Analysis and management method for industrial production and assembly of products
By constructing a module for predicting the production and assembly cycle of components, and utilizing deep feature extractors and residual networks, the problem of low intelligence in product production and assembly scheduling was solved, achieving accurate prediction of the entire product production cycle and improving resource utilization.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHANGHAI WEISHI BUSINESS CONSULTING CO LTD
- Filing Date
- 2026-01-30
- Publication Date
- 2026-05-12
AI Technical Summary
In existing technologies, product production and assembly scheduling relies on expert experience, resulting in a low level of intelligence. Furthermore, there is uncertainty when production line equipment is replaced or production capacity changes, and there is a lack of collaborative scheduling and prediction schemes for production lines and assembly lines.
By constructing a component production cycle prediction module, a component assembly cycle prediction module, and a product full production cycle prediction module, and utilizing deep feature extractors and residual networks, combined with the inherent correlation between component production and assembly, accurate prediction of the entire product production cycle can be achieved.
It enables accurate and efficient forecasting of the entire product production cycle, improves the intelligence level of production planning and scheduling and resource allocation, and reduces reliance on human experience.
Smart Images
Figure CN122022337A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent production scheduling technology, specifically to analytical management methods for product industrial production and assembly. Background Technology
[0002] Product production and assembly scheduling refers to arranging the production sequence, assembly sequence, production lines, and assembly lines of a product within limited resources to shorten the production cycle and maximize resource utilization. However, in existing technologies, product production and assembly scheduling largely relies on expert experience. However, production variations may exist between different production lines and assembly lines at different times (e.g., equipment changes on a production line significantly increase production capacity), leading to uncertainty in judgments based on human experience, low levels of automation, and high requirements for personnel expertise. Currently, there are very few predictive solutions on the market that allow for the coordinated scheduling of products to be produced on various production lines and assembly lines, organized by component. Summary of the Invention
[0003] The present invention aims to provide an analytical management method for the industrial production and assembly of products, in order to achieve accurate and efficient prediction of the entire production cycle of the products to be produced.
[0004] The analysis and management method for industrial production and assembly of products is applied to the scheduling management of several production lines and several assembly lines for products to be produced. The products to be produced include several components. The method includes S1, a component production cycle prediction step, in which the component production cycle prediction is based on each component and multiple production lines adapted to it to predict the production cycle of the component on each production line.
[0005] S2, Component assembly cycle prediction step, wherein the component assembly cycle prediction step is used to predict the assembly cycle of at least two components on an assembly line adapted to them.
[0006] S3. Product full production cycle prediction step: The product full production cycle prediction step is based on the predicted production cycle of each component on each production line, the predicted assembly cycle, and the assembly waiting time as the characteristics of the production and assembly scheme of the product to be produced. Based on the optimal combination, the product production scheme with the shortest full production cycle time is finally selected, and the production line and assembly line of the product are scheduled based on the production scheme.
[0007] The product industrial production and assembly cycle prediction method proposed in this invention achieves accurate prediction of the entire product production cycle through the coordinated operation of a component production cycle prediction module, a component assembly cycle prediction module, and a product full production cycle prediction module. Attached Figure Description
[0008] Figure 1This is a flowchart illustrating the analysis and management method for the industrial production and assembly of the product of this invention. Detailed Implementation
[0009] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described below are only for explaining the present invention and do not limit the scope of protection of the present invention.
[0010] The present invention will be further described in detail below through preferred embodiments:
[0011] As attached Figure 1 The following is an analysis and management method for industrial production and assembly of products, applied to the scheduling and management of several production lines and several assembly lines for products to be produced. The products to be produced include several components, including S1, a component production cycle prediction step. The component production cycle prediction is based on each component and multiple production lines adapted to it to predict the production cycle of the component on each production line.
[0012] S2, Component assembly cycle prediction step, wherein the component assembly cycle prediction step is used to predict the assembly cycle of at least two components on an assembly line adapted to them.
[0013] S3. Product full production cycle prediction step: The product full production cycle prediction step is based on the predicted production cycle of each component on each production line, the predicted assembly cycle, and the assembly waiting time as the characteristics of the production and assembly scheme of the product to be produced. Based on the optimal combination, the product production scheme with the shortest full production cycle time is finally selected, and the production line and assembly line of the product are scheduled based on the production scheme.
[0014] The S1 component production cycle prediction step specifically includes:
[0015] S11, Component Production Data Feature Construction
[0016] Component production data characteristics consist of three parts: component task characteristics. Component execution progress characteristics and production line real-time status characteristics These three parts respectively construct data features of the component production process from production targets, historical production records of components, and real-time operating conditions of the production line, which serve as inputs to the production cycle prediction network.
[0017] Component task characteristics Characterizing components to be manufactured The demand for production resources consists of the following components:
[0018]
[0019] in, For components in the current production line The planned total production volume; The complexity of component processing on the production line;
[0020] Component execution progress characteristics Characterizing components to be manufactured The current processing stage consists of the following:
[0021]
[0022] in, For components The processes already completed on the current production line Quantity; This represents the predicted time span. In the prediction of component production cycle, by dynamically adjusting this time span, the minimum time required for the component production progress to reach the completion state is fitted.
[0023] Real-time status characteristics of the production line Characterization production line The Middle The process for components The actual production capacity is composed of the following:
[0024]
[0025]
[0026]
[0027]
[0028]
[0029] in, Characterization production line in the first The production material feeding buffer status of each process, including the material feeding queue length. Cumulative waiting time for feeding ; Characterization production line in the first The machine tool status of each process includes the machine tool status code. (0-Processing, 1-Waiting, 2-Fault, 3-Maintenance), Processing time since material input Real-time failure rate of production machine tools ; Characterization production line in the first The production output buffer status of each process, including the output queue length. Cumulative waiting time for material discharge ; Characterization production line in the first From one process to the next process The intermediate transit status, including the transit sequence length and transit time ;
[0030] S12, Component Production Data Feature Sampling
[0031] Component production data characteristics Based on component task characteristics Component execution progress characteristics and production line real-time status characteristics Together they constitute:
[0032]
[0033] The component production data characteristics were recorded at multiple points during the long-term production process. By stacking them in time series, the original dataset is formed. ,in This represents the total number of features in the component's production data. To improve feature robustness, it includes... The original dataset of 1 sample In the middle, through A second random sampling with replacement is used to construct a sample set of the same size as the original dataset. Repeat this process. Next, finally generated A set of independent sampling datasets .
[0034] S13, Feature Extraction of Component Production Data
[0035] Each sampling set This corresponds to a deep feature extractor, trained in parallel. Each deep feature extractor improves the generalization performance of feature extraction. Each deep feature extractor includes an encoder and a decoder, where the encoder performs the data feature transformation from the input layer to the hidden layer, and the decoder performs the data feature transformation from the hidden layer to the output layer. The encoder and decoder have a symmetrical structure.
[0036] In each deep feature extractor, the first Hidden layers to The mapping relationship between hidden layers follows:
[0037]
[0038] in, and These are the weight matrix and bias vector of the hidden layer in the deep feature extractor, respectively. This is the activation function.
[0039] No. The hidden layer features extracted by the deep feature extractor are denoted as sub-feature vectors. Ultimately, The outputs of several parallel deep feature extractors are concatenated to form a composite feature vector for prediction. :
[0040]
[0041] When training the deep feature extractor, define the loss function. Guide them to acquire production data characteristics:
[0042]
[0043] in, The reconstruction error term of the production data features is calculated by the encoder's first... Component production data characteristics The first one reconstructed by the decoder Component production data characteristics The goal of the reconstruction error term is to minimize the information distortion between the input component's production data features and the output component's production data features by summing the squared distances between them. Represents the total characteristics of production data; The weight decay regularization term is calculated by examining the weight matrices of each layer in the encoder and decoder. The inner elements are summed by squares, and a weight decay coefficient is introduced. To prevent overfitting of the deep feature extractor; The term maximizing feature mutual information is used to enhance the information correlation between the hidden layer features extracted by the deep feature extractor and the original input features, where... The mutual information loss weighting coefficient, The calculation method is as follows:
[0044]
[0045]
[0046] in, This represents the set of hidden layer feature vectors obtained after encoding the production data features of all components by a deep feature extractor; For the first Hidden layer feature vectors of each component; For the first The transpose of the hidden layer feature vectors of each component; This represents the set of feature vectors representing the production data of all components. For the first Characteristics of production data for each component; This is a scoring function used to measure the features of the hidden layer. Characteristics of component production data The degree of matching between them, when Time represents the matching score of positive sample pairs, when Time represents the matching score of the negative sample pair; This is the scoring weight matrix, used to establish the mapping relationship between the hidden layer feature space and the input feature space.
[0047] S14. Process the extracted composite feature vector through a residual network. A nonlinear mapping is performed to fit the production cycle prediction value. The residual network consists of 6 residual blocks and 2 fully connected layers.
[0048] In the residual block, the first Layer residual block to the first The mapping relationship between layer residual blocks follows:
[0049]
[0050] in, The weight matrix represents the residual block; Represents a nonlinear residual mapping operator; It is the sigmoid activation function.
[0051] After the residual network performs a nonlinear mapping on the composite feature vector, the feature space is further reconstructed and its dimensionality reduced through two fully connected layers. The computation method follows:
[0052]
[0053] in, and These are the weight matrix and bias vector of the fully connected layer, respectively. It is the sigmoid activation function.
[0054] S15, Component Production Cycle Forecast
[0055] Residual networks transform component production data features into predictions of production schedules. The predicted production schedule is output from the last fully connected layer. It is used to calculate the regression loss to guide the parameter updates of the residual network model; its regression loss The calculation is as follows:
[0056]
[0057] in, This refers to the actual component production schedule; To predict component production schedules; This represents the total number of training samples. The predicted production schedule is based on the output. The component production cycle is achieved through iteration. Forecasting. First, based on the predicted production schedule. In component execution progress characteristics Adjusting the forecast time span When satisfied That is, the predicted time span when component production is completed. This refers to the component production cycle on the current production line. .
[0058] The S2 component assembly cycle prediction step specifically includes:
[0059] The component assembly cycle prediction depends on the production progress of each component. Assembly can only begin when all components are in place.
[0060] S21. Component Assembly Data Feature Construction
[0061] Component assembly data characteristics It consists of four parts, and the characteristics of component assembly tasks. Component assembly execution progress characteristics Real-time status characteristics of the assembly line Matching component state features ;
[0062]
[0063] Assembly task characteristics Characterize the product to be assembled The requirements for assembly resources are as follows:
[0064]
[0065] in, For products in the current assembly line The planned total assembly volume; The product assembly complexity coefficient; The quantity of each component type.
[0066] Component assembly execution progress characteristics Characterize the product to be assembled The current assembly stage consists of the following:
[0067]
[0068] in, For products Assembly processes already completed on the current assembly line Quantity; This represents the assembly prediction time span. In the component assembly cycle prediction, the model dynamically adjusts this time span to fit the minimum time required for the assembly progress to reach the completion state.
[0069] Real-time status characteristics of assembly station Characterizing the assembly line The Middle The actual assembly capacity of each assembly process is composed of the following:
[0070]
[0071]
[0072]
[0073]
[0074] in, Characterizing the assembly line in the first The assembly component buffer status of the assembly process includes the queue sequence length of each component. Cumulative waiting time for each component Number of batches of complete sets of materials ; Characterizing the assembly line in the first The assembly machine tool status of each assembly process, including the assembly machine tool status code. (0-Assembly, 1-Waiting for Components, 2-Fault, 3-Changeover) Current assembly time elapsed Real-time failure rate of assembly machine tools ; Characterizing the assembly line in the first The output status of the assembly process, including the length of the output queue. Cumulative waiting time with output .
[0075] All components are in complete set status characteristics Characterization products The production completion status of each required component is as follows:
[0076]
[0077] in, A collection of components; This is the set of predicted production cycles for each component, and this value is obtained from component production cycle prediction calculations. Let be the readiness vector for each component, and let be the ratio of the number of components already produced to the total number of components planned for production.
[0078] S22. Feature Extraction of Component Assembly Data Based on Feature Transfer
[0079] Considering the inherent correlation between component production and assembly in the industrial manufacturing process, a feature transfer method is used to extract features from component assembly data. In component assembly cycle prediction, a deep feature encoder with the same structure as that used in component production cycle prediction is employed, while simultaneously adjusting the encoder parameters... The deep feature extractor in assembly cycle prediction is migrated and fine-tuned using component assembly data features. A loss function for fine-tuning is defined. for:
[0080]
[0081] in, The reconstruction error term of the assembly data features is calculated by the encoder's first... Component assembly data characteristics The first one reconstructed by the decoder Component production data characteristics The goal of the reconstruction error term is to minimize the information distortion between the input component assembly data features and the output component assembly data features by summing the squared distances between them. This represents the total feature quantity of the component assembly data. For the transfer regularization term, the weight matrix of the deep feature extractor in the assembly cycle is constrained. Weight matrix of deep feature extractor in production cycle Differences in characteristics enable the transfer of similar features; The weight decay regularization term is calculated by examining the weight matrices of each layer in the encoder and decoder. The inner elements are summed by squares, and a weight decay coefficient is introduced. This prevents overfitting of the deep feature extractor.
[0082] S23, Assembly completeness constraint
[0083] Before predicting the component assembly cycle, it is necessary to determine the assembly status of each component. Define the time frame for each component. Homologous judgment function for:
[0084]
[0085] in, This is an indicator function that takes the value 1 when the condition is met and 0 otherwise. For products The Middle The readiness rate of class components; For the first The minimum homogeneous threshold for class components;
[0086] when Time, i.e., product If the complete set conditions are met, the assembly cycle prediction can begin; otherwise, it is necessary to wait until the latest component is produced, at the specified time. Waiting time for all components to be assembled The calculation is as follows:
[0087]
[0088] in, For the first The predicted production cycle value of a component is calculated from the component production cycle prediction step. For indicator functions; For products The Middle The readiness rate of class components; For the first The minimum homogeneous threshold for class components.
[0089] S24, Component Assembly Cycle Prediction
[0090] The component assembly cycle prediction network uses the same residual network structure as the component production cycle prediction network, and ultimately outputs the predicted assembly schedule. Regression loss for assembly cycle prediction The calculation is as follows:
[0091]
[0092] in, This refers to the actual component assembly progress; To predict assembly progress; To train the assembly sample size, the regression loss uses the root mean square error form. By minimizing the deviation between the actual assembly progress and the predicted assembly progress, it guides the parameter updates of the residual network model, enabling the model to gradually fit the evolution of the assembly process.
[0093] Predicted assembly schedule based on output The product assembly cycle is adjusted through iteration. The forecast. Taking into account the waiting time for all components to be assembled, the product... Total assembly cycle The calculation is as follows:
[0094]
[0095] in, Waiting time for components to be fully assembled; The predicted assembly cycle value is obtained by dynamically adjusting the predicted time span in the assembly execution progress feature. When satisfied When, that is, when the component assembly is completed, the corresponding predicted time span. This is the predicted value of the component assembly cycle. .
[0096] The specific steps for predicting the entire production cycle of the S3 product include:
[0097] The product full production cycle prediction is based on the output results of component production cycle prediction and component assembly cycle prediction. It combines the production cycle and assembly cycle of each component to predict the full production cycle of the product.
[0098] For products awaiting production Its required The predicted production cycle value of a component is calculated by the component production cycle prediction step, forming a production cycle vector. :
[0099] in, For the first The predicted production cycle value for class components is output from the component production cycle prediction step. For products The required number of component types.
[0100] Because the components can be manufactured in parallel, the product Component production cycle Depends on the component with the longest production cycle:
[0101]
[0102] in, This represents the longest production cycle among all components, i.e., the total time spent in each component's production phase.
[0103] product The entire production cycle It consists of three parts: component production cycle, kitting waiting time, and assembly cycle.
[0104]
[0105] in, This represents the component production cycle, which is the maximum value of the production cycles for all components. The component kitting wait time is calculated from the kitting constraints in the component assembly cycle prediction. The product assembly cycle is predicted and output based on the component assembly cycle.
[0106] This invention first characterizes the dynamic state of the component production process from three dimensions: production target, historical progress, and production line operating conditions, by constructing component task features, component execution progress features, and production line real-time status features. This provides rich input information for production cycle prediction and enhances the prediction model's ability to perceive complex production environments. A deep feature extractor is used to encode and reconstruct production data, and the jump connection mechanism of the residual network enhances the transmission capability of the deep network, improving the fitting effect on the nonlinear evolution law of production progress. Then, considering the inherent correlation between component production and assembly in the industrial manufacturing process, the parameters of the deep feature encoder trained in the production cycle prediction are transferred to the assembly cycle prediction step, improving the feature extraction effect and the accuracy of assembly cycle prediction in component assembly prediction. Finally, the product full production cycle prediction module integrates the component production cycle, kitting waiting time, and assembly cycle, outputting the full production cycle prediction value of the product from raw material input to assembly completion. This achieves cycle prediction of the complete product production process, providing data support for enterprise production planning, order delivery commitment, and resource scheduling optimization.
[0107] The preferred embodiments of this application have been described in detail above with reference to the accompanying drawings. Typical known structures and common knowledge techniques in the preferred embodiments have not been described in detail here. Those skilled in the art can improve and implement the technical solutions of this invention based on the guidance provided in these embodiments and their own capabilities. Some typical known structures, known methods or common knowledge techniques should not be obstacles for those skilled in the art to implement this application.
[0108] The scope of protection claimed in this application shall be determined by the contents of its claims, and the contents described in the invention description, specific embodiments and drawings shall be used to interpret the claims.
[0109] Within the scope of the technical concept of this application, several modifications can be made to the specific implementation of this application, and these modified implementations should also be considered within the protection scope of this application.
Claims
1. An analytical management method for product industrial production and assembly, characterized by: It is applied to the scheduling management of several production lines and several assembly lines for products to be manufactured. The products to be manufactured include several components, including S1, a component production cycle prediction step. The component production cycle prediction is based on each component and multiple production lines adapted to it to predict the production cycle of the component on each production line. S2, Component assembly cycle prediction step, wherein the component assembly cycle prediction step is used to predict the assembly cycle of at least two components on an assembly line adapted to them. S3. Product full production cycle prediction step: The product full production cycle prediction step is based on the predicted production cycle of each component on each production line, the predicted assembly cycle, and the assembly waiting time as the characteristics of the production and assembly scheme of the product to be produced. Based on the optimal combination, the product production scheme with the shortest full production cycle time is finally selected, and the production line and assembly line of the product are scheduled based on the production scheme.
2. The analysis and management method for product industrial production and assembly according to claim 1, characterized in that: The specific steps for predicting the production cycle of S1 components include: S11, Component Production Data Feature Construction Component production data characteristics consist of three parts: component task characteristics. Component execution progress characteristics and production line real-time status characteristics ; Component task characteristics Characterizing components to be manufactured The demand for production resources consists of the following components: ， in, For components in the current production line The planned total production volume; The complexity of component processing on the production line; Component execution progress characteristics Characterizing components to be manufactured The current processing stage consists of the following: ， in, For components The processes already completed on the current production line Quantity; Indicates the predicted time span; Real-time status characteristics of the production line Characterization production line The Middle The process for components The actual production capacity is composed of the following: ， ， ， ， ， in, Characterization production line in the first The production material feeding buffer status of each process, including the material feeding queue length. Cumulative waiting time for feeding ; Characterization production line in the first The machine tool status of each process includes the machine tool status code. Material feeding and processing time Real-time failure rate of production machine tools ; Characterization production line in the first The production output buffer status of each process, including the output queue length. Cumulative waiting time for material discharge ; Characterization production line in the first From one process to the next process The intermediate transit status, including the transit sequence length and transit time ; S12, Component Production Data Feature Sampling Component production data characteristics Based on component task characteristics Component execution progress characteristics and production line real-time status characteristics Together they constitute: ， From containing The original dataset of 1 sample In the middle, through A second random sampling with replacement is used to construct a sampling set of the same size as the original dataset. Repeat this process. Next, finally generated A set of independent sampling datasets .
3. The analysis and management method for product industrial production and assembly according to claim 2, characterized in that, The S1 component production cycle forecasting steps also include: S13, Feature Extraction of Component Production Data Each sampling set Each deep feature extractor consists of an encoder and a decoder. The encoder performs the data feature transformation from the input layer to the hidden layer, and the decoder performs the data feature transformation from the hidden layer to the output layer. In each deep feature extractor, the first Hidden layers to The mapping relationship between hidden layers follows: ， in, and These are the weight matrix and bias vector of the hidden layer in the deep feature extractor, respectively. For activation functions; No. The hidden layer features extracted by the deep feature extractor are denoted as sub-feature vectors. ,Will The outputs of several parallel deep feature extractors are concatenated to form a composite feature vector for prediction. : 。 4. The analysis and management method for product industrial production and assembly according to claim 3, characterized in that, The S1 component production cycle forecasting steps also include: When training the deep feature extractor, define the loss function. Guide them to acquire production data characteristics: ， in, The reconstruction error term of the production data features is calculated by the encoder's first... Component production data characteristics The first one reconstructed by the decoder Component production data characteristics Sum of the squares of the distances between them Represents the total characteristic quantity of production data; The weight decay regularization term is calculated by examining the weight matrices of each layer in the encoder and decoder. The inner elements are summed by squares, and a weight decay coefficient is introduced. To prevent overfitting of the deep feature extractor; The term maximizing feature mutual information is used to enhance the information correlation between the hidden layer features extracted by the deep feature extractor and the original input features, where... The mutual information loss weighting coefficient.
5. The analysis and management method for product industrial production and assembly according to claim 4, characterized in that, Feature mutual information maximization term The calculation method is as follows: ， ， in, This represents the set of hidden layer feature vectors obtained after encoding the production data features of all components by a deep feature extractor; For the first Hidden layer feature vectors of each component; For the first The transpose of the hidden layer feature vectors of each component; This represents the set of feature vectors representing the production data of all components. For the first Characteristics of production data for each component; This is a scoring function used to measure the features of the hidden layer. Characteristics of component production data The degree of matching between them, when Time represents the matching score of positive sample pairs, when Time represents the matching score of the negative sample pair; This is the scoring weight matrix, used to establish the mapping relationship between the hidden layer feature space and the input feature space.
6. The analysis and management method for product industrial production and assembly according to claim 5, characterized in that, The S1 component production cycle forecasting steps also include: S14. Process the extracted composite feature vector through a residual network. Perform a nonlinear mapping to fit the production cycle forecast; In the residual block, the first Layer residual block to the first The mapping relationship between layer residual blocks follows: ， in, The weight matrix represents the residual block; Represents a nonlinear residual mapping operator; It is the sigmoid activation function; After the residual network performs a nonlinear mapping on the composite feature vector, the feature space is further reconstructed and its dimensionality reduced through two fully connected layers. The computation method follows: ， in, and These are the weight matrix and bias vector of the fully connected layer, respectively. It is the sigmoid activation function; S15, Component Production Cycle Forecast Residual networks transform component production data features into predictions of production schedules. The predicted production schedule is output from the last fully connected layer. It is used to calculate the regression loss to guide the parameter updates of the residual network model; its regression loss The calculation is as follows: ， in, This refers to the actual component production schedule; To predict component production schedules; The total number of training component samples, and the predicted production schedule based on the output. The component production cycle is achieved through iteration. Forecasting, firstly, is based on the predicted production schedule. In component execution progress characteristics Adjusting the forecast time span When satisfied That is, the predicted time span when component production is completed. This refers to the component production cycle on the current production line. .
7. The analysis and management method for product industrial production and assembly according to claim 1, characterized in that, The specific steps for predicting the assembly cycle of S2 components include: S21. Component Assembly Data Feature Construction Component assembly data characteristics It consists of four parts, and the characteristics of component assembly tasks. Component assembly execution progress characteristics Real-time status characteristics of the assembly line Matching component state features ; ， Assembly task characteristics Characterize the product to be assembled The requirements for assembly resources are as follows: ， in, For products in the current assembly line The planned total assembly volume; The product assembly complexity coefficient; The quantity of each component type; Component assembly execution progress characteristics Characterize the product to be assembled The current assembly stage consists of the following: ， in, For products Assembly processes already completed on the current assembly line Quantity; This represents the assembly prediction time span. In the component assembly cycle prediction, by dynamically adjusting this time span, the minimum time required for the assembly progress to reach the completion state is fitted. Real-time status characteristics of assembly station Characterizing the assembly line The Middle The actual assembly capacity of each assembly process is composed of the following: ， ， ， ， in, Characterizing the assembly line in the first The assembly component buffer status of the assembly process includes the queue sequence length of each component. Cumulative waiting time for each component Number of batches of complete sets of materials ; Characterizing the assembly line in the first The assembly machine tool status of each assembly process, including the assembly machine tool status code. The current assembly process has taken a long time. Real-time failure rate of assembly machine tools ; Characterizing the assembly line in the first The output status of the assembly process, including the length of the output queue. Cumulative waiting time with output ; All components are in complete set status characteristics Characterization products The production completion status of each required component is as follows: ， in, A collection of components; This is the set of predicted production cycles for each component, and this value is obtained from component production cycle prediction calculations. Let be the readiness vector for each component, and let be the ratio of the number of components already produced to the total number of components planned for production.
8. The analysis and management method for product industrial production and assembly according to claim 7, characterized in that, The S2 component assembly cycle prediction step also includes: S22. Feature Extraction of Component Assembly Data Based on Feature Transfer In component assembly cycle prediction, a deep feature encoder with the same structure as that used in component production cycle prediction is employed. Simultaneously, the encoder parameters are transferred to the deep feature extractor in assembly cycle prediction, and fine-tuning is achieved by inputting component assembly data features. A loss function for fine-tuning is defined. for: ， in, For the reconstruction error term of the assembly data features, the encoder's first... Component assembly data characteristics The first one reconstructed by the decoder Component production data characteristics Sum of the squares of the distances between them This represents the total feature quantity of the component assembly data. For the transfer regularization term, the weight matrix of the deep feature extractor in the assembly cycle is constrained. Weight matrix of deep feature extractor in production cycle Differences in features allow for the transfer of similar characteristics; The weight decay regularization term is calculated by examining the weight matrices of each layer in the encoder and decoder. The inner elements are summed by squares, and a weight decay coefficient is introduced. This prevents overfitting of the deep feature extractor.
9. The analysis and management method for product industrial production and assembly according to claim 8, characterized in that, The S2 component assembly cycle prediction step also includes: S23, Assembly completeness constraint Before predicting the component assembly cycle, it is necessary to determine the completeness of each component and define the time of each component. Homologous judgment function for: ， in, This is an indicator function that takes the value 1 when the condition is met and 0 otherwise. For products The Middle The readiness rate of class components; For the first The minimum homogeneous threshold for class components; when Time, i.e., product If the complete set conditions are met, the assembly cycle prediction can begin; otherwise, it is necessary to wait until the latest component is produced, at the specified time. Waiting time for all components to be assembled The calculation is as follows: ， in, For the first The predicted production cycle value of a component is calculated from the component production cycle prediction step. For indicator functions; For products The Middle The readiness rate of class components; For the first The minimum homogeneous threshold for class components.
10. The analysis and management method for product industrial production and assembly according to claim 9, characterized in that, The S2 component assembly cycle prediction step also includes: S24, Component Assembly Cycle Prediction The component assembly cycle prediction network uses the same residual network structure as the component production cycle prediction network, and ultimately outputs the predicted assembly schedule. Regression loss for assembly cycle prediction The calculation is as follows: ， in, This refers to the actual component assembly progress; To predict assembly progress; The total number of training assembly samples; Predicted assembly schedule based on output The product assembly cycle is adjusted through iteration. Based on the forecast and taking into account the waiting time for all components to be assembled, the product Total assembly cycle The calculation is as follows: ， in, Waiting time for components to be fully assembled; The predicted assembly cycle value is obtained by dynamically adjusting the predicted time span in the assembly execution progress feature. When satisfied When, that is, when the component assembly is completed, the corresponding predicted time span. This is the predicted value of the component assembly cycle. .