Architecture system for digital factory intelligent production and implementation method thereof

Abstract

The application relates to an architecture system for digital factory intelligent production and an implementation method thereof. The architecture system comprises a data sensing layer, an edge computing layer, a cloud computing layer and an application layer. The data sensing layer sends real-time collected equipment state information of corresponding production equipment to the edge computing layer for abnormal signal analysis processing, workshop production plan scheduling optimization and workshop environment analysis. The cloud computing layer performs multi-modal data fusion on data provided by the edge computing layer and different downstream tasks, trains inherent characteristics extracted from each modal data after multi-modal data fusion, and performs incremental update processing on different downstream task models according to different extracted inherent characteristics. The application layer calls a service interface provided by the cloud computing layer, realizes application of the architecture to different application scenarios, and meets the requirement of making a quick response to changes in factory production and improving the overall production efficiency of the digital factory.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of digital factory intelligent production, and in particular to an architecture system for digital factory intelligent production and an implementation method thereof. BACKGROUND

[0002] With the rapid development of Industry 4.0 and intelligent manufacturing, digital factory has become the core of modern industrial production. In order to improve production efficiency, reduce cost and ensure product quality, a high-efficiency and reliable production management platform is needed for digital factory to schedule and process data in each department of the factory. However, the existing production management platform often has problems such as information island, system integration difficulty and low data processing efficiency, which cannot meet the needs of rapid response and real-time monitoring of modern industrial production.

[0003] Chinese patent CN116880412B discloses a cloud-based visual production management platform, which uses an abnormality detection method based on mean variance method for abnormality monitoring and early warning, sets reasonable abnormality threshold according to historical data and actual production situation, makes abnormality detection more targeted, avoids false positives and false negatives, ensures that real abnormalities can be discovered and solved in time, ensures that the abnormality monitoring and early warning module can monitor data in real time, and immediately triggers an early warning notification to relevant staff when an abnormality occurs or exceeds the threshold, implements an abnormality processing mechanism in the data acquisition module, and sends an alarm to the staff in time when an abnormality occurs, prompts the abnormal situation of data acquisition, ensures stable data acquisition and transmission, the platform can discover abnormal situations in production more timely, provides more accurate data analysis and decision support, and makes production management more efficient and intelligent.

[0004] However, the cloud-based visual production management platform disclosed in the above-mentioned patent CN116880412B has the following deficiencies: because it sets an abnormality detection model according to historical data and uses a fixed production scheduling algorithm, it lacks flexibility and scalability, and cannot respond quickly to changes in factory production, making it difficult to improve the overall production efficiency of the digital factory. SUMMARY

[0005] The first technical problem to be solved by the present application is to provide an architecture system for digital factory intelligent production that can monitor and analyze real-time data of each plant equipment to optimize production scheduling efficiency.

[0006] The second technical problem to be solved by the present application is to provide an implementation method for the above-mentioned architecture system.

[0007] The technical solution adopted by the present application to solve the first technical problem is: an architecture system for digital factory intelligent production, characterized by comprising:

[0008] The data perception layer is formed by a plurality of industrial data acquisition modules deployed in a plurality of production devices in the digital factory, and the industrial data acquisition modules are configured to collect device state information of the corresponding production devices in real time, and to adjust the display content of the control devices of the digital factory in real time according to the device state information; wherein the device state information of the production devices includes device operation mode, device running state, spindle speed, spindle load and spindle temperature, the device running state includes running state, standby state and abnormal state, and each production device corresponds to at least one industrial data acquisition module deployed;

[0009] The edge computing layer is formed by a plurality of edge computing devices located at the network edge, and is configured to perform abnormal signal analysis processing on the device state information sent by the data perception layer, to perform scheduling optimization on the production plan in the workshop, and to analyze the workshop environment; wherein the edge computing devices are located in the vicinity of the industrial data source, the edge computing layer includes an abnormal detection analysis module and a workshop scheduling analysis module, the abnormal detection analysis module is configured to receive the device state information sent by the data perception layer, and to perform abnormal signal analysis processing on the device state information; the workshop scheduling analysis module is configured to perform real-time analysis and processing on the production information in the workshop to complete the scheduling optimization of the production plan; the production information in the workshop includes the available production device resources in the current workshop and the production tasks in the production queue;

[0010] The cloud computing layer includes a cloud computing center and a plurality of model nodes, and is configured to perform multi-modal data fusion according to the device state information provided by the edge computing layer and different downstream tasks, extract the internal features of each modal data after multi-modal data fusion, and then train the extracted internal features, and after the training is completed, incrementally update the different downstream task models according to the different extracted internal features;

[0011] The application layer realizes the application of the architecture to different application scenarios by calling the service interface provided by the cloud computing layer.

[0012] In the improved digital factory intelligent production-oriented architecture system, the process of the industrial data acquisition module in the data perception layer collecting device state information of the corresponding production device in real time includes the following steps:

[0013] The industrial data acquisition module performs networking under the condition that itself and the production devices in the workshop are normally connected, so as to obtain the device state information of the corresponding production devices;

[0014] According to the obtained device state information, the device running state is determined, and the data transmission frequency is adjusted based on the device running state, and the device state information is sent to the edge server for analysis and processing according to the adjusted data transmission frequency corresponding to the running state; wherein each device running state has a corresponding data transmission frequency.

[0015] Further, in the architecture system for digital factory intelligent production, the abnormal signal analysis processing of the abnormal detection analysis module in the edge computing layer is performed in the following manner:

[0016] Step a1, the abnormal detection analysis module makes abnormal analysis and judgment on the device state information sent by the data perception layer according to the preset abnormal detection model:

[0017] When the abnormal detection analysis module judges that an abnormal signal is obtained, the production device corresponding to the abnormal signal enters an abnormal state and stops production work, and step a2 is entered; otherwise, the production device continues production work;

[0018] Step a2, the abnormal detection analysis module analyzes the abnormal reason and type of the abnormal signal to obtain an abnormal analysis result; wherein the abnormal analysis result includes the abnormal reason and type of the abnormal signal;

[0019] Step a3, the abnormal detection analysis module takes corresponding abnormal processing measures according to the abnormal analysis result, and performs multi-index testing on the production device processed by the abnormal processing measures;

[0020] Step a4, the abnormal detection analysis module makes judgment and processing according to the multi-index testing result:

[0021] When the production device successfully passes the multi-index testing, the production device returns to the normal state and the running state of the production device is updated; otherwise, the multi-index testing is performed again on the production device until the production device successfully passes the multi-index testing.

[0022] Further, in the architecture system for digital factory intelligent production, the workshop scheduling analysis module in the edge computing layer analyzes and processes the production information in the workshop in real time to complete the scheduling optimization process of the production plan, including the following steps:

[0023] When the workshop scheduling analysis module judges that a new production task appears or a production device is in an abnormal state, the current workshop environment is analyzed and the available production device resources in the current workshop and the production tasks in the production queue are counted;

[0024] The workshop scheduling analysis module outputs a scheduling scheme suitable for the current workshop condition according to the counted available production device resources in the current workshop and the production tasks in the production queue.

[0025] The workshop scheduling analysis module verifies the feasibility of the scheduling scheme according to the task priority of each production task in the scheduling scheme and the production constraint condition corresponding to each production task:

[0026] When the scheduling scheme is feasible, the current workshop production equipment executes the scheduling scheme; otherwise, the scheduling scheme is adjusted to obtain a new scheduling scheme with feasibility.

[0027] Further, in the architecture system for digital factory intelligent production, the process of analyzing the workshop environment by the edge computing layer comprises the following steps:

[0028] Step b1, the running states of all production equipment in the workshop are traversed, and the distribution number of production equipment in each running state is calculated respectively; wherein the number of running states of all production equipment in the workshop is marked as N, the i-th running state in N running states is marked as S i , and the distribution number of production equipment in running state S i is marked as

[0029] 1≤i≤N;

[0030]

[0031] Wherein J represents the total number of production equipment in the workshop, s j represents the running state of the j-th production equipment in the workshop, and δ(s j =S i ) is an indicator function about running state s j and running state S i .

[0032] Step b2, according to the running state of each production equipment obtained by traversal, the total number of production equipment in fault state is judged, and the equipment failure rate in the current workshop is calculated; wherein the equipment failure rate in the workshop is calculated as follows:

[0033]

[0034] Wherein λ represents the equipment failure rate in the current workshop, S b represents that the production equipment is in fault state.

[0035] Step b3, the running time of each production equipment in different running states is obtained, and the running state time length proportion of each running state corresponding to each production equipment is calculated; wherein the running state time length proportion is calculated as follows:

[0036]

[0037] wherein f ji represents the time length ratio of the jth production equipment in the running state S i ; j,i represents the time length of the jth production equipment in the running state S i ;

[0038] Step b4, the product quality of each type of workpiece produced in the factory is counted, the number of excellent products of each type of workpiece is calculated, and the excellent product rate of each type of workpiece is calculated; wherein the calculation method of the excellent product rate of each type of workpiece is as follows:

[0039]

[0040] wherein e u represents the excellent product rate of the u-th type of workpiece, P u represents the total number of the u-th type of workpiece produced in the workshop, p uv represents the quality of the v-th workpiece of the u-th type of workpiece produced in the workshop;

[0041] Step b5, the product quality of each type of workpiece required by each production order is counted, the number of excellent products of each type of workpiece is obtained, and the production progress of each production task corresponding to the production order is calculated; wherein the total number of production tasks received in the workshop is marked as K, and the production progress of the k-th production task in the K production tasks is marked as r k :

[0042]

[0043] wherein Q k represents the target production quantity of the k-th production task, P u represents the total number of the u-th type of workpiece produced in the workshop, p uv represents the quality of the v-th workpiece of the u-th type of workpiece produced in the workshop;

[0044] Step b6, the number of workpieces of all types of workpieces produced in the workshop in each day is counted, and the workshop capacity of the workshop is calculated; wherein the workshop capacity of the workshop in the d-th day is marked as c d :

[0045] d≥1;

[0046] wherein c d,u represents the total number of the u-th type of workpiece produced in the d-th day of the workshop, and U represents the total number of types of workpieces produced in the workshop.

[0047] Improvedly, in the architecture system facing the digital factory intelligent production, the cloud computing layer respectively makes multi-modal data fusion according to the data provided by the edge computing layer and different downstream tasks, and the method comprises the following steps c1-c5:

[0048] Step c1, according to the modal data type after the fusion of each modal data, a modal feature extractor corresponding to the modal data type is established by a pre-training method;

[0049] Step c2, the established modal feature extractor is retrained according to the production task, so that the trained modal feature extractor meets the demand of the specific downstream task in the application layer;

[0050] Step c3, according to the demand of the downstream task in the application layer, one or more model data are selected and feature extraction processing is performed by using the trained modal feature extractor corresponding to the model data, to obtain extracted modal features, and the output dimension of each trained modal feature extractor is trained using a full-automatic encoder to obtain reduced modal features;

[0051] Step c4, each reduced modal feature is put into a contrast learning network to learn the corresponding relationship between different modal features;

[0052] Step c5, according to the difference of the downstream task, the corresponding modal is selected through a multi-modal selection gate to select the multi-modal feature, and the multi-modal feature is transmitted to the downstream model in the application layer which needs to update the weight for training processing;

[0053] Step c6, after the training of the downstream model in the application layer is completed, the updated parameters and weights are returned to the modal fusion model to train the multi-modal fusion model.

[0054] The technical scheme adopted by the present application to solve the second technical problem is: an architecture implementation method for digital factory intelligent production, which realizes any of the architecture systems for digital factory intelligent production, and is characterized by comprising the following steps:

[0055] Step 1, the data perception layer collects the device state information of the corresponding production equipment in the workshop in real time, and adjusts the display content of the control equipment of the digital factory in real time according to the device state information; wherein the device state information of the production equipment includes device operation mode, device running state, main shaft speed, main shaft load and main shaft temperature, the device running state includes running state, standby state and abnormal state, and each production equipment is at least corresponding to one deployed industrial data acquisition module;

[0056] Step 2, the edge computing layer analyzes and processes the abnormal signal of the device state information sent by the data perception layer, optimizes the production plan in the workshop, and analyzes the workshop environment;

[0057] Step 3, the cloud computing layer respectively makes multi-modal data fusion according to the data provided by the edge computing layer and different downstream tasks, extracts the internal features of each modal data after multi-modal data fusion, then trains the extracted internal features, and after the training is completed, incrementally updates different downstream task models according to the different internal features extracted.

[0058] Step 4, the application layer realizes the application of the architecture to different application scenarios by calling the service interface provided by the cloud computing layer.

[0059] Compared with the prior art, the advantages of the present application are that the architecture system for digital factory intelligent production of the present application is provided with a data sensing layer, an edge computing layer, a cloud computing layer and an application layer, the data sensing layer sends the device state information of the corresponding production equipment collected in real time to the edge computing layer, and the data sensing layer adjusts the display content of the control device of the digital factory in real time according to the device state information, then the edge computing layer analyzes the abnormal signal, optimizes the production plan in the workshop and analyzes the workshop environment according to the device state information sent by the data sensing layer, the cloud computing layer respectively makes multi-modal data fusion according to the data provided by the edge computing layer and different downstream tasks, extracts the internal features of each modal data after multi-modal data fusion, then trains the extracted internal features, and after the training is completed, incrementally updates different downstream task models according to the different internal features extracted, and the application layer realizes the application of the architecture to different application scenarios by calling the service interface provided by the cloud computing layer. In this way, the architecture system can not only acquire the device state information of the production equipment in the workshop in real time, but also analyze the abnormal signal, the workshop environment and dynamically optimize the production plan in the workshop according to the device state information, so as to meet the requirement of responding quickly to the changes in the factory production and improve the overall production efficiency of the digital factory. BRIEF DESCRIPTION OF DRAWINGS

[0060] Figure 1 Fig. 1 is a schematic diagram of the architecture system for digital factory intelligent production in the embodiment of the present application;

[0061] Figure 2 Fig. 2 is a flowchart of the implementation method of the architecture system shown in Fig. 1. Figure 1 DETAILED DESCRIPTION

[0062] The present application will be further described in detail below with reference to the embodiments of the drawings.

[0063] The present embodiment provides an architecture system for digital factory intelligent production. Specifically, referring to Fig. 1, the architecture system for digital factory intelligent production comprises a data sensing layer, an edge computing layer, a cloud computing layer and an application layer. Figure 1 ​As shown, the architecture system of the embodiment facing the intelligent production of digital factory includes a data perception layer 1, an edge computing layer 2, a cloud computing layer 3 and an application layer 4. Among them:

[0064] The data perception layer is formed by a plurality of industrial data acquisition modules deployed on a plurality of production devices in the digital factory. The industrial data acquisition modules are configured to collect device state information of the corresponding production devices in real time, and to adjust the display content of the control devices of the digital factory in real time according to the device state information. The device state information of the production device includes device operation mode, device running state, spindle speed, spindle load and spindle temperature. The device running state includes running state, standby state and abnormal state. Each production device is associated with at least one industrial data acquisition module. For example, in the data perception layer of the embodiment, the industrial data acquisition modules include cameras, laser radars, temperature sensors and air pressure sensors, etc.

[0065] The edge computing layer is formed by a plurality of edge computing devices located at the edge of the network, and is configured to analyze abnormal signals of the device state information sent by the data perception layer, to optimize the production plan in the workshop, and to analyze the workshop environment. The edge computing devices are located near the source of industrial data. The edge computing layer includes an abnormal detection analysis module and a workshop scheduling analysis module. The abnormal detection analysis module is configured to receive the device state information sent by the data perception layer and analyze abnormal signals of the device state information. The workshop scheduling analysis module is configured to analyze and process the production information in the workshop in real time to optimize the production plan. The production information in the workshop includes available production device resources in the current workshop and production tasks in the production queue.

[0066] The cloud computing layer includes a cloud computing center and a plurality of model nodes, and is configured to respectively perform multi-modal data fusion on the device state information provided by the edge computing layer and different downstream tasks, extract the intrinsic features of each modal data after multi-modal data fusion, and then train the extracted intrinsic features. After training, the different downstream task models are incrementally updated according to the different intrinsic features extracted.

[0067] The application layer realizes the application of the architecture to different application scenarios by calling the service interface provided by the cloud computing layer.

[0068] Specifically, the process of the industrial data acquisition module in the data perception layer collecting the device state information of the corresponding production device in real time includes the following steps:

[0069] The industrial data acquisition module is networked under the condition that it is normally connected with the production devices in the workshop to obtain the device state information of the corresponding production devices.

[0070] According to the obtained device state information, the device running state is determined, and the data transmission frequency is adjusted based on the device running state, and the device state information is sent to the edge server for analysis and processing according to the adjusted data transmission frequency corresponding to the running state; wherein each device running state has a corresponding data transmission frequency.

[0071] More specifically, in this embodiment, the abnormal signal analysis processing of the abnormal detection analysis module in the edge computing layer is performed in the following steps a1-a4:

[0072] Step a1, the abnormal detection analysis module makes an abnormal analysis judgment on the device state information sent by the data perception layer according to the preset abnormal detection model:

[0073] When the abnormal detection analysis module judges that an abnormal signal is obtained, the production device corresponding to the abnormal signal enters an abnormal state and stops production work, and step a2 is entered; otherwise, the production device continues production work;

[0074] Step a2, the abnormal detection analysis module analyzes the abnormal reason and type of the abnormal signal to obtain an abnormal analysis result; wherein the abnormal analysis result includes the abnormal reason and type of the abnormal signal;

[0075] Step a3, the abnormal detection analysis module takes corresponding abnormal processing measures according to the abnormal analysis result, and performs multi-index testing on the production device processed by the abnormal processing measures;

[0076] Step a4, the abnormal detection analysis module makes a judgment according to the multi-index test result:

[0077] When the production device successfully passes the multi-index test, the production device returns to the normal state and updates the running state of the production device; otherwise, the multi-index test is re-executed on the production device until the production device successfully passes the multi-index test.

[0078] For the workshop scheduling analysis module in this embodiment, the workshop scheduling analysis module in the edge computing layer analyzes and processes the production information in the workshop in real time to complete the scheduling optimization process of the production plan, including the following steps:

[0079] When the workshop scheduling analysis module determines that a new production task appears or a production device is in an abnormal state, the current workshop environment is analyzed and the available production device resources in the current workshop and the production tasks in the production queue are counted;

[0080] The workshop scheduling analysis module outputs a scheduling scheme suitable for the current workshop condition according to the counted available production device resources in the current workshop and the production tasks in the production queue.

[0081] The workshop scheduling analysis module verifies the feasibility of the scheduling scheme according to the task priority of each production task in the scheduling scheme and the production constraint condition corresponding to each production task:

[0082] When the scheduling scheme is feasible, the current workshop production equipment executes the scheduling scheme; otherwise, the scheduling scheme is adjusted to obtain a new scheduling scheme with feasibility.

[0083] As a way for the edge computing layer to analyze the workshop environment, in this embodiment, the process of the edge computing layer analyzing the workshop environment includes steps b1-b6 as follows:

[0084] Step b1, traverse the running states of all production equipment in the workshop, and calculate the distribution number of production equipment in each running state respectively; wherein the number of running states of all production equipment in the workshop is marked as N, the i-th running state in the N running states is marked as S i , and the distribution number of production equipment in the running state S i is marked as

[0085] 1≤i≤N;

[0086]

[0087] Wherein J represents the total number of production equipment in the workshop, s j represents the running state of the j-th production equipment in the workshop, and δ(s j =S i ) is an indicator function about running state s j and running state S i .

[0088] Step b2, according to the running state of each production equipment obtained by traversing, judge the total number of production equipment in the fault state, and calculate the equipment failure rate in the current workshop; wherein the equipment failure rate in the workshop is calculated as follows:

[0089]

[0090] Wherein λ represents the equipment failure rate in the current workshop, S b represents that the production equipment is in the fault state.

[0091] Step b3, obtain the running time of each production equipment in different running states, and calculate the running state time length proportion of each running state corresponding to each production equipment; wherein the running state time length proportion is calculated as follows:

[0092]

[0093] wherein f ji represents the time length ratio of the jth production equipment in the running state S i ; j,i represents the time length of the jth production equipment in the running state S i ;

[0094] Step b4, the product quality of each type of workpiece produced in the factory is counted, the number of excellent products of each type of workpiece is calculated, and the excellent product rate of each type of workpiece is calculated; wherein the calculation method of the excellent product rate of each type of workpiece is as follows:

[0095]

[0096] wherein e u represents the excellent product rate of the u-th type of workpiece, P u represents the total number of the u-th type of workpiece produced in the workshop, p uv represents the quality of the v-th workpiece of the u-th type of workpiece produced in the workshop;

[0097] Step b5, the product quality of each type of workpiece required by each production order is counted, the number of excellent products of each type of workpiece is obtained, and the production progress of each production task corresponding to the production order is calculated; wherein the total number of production tasks received in the workshop is marked as K, and the production progress of the k-th production task in the K production tasks is marked as r k :

[0098]

[0099] wherein Q k represents the target production quantity of the k-th production task, P u represents the total number of the u-th type of workpiece produced in the workshop, p uv represents the quality of the v-th workpiece of the u-th type of workpiece produced in the workshop;

[0100] Step b6, the number of workpieces of all types of workpieces produced in the workshop in each day is counted, and the workshop capacity of the workshop is calculated; wherein the workshop capacity of the workshop in the d-th day is marked as c d :

[0101] d≥1;

[0102] wherein c d,u represents the total number of the u-th type of workpiece produced in the d-th day of the workshop, and U represents the total number of types of workpieces produced in the workshop.

[0103] It should be noted that in this embodiment, the cloud computing layer respectively makes multi-modal data fusion according to the data provided by the edge computing layer and different downstream tasks, including the following steps c1-c5:

[0104] Step c1, according to the modal data type after the fusion of each modal data, a modal feature extractor corresponding to the modal data type is established by a pre-training method; wherein in this embodiment, the model of the modal feature extractor is as follows:

[0105] F c =M c (X c );

[0106] Wherein, X c represents modal data of type c, M c represents a modal feature extractor for modal data of type c, F c represents the modal feature extracted from the modal data X c of type c.

[0107] Step c2, retrain the established modal feature extractor according to the production task, so that the trained modal feature extractor meets the specific downstream task requirements in the application layer; wherein the trained modal feature extraction model that meets the specific downstream task requirements in the application layer is as follows:

[0108] F task =M task (X task ,Y task );

[0109] Wherein, Y task represents the label of the specific downstream task, X task represents the training data of the specific downstream task, M task represents the trained modal feature extractor corresponding to the specific downstream task, F task represents the modal feature extracted by the trained modal feature extractor and meeting the specific downstream task requirements in the application layer.

[0110] Step c3, select one or more model data according to the downstream task requirements in the application layer and use the trained modal feature extractor corresponding to the model data for feature extraction processing to obtain the extracted modal feature, and use the full-automatic encoder to train the output dimension of each trained modal feature extractor to obtain the reduced modal feature; wherein the model for obtaining the reduced modal feature is as follows:

[0111] X encode =A(X,D);

[0112] Wherein, X encode, A represents an autoencoder, X represents input data of the autoencoder, and D represents a dimension of the input data X;

[0113] Step c4, each reduced dimension modal feature is put into a contrast learning network to learn the correspondence between different modal features; wherein the model of the contrast learning network is as follows:

[0114]

[0115] , wherein z i and z j are reduced dimension modal features, τ is a temperature parameter, and sim(z i , z j ) represents the cosine similarity of the modal features z i and z j ;

[0116] Step c5, according to different downstream tasks, a multimodal selection gate is used to select the corresponding modal to select the multimodal feature, and the multimodal feature is transmitted to the downstream model in the application layer which needs to update the weight for training processing; wherein the selection model of the multimodal feature is as follows:

[0117] F select = C(F i , Task; θ i ), F i ∈{F1,F2,…,F n}, 1≤i≤n;

[0118] , wherein F select is the selected modal feature, C is the multimodal selection gate, F i represents the feature of the i-th modal, n is the total number of modal, Task is the type of downstream task, and θ i is the weight of the i-th modal;

[0119] Step c6, after the downstream model in the application layer is trained, the updated parameters and weights are returned to the modal fusion model to train the multimodal fusion model. Wherein: the model of training the multimodal fusion model is as follows:

[0120] W i ∈{W1,W2,…,W n}, n∈R, 1≤i≤n;

[0121] , wherein W fusion represents the weight of the multimodal fusion model, L fusion represents the loss function of the multimodal fusion model, and W idenotes the weight from the i-th downstream model. Wherein, in this embodiment, the modal fusion model is located in the cloud computing layer. It is responsible for fusing data from different modalities to generate more comprehensive and accurate feature representations. Specifically, the modal fusion model receives the updated parameters and weights from the application layer downstream model after training, and uses this information (i.e. the updated parameters and weights after training) to train the multi-modal fusion model. The existence and function of this modal fusion model is to improve the performance of downstream tasks, and to enhance the learning ability and generalization ability of the model by fusing information from different modalities.

[0122] In addition, the embodiment also provides an architecture implementation method for digital factory intelligent production to implement the architecture system for digital factory intelligent production described above. Specifically, the architecture implementation method for digital factory intelligent production includes the following steps 1-4:

[0123] Step 1, the data perception layer collects the device state information of the corresponding production equipment in the workshop in real time, and adjusts the display content of the control equipment of the digital factory in real time according to the device state information; wherein the device state information of the production equipment includes device operation mode, device running state, spindle speed, spindle load and spindle temperature, the device running state includes running state, standby state and abnormal state, and each production equipment is at least corresponding to one deployed industrial data acquisition module;

[0124] Step 2, the edge computing layer analyzes the abnormal signal of the device state information sent by the data perception layer, optimizes the production plan in the workshop, and analyzes the workshop environment;

[0125] Step 3, the cloud computing layer respectively performs multi-modal data fusion according to the data provided by the edge computing layer and different downstream tasks, extracts the internal features of each modal data after multi-modal data fusion, and then trains the extracted internal features, and after training, incrementally updates the different downstream task models according to the different internal features extracted;

[0126] Step 4, the application layer realizes the application of the architecture to different application scenarios by calling the service interface provided by the cloud computing layer.

[0127] Wherein, the process of the industrial data acquisition module in the data perception layer collecting the device state information of the corresponding production equipment, the way of the abnormal detection analysis module in the edge computing layer performing abnormal signal analysis processing, the process of the workshop scheduling analysis module in the edge computing layer analyzing and processing the production information in the workshop to complete the scheduling optimization of the production plan, and the process of the edge computing layer analyzing the workshop environment can be referred to the contents described in the architecture system for digital factory intelligent production described above, which will not be repeated here.

[0128] While the preferred embodiments of the application have been described above in detail, it is to be understood that various modifications and alterations to the preferred embodiments will occur to persons skilled in the art. Any such modifications or alterations are intended to fall within the scope of the application.

Claims

1. An architecture system for intelligent production in digital factories, characterized in that: include: The data perception layer consists of several industrial data acquisition modules deployed on multiple production devices within the digital factory. These modules are configured to collect real-time equipment status information of the corresponding production devices and adjust the display content of the digital factory's control devices based on this information. The equipment status information includes equipment operation mode, equipment running status, spindle speed, spindle load, and spindle temperature. The equipment running status includes running status, standby status, and abnormal status. Each production device has at least one corresponding industrial data acquisition module. The edge computing layer, formed by multiple edge computing devices located at the network edge, is configured to perform anomaly signal analysis and processing on device status information sent by the data sensing layer, optimize production scheduling within the workshop, and analyze the workshop environment. The edge computing devices are located close to the industrial data generation source. This edge computing layer includes an anomaly detection and analysis module and a workshop scheduling and analysis module. The anomaly detection and analysis module is configured to receive device status information sent by the data sensing layer and perform anomaly signal analysis and processing on that information. The workshop scheduling and analysis module is configured to perform real-time analysis and processing of production information within the workshop to optimize production scheduling. The production information within the workshop includes currently available production equipment resources and production tasks in the production queue. The cloud computing layer, including the cloud computing center and multiple model nodes, is configured to perform multimodal data fusion based on the device status information provided by the edge computing layer and different downstream tasks, extract the intrinsic features of each modality after multimodal data fusion, then train the extracted intrinsic features, and after training, incrementally update the models of different downstream tasks based on the differences in the extracted intrinsic features. The application layer utilizes the service interfaces provided by the cloud computing layer to apply this architecture to different application scenarios.

2. The architecture system for intelligent production in digital factories according to claim 1, characterized in that, The process by which the industrial data acquisition module within the data perception layer collects real-time equipment status information of the corresponding production equipment includes the following steps: The industrial data acquisition module forms a network when it is normally connected to the production equipment in the workshop in order to obtain the equipment status information of the corresponding production equipment. The device's operating status is determined based on the acquired device status information, and the data transmission frequency is adjusted accordingly. The device status information is then sent to the edge server for analysis and processing according to the adjusted data transmission frequency corresponding to its operating status. Each device operating status has a corresponding data transmission frequency.

3. The architecture system for intelligent production in digital factories according to claim 1, characterized in that, The anomaly detection and analysis module within the edge computing layer performs anomaly signal analysis and processing in the following manner: Step a1: The anomaly detection and analysis module performs anomaly analysis and judgment on the device status information sent by the data perception layer according to the preset anomaly detection model: When the anomaly detection and analysis module determines that an anomaly signal has been obtained, it will cause the production equipment corresponding to the anomaly signal to enter an abnormal state, stop production, and proceed to step a2; otherwise, it will cause the production equipment to continue production. Step a2: The anomaly detection and analysis module analyzes the cause and type of the abnormal signal to obtain the anomaly analysis results; wherein, the anomaly analysis results include the cause and type of the abnormal signal. Step a3: The anomaly detection and analysis module takes corresponding anomaly handling measures based on the anomaly analysis results, and performs multi-index tests on the production equipment after the anomaly handling measures have been taken. Step a4: The anomaly detection and analysis module makes judgments and processes based on the results of multi-index tests. When the production equipment successfully passes the multi-indicator test, restore the production equipment to normal status and update the operating status of the production equipment; otherwise, re-execute the multi-indicator test on the production equipment until the production equipment successfully passes the multi-indicator test.

4. The architecture system for intelligent production in digital factories according to claim 3, characterized in that, The workshop scheduling and analysis module within the edge computing layer performs real-time analysis and processing of production information within the workshop to complete the scheduling optimization process of the production plan, including the following steps: When the workshop scheduling and analysis module determines that a new production task has occurred or that the production equipment is in an abnormal state, it analyzes the current workshop environment and counts the available production equipment resources in the current workshop as well as the production tasks in the production queue. The workshop scheduling analysis module outputs a scheduling plan suitable for the current workshop situation based on the statistical data of available production equipment resources and production tasks in the production queue. The workshop scheduling analysis module verifies the feasibility of the scheduling scheme based on the task priority of each production task and the corresponding production constraints in the scheduling scheme: If the scheduling plan is feasible, then all production equipment in the current workshop shall execute the scheduling plan; otherwise, the scheduling plan shall be readjusted to obtain a new scheduling plan that is feasible.

5. The architecture system for intelligent production in digital factories according to claim 4, characterized in that, The process of analyzing the workshop environment using the edge computing layer includes the following steps: Step b1: Iterate through the operating states of all production equipment in the workshop and calculate the distribution of production equipment in each operating state; where the number of operating states of all production equipment in the workshop is labeled N, and the i-th operating state among the N operating states is labeled S. i In running state S i The distribution quantity of the corresponding production equipment is marked as Where J represents the total number of production equipment in the workshop, s j δ(s) represents the operating status of the j-th production equipment in the workshop. j =S i (This refers to the running state s) j With running state S i Indicator functions; Step b2: Based on the operating status of each production device obtained through traversal, determine the total number of production devices in a faulty state, and calculate the current equipment failure rate in the workshop; the equipment failure rate in the workshop is calculated as follows: Where λ represents the current equipment failure rate in the workshop, S b This indicates that the production equipment is in a faulty state; Step b3: Obtain the runtime of each production device in different operating states, and calculate the percentage of runtime for each operating state for each production device; the calculation method for the percentage of runtime for each operating state is as follows: Among them, f ji S indicates that the j-th production equipment is in an operating state. i The percentage of time corresponding to each hour, T j,i S indicates that the j-th production equipment is in an operating state. i The duration of time occupied; Step b4: Analyze the production quality of each type of workpiece produced within the factory, calculate the number of high-quality workpieces for each type, and calculate the yield rate for each type of workpiece. The yield rate for each type of workpiece is calculated as follows: Among them, e u P represents the yield rate of the u-th type of workpiece. u p represents the total number of workpieces of type u that have been produced in the workshop. uv This indicates the quality of the v-th workpiece of type u produced in the workshop; Step b5: Calculate the production quality of all produced workpieces of each type required for each production order, obtain the quantity of high-quality workpieces of each type, and calculate the production progress of the corresponding production task for each production order; where the total number of production tasks received in the workshop is denoted as K, and the production progress of the k-th production task out of the K production tasks is denoted as r. k : Among them, Q k Let P represent the target production quantity for the k-th production task. u p represents the total number of workpieces of type u that have been produced in the workshop. uv This indicates the quality of the v-th workpiece of type u produced in the workshop; Step b6: Calculate the daily production quantity of all workpiece types in the workshop and determine the workshop's capacity; the workshop capacity on day d is denoted as c. d : Among them, c d,u This represents the total number of workpieces of type u produced by the workshop on day d, where U represents the total number of different types of workpieces produced by the workshop.

6. An architecture implementation method for intelligent production in digital factories, implementing the architecture system for intelligent production in digital factories as described in any one of claims 1 to 5, characterized in that, Includes the following steps: Step 1: The data perception layer collects the equipment status information of the corresponding production equipment in the workshop in real time, and adjusts the display content of the control equipment of the digital factory in real time according to the equipment status information. The equipment status information of the production equipment includes equipment operation mode, equipment running status, spindle speed, spindle load and spindle temperature. The equipment running status includes running status, standby status and abnormal status. Each production equipment has at least one corresponding industrial data acquisition module. Step 2: The edge computing layer performs abnormal signal analysis and processing on the device status information sent by the data perception layer, optimizes the production plan in the workshop, and analyzes the workshop environment. Step 3: The cloud computing layer performs multimodal data fusion based on the data provided by the edge computing layer and different downstream tasks, and extracts the intrinsic features of each modality after multimodal data fusion. Then, the extracted intrinsic features are used for training. After training, the different downstream task models are incrementally updated based on the different extracted intrinsic features. Step 4: The application layer calls the service interfaces provided by the cloud computing layer to apply this architecture to different application scenarios.

7. The architecture implementation method for intelligent production in digital factories according to claim 6, characterized in that, The process by which the industrial data acquisition module within the data perception layer collects real-time equipment status information of the corresponding production equipment includes the following steps: The industrial data acquisition module forms a network when it is normally connected to the production equipment in the workshop in order to obtain the equipment status information of the corresponding production equipment. The device's operating status is determined based on the acquired device status information, and the data transmission frequency is adjusted accordingly. The device status information is then sent to the edge server for analysis and processing according to the adjusted data transmission frequency corresponding to its operating status. Each device operating status has a corresponding data transmission frequency.

8. The architecture implementation method for intelligent production in digital factories according to claim 7, characterized in that, The anomaly detection and analysis module within the edge computing layer performs anomaly signal analysis and processing in the following manner: Step a1: The anomaly detection and analysis module performs anomaly analysis and judgment on the device status information sent by the data perception layer according to the preset anomaly detection model: When the anomaly detection and analysis module determines that an anomaly signal has been obtained, it will cause the production equipment corresponding to the anomaly signal to enter an abnormal state, stop production, and proceed to step a2; otherwise, it will cause the production equipment to continue production. Step a2: The anomaly detection and analysis module analyzes the cause and type of the abnormal signal to obtain the anomaly analysis results; wherein, the anomaly analysis results include the cause and type of the abnormal signal. Step a3: The anomaly detection and analysis module takes corresponding anomaly handling measures based on the anomaly analysis results, and performs multi-index tests on the production equipment after the anomaly handling measures have been taken. Step a4: The anomaly detection and analysis module makes judgments and processes based on the results of multi-index tests. When the production equipment successfully passes the multi-indicator test, restore the production equipment to normal status and update the operating status of the production equipment; otherwise, re-execute the multi-indicator test on the production equipment until the production equipment successfully passes the multi-indicator test. The workshop scheduling and analysis module within the edge computing layer performs real-time analysis and processing of production information within the workshop to complete the scheduling optimization process of the production plan, including the following steps: When the workshop scheduling and analysis module determines that a new production task has occurred or that the production equipment is in an abnormal state, it analyzes the current workshop environment and counts the available production equipment resources in the current workshop as well as the production tasks in the production queue. The workshop scheduling analysis module outputs a scheduling plan suitable for the current workshop situation based on the statistical data of available production equipment resources and production tasks in the production queue. The workshop scheduling analysis module verifies the feasibility of the scheduling scheme based on the task priority of each production task and the corresponding production constraints in the scheduling scheme: If the scheduling plan is feasible, then all production equipment in the current workshop shall execute the scheduling plan; otherwise, the scheduling plan shall be readjusted to obtain a new scheduling plan that is feasible.

9. The architecture implementation method for intelligent production in digital factories according to claim 8, characterized in that, The process of analyzing the workshop environment using the edge computing layer includes the following steps: Step b1: Iterate through the operating states of all production equipment in the workshop and calculate the distribution of production equipment in each operating state; where the number of operating states of all production equipment in the workshop is labeled N, and the i-th operating state among the N operating states is labeled S. i In running state S i The distribution quantity of the corresponding production equipment is marked as Where J represents the total number of production equipment in the workshop, s j δ(s) represents the operating status of the j-th production equipment in the workshop. j =S i (This refers to the running state s) j With running state S i Indicator functions; Step b2: Based on the operating status of each production device obtained through traversal, determine the total number of production devices in a faulty state, and calculate the current equipment failure rate in the workshop; the equipment failure rate in the workshop is calculated as follows: Where λ represents the current equipment failure rate in the workshop, S b This indicates that the production equipment is in a faulty state; Step b3: Obtain the runtime of each production device in different operating states, and calculate the percentage of runtime for each operating state for each production device; the calculation method for the percentage of runtime for each operating state is as follows: Among them, f ji S indicates that the j-th production equipment is in an operating state. i The percentage of time corresponding to each hour, T j,i S indicates that the j-th production equipment is in an operating state. i The duration of time occupied; Step b4: Analyze the production quality of each type of workpiece produced within the factory, calculate the number of high-quality workpieces for each type, and calculate the yield rate for each type of workpiece. The yield rate for each type of workpiece is calculated as follows: Among them, e u P represents the yield rate of the u-th type of workpiece. u p represents the total number of workpieces of type u that have been produced in the workshop. uv This indicates the quality of the v-th workpiece of type u produced in the workshop; Step b5: Calculate the production quality of all produced workpieces of each type required for each production order, obtain the quantity of high-quality workpieces of each type, and calculate the production progress of the corresponding production task for each production order; where the total number of production tasks received in the workshop is denoted as K, and the production progress of the k-th production task out of the K production tasks is denoted as r. k : Among them, Q k Let P represent the target production quantity for the k-th production task. u p represents the total number of workpieces of type u that have been produced in the workshop. uv This indicates the quality of the v-th workpiece of type u produced in the workshop; Step b6: Calculate the daily production quantity of all workpiece types in the workshop and determine the workshop's capacity; the workshop capacity on day d is denoted as c. d : Among them, c d,u This represents the total number of workpieces of type u produced by the workshop on day d, where U represents the total number of different types of workpieces produced by the workshop.

Images