Production process optimization method and system based on artificial intelligence

Abstract

The embodiment of the invention provides a production process optimization method and system based on artificial intelligence. According to the method, firstly, multi-source heterogeneous production data generated by each piece of Internet of Things equipment is collected through a data sensing layer, and the defects that traditional data collaboration is insufficient and data islands exist are overcome; then, based on the multi-source heterogeneous production data, modeling is conducted on the topological relation between the Internet of Things devices by means of a graph neural network model, meanwhile, a target scheduling strategy adaptive to the real-time working condition is generated with a preset production target in combination with the deep reinforcement learning technology, and the problem that traditional scheduling cannot respond to the change of the real-time working condition is solved; the target scheduling strategy is imported into the digital twinborn virtual environment rehearsal, the adaptability problem of the target scheduling strategy under different working conditions can be predicted in advance, meanwhile, a final execution scheme is determined through a strategy verification link, and after strategy verification is passed, an execution control layer issues an execution instruction, so that the adaptability of the production process is remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a production process optimization method and system based on artificial intelligence. Background Technology

[0002] With the rapid development of intelligent manufacturing technology, production scheduling, as a core link to ensure the efficient operation of production lines and balance capacity, energy consumption and quality, directly determines the overall production benefits through its adaptability and response efficiency.

[0003] Currently, most traditional production processes use static scheduling rules, which take historical production data as the core input and then use experience to transform the scheduling logic into fixed static scheduling rules.

[0004] However, traditional production processes that rely on static scheduling rules have long-term limitations once the scheduling scheme is generated. If sudden changes occur, such as equipment failure or urgent orders, the process cannot respond to real-time changes in operating conditions in a timely manner, resulting in poor adaptability of the production process. Summary of the Invention

[0005] This invention provides a production process optimization method and system based on artificial intelligence, which can improve the adaptability of the production process.

[0006] In a first aspect, embodiments of the present invention provide a production process optimization method based on artificial intelligence, the method comprising:

[0007] The data sensing layer is used to collect multi-source heterogeneous production data generated by at least one pre-deployed Internet of Things (IoT) device in real time, wherein each IoT device includes: a vibration sensor, a machine vision unit, and a material tracking tag;

[0008] Based on the multi-source heterogeneous production data, the topological relationship of each IoT device is modeled using a graph neural network model, and a target scheduling strategy is generated using deep reinforcement learning technology based on preset production targets.

[0009] The target scheduling strategy is imported into the virtual environment constructed by the digital twin platform for pre-simulation, and it is determined whether the strategy verification can pass.

[0010] When the strategy verification is successful, the target scheduling strategy is executed by issuing a target execution instruction from the control layer.

[0011] Preferably,

[0012] After the real-time acquisition of multi-source heterogeneous production data generated by at least one pre-deployed IoT device using the data sensing layer, and before the step of modeling the topological relationships of each IoT device based on the multi-source heterogeneous production data using a graph neural network model, and generating a target scheduling strategy based on a preset production target using deep reinforcement learning technology, the process further includes:

[0013] The distributed edge computing node clusters pre-deployed in the edge computing layer are used to perform data cleaning operations on the multi-source heterogeneous production data to remove abnormal data from the multi-source heterogeneous production data.

[0014] Mechanical state features of the equipment are extracted from vibration signal data contained in the multi-source heterogeneous production data after cleaning, and defect features are simultaneously extracted from workpiece appearance data contained in the multi-source heterogeneous production data after cleaning.

[0015] The dynamic time warping algorithm is used to perform time series alignment on the multi-source heterogeneous production data after feature extraction, so as to construct a standardized production dataset.

[0016] Preferably,

[0017] The process of modeling the topological relationships of various IoT devices based on the multi-source heterogeneous production data using a graph neural network model, and generating a target scheduling strategy using deep reinforcement learning technology based on preset production targets, includes:

[0018] Based on the standardized production dataset, a dynamic weighted graph structure is constructed with production units as nodes and material flows as edges;

[0019] The core state parameters of each IoT device are extracted from the standardized production dataset to construct a node feature vector, wherein the node feature vector includes: device state, queue length and peak energy consumption;

[0020] Based on the dynamic weighted graph structure and the node feature vectors, the temporal evolution trend of the device state is determined by the graph convolution algorithm of the graph neural network model, and the associated features of adjacent nodes are aggregated through the graph convolution layer to generate a fusion feature matrix.

[0021] Based on the preset production target, a multi-objective optimization function is constructed using the first formula, and a target scheduling strategy is generated using deep reinforcement learning technology based on the fused feature matrix.

[0022] The first formula is: F = ω1P - ω2E - ω3D;

[0023] Where F is the comprehensive optimization objective function value, P is the effective production capacity per unit time, E is the energy consumption per unit product, D is the product defect rate, ω1 is the production capacity weight coefficient, ω2 is the energy consumption weight coefficient, ω3 is the defect rate weight coefficient, and ω1+ω2+ω3=1.

[0024] Preferably,

[0025] The step of importing the target scheduling strategy into the virtual environment constructed by the digital twin platform for pre-performance and determining whether the strategy verification can pass includes:

[0026] The target scheduling strategy is imported into the virtual environment constructed by the digital twin platform for pre-simulation;

[0027] The execution control layer converts the target scheduling strategy into standardized format instructions.

[0028] The standardized format instructions are double-signed using an asymmetric encryption algorithm, and the encrypted standardized format instructions are sent to the security sandbox built into the execution control layer.

[0029] The monitoring module is used to monitor the actual performance indicator values ​​in real time, and the performance indicator values ​​are predicted based on the digital twin platform to determine whether the strategy verification can pass.

[0030] Secondly, embodiments of the present invention provide an artificial intelligence-based production process optimization system, the system comprising:

[0031] Data acquisition module: used to collect multi-source heterogeneous production data generated by at least one pre-deployed IoT device in real time using the data perception layer, wherein each IoT device includes: a vibration sensor, a machine vision unit and a material tracking tag;

[0032] Strategy generation module: Based on the multi-source heterogeneous production data collected by the data acquisition module, it models the topological relationship of each IoT device through a graph neural network model, and generates a target scheduling strategy based on a preset production target using deep reinforcement learning technology.

[0033] Policy verification module: used to import the target scheduling policy generated by the policy generation module into the virtual environment constructed by the digital twin platform for pre-rehearsal, and to determine whether the policy verification can pass;

[0034] Execution module: When the policy verification module determines that the policy verification is successful, it executes the target scheduling policy by issuing a target execution instruction from the control layer.

[0035] Preferably,

[0036] Following the data acquisition module and preceding the strategy generation module, the system further includes: a data processing module;

[0037] The data processing module is used to perform:

[0038] The distributed edge computing node clusters pre-deployed in the edge computing layer are used to perform data cleaning operations on the multi-source heterogeneous production data to remove abnormal data from the multi-source heterogeneous production data.

[0039] Mechanical state features of the equipment are extracted from vibration signal data contained in the multi-source heterogeneous production data after cleaning, and defect features are simultaneously extracted from workpiece appearance data contained in the multi-source heterogeneous production data after cleaning.

[0040] The dynamic time warping algorithm is used to perform time series alignment on the multi-source heterogeneous production data after feature extraction, so as to construct a standardized production dataset.

[0041] Preferably,

[0042] The strategy generation module is also used to execute:

[0043] Based on the standardized production dataset, a dynamic weighted graph structure is constructed with production units as nodes and material flows as edges;

[0044] The core state parameters of each IoT device are extracted from the standardized production dataset to construct a node feature vector, wherein the node feature vector includes: device state, queue length and peak energy consumption;

[0045] Based on the dynamic weighted graph structure and the node feature vectors, the temporal evolution trend of the device state is determined by the graph convolution algorithm of the graph neural network model, and the associated features of adjacent nodes are aggregated through the graph convolution layer to generate a fusion feature matrix.

[0046] Based on the preset production target, a multi-objective optimization function is constructed using the first formula, and a target scheduling strategy is generated using deep reinforcement learning technology based on the fused feature matrix.

[0047] The first formula is: F = ω1P - ω2E - ω3D;

[0048] Where F is the comprehensive optimization objective function value, P is the effective production capacity per unit time, E is the energy consumption per unit product, D is the product defect rate, ω1 is the production capacity weight coefficient, ω2 is the energy consumption weight coefficient, ω3 is the defect rate weight coefficient, and ω1+ω2+ω3=1.

[0049] Preferably,

[0050] The policy verification module is also used to perform:

[0051] The target scheduling strategy is imported into the virtual environment constructed by the digital twin platform for pre-simulation;

[0052] The execution control layer converts the target scheduling strategy into standardized format instructions.

[0053] The standardized format instructions are double-signed using an asymmetric encryption algorithm, and the encrypted standardized format instructions are sent to the security sandbox built into the execution control layer.

[0054] The monitoring module is used to monitor the actual performance indicator values ​​in real time, and the performance indicator values ​​are predicted based on the digital twin platform to determine whether the strategy verification can pass.

[0055] Thirdly, embodiments of the present invention provide an artificial intelligence-based production process optimization system, comprising: at least one memory and at least one processor;

[0056] The at least one memory is used to store a machine-readable program;

[0057] The at least one processor is configured to invoke the machine-readable program to execute any of the methods described in the first aspect.

[0058] Fourthly, embodiments of the present invention provide a computer-readable medium storing computer instructions, which, when executed by a processor, cause the processor to perform any of the methods described in the first aspect.

[0059] This invention provides a production process optimization method and system based on artificial intelligence. The method first collects multi-source heterogeneous production data generated by various IoT devices through a data perception layer, addressing the shortcomings of insufficient data collaboration and data silos in traditional methods. Then, based on this multi-source heterogeneous production data, a graph neural network model is used to model the topological relationships between IoT devices. Simultaneously, deep reinforcement learning technology is combined to generate a target scheduling strategy adapted to real-time operating conditions based on preset production goals, solving the problem that traditional scheduling cannot respond to changes in real-time operating conditions. Next, the target scheduling strategy is imported into a digital twin virtual environment for pre-simulation, allowing for advance prediction of the strategy's adaptability under different operating conditions. The final execution plan is determined through a strategy verification stage, and after successful strategy verification, the execution control layer issues execution instructions, thereby significantly improving the adaptability of the production process. Attached Figure Description

[0060] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0061] Figure 1 This is a flowchart of a production process optimization method based on artificial intelligence provided in an embodiment of the present invention;

[0062] Figure 2 This is a flowchart of another production process optimization method based on artificial intelligence provided in an embodiment of the present invention;

[0063] Figure 3 This is a schematic diagram of a production process optimization system based on artificial intelligence provided in an embodiment of the present invention;

[0064] Figure 4 This is a schematic diagram of another production process optimization system based on artificial intelligence provided in an embodiment of the present invention. Detailed Implementation

[0065] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0066] like Figure 1 As shown, this embodiment of the invention provides a production process optimization method based on artificial intelligence, which may include the following steps:

[0067] Step 101: Real-time acquisition of multi-source heterogeneous production data generated by at least one pre-deployed IoT device using the data sensing layer, wherein each IoT device includes: a vibration sensor, a machine vision unit, and a material tracking tag;

[0068] Step 102: Based on multi-source heterogeneous production data, model the topological relationship of each IoT device using a graph neural network model, and generate a target scheduling strategy based on preset production targets using deep reinforcement learning technology.

[0069] Step 103: Import the target scheduling strategy into the virtual environment built by the digital twin platform for pre-running and determine whether the strategy verification can pass;

[0070] Step 104: When the strategy verification is successful, execute the target scheduling strategy by issuing the target execution instruction from the control layer.

[0071] This invention provides an artificial intelligence-based production process optimization method. First, the method collects multi-source heterogeneous production data generated by various IoT devices through a data perception layer, addressing the shortcomings of insufficient traditional data collaboration and the existence of data silos. Then, based on the multi-source heterogeneous production data, a graph neural network model is used to model the topological relationships between IoT devices. Simultaneously, deep reinforcement learning technology is combined to generate a target scheduling strategy adapted to real-time operating conditions based on preset production goals, solving the problem that traditional scheduling cannot respond to changes in real-time operating conditions. Next, the target scheduling strategy is imported into a digital twin virtual environment for pre-simulation, allowing for advance prediction of the adaptability of the target scheduling strategy under different operating conditions. Simultaneously, the final execution plan is determined through a strategy verification stage. After the strategy verification is passed, the execution control layer issues execution instructions, thereby significantly improving the adaptability of the production process.

[0072] To improve the reliability of production data, in one embodiment of the present invention, the above embodiment further includes the following steps after step 101 and before step 102:

[0073] The distributed edge computing node clusters pre-deployed in the edge computing layer are used to perform data cleaning operations on the multi-source heterogeneous production data to remove abnormal data from the multi-source heterogeneous production data.

[0074] Mechanical state features of the equipment are extracted from vibration signal data contained in the multi-source heterogeneous production data after cleaning, and defect features are simultaneously extracted from workpiece appearance data contained in the multi-source heterogeneous production data after cleaning.

[0075] The dynamic time warping algorithm is used to perform time series alignment on the multi-source heterogeneous production data after feature extraction, so as to construct a standardized production dataset.

[0076] In this embodiment of the invention, to ensure the reliability of production data, after collecting multi-source heterogeneous production data from various IoT devices, a pre-deployed distributed edge computing node cluster can be used to perform data cleaning operations on the production data (for example, outliers can be removed using the 3σ criterion), thereby effectively filtering outliers and other invalid data. Secondly, feature extraction is performed on the cleaned data. Mechanical state features of the equipment are extracted from vibration signal data to accurately capture mechanical anomaly-related features during equipment operation (such as bearing wear, spindle imbalance, etc.). Defect features are extracted from workpiece appearance data (for example, using a deployed TensorRT-optimized YOLOv5s model (INT8 quantization)) to efficiently obtain workpiece quality-related features. Finally, a time series alignment operation is performed using a dynamic time warping algorithm to solve the problem of asynchronous time dimensions of multi-source heterogeneous production data, ultimately constructing a standardized production dataset with spatiotemporal consistency and feature integrity.

[0077] In order to generate a target scheduling strategy, in one embodiment of the present invention, step 102 in the above embodiment may specifically include the following steps:

[0078] Based on the standardized production dataset, a dynamic weighted graph structure is constructed with production units as nodes and material flows as edges;

[0079] The core state parameters of each IoT device are extracted from the standardized production dataset to construct a node feature vector, wherein the node feature vector includes: device state, queue length and peak energy consumption;

[0080] Based on the dynamic weighted graph structure and the node feature vectors, the temporal evolution trend of the device state is determined by the graph convolution algorithm of the graph neural network model, and the associated features of adjacent nodes are aggregated through the graph convolution layer to generate a fusion feature matrix.

[0081] Based on the preset production target, a multi-objective optimization function is constructed using the first formula, and a target scheduling strategy is generated using deep reinforcement learning technology based on the fused feature matrix.

[0082] The first formula is: F = ω1P - ω2E - ω3D;

[0083] Where F is the comprehensive optimization objective function value, P is the effective production capacity per unit time, E is the energy consumption per unit product, D is the product defect rate, ω1 is the production capacity weight coefficient, ω2 is the energy consumption weight coefficient, ω3 is the defect rate weight coefficient, and ω1+ω2+ω3=1.

[0084] In this embodiment of the invention, to generate a target scheduling strategy, firstly, a dynamic weighted graph structure with production units as nodes and material flows as edges is constructed based on the aforementioned standardized production dataset to ensure that the graph structure accurately reflects the real-time relationships between production units. Secondly, core state parameters of IoT devices are extracted from the standardized dataset to construct node feature vectors (e.g., device status, queue length, peak energy consumption), which can provide accurate node state input for the graph neural network model. Next, the graph convolution algorithm of the graph neural network model is used to determine the temporal evolution trend of device status and aggregate the association features of adjacent nodes to generate a fusion feature matrix. By aggregating the features of adjacent nodes, the collaborative influence between devices is captured, breaking through the limitation of traditional models ignoring topological relationships. Finally, a multi-objective optimization function is constructed based on the preset production target, and a target scheduling strategy is generated by combining the fusion feature matrix with deep reinforcement learning technology. This can solve the problem of a single scheduling target in traditional models while achieving coordinated optimization of production capacity, energy consumption, and quality.

[0085] To ensure that the target scheduling strategy can adapt to changes in real-time operating conditions, in one embodiment of the present invention, step 103 in the above embodiment may specifically include the following steps:

[0086] The target scheduling strategy is imported into the virtual environment constructed by the digital twin platform for pre-simulation;

[0087] The execution control layer converts the target scheduling strategy into standardized format instructions.

[0088] The standardized format instructions are double-signed using an asymmetric encryption algorithm, and the encrypted standardized format instructions are sent to the security sandbox built into the execution control layer.

[0089] The monitoring module is used to monitor the actual performance indicator values ​​in real time, and the performance indicator values ​​are predicted based on the digital twin platform to determine whether the strategy verification can pass.

[0090] In this embodiment of the invention, to improve the adaptability of the target scheduling strategy, the target scheduling strategy can first be imported into a virtual environment constructed by a digital twin platform for pre-simulation. By synchronizing the physical production line status and pre-simulating the strategy in the virtual replica, potential constraint violations (such as insufficient production capacity or excessive energy consumption) can be identified in advance, while also avoiding production risks caused by direct implementation of the strategy. Secondly, the target scheduling strategy is converted into standardized format instructions by the execution control layer, ensuring that the instructions can be recognized and compatible by controllers from different manufacturers, further solving the execution obstacles caused by the inconsistency of traditional instruction formats. Then, asymmetric addition is utilized... The encryption algorithm double-signs the standardized format instructions and sends them to the security sandbox built into the execution control layer. Through the isolation of the encryption signature and the security sandbox, the instructions can be prevented from being tampered with or leaked during transmission and execution, ensuring the security and integrity of the scheduling instructions. Finally, the monitoring module monitors the actual performance indicator values ​​in real time and combines them with the predicted performance indicator values ​​from the digital twin platform to determine whether the strategy verification has passed. When the deviation between the actual performance indicator value and the predicted performance indicator value exceeds the threshold (for example, the capacity deviation > 5%, where 5% is the threshold), the strategy verification can be determined in a timely manner and optimization adjustments can be initiated to ensure that the target scheduling strategy can adapt to real-time changes in operating conditions.

[0091] like Figure 2 As shown, in order to more clearly illustrate the technical solution and advantages of the present invention, the following provides a detailed description of the production process optimization method based on artificial intelligence, which may include the following steps:

[0092] Step 201: Real-time acquisition of multi-source heterogeneous production data generated by at least one pre-deployed IoT device using the data sensing layer, wherein each IoT device includes: a vibration sensor, a machine vision unit, and a material tracking tag;

[0093] Step 202: Use the pre-deployed distributed edge computing node cluster in the edge computing layer to perform data cleaning operations on the multi-source heterogeneous production data and remove abnormal data from the multi-source heterogeneous production data;

[0094] Step 203: Extract the mechanical state features of the equipment from the vibration signal data contained in the multi-source heterogeneous production data after cleaning, and simultaneously extract the defect features from the workpiece appearance data contained in the multi-source heterogeneous production data after cleaning.

[0095] Step 204: Use the dynamic time warping algorithm to perform time series alignment on the multi-source heterogeneous production data after feature extraction to construct a standardized production dataset;

[0096] Step 205: Based on the standardized production dataset, construct a dynamic weighted graph structure with production units as nodes and material flows as edges;

[0097] Step 206: Extract the core state parameters of each IoT device from the standardized production dataset and construct a node feature vector, where the node feature vector includes: device state, queue length and peak energy consumption;

[0098] Step 207: Based on the dynamic weighted graph structure and node feature vectors, the graph convolution algorithm of the graph neural network model is used to determine the temporal evolution trend of the device state, and the associated features of adjacent nodes are aggregated through the graph convolution layer to generate a fused feature matrix;

[0099] Step 208: Based on the preset production target, construct a multi-objective optimization function using the first formula, and generate a target scheduling strategy based on the fused feature matrix using deep reinforcement learning technology;

[0100] Specifically, the first formula is: F = ω1P - ω2E - ω3D;

[0101] Where F is the comprehensive optimization objective function value, P is the effective production capacity per unit time, E is the energy consumption per unit product, D is the product defect rate, ω1 is the production capacity weight coefficient, ω2 is the energy consumption weight coefficient, ω3 is the defect rate weight coefficient, and ω1+ω2+ω3=1.

[0102] Step 209: Import the target scheduling strategy into the virtual environment built by the digital twin platform for pre-simulation;

[0103] Step 210: Convert the target scheduling policy into standardized format instructions by executing the control layer;

[0104] Step 211: Double-sign the standardized format instruction using an asymmetric encryption algorithm, and then send the encrypted standardized format instruction to the security sandbox built into the execution control layer;

[0105] Step 212: Use the monitoring module to monitor the actual performance indicator values ​​in real time, and use the digital twin platform to predict the performance indicator values ​​to determine whether the strategy verification can pass;

[0106] Step 213: When the strategy verification is successful, execute the target scheduling strategy by issuing the target execution instruction from the control layer.

[0107] like Figure 3 As shown, this embodiment of the invention provides an artificial intelligence-based production process optimization system, which includes:

[0108] Data acquisition module 301: used to acquire multi-source heterogeneous production data generated by at least one pre-deployed Internet of Things (IoT) device in real time using the data perception layer, wherein each IoT device includes: a vibration sensor, a machine vision unit, and a material tracking tag;

[0109] Strategy generation module 302: Based on the multi-source heterogeneous production data collected by the data acquisition module 301, it models the topological relationship of each IoT device through a graph neural network model, and generates a target scheduling strategy based on a preset production target using deep reinforcement learning technology.

[0110] Strategy verification module 303: used to import the target scheduling strategy generated by the strategy generation module 302 into the virtual environment constructed by the digital twin platform for pre-rehearsal, and to determine whether the strategy verification can pass;

[0111] Execution module 304: When the policy verification module 303 determines that the policy verification is successful, it executes the target scheduling policy by issuing a target execution instruction from the control layer.

[0112] based on Figure 3 The AI-based production process optimization system shown is as follows: Figure 4 As shown, after the data acquisition module 301 and before the strategy generation module 302, it further includes: a data processing module 305;

[0113] The data processing module 305 is used to perform:

[0114] The distributed edge computing node clusters pre-deployed in the edge computing layer are used to perform data cleaning operations on the multi-source heterogeneous production data to remove abnormal data from the multi-source heterogeneous production data.

[0115] Mechanical state features of the equipment are extracted from vibration signal data contained in the multi-source heterogeneous production data after cleaning, and defect features are simultaneously extracted from workpiece appearance data contained in the multi-source heterogeneous production data after cleaning.

[0116] The dynamic time warping algorithm is used to perform time series alignment on the multi-source heterogeneous production data after feature extraction, so as to construct a standardized production dataset.

[0117] like Figure 4 As shown, the strategy generation module 302 is further configured to perform:

[0118] Based on the standardized production dataset, a dynamic weighted graph structure is constructed with production units as nodes and material flows as edges;

[0119] The core state parameters of each IoT device are extracted from the standardized production dataset to construct a node feature vector, wherein the node feature vector includes: device state, queue length and peak energy consumption;

[0120] Based on the dynamic weighted graph structure and the node feature vectors, the temporal evolution trend of the device state is determined by the graph convolution algorithm of the graph neural network model, and the associated features of adjacent nodes are aggregated through the graph convolution layer to generate a fusion feature matrix.

[0121] Based on the preset production target, a multi-objective optimization function is constructed using the first formula, and a target scheduling strategy is generated using deep reinforcement learning technology based on the fused feature matrix.

[0122] The first formula is: F = ω1P - ω2E - ω3D;

[0123] Where F is the comprehensive optimization objective function value, P is the effective production capacity per unit time, E is the energy consumption per unit product, D is the product defect rate, ω1 is the production capacity weight coefficient, ω2 is the energy consumption weight coefficient, ω3 is the defect rate weight coefficient, and ω1+ω2+ω3=1.

[0124] like Figure 4 As shown, the policy verification module 303 is also used to perform:

[0125] The target scheduling strategy is imported into the virtual environment constructed by the digital twin platform for pre-simulation;

[0126] The execution control layer converts the target scheduling strategy into standardized format instructions.

[0127] The standardized format instructions are double-signed using an asymmetric encryption algorithm, and the encrypted standardized format instructions are sent to the security sandbox built into the execution control layer.

[0128] The monitoring module is used to monitor the actual performance indicator values ​​in real time, and the performance indicator values ​​are predicted based on the digital twin platform to determine whether the strategy verification can pass.

[0129] It is understood that the structures illustrated in the embodiments of the present invention do not constitute a specific limitation on the AI-based production process optimization system. In other embodiments of the present invention, the AI-based production process optimization system may include more or fewer components than illustrated, or combine some components, or split some components, or have different component arrangements. The illustrated components may be implemented in hardware, software, or a combination of software and hardware.

[0130] The information interaction and execution process between the various units in the above-mentioned device are based on the same concept as the method embodiment of the present invention, and the specific details can be found in the description of the method embodiment of the present invention, and will not be repeated here.

[0131] This invention also provides an artificial intelligence-based production process optimization system, comprising: at least one memory and at least one processor;

[0132] At least one memory for storing machine-readable programs;

[0133] At least one processor is configured to invoke a machine-readable program to execute the AI-based production process optimization method in any embodiment of the present invention.

[0134] This invention also provides a computer-readable medium storing computer instructions, which, when executed by a processor, cause the processor to perform the artificial intelligence-based production process optimization method according to any embodiment of this invention.

[0135] Specifically, a system or apparatus equipped with a storage medium may be provided, on which software program code implementing the functions of any of the embodiments described above is stored, and the computer (or CPU or MPU) of the system or apparatus may read and execute the program code stored in the storage medium.

[0136] In this case, the program code read from the storage medium can itself implement the function of any of the above embodiments, and therefore the program code and the storage medium storing the program code constitute part of the present invention.

[0137] Examples of storage media used to provide program code include floppy disks, hard disks, magneto-optical disks, optical disks (such as CD-ROM, CD-R, CD-RW, DVD-ROM, DVD-RAM, DVD-RW, DVD+RW), magnetic tapes, non-volatile memory cards, and ROMs. Alternatively, program code can be downloaded from a server computer via a communication network.

[0138] Furthermore, it should be clear that not only can the program code read by the computer be executed, but also the operating system or other components operating on the computer can be instructed based on the program code to perform some or all of the actual operations, thereby realizing the function of any of the embodiments described above.

[0139] Furthermore, it is understood that the program code read from the storage medium is written to the memory set in the expansion board inserted into the computer or to the memory set in the expansion unit connected to the computer. Then, based on the instructions of the program code, the CPU or other components installed on the expansion board or expansion unit execute some and all of the actual operations, thereby realizing the function of any of the embodiments described above.

[0140] The various embodiments of the present invention have at least the following beneficial effects:

[0141] 1. In this embodiment of the invention, an artificial intelligence-based production process optimization method is provided. This method first collects multi-source heterogeneous production data generated by various IoT devices through a data perception layer, solving the shortcomings of insufficient traditional data collaboration and the existence of data silos. Then, based on the multi-source heterogeneous production data, a graph neural network model is used to model the topological relationships between IoT devices. Simultaneously, deep reinforcement learning technology is combined to generate a target scheduling strategy adapted to real-time operating conditions based on preset production goals, solving the problem that traditional scheduling cannot respond to real-time changes in operating conditions. The target scheduling strategy is then imported into a digital twin virtual environment for pre-simulation, which can predict the adaptability of the target scheduling strategy under different operating conditions. Simultaneously, the final execution plan is determined through a strategy verification stage. After the strategy verification is passed, the execution control layer issues execution instructions, thereby significantly improving the adaptability of the production process.

[0142] 2. In this embodiment of the invention, to ensure the reliability of production data, after collecting multi-source heterogeneous production data from various IoT devices, a pre-deployed distributed edge computing node cluster can be used to perform data cleaning operations on the production data, thereby effectively filtering out invalid data such as outliers. Secondly, feature extraction is performed on the cleaned data. Mechanical state features of the equipment are extracted from vibration signal data to accurately capture mechanical anomaly-related features during equipment operation. Defect features are extracted from workpiece appearance data to efficiently obtain workpiece quality-related features. Finally, a dynamic time warping algorithm is used to perform time series alignment operations to solve the problem of asynchronous time dimensions of multi-source heterogeneous production data, ultimately constructing a standardized production dataset with spatiotemporal consistency and feature integrity.

[0143] 3. In this embodiment of the invention, in order to generate a target scheduling strategy, firstly, a dynamic weighted graph structure with production units as nodes and material flows as edges needs to be constructed based on the aforementioned standardized production dataset to ensure that the graph structure accurately reflects the real-time association between production units; secondly, the core state parameters of IoT devices are extracted from the standardized dataset to construct node feature vectors, which can provide accurate node state input for the graph neural network model; then, the graph convolution algorithm of the graph neural network model is used to determine the temporal evolution trend of device states and aggregate the association features of adjacent nodes to generate a fusion feature matrix. By aggregating the features of adjacent nodes, the collaborative influence between devices is captured, breaking through the limitation of traditional models ignoring topological associations; finally, a multi-objective optimization function is constructed based on the preset production target, and a target scheduling strategy is generated by combining the fusion feature matrix with deep reinforcement learning technology. This can solve the problem of single scheduling target in traditional models while achieving coordinated optimization of production capacity, energy consumption, and quality.

[0144] It should be noted that not all steps and modules in the above processes and system structure diagrams are mandatory; some steps or modules can be omitted as needed. The execution order of each step is not fixed and can be adjusted as required. The system structure described in the above embodiments can be a physical structure or a logical structure. That is, some modules may be implemented by the same physical entity, or some modules may be implemented by multiple physical entities, or they may be jointly implemented by certain components in multiple independent devices.

[0145] In the above embodiments, the hardware units can be implemented mechanically or electrically. For example, a hardware unit may include permanent dedicated circuitry or logic (such as a dedicated processor, FPGA, or ASIC) to perform the corresponding operation. The hardware unit may also include programmable logic or circuitry (such as a general-purpose processor or other programmable processor), which can be temporarily configured by software to perform the corresponding operation. The specific implementation method (mechanical, dedicated permanent circuitry, or temporarily configured circuitry) can be determined based on cost and time considerations.

[0146] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A production process optimization method based on artificial intelligence, characterized in that, The method includes: using a data sensing layer to collect multi-source heterogeneous production data generated by at least one pre-deployed Internet of Things (IoT) device in real time, wherein each IoT device includes: a vibration sensor, a machine vision unit, and a material tracking tag; Based on the multi-source heterogeneous production data, the topological relationship of each IoT device is modeled using a graph neural network model, and a target scheduling strategy is generated using deep reinforcement learning technology based on preset production targets. The target scheduling strategy is imported into the virtual environment constructed by the digital twin platform for pre-simulation, and it is determined whether the strategy verification can pass. When the strategy verification is successful, the target scheduling strategy is executed by issuing a target execution instruction from the control layer.

2. The method according to claim 1, characterized in that, After the real-time acquisition of multi-source heterogeneous production data generated by at least one pre-deployed IoT device using the data sensing layer, and before the step of modeling the topological relationships of each IoT device based on the multi-source heterogeneous production data using a graph neural network model, and generating a target scheduling strategy based on a preset production target using deep reinforcement learning technology, the process further includes: The distributed edge computing node clusters pre-deployed in the edge computing layer are used to perform data cleaning operations on the multi-source heterogeneous production data to remove abnormal data from the multi-source heterogeneous production data. Mechanical state features of the equipment are extracted from vibration signal data contained in the multi-source heterogeneous production data after cleaning, and defect features are simultaneously extracted from workpiece appearance data contained in the multi-source heterogeneous production data after cleaning. The dynamic time warping algorithm is used to perform time series alignment on the multi-source heterogeneous production data after feature extraction, so as to construct a standardized production dataset.

3. The method according to claim 2, characterized in that, The process of modeling the topological relationships of various IoT devices based on the multi-source heterogeneous production data using a graph neural network model, and generating a target scheduling strategy using deep reinforcement learning technology based on preset production targets, includes: Based on the standardized production dataset, a dynamic weighted graph structure is constructed with production units as nodes and material flows as edges; The core state parameters of each IoT device are extracted from the standardized production dataset to construct a node feature vector, wherein the node feature vector includes: device state, queue length and peak energy consumption; Based on the dynamic weighted graph structure and the node feature vectors, the temporal evolution trend of the device state is determined by the graph convolution algorithm of the graph neural network model, and the associated features of adjacent nodes are aggregated through the graph convolution layer to generate a fusion feature matrix. Based on the preset production target, a multi-objective optimization function is constructed using the first formula, and a target scheduling strategy is generated using deep reinforcement learning technology based on the fused feature matrix. The first formula is: F = ω1P - ω2E - ω3D; Where F is the comprehensive optimization objective function value, P is the effective production capacity per unit time, E is the energy consumption per unit product, D is the product defect rate, ω1 is the production capacity weight coefficient, ω2 is the energy consumption weight coefficient, ω3 is the defect rate weight coefficient, and ω1+ω2+ω3=1.

4. The method according to claim 1, characterized in that, The step of importing the target scheduling strategy into the virtual environment constructed by the digital twin platform for pre-performance and determining whether the strategy verification can pass includes: The target scheduling strategy is imported into the virtual environment constructed by the digital twin platform for pre-simulation; The execution control layer converts the target scheduling strategy into standardized format instructions. The standardized format instructions are double-signed using an asymmetric encryption algorithm, and the encrypted standardized format instructions are sent to the security sandbox built into the execution control layer. The monitoring module is used to monitor actual performance indicator values ​​in real time, and the performance indicator values ​​are predicted based on the digital twin platform to determine whether the strategy verification can pass.

5. A production process optimization system based on artificial intelligence, characterized in that: The system includes: Data acquisition module: used to collect multi-source heterogeneous production data generated by at least one pre-deployed IoT device in real time using the data perception layer, wherein each IoT device includes: a vibration sensor, a machine vision unit and a material tracking tag; Strategy generation module: Based on the multi-source heterogeneous production data collected by the data acquisition module, it models the topological relationship of each IoT device through a graph neural network model, and generates a target scheduling strategy based on a preset production target using deep reinforcement learning technology. Policy verification module: used to import the target scheduling policy generated by the policy generation module into the virtual environment constructed by the digital twin platform for pre-rehearsal, and to determine whether the policy verification can pass; Execution module: When the policy verification module determines that the policy verification is successful, it executes the target scheduling policy by issuing a target execution instruction from the control layer.

6. The system according to claim 5, characterized in that, Following the data acquisition module and preceding the strategy generation module, the system further includes: a data processing module; The data processing module is used to perform: The distributed edge computing node clusters pre-deployed in the edge computing layer are used to perform data cleaning operations on the multi-source heterogeneous production data to remove abnormal data from the multi-source heterogeneous production data. Mechanical state features of the equipment are extracted from vibration signal data contained in the multi-source heterogeneous production data after cleaning, and defect features are simultaneously extracted from workpiece appearance data contained in the multi-source heterogeneous production data after cleaning. The dynamic time warping algorithm is used to perform time series alignment on the multi-source heterogeneous production data after feature extraction, so as to construct a standardized production dataset.

7. The system according to claim 6, characterized in that, The strategy generation module is also used to execute: Based on the standardized production dataset, a dynamic weighted graph structure is constructed with production units as nodes and material flows as edges; The core state parameters of each IoT device are extracted from the standardized production dataset to construct a node feature vector, wherein the node feature vector includes: device state, queue length and peak energy consumption; Based on the dynamic weighted graph structure and the node feature vectors, the temporal evolution trend of the device state is determined by the graph convolution algorithm of the graph neural network model, and the associated features of adjacent nodes are aggregated through the graph convolution layer to generate a fusion feature matrix. Based on the preset production target, a multi-objective optimization function is constructed using the first formula, and a target scheduling strategy is generated using deep reinforcement learning technology based on the fused feature matrix. The first formula is: F = ω1P - ω2E - ω3D; Where F is the comprehensive optimization objective function value, P is the effective production capacity per unit time, E is the energy consumption per unit product, D is the product defect rate, ω1 is the production capacity weight coefficient, ω2 is the energy consumption weight coefficient, ω3 is the defect rate weight coefficient, and ω1+ω2+ω3=1.

8. The system according to claim 5, characterized in that, The policy verification module is also used to perform: The target scheduling strategy is imported into the virtual environment constructed by the digital twin platform for pre-simulation; The execution control layer converts the target scheduling strategy into standardized format instructions. The standardized format instructions are double-signed using an asymmetric encryption algorithm, and the encrypted standardized format instructions are sent to the security sandbox built into the execution control layer. The monitoring module is used to monitor actual performance indicator values ​​in real time, and the performance indicator values ​​are predicted based on the digital twin platform to determine whether the strategy verification can pass.

9. A production process optimization system based on artificial intelligence, characterized in that, include: At least one memory and at least one processor; The at least one memory is used to store a machine-readable program; The at least one processor is configured to invoke the machine-readable program to perform the method according to any one of claims 1 to 4.

10. A computer-readable medium, characterized in that, The computer-readable medium stores computer instructions that, when executed by a processor, cause the processor to perform the method described in any one of claims 1 to 4.

Images