Industrial defective product recovery management system for industrial internet data processing

Abstract

The invention relates to the technical field of defective product recovery route management, in particular to an industrial defective product recovery management system for industrial internet data processing, which comprises an intelligent monitoring classification unit, a traceability and route planning unit and a maintenance closed loop optimization unit, the intelligent monitoring classification unit adopts a multispectral imaging technology and a convolutional neural network image enhancement algorithm, combines a fuzzy clustering and support vector machine hybrid classification method, identifies and classifies defective products, triggers the recovery, traceability and path planning unit to generate a block chain ID through a Hash algorithm and a timestamp, fuses real-time traffic data, and improves the real-time traffic quality. An optimal recovery route is planned by using a genetic and ant colony hybrid algorithm, a real-time road condition prediction model and a multi-agent system are introduced to improve the planning effect, a closed-loop optimization unit is maintained to predict a defect trend based on a deep belief network and a random forest hybrid model, and production process parameters are adjusted by using a model prediction control algorithm. And the recovery management efficiency of the industrial defective products is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of defective product recycling route management. Specifically, it relates to an industrial defective product recycling management system for industrial Internet data processing. Background Art

[0002] The management of defective product recycling routes is an important technology, but the traditional defective product recycling route planning method has significant deficiencies.

[0003] Currently, the recycling route planning often determines the route based on experience or simple distance calculation, without fully considering the actual traffic conditions and the complexity of product recycling. With the increasing severity of urban traffic congestion problems and the ever-changing road conditions, such as frequent occurrences of road congestion, traffic accidents, and road construction, the traditional planning method mostly relies on fixed geographical information and past experience to plan the recycling route, making it difficult to adapt to these dynamic changes. As a result, the recycling vehicles often encounter long-term congestion, which not only increases the transportation cost and time cost but also may cause delays in the processing of recycled products, affecting the production rhythm and economic benefits of the enterprise. In addition, the traditional method does not consider the load limit and driving time limit of the recycling vehicles, which may result in overloading or illegal driving of the vehicles, posing potential safety hazards. This lack of scientific planning for recycling routes seriously restricts the efficiency and quality of industrial defective product recycling management. To solve this technical problem, we have provided an industrial defective product recycling management system for industrial Internet data processing. Summary of the Invention

[0004] The purpose of the present invention is to provide an industrial defective product recycling management system for industrial Internet data processing to solve the problems raised in the above background art.

[0005] To achieve the above purpose, an industrial defective product recycling management system for industrial Internet data processing is provided, including an intelligent monitoring and classification unit, a traceability and path planning unit, and a maintenance closed-loop optimization unit;

[0006] The intelligent monitoring and classification unit scans the surface and internal defects of the product through a vision sensor and classifies the defects based on edge computing. After the classification is completed, the recycling process is triggered;

[0007] The traceability and path planning unit generates a unique blockchain ID for each defective product, records the original information of the product, and then plans the optimal recycling route for each defective product in combination with real-time traffic data and reverse logistics algorithms;

[0008] The maintenance closed-loop optimization unit is trained based on the AI prediction model using historical defective data, predicts the defect trend, adjusts the production process parameters according to the defect trend, and finally feeds back the repair data of the recycled product to the production end.

[0009] As a further improvement of this technical solution, the visual sensor in the intelligent monitoring and classification unit adopts multi-spectral imaging technology:

[0010] Collect image information of the product in different spectral bands, obtain the defect characteristics on the surface and inside of the product, and use an image enhancement algorithm based on a convolutional neural network to preprocess the collected multi-spectral images, and learn the mapping relationship between the defect characteristics in the image and the enhanced image.

[0011] As a further improvement of this technical solution, when the intelligent monitoring and classification unit classifies defects based on edge computing, a hybrid classification method combining fuzzy clustering algorithm and support vector machine is adopted:

[0012] Use the fuzzy clustering algorithm to preliminarily cluster the collected defect feature data, divide different fuzzy categories, and each category has a certain membership degree;

[0013] Take the clustering result as the input and use the support vector machine to classify each fuzzy category, determine the specific type of the defect, and adopt an adaptive parameter adjustment mechanism to adjust the parameters of the fuzzy clustering algorithm and the support vector machine according to the historical defect data of the product and the defect characteristics collected in real time.

[0014] As a further improvement of this technical solution, when the traceability and path planning unit generates a unique blockchain ID for each defective product, a method combining a hash algorithm and a timestamp is adopted:

[0015] Perform a hash operation on the production batch number, production date, and workstation number on the production line of the product to obtain a hash value, and then combine this hash value with the current timestamp to generate the final blockchain ID.

[0016] As a further improvement of this technical solution, when the traceability and path planning unit records the original information of the product, a distributed ledger technology is used to store the information on multiple nodes, specifically as follows:

[0017] Each node stores a complete copy of the original product information, and maintains the consistency of the information on each node through a consensus mechanism. When recording information, the information is encrypted. A symmetric encryption algorithm is used to encrypt the original information, and the encryption key is managed by an asymmetric encryption algorithm.

[0018] As a further improvement of this technical solution, when the traceability and path planning unit plans the optimal recovery route for each defective product by combining real-time traffic data and reverse logistics algorithms, the real-time traffic data is obtained through data interaction with the traffic information platform, including road congestion conditions, traffic accident information, and road construction information. The reverse logistics algorithm uses a hybrid optimization algorithm that combines genetic algorithms and ant colony algorithms;

[0019] The genetic algorithm is used for global search to find a solution space for a set of recovery routes, and then the ant colony algorithm is used for local optimization within this solution space to obtain the optimal recovery route. At the same time, considering the load limit and driving time limit of the recovery vehicle, constraint conditions are added to the algorithm.

[0020] As a further improvement of this technical solution, when the traceability and path planning unit plans the recovery route, a real-time road condition prediction model is introduced, as follows:

[0021] The real-time road condition prediction model is trained using a long short-term memory network based on historical traffic data and real-time traffic information to predict road congestion conditions. When planning the route, the route is dynamically adjusted according to the predicted road condition information.

[0022] As a further improvement of this technical solution, the traceability and path planning unit uses a multi-agent system (MAS) for route planning, as follows:

[0023] Each defective product is regarded as an agent. The agents complete the planning of the recovery route through information interaction and collaboration, and at the same time, a reinforcement learning algorithm is introduced to optimize the decision-making of the agents to obtain the optimal recovery route.

[0024] As a further improvement of this technical solution, when the maintenance closed-loop optimization unit trains using historical defect data based on the AI prediction model, the AI prediction model uses a hybrid model that combines a deep belief network and a random forest:

[0025] The deep belief network is used to extract features and reduce the dimension of the historical defect data, learn the deep feature representation of the data, and then the extracted features are input into the random forest for training to establish a defect trend prediction model. Cross-validation and regularization methods are used to optimize the model.

[0026] As a further improvement of this technical solution, when the maintenance closed-loop optimization unit adjusts the production process parameters according to the defect trend, a model predictive control algorithm is used:

[0027] The model predictive control algorithm establishes a dynamic model between production process parameters and defect indicators according to the defect trend predicted by the AI prediction model, then solves the optimal production process parameter adjustment strategy within the prediction time domain, and adds constraint equations to the algorithm in combination with the constraints of the production process. Finally, when the repair data of the recycled products is fed back to the production end, a data mining algorithm is used to analyze the repair data, extract the analysis results, and conduct correlation analysis between the analysis results and the production process parameters.

[0028] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0029] In an industrial defective product recycling management system for industrial Internet data processing, the traceability and path planning unit generates a unique blockchain ID through a hash algorithm combined with a timestamp, and uses distributed ledger technology to store the original information to ensure the security and reliability of the information. Then, a hybrid optimization algorithm of genetic algorithm and ant colony algorithm is used to plan the optimal recycling route in combination with real-time traffic data, and a real-time road condition prediction model and a multi-agent system are introduced. Considering the load and travel time limits, the transportation cost is reduced, the recycling efficiency is improved, and the maintenance closed-loop optimization unit trains and predicts the defect trend based on a hybrid model of deep belief network and random forest, adjusts the production process parameters through the model predictive control algorithm, analyzes the repair data using the data mining algorithm and correlates the production process parameters, continuously optimizes production, reduces the generation of defective products, forms a complete management closed-loop, and improves the overall efficiency of industrial production. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 is the overall block diagram of the present invention;

[0031] Figure 2 is the schematic diagram of the defect classification process;

[0032] Figure 3 is the flowchart of the recycling route planning.

[0033] The meanings of the various labels in the figure are as follows:

[0034] 1. Intelligent monitoring and classification unit; 2. Traceability and path planning unit; 3. Maintenance closed-loop optimization unit. SPECIFIC EMBODIMENTS

[0035] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0036] The present invention provides an industrial defective product recycling management system for industrial Internet data processing. Please refer toFigures 1-3 As shown in the figure, it includes an intelligent monitoring and classification unit 1, a traceability and path planning unit 2, and a maintenance closed-loop optimization unit 3;

[0037] The intelligent monitoring and classification unit 1 scans the surface and internal defects of the product through a vision sensor, classifies the defects based on edge computing, and triggers the recycling process after classification;

[0038] The vision sensor in the intelligent monitoring and classification unit 1 adopts multi-spectral imaging technology:

[0039] Collect image information of the product in different spectral bands to obtain the surface and internal defect characteristics of the product. Different defects may have different manifestation characteristics in different spectral bands. A single spectral image may not be able to comprehensively capture the defect information on the surface and inside of the product. For the collected multi-spectral images, use an image enhancement algorithm based on a convolutional neural network to preprocess the images, learn the mapping relationship between the defect characteristics in the images and the enhanced images, without the need to manually design complex enhancement rules. At the same time, through a large amount of data training, the model has good generalization ability and can adapt to different types of defects and image conditions, which is beneficial to subsequent defect classification and recognition tasks and improves the performance of the entire intelligent monitoring and classification unit;

[0040] When the intelligent monitoring and classification unit 1 classifies the defects based on edge computing, it adopts a hybrid classification method combining fuzzy clustering algorithm and support vector machine:

[0041] The collected defect feature data may have complex data distributions and unclear boundaries. Traditional hard clustering methods are difficult to accurately classify categories. Use the fuzzy clustering algorithm to preliminarily cluster the collected defect feature data, divide it into different fuzzy categories, and each category has a certain membership degree, providing more targeted input for the subsequent accurate classification of the support vector machine, improving the efficiency and accuracy of the overall classification;

[0042] Although the fuzzy clustering algorithm can perform preliminary classification, it cannot accurately determine the specific type of the defect. Use the support vector machine to classify each fuzzy category with the clustering result as the input, which can find the optimal classification hyperplane in the high-dimensional space, accurately classify each fuzzy category, and determine the specific type of the defect. For each fuzzy category, regard its data as an independent data set and use the support vector machine for classification. Let the data set of the th fuzzy category be , where is a preset membership degree threshold, is expressed as the th data point, is the belonging to the Membership degrees of each category, and select a kernel function Construct a support vector machine classification model, where is denoted as the th data point. Determine the classification hyperplane by solving the following optimization problem, that is:

[0043] ; , , where , are Lagrange multipliers, , is the data point , 's class label, is the kernel function, is the penalty parameter, is the number of defect feature data. Solving the above optimization problem gives the optimal Lagrange multiplier , and then determine the classification decision function , where is the bias term. Further refine the preliminary classification results obtained by fuzzy clustering, accurately determine the specific type of defect, and provide more explicit information for subsequent processing and decision-making. According to the distribution of historical defect data and the changes in defect features collected in real time, dynamically adjust the fuzzy weighting index and the number of clusters. For the support vector machine algorithm, dynamically adjust the width parameter and penalty parameter of the kernel function according to the characteristics of historical data and real-time data, so that it can maintain a high classification accuracy under different products and production conditions, improving the overall performance and reliability of the intelligent monitoring classification unit.

[0044] The traceability and path planning unit 2 generates a unique blockchain ID for each defective product, records the original information of the product, and then plans the optimal recycling route for each defective product in combination with real-time traffic data and reverse logistics algorithms;

[0045] When the traceability and path planning unit 2 generates a unique blockchain ID for each defective product, it uses the method of combining the hash algorithm with the timestamp:

[0046] The production batch number, production date, and workstation number on the production line of the product contain key information during the product production process. By performing a hash operation on the production batch number, production date, and workstation number of the product, a hash value is obtained, thereby providing a unique identifier basis based on production information for each product, which is an important part of generating the final blockchain ID subsequently, improving the accuracy and reliability of product identification. Using the system clock to obtain the current timestamp avoids the problem of identification conflicts caused by the same production information, enabling each defective product to have a unique identifier. Concatenating the previously obtained hash value and the timestamp to get a new string, and then performing a hash operation on the new string to generate the final blockchain ID, which provides strong support for the accurate traceability and management of defective products on the blockchain;

[0047] When the traceability and path planning unit 2 records the original information of the product, it uses distributed ledger technology to store the information on multiple nodes, specifically as follows:

[0048] In the distributed ledger system, having each node save a complete copy of the product's original information can enhance the redundancy and reliability of the data. Since multiple nodes have copies of the product's original information, when the data is updated or new information is added, it is necessary to ensure that the information on each node is consistent. Otherwise, it will lead to data inconsistency problems, affecting the normal operation of the system and the accuracy of the data. The consensus mechanism is used to maintain the consistency of the information on each node. The consensus mechanism can quickly reach a consensus in the presence of a certain number of faulty nodes, ensuring data consistency. The original information of the product may contain sensitive data. To protect the security and privacy of this data, it needs to be encrypted. When recording the information, the information is encrypted. A symmetric encryption algorithm is used to encrypt the original information to prevent the data from being stolen or tampered with during transmission and storage. The encryption key is managed using an asymmetric encryption algorithm, that is, each node generates a pair of public and private keys , where is the public key, is the private key. When node wants to send the symmetric encryption key to node , it uses the public key of node to encrypt , that is . Node receives and uses its own private key to decrypt it, that is , improving the management security of the symmetric encryption key and further enhancing the reliability of the encryption of the product's original information;

[0049] When the traceability and path planning unit 2 plans the optimal recycling route for each defective product by combining real-time traffic data and reverse logistics algorithms, the real-time traffic data is obtained through data interaction with the traffic information platform, including road congestion conditions, traffic accident information, and road construction information, and the road congestion coefficient is extracted. Whether there is a traffic accident , if it exists, it is 1, if it does not exist, it is 0, and whether there is road construction , if it exists, it is 1, if it does not exist, it is 0, ensuring that the planned recycling route is based on the latest traffic conditions, improving the timeliness and feasibility of the route. The reverse logistics algorithm uses a hybrid optimization algorithm that combines genetic algorithms and ant colony algorithms. The search space for recycling routes is usually very large, and the computational complexity of directly finding the optimal route is very high. Genetic algorithms have strong global search capabilities and can quickly find a set of relatively optimal recycling route solution spaces in a large search space, providing a basis for subsequent local optimization.

[0050] Use genetic algorithms for global search. Represent the recycling route as a chromosome, where each gene represents the number of a location to be visited, such as the location of the defective product or the recycling center, and randomly generate a certain number of chromosomes to form an initial population , and define a fitness function to evaluate the quality of each chromosome (i.e., each recycling route). Use the roulette wheel selection method to select a certain number of chromosomes from the current population as the parent generation. Perform crossover operations on the parent chromosomes to generate offspring chromosomes, and then perform mutation operations on the offspring chromosomes to introduce new gene combinations. After a certain number of iterations, a set of recycling route solution spaces is obtained, providing a good basis for subsequent local optimization. Then, use the ant colony algorithm for local optimization within this solution space. Initialize the pheromone concentration on each route in the solution space , where and represent two adjacent locations on the route. In each iteration, let a certain number of ants build their own recycling routes in the solution space according to the pheromone concentration and distance. After the ants complete route construction, update the pheromone concentration according to the quality of the route. Repeat the above steps until the termination condition is met to obtain the optimal recycling route, making the route more reasonable and reducing the recycling cost and time. In the actual recycling process, the recycling vehicle has load limits and driving time limits. At the same time, combining the load limits and driving time limits of the recycling vehicle, add constraint conditions to the algorithm to make the planned recycling route more in line with the actual transportation situation, improving the feasibility and practicality of the route.

[0051] When the traceability and path planning unit 2 plans the recycling route, it introduces a real-time road condition prediction model, as follows:

[0052] The real-time traffic condition prediction model is trained using a long short-term memory network based on historical traffic data and real-time traffic information to predict road congestion. When planning a route, the route is dynamically adjusted according to the predicted traffic condition information. When initially planning a recycling route, a path planning algorithm is used to generate an initial route in combination with real-time traffic information and predicted traffic condition information. During vehicle driving, the current location and predicted traffic condition information are obtained in real time. If it is found that there is a section of the current route that is about to be congested, the route is recalculated. Let the current location be , and the target location be . The predicted traffic condition information is converted into the weight of the road section . The path planning algorithm is used again with as the starting point as the end point, and a new route is calculated according to the new road section weights, reducing the vehicle's driving time and fuel consumption and lowering the recycling cost;

[0053] The traceability and path planning unit 2 uses a multi-agent system for route planning, as follows:

[0054] Regarding each defective product as an agent, it can fully consider the characteristics and requirements of each product, making the route planning more personalized and accurate. Through information interaction and collaboration among agents, the recycling route planning is jointly completed. It can integrate global information and jointly explore a better recycling route plan, avoiding the limitations of single decision-making. At the same time, a reinforcement learning algorithm is introduced to optimize the decisions of the agents, defining the state space . The state of the agent includes its own position, the positions and states of other surrounding agents, and the current traffic condition information. The action space is defined . The action of the agent represents the selection of the next location to go to. The reward function is defined . The reward function is used to evaluate the quality of the agent taking action in state . The deep Q-network learning algorithm is used for reinforcement learning. The agent regularly shares its own state information, learned strategies, and planned route segments with other agents. When conflicts occur in the decisions of multiple agents, a negotiation mechanism is used to solve them. Based on sufficient information interaction and collaboration among agents, the decisions of each agent are integrated to generate a global recycling route, improving the efficiency and performance of the entire recycling system.

[0055] The maintenance closed-loop optimization unit 3 is trained using historical defect data based on an AI prediction model to predict defect trends, and adjusts production process parameters according to the defect trends. Finally, the repair data of the recycled products is fed back to the production end;

[0056] When the maintenance closed-loop optimization unit 3 is trained using historical defect data based on the AI prediction model, the AI prediction model adopts a hybrid model combining a deep belief network and a random forest:

[0057] Historical defect data may have high-dimensional and complex feature relationships. Directly using it for training the model may lead to high computational complexity and overfitting problems. The deep belief network is used to extract features and reduce the dimension of the historical defect data, learning the deep feature representation of the data. By means of layer-by-layer training, the essential features of the data are extracted, and at the same time, dimensionality reduction is achieved. The deep belief network is composed of multiple restricted Boltzmann machines stacked together. Each restricted Boltzmann machine consists of a visible layer and a hidden layer. There are no connections between nodes within the layer, and all nodes between layers are fully connected. The contrastive divergence algorithm is used to train each restricted Boltzmann machine. Taking the first restricted Boltzmann machine as an example, the input data is sent to the visible layer, the activation probability of the hidden layer nodes is calculated through forward propagation, and then the visible layer is reconstructed through backward propagation. The weights and biases are updated according to the reconstruction error. Each restricted Boltzmann machine is trained in turn. The output of the hidden layer of the previous restricted Boltzmann machine is used as the input of the next restricted Boltzmann machine. The trained deep belief network is applied to the historical defect data, and the deep feature representation of the data is obtained through forward propagation. Then the extracted features are input into the random forest for training. The random forest is a powerful ensemble learning algorithm with good generalization ability and anti-overfitting ability. Inputting the features extracted by the deep belief network into the random forest can further utilize the advantages of the random forest to establish an accurate defect trend prediction model. The random forest is composed of multiple decision trees. During the training process, samples are randomly drawn from the training data with replacement, and some features are randomly selected to construct each decision tree. For each decision tree, the information gain criterion is used to select the optimal features and splitting points for node division. For new samples, after extracting features through the deep belief network, they are input into the random forest. Each decision tree makes a prediction, and finally the final prediction result is obtained through voting to effectively predict the future defect trend. In order to evaluate the performance of the model and avoid overfitting, cross-validation and regularization methods need to be used for optimization. Cross-validation can more accurately evaluate the generalization ability of the model on different data sets, and regularization can constrain the complexity of the model to prevent the model from overfitting the training data;

[0058] When the maintenance closed-loop optimization unit 3 adjusts the production process parameters according to the defect trend, the model predictive control algorithm is adopted:

[0059] In order to accurately adjust the production process parameters according to the defect trend, it is necessary to understand the internal dynamic relationship between the two. Establishing a dynamic model can quantify this relationship and provide a basis for subsequent solving of the optimal adjustment strategy. The model predictive control algorithm establishes a dynamic model between the production process parameters and the defect index according to the defect trend predicted by the AI prediction model. Let the production process parameter vector be , where represents the discrete time step, is the number of process parameters, and the defect index vector is , is the number of defect indexes. A linear state space model is used to describe the dynamic relationship between them. The state equation is , and the output equation is , where is the state vector of the system, reflecting the internal state of the production process, is the corresponding coefficient matrix, which is estimated by using the system identification method through historical production data and defect data, and are the process noise and measurement noise respectively. It is assumed that they are Gaussian white noises with zero mean, providing an accurate mathematical description for subsequent parameter optimization, making the adjustment strategy more targeted, effectively reducing the generation of defects, then solving the optimal production process parameter adjustment strategy within the prediction time domain, defining the objective function, comprehensively considering the tracking error of the defect index and the adjustment range of the process parameters, improving the product quality stability, reducing the defective product rate caused by process fluctuations, and adding constraint equations to the algorithm in combination with the constraint conditions of the production process. Adding constraint equations can ensure that the optimal process parameter adjustment strategy obtained is feasible in reality and avoid obtaining unimplementable solutions. Finally, when the repair data of the recycled products is fed back to the production end, a data mining algorithm is used to analyze the repair data, extract the analysis results and conduct correlation analysis between the analysis results and the production process parameters to find the correlation with the production process parameters, providing a basis for further optimizing the process and enhancing the overall efficiency of the enterprise.

[0060] In the present invention, the intelligent monitoring and classification unit 1 uses multi-spectral imaging technology and convolutional neural network image enhancement algorithm, combines fuzzy clustering and support vector machine hybrid classification method to identify and classify defective products and trigger recycling. The traceability and path planning unit 2 generates blockchain IDs through hash algorithms and timestamps, integrates real-time traffic data, uses genetic and ant colony hybrid algorithms to plan the optimal recycling route, and introduces real-time road condition prediction models and multi-agent systems to improve the planning effect. The maintenance closed-loop optimization unit 3 predicts defect trends based on a deep belief network and random forest hybrid model, and uses model predictive control algorithms to adjust production process parameters, improving the recycling management efficiency of industrial defective products.

[0061] The foregoing has shown and described the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments, and the above embodiments and the descriptions in the specification are only preferred examples of the present invention, which are not used to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements fall within the scope of the present invention claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. An industrial defective product recycling management system for industrial Internet data processing, characterized in that, It includes an intelligent monitoring and classification unit (1), a traceability and path planning unit (2), and a maintenance closed-loop optimization unit (3); The intelligent monitoring and classification unit (1) scans the surface and internal defects of the product through a vision sensor, classifies the defects based on edge computing, and triggers the recycling process after classification; The traceability and path planning unit (2) generates a unique blockchain ID for each defective product, records the original information of the product, and then plans the optimal recycling route for each defective product in combination with real-time traffic data and reverse logistics algorithms; The maintenance closed-loop optimization unit (3) is trained using historical defect data based on an AI prediction model, predicts the defect trend, adjusts the production process parameters according to the defect trend, and finally feeds back the repair data of the recycled products to the production end.

2. The industrial defective product recycling management system for industrial Internet data processing according to claim 1, wherein, The vision sensor in the intelligent monitoring and classification unit (1) adopts multi-spectral imaging technology: Collects image information of the product in different spectral bands, obtains the surface and internal defect characteristics of the product, and preprocesses the collected multi-spectral images using an image enhancement algorithm based on a convolutional neural network to learn the mapping relationship between the defect characteristics in the image and the enhanced image.

3. The industrial defective product recycling management system for industrial Internet data processing according to claim 2, characterized in that, When the intelligent monitoring and classification unit (1) classifies the defects based on edge computing, it uses a hybrid classification method combining fuzzy clustering algorithm and support vector machine: Uses the fuzzy clustering algorithm to preliminarily cluster the collected defect feature data, divides different fuzzy categories, and each category has a certain membership degree; Takes the clustering result as input and uses the support vector machine to classify each fuzzy category, determines the specific type of the defect, and adopts an adaptive parameter adjustment mechanism to adjust the parameters of the fuzzy clustering algorithm and the support vector machine according to the historical defect data of the product and the defect characteristics collected in real time.

4. An industrial defective product recycling management system for industrial Internet data processing according to claim 1, characterized in that, When the traceability and path planning unit (2) generates a unique blockchain ID for each defective product, it uses a method combining the hash algorithm and the timestamp: Performs a hash operation on the production batch number, production date, and workstation number on the production line of the product to obtain a hash value, and then combines the hash value with the current timestamp to generate the final blockchain ID.

5. The industrial defective product recycling management system for industrial Internet data processing according to claim 4, characterized in that, When the traceability and path planning unit (2) records the original information of the product, it uses distributed ledger technology to store the information on multiple nodes, specifically as follows: Each node stores a complete copy of the original product information, and maintains the consistency of the information on each node through a consensus mechanism. When recording the information, the information is encrypted, and the symmetric encryption algorithm is used to encrypt the original information, and the encryption key is managed using the asymmetric encryption algorithm.

6. The industrial defective product recycling management system for industrial Internet data processing according to claim 5, characterized in that, When the traceability and path planning unit (2) plans the optimal recycling route for each defective product in combination with real-time traffic data and reverse logistics algorithms, the real-time traffic data is obtained through data interaction with the traffic information platform, including road congestion conditions, traffic accident information, and road construction information. The reverse logistics algorithm uses a hybrid optimization algorithm combining the genetic algorithm and the ant colony algorithm; Use a genetic algorithm for global search to find a set of solution spaces for the recycling routes, and then use an ant colony algorithm for local optimization within this solution space to obtain the optimal recycling route. At the same time, considering the load limit and driving time limit of the recycling vehicles, add constraint conditions to the algorithm.

7. An industrial defective product recycling management system for industrial Internet data processing according to claim 6, characterized in that When planning the recycling route, the traceability and path planning unit (2) introduces a real-time traffic condition prediction model, specifically as follows: The real-time traffic condition prediction model is trained using a long short-term memory network based on historical traffic data and real-time traffic information to predict road congestion. When planning the route, dynamically adjust the route according to the predicted traffic condition information.

8. An industrial defective product recycling management system for industrial Internet data processing according to claim 7, characterized in that, The traceability and path planning unit (2) uses a multi-agent system for route planning, specifically as follows: Regard each defective product as an agent. Through information interaction and cooperation among the agents, jointly complete the planning of the recycling route. At the same time, introduce a reinforcement learning algorithm to optimize the decision-making of the agents to obtain the optimal recycling route.

9. An industrial defective product recycling management system for industrial Internet data processing according to claim 1, characterized in that, When the maintenance closed-loop optimization unit (3) uses historical defect data for training based on the AI prediction model, the AI prediction model adopts a hybrid model combining a deep belief network and a random forest: Use a deep belief network to extract features and reduce the dimension of the historical defect data, learn the deep feature representation of the data, and then input the extracted features into the random forest for training to establish a defect trend prediction model. And use cross-validation and regularization methods to optimize the model.

10. An industrial defective product recycling management system for industrial Internet data processing according to claim 9, characterized in that, When the maintenance closed-loop optimization unit (3) adjusts the production process parameters according to the defect trend, it adopts a model predictive control algorithm: The model predictive control algorithm establishes a dynamic model between the production process parameters and the defect index according to the defect trend predicted by the AI prediction model, then solves the optimal production process parameter adjustment strategy within the prediction time domain, and adds constraint equations to the algorithm in combination with the constraints of the production process. Finally, when feeding back the repair data of the recycled products to the production end, use a data mining algorithm to analyze the repair data, extract the analysis results and conduct correlation analysis between the analysis results and the production process parameters.

Images