discount factor, Indicates the next state Determine the optimal action The maximum Q value, It represents the next state reached after taking action a in the current state s. represents the optimal action.

[0037] The function of the state-action value function update formula is to learn the long-term value of taking different maintenance actions under different equipment states and optimize the maintenance strategy; The immediate reward represents the direct feedback from the current action, while the discount factor takes into account the long-term benefits. It is used to weigh the importance of short-term and long-term rewards and balance the short-term and long-term maintenance benefits. The learning rate determines the speed at which new experience covers old experience. During equipment operation, the equipment status and the corresponding optimal maintenance actions will change due to changes in time and environment. Through reasonable settings, the system can quickly adjust the maintenance strategy according to new experience, so that the maintenance plan can adapt to changes and always maintain the optimal or near-optimal state.

[0038] The intelligent maintenance decision module optimizes the maintenance task scheduling through genetic algorithm, and the fitness function is: , in, is the cost of the ith maintenance, Maintenance is time-consuming. is the balance coefficient, represents the total cost, and N represents the total number of cost items or time items.

[0039] The role of the fitness function formula in equipment maintenance and repair operation management. Focusing on cost or time consumption is not comprehensive, which will cause equipment to shut down for a long time and affect production efficiency; When optimizing maintenance task scheduling, the genetic algorithm uses the fitness function to measure the quality of each individual. A high fitness function value indicates that the solution is good. During the evolution process, the genetic algorithm continuously generates new individuals through operations such as selection, crossover, and mutation, and evaluates and screens them according to the fitness function. The balance coefficient plays a role in balancing the maintenance cost and maintenance time in the fitness function. By adjusting the values ​​of and, the relative importance of cost and time in the optimization process is determined according to actual needs.

[0040] A method for equipment maintenance, repair and operation management based on industrial visual intelligence, comprising the following steps: S1. First, use the RGB camera in the industrial vision acquisition module to collect the surface image of the equipment, then use the infrared camera to detect the abnormal thermal radiation of the equipment and detect the overheating parts in time. Finally, use the 3D camera to obtain the three-dimensional deformation information of the equipment surface; S2, using a multi-sensor fusion module, using a weighted fusion algorithm to integrate data from different types of sensors to obtain multi-dimensional perception data of the device status; S3, deploy edge computing processing modules at the equipment site. First, a lightweight deep learning model is used for real-time fault detection to detect potential faults. Then, a Sobel edge detector is used to extract image features and key features of the image. S4, conducts in-depth analysis of equipment data based on the convolutional neural network model, outputs the fault probability through the Softmax layer after identifying potential problems, and determines the probability of different types of equipment faults according to the calculation formula; S5. Comprehensively evaluate the analysis results of the deep learning module to accurately identify the fault status of the current equipment; S6. Generate an adaptive maintenance plan based on a reinforcement learning algorithm, use a Q learning algorithm to update the state-action value function, optimize the maintenance strategy, and optimize the maintenance task scheduling through a genetic algorithm to determine the optimal maintenance task schedule; S7. Using augmented reality and user interaction modules, maintenance personnel can obtain equipment status and visualize fault information in real time, providing remote collaboration functions; S8. Responsible for system integration through system integration and security module to ensure the collaborative work among modules.

[0041] In step S4, an attention mechanism is introduced to the output of the convolutional neural network model to highlight important features and fault types; Assume that the original output of the convolutional neural network is , the attention weight is , the improved output Calculated by the following formula: , in, Represents element-wise multiplication; Attention Weight Calculated through a separate network or algorithm, it automatically adjusts the attention paid to different features and fault types based on different input data.

[0042] By multiplying the original output of the convolutional neural network by the attention weight element by element, the important features and fault types in the original output can be effectively highlighted. In equipment maintenance and repair, different features and fault types have different impacts on the equipment operation status. A specific fault type will affect the safety and reliability of the equipment, while other fault types are relatively minor. By introducing the attention mechanism, the attention to different features and fault types can be automatically adjusted, and more resources can be concentrated on key information, thus improving the accuracy of fault detection and diagnosis. The attention weights are calculated by a separate network or algorithm and can be automatically adjusted according to the input data, which means that the model can dynamically adjust the degree of attention to different features according to different device states, working environments and fault conditions; In equipment maintenance and repair, it is crucial to accurately diagnose the type of fault. By highlighting important features and fault types, the improved output provides maintenance personnel with more targeted information, helping them to determine the cause of the fault more quickly and take appropriate maintenance measures.

[0043] In step S6, a more comprehensive maintenance decision is made based on the results of the reinforcement learning algorithm and the genetic algorithm; Assume that the maintenance plan value obtained by the reinforcement learning algorithm is , the maintenance task scheduling fitness obtained by genetic algorithm is The maintenance cost is , the equipment importance coefficient is , the comprehensive evaluation value is calculated by the following formula:

[0044] in, , , , is the corresponding weight coefficient, which is adjusted according to actual conditions.

[0045] The role of the comprehensive evaluation value formula is to combine the maintenance plan value obtained by the reinforcement learning algorithm, the maintenance task scheduling fitness obtained by the genetic algorithm, the maintenance cost and the equipment importance coefficient. In the equipment maintenance and repair decision-making, relying solely on reinforcement learning or genetic algorithms cannot fully consider all key factors. Through the formula, the advantages of different algorithms are combined, and comprehensive evaluation is performed from multiple angles to make more comprehensive and reasonable maintenance decisions; The weight coefficient in the formula is adjusted according to the actual situation, so that decision makers can flexibly balance the relationship between different goals according to specific application scenarios and needs.

[0046] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An equipment maintenance and repair operation management system based on industrial visual intelligence, characterized in that: include: The industrial vision acquisition module is responsible for obtaining the working status and appearance data of the equipment in real time through industrial-grade visual sensors; The multi-sensor fusion module is used to integrate the working status and appearance data of the equipment obtained in real time from the industrial vision acquisition module, providing multi-dimensional perception capabilities of the equipment status; The edge computing processing module is responsible for local processing and real-time analysis of the working status and appearance data of the equipment acquired by the industrial vision acquisition module in real time; The deep learning analysis module is responsible for in-depth analysis of the working status and appearance data of the equipment acquired in real time through deep learning algorithms, extracting the complex characteristics of equipment failures and identifying potential problems; The fault detection module is responsible for comprehensively evaluating the analysis results of the deep learning analysis module and identifying the fault status of the current equipment; Intelligent maintenance decision module, which generates adaptive maintenance plans based on the reinforcement learning algorithm of the deep learning analysis module; Augmented reality and user interaction module, which is used by maintenance personnel to obtain equipment status in real time, visualize fault information and provide remote collaboration functions; The system integration and security module is responsible for system integration and data transmission security.

2. According to claim 1, an equipment maintenance and repair operation management system based on industrial visual intelligence is characterized in that: The industrial vision acquisition module includes: RGB camera unit, used to capture images of the device surface; Infrared camera unit to detect thermal radiation anomalies of the equipment; A 3D camera unit, used to obtain three-dimensional deformation information of the device surface; The multi-sensor fusion module fuses the sensor data through a weighted fusion algorithm, and the fusion formula is: ， in, is the data of the i-th sensor, is the corresponding weight, n is the number of sensors, Represents the fused features or results.

3. According to claim 1, the equipment maintenance and repair operation management system based on industrial visual intelligence is characterized in that: The edge computing processing module includes: The lightweight deep learning model is used for real-time fault detection. The lightweight deep learning model adopts deep separable convolution, and its convolution operation formula is: ， in, is the input feature map, is the depth convolution kernel, is the convolution kernel, Represents the output feature map or result matrix.

4. The equipment maintenance and repair operation management system based on industrial visual intelligence according to claim 3 is characterized in that: The edge computing processing module further includes a Sobel edge detector for image feature extraction, and its calculation formula is: ， in, and Represents the gradient of the image in the horizontal and vertical directions. represents the image gradient magnitude, Represents the gray value distribution function of the image, item.

5. According to the equipment maintenance and repair operation management system based on industrial visual intelligence in claim 1, it is characterized in that: The deep learning analysis module performs fault classification based on a convolutional neural network model, and the output of the convolutional layer is calculated by the following formula: ， in, is the convolution kernel of the lth layer, is the input feature map, is the bias, is the nonlinear activation function, represents the hidden state of the i+1th layer or time step; The deep learning analysis module includes a Softmax layer for fault classification, and the fault probability of its output is calculated by the following formula: ， in, Indicates that given an input feature vector The probability of category j under the condition of ; is the input feature vector, For input features The score corresponding to category j; K represents the total number of categories, The denominator represents the sum of the index scores of all categories, which is used for normalization so that the sum of the probabilities of all categories is 1.

6. The equipment maintenance and repair operation management system based on industrial visual intelligence according to claim 1 is characterized in that: The intelligent maintenance decision module generates a maintenance plan based on a reinforcement learning algorithm and uses a Q learning algorithm to update the state-action value function , and its update formula is: ， in, represents the Q value of taking action a in state s, represents the learning rate, Indicates immediate reward, represents the discount factor, Indicates the next state Determine the best action The maximum Q value, It represents the next state reached after taking action a in the current state s. represents the optimal action.

7. The equipment maintenance and repair operation management system based on industrial visual intelligence according to claim 1 is characterized in that: The intelligent maintenance decision module optimizes maintenance task scheduling through genetic algorithm, and the fitness function is: ， in, is the cost of the ith maintenance, Maintenance is time-consuming. is the balance coefficient, represents the total cost, and N represents the total number of cost items or time items.

8. A method for equipment maintenance, repair and operation management based on industrial visual intelligence, according to the equipment maintenance, repair and operation management system based on industrial visual intelligence according to any one of claims 1 to 7, characterized in that: The method comprises the following steps: S1. First, the RGB camera in the industrial vision acquisition module is used to collect the surface image of the equipment, and then the thermal radiation anomaly of the equipment is detected by the infrared camera to detect the overheated parts in time, and finally, the three-dimensional deformation information of the surface of the equipment is obtained by the 3D camera; S2, using a multi-sensor fusion module, using a weighted fusion algorithm to integrate data from different types of sensors to obtain multi-dimensional perception data of the device status; S3, deploy edge computing processing modules at the equipment site. First, a lightweight deep learning model is used for real-time fault detection to detect potential faults. Then, a Sobel edge detector is used to extract image features and key features of the image. S4, conducts in-depth analysis of equipment data based on the convolutional neural network model, outputs the fault probability through the Softmax layer after identifying potential problems, and determines the probability of different types of equipment faults according to the calculation formula; S5. Comprehensively evaluate the analysis results of the deep learning module to accurately identify the fault status of the current equipment; S6. Generate an adaptive maintenance plan based on a reinforcement learning algorithm, use a Q learning algorithm to update the state-action value function, optimize the maintenance strategy, and optimize the maintenance task scheduling through a genetic algorithm to determine the optimal maintenance task schedule; S7. Using augmented reality and user interaction modules, maintenance personnel can obtain equipment status and visualize fault information in real time, providing remote collaboration functions; S8. Responsible for system integration through system integration and security module to ensure the collaborative work among modules.

9. The equipment maintenance and repair operation management method based on industrial visual intelligence according to claim 8 is characterized in that: In the step S4, an attention mechanism is introduced to the output of the convolutional neural network model to highlight important features and fault types; Assume that the original output of the convolutional neural network is , the attention weight is , the improved output Calculated by the following formula: ， in, Represents element-wise multiplication; Attention Weight Calculated through a separate network or algorithm, it automatically adjusts the attention paid to different features and fault types based on different input data.

10. The equipment maintenance and repair operation management method based on industrial visual intelligence according to claim 8 is characterized in that: In step S6, a more comprehensive maintenance decision is made based on the results of the reinforcement learning algorithm and the genetic algorithm; Assume that the maintenance plan value obtained by the reinforcement learning algorithm is , the maintenance task scheduling fitness obtained by genetic algorithm is The maintenance cost is , the equipment importance coefficient is , the comprehensive evaluation value is calculated by the following formula: in, , , , is the corresponding weight coefficient, which is adjusted according to actual conditions.

Images