Intelligent all-in-one machine control method and system

Abstract

The present invention relates to the field of equipment control technology, and specifically to an intelligent all-in-one machine control method and system. First, a collection frequency is set for different types of production parameters, raw data is collected in real time and preliminarily processed, a unified timestamp is established, and multi-source production parameters are obtained; then, fusion analysis is performed on the multi-source production parameters to obtain fused operation mode information, and the real-time operation mode is identified; a multi-objective defect prediction model is constructed, and when an abnormality occurs in the real-time operation mode, the potential defect status is predicted according to the fused operation mode information and the multi-source production parameters; then, an intelligent decision-making model based on reinforcement learning is introduced, and an optimal production parameter adjustment plan is generated based on the potential defect status and the multi-source production parameters; finally, automatic control of the intelligent all-in-one machine is realized based on the optimal production parameter adjustment plan; the present invention realizes adaptive optimization of the production process and improves the adaptability of the intelligent all-in-one machine control method to changes in the production process.

Description

Technical Field

[0001] The present invention relates to the technical field of equipment control, and in particular to an intelligent all-in-one machine control method and system. Background Art

[0002] With the rapid development of intelligent manufacturing, traditional manufacturing is gradually transforming towards intelligence and automation. Especially in the field of light industry, the application of intelligent integrated equipment is becoming more and more widespread in the manufacturing process of daily consumer goods such as clothing and socks. These intelligent all-in-one machines usually integrate multiple functions of the production process, including raw material processing, finished product weaving, testing and packaging, which significantly improves the continuity and efficiency of the overall production process and further reduces the participation of manual operations and production costs.

[0003] In the application of intelligent all-in-one machines, the degree of automation of the equipment and the level of intelligent operation play a vital role in the quality and production efficiency of the final product. Therefore, the control method of the intelligent all-in-one machine is particularly critical. For example, the production process of socks involves multiple steps such as yarn supply, weaving, and molding. The parameter adjustment of each step and the detection and processing of abnormal situations all require efficient management of intelligent control methods, which can not only effectively reduce quality fluctuations caused by human errors, but also realize refined and standardized operations in the production process.

[0004] Most of the existing intelligent all-in-one machine control methods are based on traditional program control and preset logic. These methods usually rely on pre-set parameters and fixed processes, and are often difficult to adapt to complex changes in the production process. They are not flexible enough to respond to complex and changing production needs, and it is difficult to achieve true adaptive and dynamic optimization control.

[0005] To this end, a smart all-in-one machine control method and system are proposed. Summary of the invention

[0006] The purpose of the present invention is to provide an intelligent all-in-one machine control method and system. First, the collection frequency is set for different types of production parameters, the original data is collected in real time and preliminarily processed, and a unified timestamp is established; then, the multi-source production parameters are fused and analyzed to identify the real-time operation mode, including normal, slightly abnormal and seriously abnormal modes; a multi-objective defect prediction model is constructed to predict the type, generation time and severity of potential defects. The model adopts a structure of a shared feature extraction layer and a task-specific output layer, and is trained using a weighted cross entropy loss function and an Adam optimization algorithm; then, an intelligent decision-making model based on reinforcement learning is introduced to generate an optimal production parameter adjustment plan. The decision-making model adopts a dual network structure and an experience replay mechanism, and continuously optimizes the decision-making strategy through continuous learning; finally, the automatic control of the intelligent all-in-one machine is realized based on the optimal production parameter adjustment plan; the present invention realizes intelligent monitoring, prediction and adaptive optimization of the production process, and improves production efficiency and product quality.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A method for controlling an intelligent all-in-one machine, comprising:

[0009] Set the collection frequency for different types of production parameters to collect raw data in real time during the product production process;

[0010] The raw data is preliminarily processed, specifically including: determining a reference frequency, downsampling the raw data with a sampling frequency higher than the reference frequency, interpolating the raw data with a sampling frequency lower than the reference frequency, establishing a unified timestamp, and obtaining continuous multi-source production parameters. ,in Indicates that from Production parameters of each data source;

[0011] Performing fusion analysis on the multi-source production parameters, identifying the real-time operation mode and constructing fusion operation mode information; the real-time operation mode includes a normal operation mode, a slight abnormality mode and a serious abnormality mode;

[0012] Combining multi-source production parameters with the fused operation mode information to generate fused time series data;

[0013] Build a multi-objective defect prediction model and train and optimize it through historical data;

[0014] When the real-time operation mode is not the normal operation mode, inputting the fused time series data into the multi-objective defect prediction model to predict potential defect conditions; the potential defect conditions include defect type, defect generation time and defect severity;

[0015] According to the potential defect conditions, an intelligent decision-making model is constructed to generate an optimal production parameter adjustment plan;

[0016] The optimal production parameter adjustment plan is converted into specific control instructions, the control instructions are executed, and the production equipment is adjusted in real time.

[0017] Preferably, performing fusion analysis on the multi-source production parameters to identify the real-time operation mode comprises the following specific steps:

[0018] Define a set of run modes ={normal operation mode, slight abnormality mode, severe abnormality mode};

[0019] Define each data source and run mode subset The distribution function between them is used to obtain the distribution probability of each data source to each of the operating mode subsets. ;

[0020] Calculate the fusion assignment probability of all data sources to the subset of operating modes:

[0021] ;

[0022] in, Represents all data sources for the specified subset of operation modes The fusion allocation probability of Indicates The operating mode subset corresponding to the data source; express The intersection of the operating mode subsets corresponding to the data sources is ; Indicates the number of data sources; Indicates Data sources for the subset of operating modes The distribution probability of Indicates The weight of each data source; Represents the normalization factor, and the calculation formula is:

[0023] ;

[0024] in, express The intersection of the operating mode subsets corresponding to the data sources is an empty set;

[0025] Calculate the fusion allocation probability of all data sources for the operation mode subset {normal operation mode}, {slight abnormal mode} and {severe abnormal mode} respectively , and , determine the real-time operation mode ,in, represents a parameter maximization function.

[0026] Preferably, the fused time series data ;in, express The multi-source production parameters at the time; express The fusion operation mode information at the time; Indicates the time window size; the fusion operation mode information ;in, Indicates uncertainty; Indicates the degree of conflict; Indicates the real-time operating mode.

[0027] Preferably, the specific steps of constructing the multi-objective defect prediction model include:

[0028] Construct a shared feature extraction layer to extract the features of the fused time series data to obtain shared features; the shared feature extraction layer is composed of The residual blocks are composed of a standard convolution layer and a hole convolution layer;

[0029] Based on the shared features, an independent output layer is designed for each task objective, wherein the task objectives include predicting defect type, predicting defect occurrence time, and predicting defect severity;

[0030] Based on the cross entropy loss function, the loss of each task target is calculated, and the total loss function is defined to calculate :

[0031] Use the adaptive optimization algorithm Adam to optimize the total loss function and update the model parameters:

[0032] ;

[0033] in, Indicates the current parameters of the model; Represents the model update parameters; Represents the loss function Current parameters of the model gradient.

[0034] Preferably, the specific steps of constructing the intelligent decision-making model to generate the optimal production parameter adjustment plan include:

[0035] Define a state space, an action space and a reward function; the state space includes current production parameters and potential defect conditions; the action space includes all adjustable production parameters;

[0036] Creating a main network and a target network; the main network and the target network are two neural networks with the same structure but independent parameters;

[0037] The intelligent decision model is pre-trained based on historical data. The training process includes: for each time step, obtaining the current state ,use Greedy strategy selects actions and executes them , get rewarded and the new state ; Create transfer sample and store it in the experience replay buffer; randomly sample a batch of transfer samples from the experience replay buffer, and calculate the target for each transfer sample value; according to the target Calculate the loss function and update the main network parameters using gradient descent ;Every Step 1 Update the target network parameters ;

[0038] The pre-trained intelligent decision-making model is applied, and the intelligent decision-making model is fine-tuned through continuous learning in the application, the optimal action is calculated, and the optimal action is converted into the optimal production parameter adjustment plan.

[0039] An intelligent all-in-one machine control system, comprising:

[0040] Multi-source data acquisition module, real-time acquisition of raw data during product production;

[0041] A data preprocessing module preprocesses the raw data to obtain multi-source production parameters;

[0042] A fusion data analysis module performs fusion analysis on the multi-source production parameters to identify the real-time operation mode;

[0043] A fusion data construction module, which constructs fusion operation mode information according to the result of fusion analysis, and generates fusion time series data by combining the multi-source production parameters and the fusion operation mode information;

[0044] A defect condition prediction module, which constructs a multi-objective defect prediction model based on historical data, and when the real-time operation mode is not the normal operation mode, inputs the fused time series data into the multi-objective defect prediction model to predict potential defect conditions;

[0045] An intelligent decision-making module, which builds an intelligent decision-making model to generate an optimal production parameter adjustment plan based on the potential defect conditions;

[0046] The decision-making execution module converts the optimal production parameter adjustment plan into specific control instructions, executes the control instructions, and adjusts the production equipment in real time.

[0047] Preferably, performing fusion analysis on the multi-source production parameters to identify the real-time operation mode comprises the following specific steps:

[0048] Define a set of run modes ={normal operation mode, slight abnormality mode, severe abnormality mode};

[0049] Define each data source and run mode subset The distribution function between them is used to obtain the distribution probability of each data source to each of the operating mode subsets. ;

[0050] Calculate the fusion assignment probability of all data sources to the subset of operating modes:

[0051] ;

[0052] in, Represents all data sources for the specified subset of operation modes The fusion allocation probability of Indicates The operating mode subset corresponding to the data source; express The intersection of the operating mode subsets corresponding to the data sources is ; Indicates the number of data sources; Indicates Data sources for the subset of operating modes The distribution probability of Indicates The weight of each data source; Represents the normalization factor, and the calculation formula is:

[0053] ;

[0054] in, express The intersection of the operating mode subsets corresponding to the data sources is an empty set;

[0055] Calculate the fusion allocation probability of all data sources for the operation mode subset {normal operation mode}, {slight abnormal mode} and {severe abnormal mode} respectively , and , determine the real-time operation mode ,in, represents a parameter maximization function.

[0056] Preferably, the fused time series data ;in, express The multi-source production parameters at the time; express The fusion operation mode information at the time; Indicates the time window size; the fusion operation mode information ;in, Indicates uncertainty; Indicates the degree of conflict; Indicates the real-time operating mode.

[0057] Preferably, the specific steps of constructing the multi-objective defect prediction model include:

[0058] Construct a shared feature extraction layer to extract the features of the fused time series data to obtain shared features; the shared feature extraction layer is composed of The residual blocks are composed of a standard convolution layer and a hole convolution layer;

[0059] Based on the shared features, an independent output layer is designed for each task objective, wherein the task objectives include predicting defect type, predicting defect occurrence time, and predicting defect severity;

[0060] Based on the cross entropy loss function, the loss of each task target is calculated, and the total loss function is defined to calculate :

[0061] Use the adaptive optimization algorithm Adam to optimize the total loss function and update the model parameters:

[0062] ;

[0063] in, Indicates the current parameters of the model; Represents the model update parameters; Represents the loss function Current parameters of the model gradient.

[0064] Preferably, the specific steps of constructing the intelligent decision-making model to generate the optimal production parameter adjustment plan include:

[0065] Define a state space, an action space and a reward function; the state space includes current production parameters and potential defect conditions; the action space includes all adjustable production parameters;

[0066] Creating a main network and a target network; the main network and the target network are two neural networks with the same structure but independent parameters;

[0067] The intelligent decision model is pre-trained based on historical data. The training process includes: for each time step, obtaining the current state ,use Greedy strategy selects actions and executes them , get rewarded and the new state ; Create transfer sample and store it in the experience replay buffer; randomly sample a batch of transfer samples from the experience replay buffer, and calculate the target for each transfer sample value; according to the target Calculate the loss function and update the main network parameters using gradient descent ;Every Step 1 Update the target network parameters ;

[0068] The pre-trained intelligent decision-making model is applied, and the intelligent decision-making model is fine-tuned through continuous learning in the application, the optimal action is calculated, and the optimal action is converted into the optimal production parameter adjustment plan.

[0069] Compared with the prior art, the present invention has the following beneficial effects:

[0070] 1. The present invention conducts fusion analysis on multi-source data to identify the real-time operation mode. First, the acquisition frequency is set for different types of production parameters to collect raw data in real time; then, a unified timestamp is established through downsampling and interpolation processing to obtain continuous multi-source production parameters; the present invention introduces a data fusion algorithm based on evidence theory to calculate the fusion allocation probability of different operation modes, thereby identifying the current real-time operation mode, and can comprehensively consider the information of multiple data sources to improve the accuracy and reliability of operation mode identification; by timely identifying abnormal operation modes, a quick response can be made to prevent the problem from further deteriorating, thereby improving the adaptability of the intelligent all-in-one machine control method to complex and changeable production processes, and laying the foundation for dynamic optimization control.

[0071] 2. The present invention constructs a multi-objective defect prediction model for predicting potential defect conditions. The model adopts a structure of a shared feature extraction layer and a task-specific output layer, which can simultaneously predict the defect type, generation time and severity; the shared feature extraction layer is composed of residual blocks, including standard convolution and hole convolution, which helps to capture features of different scales; by accurately predicting potential defects, preventive measures can be taken to reduce the probability of actual defects, ensure product quality and reduce the waste of raw materials, improve the control ability of the intelligent all-in-one control method in the face of emergencies, and improve the accuracy of dynamic optimization control.

[0072] 3. The present invention designs an intelligent decision-making model based on reinforcement learning to generate the optimal production parameter adjustment plan. The model defines a state space containing current production parameters and potential defect conditions, as well as an action space for adjusting production parameters; the dual network structure and experience playback mechanism are adopted to improve the stability and efficiency of learning; after the model is pre-trained with historical data, it continues to learn and fine-tune in practical applications and can adapt to dynamically changing production environments; this method can automatically generate the optimal parameter adjustment plan based on current production parameters and predicted defect conditions, and through continuous learning and optimization, the model can gradually improve the accuracy and effect of decision-making, enhance the adaptive ability of the intelligent all-in-one control method and the flexible response to changes in the production process, and realize flexible dynamic optimization control. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] Figure 1 A schematic diagram of a flow chart of a smart all-in-one machine control method provided by an embodiment of the present invention;

[0074] Figure 2 A schematic diagram of the structure of an intelligent all-in-one machine control system provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0075] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0076] In the field of sock production, the intelligent all-in-one machine integrates multiple processes such as weaving, sewing, and turning socks. The production of socks involves a complex weaving process, which greatly shortens the entire sock knitting time and reduces the company's employment; however, the specific sock production steps involve a complex weaving process, which requires high accuracy in parameter control. Through the intelligent all-in-one machine control method, key parameters such as yarn tension, needle pitch, and temperature can be monitored in real time, and abnormal conditions such as broken threads, skipped stitches, or temperature deviations can be quickly discovered. The multi-objective defect prediction model can provide early warning of quality problems such as uneven weaving, looseness, or size deviations. The intelligent decision-making model can automatically optimize parameters such as needle speed and yarn feeding. This not only significantly reduces the defective rate and improves product quality stability, but also realizes automated adjustment of the production process, promoting the sock production industry to develop in the direction of intelligent manufacturing and Industry 4.0, and achieving a comprehensive improvement in production efficiency and product quality.

[0077] The intelligent all-in-one machine control method and system proposed in the present invention can realize rapid identification of abnormal conditions, early warning of potential problems, and automatic adjustment of production parameters through real-time data collection, multi-source parameter fusion analysis, multi-target defect prediction and intelligent decision-making model, thereby significantly improving production efficiency, product quality and resource utilization. In order to illustrate the role of the present invention, the effectiveness of the present invention will be illustrated from the following two embodiments.

[0078] Embodiment 1

[0079] Sock manufacturers use a knitting, sewing and turning integrated intelligent sock machine to achieve automation transformation in the production process, and apply an intelligent integrated machine control method, see Figure 1 ,include:

[0080] Set the collection frequency for different types of production parameters to collect raw data in real time during the product production process;

[0081] The raw data is preliminarily processed, specifically including: determining a reference frequency, downsampling the raw data with a sampling frequency higher than the reference frequency, interpolating the raw data with a sampling frequency lower than the reference frequency, establishing a unified timestamp, and obtaining continuous multi-source production parameters. ,in Indicates that from Production parameters of each data source;

[0082] Performing fusion analysis on the multi-source production parameters, identifying the real-time operation mode and constructing fusion operation mode information; the real-time operation mode includes a normal operation mode, a slight abnormality mode and a serious abnormality mode;

[0083] Combining multi-source production parameters with the fused operation mode information to generate fused time series data;

[0084] Build a multi-objective defect prediction model and train and optimize it through historical data;

[0085] When the real-time operation mode is not the normal operation mode, inputting the fused time series data into the multi-objective defect prediction model to predict potential defect conditions; the potential defect conditions include defect type, defect generation time and defect severity;

[0086] According to the potential defect conditions, an intelligent decision-making model is constructed to generate an optimal production parameter adjustment plan;

[0087] The optimal production parameter adjustment plan is converted into specific control instructions, the control instructions are executed, and the production equipment is adjusted in real time.

[0088] In the embodiment of the present application, the reference frequency is selected as 10 Hz, and an example of multi-source production parameters obtained after preliminary processing is shown in Table 1.

[0089] Table 1 Example of multi-source production parameters

[0090]

[0091] Comprehensive production process monitoring and optimization are achieved through multiple steps. First, the collection frequency is set for different types of production parameters to ensure the comprehensiveness and pertinence of data collection; the initial processing of raw data achieves data standardization and consistency, laying the foundation for subsequent analysis; the fusion analysis of multi-source production parameters improves the accuracy of operation mode recognition; the construction of a multi-objective defect prediction model can timely warn of potential problems; finally, the application of the intelligent decision-making model realizes the automatic optimization and adjustment of production parameters; the proposed intelligent all-in-one machine control method significantly improves the intelligence level of the production process, realizes rapid response and processing of abnormal situations, effectively reduces product defect rates, and improves production efficiency and product quality.

[0092] Furthermore, performing fusion analysis on the multi-source production parameters to identify the real-time operation mode includes the following specific steps:

[0093] Define a set of run modes ={normal operation mode, slight abnormality mode, severe abnormality mode};

[0094] Define each data source and run mode subset The distribution function between them is used to obtain the distribution probability of each data source to each of the operating mode subsets. ;

[0095] Calculate the fusion assignment probability of all data sources to the subset of operating modes:

[0096] ;

[0097] in, Represents all data sources for the specified subset of operation modes The fusion allocation probability of Indicates The operating mode subset corresponding to the data source; express The intersection of the operating mode subsets corresponding to the data sources is ; Indicates the number of data sources; Indicates Data sources for the subset of operating modes The probability of the allocation; Indicates The weight of each data source; Represents the normalization factor, and the calculation formula is:

[0098] ;

[0099] in, express The intersection of the operating mode subsets corresponding to the data sources is an empty set;

[0100] Calculate the fusion allocation probability of all data sources for the operation mode subset {normal operation mode}, {slight abnormal mode} and {severe abnormal mode} respectively , and , determine the real-time operation mode ,in, Represents a parameter maximization function, that is, the operation mode with the highest fusion allocation probability is selected as the current real-time operation mode.

[0101] In the embodiment of the present application, each data source and run mode subset The allocation function between them is a basic assignment function based on evidence theory. In high-frequency data scenarios, manually setting the allocation function is obviously not applicable. The allocation function can be dynamically generated through automation methods: pre-set the threshold range of the multi-source production parameter values, corresponding to different operating modes (normal operating mode, micro-abnormal mode, severe abnormal mode); according to the value of the multi-source production parameter, the assignment rules are automatically triggered and the corresponding distribution function is generated.

[0102] The multi-source data fusion method based on evidence theory provides an efficient and reliable solution for operation mode identification. By defining the operation mode set, calculating the allocation probability and fusing the allocation probability, this method can effectively process heterogeneous information from different data sources, and consider the reliability and importance of each data source to provide more reliable input information for subsequent defect prediction and intelligent decision-making.

[0103] Furthermore, the fused time series data ;in, express The multi-source production parameters at the time; express The fusion operation mode information at the time; Indicates the time window size; the fusion operation mode information ;in, Indicates uncertainty; Indicates the degree of conflict; Indicates the real-time operating mode.

[0104] The design of the time series data structure integrates multi-source production parameters and fused operation mode information, providing a rich and structured data basis for subsequent analysis and decision-making; by introducing uncertainty indicators and conflict degree, it can more comprehensively describe the operation mode and help improve the reliability of decision-making.

[0105] Furthermore, the specific steps of constructing the multi-objective defect prediction model include:

[0106] Construct a shared feature extraction layer to extract the features of the fused time series data to obtain shared features; the shared feature extraction layer is composed of The residual blocks are composed of a standard convolution layer and a hole convolution layer;

[0107] Based on the shared features, an independent output layer is designed for each task objective, wherein the task objectives include predicting defect type, predicting defect occurrence time, and predicting defect severity;

[0108] The loss of each task objective is calculated based on the cross entropy loss function , define the calculation of the total loss function :

[0109] ;

[0110] in, Indicates The cross entropy loss of the task objectives; Indicates The weight of each of the task objectives; represents the number of objectives of the task; Indicates The true labels of samples; Indicates The predicted value of samples; represents the number of samples;

[0111] Use the adaptive optimization algorithm Adam to optimize the total loss function and update the model parameters:

[0112] ;

[0113] in, Indicates the current parameters of the model; Represents the model update parameters; Represents the loss function Current parameters of the model gradient.

[0114] The use of residual blocks and dilated convolutions can effectively capture feature information of different scales and improve the effect of feature extraction. An independent output layer is designed for each task objective, so that the model can simultaneously predict multiple defect-related indicators and provide a comprehensive defect risk assessment. The Adam optimization algorithm is used for parameter update to improve the efficiency and stability of model training. Overall, it not only improves the accuracy and comprehensiveness of defect prediction, but also enhances the adaptability of the model.

[0115] Furthermore, the specific steps of constructing the intelligent decision-making model to generate the optimal production parameter adjustment plan include:

[0116] Define a state space, an action space and a reward function; the state space includes current production parameters and potential defect conditions; the action space includes all adjustable production parameters;

[0117] Create the main network and target network ; The main network and the target network are two neural networks with the same structure but independent parameters;

[0118] The intelligent decision model is pre-trained based on historical data. The training process includes: for each time step, obtaining the current state ,use Greedy strategy selects actions and executes them , get rewarded and the new state ; Create transfer sample and store it in the experience replay buffer; randomly sample a batch of transfer samples from the experience replay buffer, and calculate the target for each transfer sample value ,in, represents the discount factor; according to the target Value calculation loss function , use gradient descent to update the main network parameters ;Every Step 1 Update the target network parameters ;

[0119] Apply the pre-trained intelligent decision model and fine-tune the intelligent decision model through continuous learning in the application to calculate the optimal action , converting the optimal action into the optimal production parameter adjustment plan.

[0120] In the embodiment of the present application, the adjustable production parameters include yarn tension, needle speed and yarn feeding speed. Table 2 gives an example of the optimal production parameter adjustment scheme corresponding to the potential defect conditions.

[0121] Table 2 Example of optimal production parameter adjustment plan

[0122]

[0123] By defining a state space that includes current production parameters and potential defect conditions, as well as an action space for adjusting production parameters, the model can fully consider the complexity of the production environment; the application of a dual network structure improves the stability of the learning process, while the experience replay mechanism improves the efficiency of sample utilization; the model has strong adaptive capabilities through pre-training with historical data and continuous learning in actual applications, and can continuously optimize decision-making strategies; the present invention realizes the intelligent and automated adjustment of production parameters, and improves the timeliness and accuracy of decision-making.

[0124] Embodiment 2

[0125] The present invention also provides an intelligent integrated machine control system, see Figure 2 ,include:

[0126] Multi-source data acquisition module, real-time acquisition of raw data during product production;

[0127] A data preprocessing module preprocesses the raw data to obtain multi-source production parameters;

[0128] A fusion data analysis module performs fusion analysis on the multi-source production parameters to identify the real-time operation mode;

[0129] A fusion data construction module, which constructs fusion operation mode information according to the result of fusion analysis, and generates fusion time series data by combining the multi-source production parameters and the fusion operation mode information;

[0130] A defect condition prediction module, which constructs a multi-objective defect prediction model based on historical data, and when the real-time operation mode is not the normal operation mode, inputs the fused time series data into the multi-objective defect prediction model to predict potential defect conditions;

[0131] An intelligent decision-making module, which builds an intelligent decision-making model to generate an optimal production parameter adjustment plan based on the potential defect conditions;

[0132] The decision-making execution module converts the optimal production parameter adjustment plan into specific control instructions, executes the control instructions, and adjusts the production equipment in real time.

[0133] Furthermore, performing fusion analysis on the multi-source production parameters to identify the real-time operation mode includes the following specific steps:

[0134] Define a set of run modes ={normal operation mode, slight abnormality mode, severe abnormality mode};

[0135] Define each data source and run mode subset The distribution function between them is used to obtain the distribution probability of each data source to each of the operating mode subsets. ;

[0136] Calculate the fusion assignment probability of all data sources to the subset of operating modes:

[0137] ;

[0138] in, Represents all data sources for the specified subset of operation modes The fusion allocation probability of Indicates The operating mode subset corresponding to the data source; express The intersection of the operating mode subsets corresponding to the data sources is ; Indicates the number of data sources; Indicates Data sources for the subset of operating modes The probability of the allocation; Indicates The weight of each data source; Represents the normalization factor, and the calculation formula is:

[0139] ;

[0140] in, express The intersection of the operating mode subsets corresponding to the data sources is an empty set;

[0141] Calculate the fusion allocation probability of all data sources for the operation mode subset {normal operation mode}, {slight abnormal mode} and {severe abnormal mode} respectively , and , determine the real-time operation mode ,in, represents a parameter maximization function.

[0142] Furthermore, the fused time series data ;in, express The multi-source production parameters at the time; express The fusion operation mode information at the time; Indicates the time window size; the fusion operation mode information ;in, Indicates uncertainty; Indicates the degree of conflict; Indicates the real-time operating mode.

[0143] Furthermore, the specific steps of constructing the multi-objective defect prediction model include:

[0144] Construct a shared feature extraction layer to extract the features of the fused time series data to obtain shared features; the shared feature extraction layer is composed of The residual blocks are composed of a standard convolution layer and a hole convolution layer;

[0145] Based on the shared features, an independent output layer is designed for each task objective, wherein the task objectives include predicting defect type, predicting defect occurrence time, and predicting defect severity;

[0146] The loss of each task objective is calculated based on the cross entropy loss function , define the calculation of the total loss function:

[0147] ;

[0148] in, Indicates The cross entropy loss of the task objectives; Indicates The weight of each of the task objectives; represents the number of objectives of the task; Indicates The true labels of samples; Indicates The predicted value of samples; represents the number of samples;

[0149] Use the adaptive optimization algorithm Adam to optimize the total loss function and update the model parameters:

[0150] ;

[0151] in, Indicates the current parameters of the model; Represents the model update parameters; Represents the loss function Current parameters of the model gradient.

[0152] Furthermore, the specific steps of constructing the intelligent decision-making model to generate the optimal production parameter adjustment plan include:

[0153] Define a state space, an action space and a reward function; the state space includes current production parameters and potential defect conditions; the action space includes all adjustable production parameters;

[0154] Create the main network and target network ; The main network and the target network are two neural networks with the same structure but independent parameters;

[0155] The intelligent decision model is pre-trained based on historical data. The training process includes: for each time step, obtaining the current state ,use Greedy strategy selects actions and executes them , get rewarded and the new state ; Create transfer sample and store it in the experience replay buffer; randomly sample a batch of transfer samples from the experience replay buffer, and calculate the target for each transfer sample value ,in, represents the discount factor; according to the target Value calculation loss function , use gradient descent to update the main network parameters ;Every Step 1 Update the target network parameters ;

[0156] Apply the pre-trained intelligent decision model and fine-tune the intelligent decision model through continuous learning in the application to calculate the optimal action , converting the optimal action into the optimal production parameter adjustment plan.

[0157] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for controlling an intelligent all-in-one machine, characterized in that: include: Set the collection frequency for different types of production parameters to collect raw data in real time during the product production process; The raw data is preliminarily processed, specifically including: determining a reference frequency, downsampling the raw data with a sampling frequency higher than the reference frequency, interpolating the raw data with a sampling frequency lower than the reference frequency, establishing a unified timestamp, and obtaining continuous multi-source production parameters. in Indicates that from Production parameters of each data source; Performing fusion analysis on the multi-source production parameters, identifying the real-time operation mode and constructing fusion operation mode information; the real-time operation mode includes a normal operation mode, a slight abnormality mode and a serious abnormality mode; Combining the multi-source production parameters with the fused operation mode information to generate fused time series data; Build a multi-objective defect prediction model and train and optimize it through historical data; When the real-time operation mode is not the normal operation mode, inputting the fused time series data into the multi-objective defect prediction model to predict potential defect conditions; the potential defect conditions include defect type, defect generation time and defect severity; According to the potential defect conditions, an intelligent decision-making model is constructed to generate an optimal production parameter adjustment plan; Convert the optimal production parameter adjustment plan into specific control instructions, execute the control instructions, and adjust the production equipment in real time; The fusion analysis of the multi-source production parameters to identify the real-time operation mode includes the following specific steps: defining an operation mode set ={Normal operation mode, Slight abnormality mode, Severe abnormality mode}; define each data source and run mode subset The distribution function between them is used to obtain the distribution probability of each data source to each of the operating mode subsets. ; Calculate the fusion assignment probability of all data sources to the subset of operating modes: in, Represents all data sources for the specified subset of operation modes The fusion allocation probability of Indicates The operating mode subset corresponding to the data source; express The intersection of the operating mode subsets corresponding to the data sources is Indicates the number of data sources; Indicates Data sources for the subset of operating modes The probability of the allocation; Indicates The weight of each data source; Represents the normalization factor, and the calculation formula is: in, express The intersection of the operating mode subsets corresponding to the data sources is an empty set; the fusion allocation probability of all data sources to the operating mode subsets {normal operating mode}, {slight abnormal mode} and {severe abnormal mode} is calculated respectively. , and , determine the real-time operation mode ,in, represents the parameter maximization function; The fused time series data ;in, express The multi-source production parameters at the time; express The fusion operation mode information at the time; Indicates the time window size; the fusion operation mode information ;in, Indicates uncertainty; Indicates the degree of conflict; Indicates the real-time operating mode.

2. The intelligent all-in-one machine control method according to claim 1, characterized in that: The specific steps of constructing the multi-objective defect prediction model include: Construct a shared feature extraction layer to extract the features of the fused time series data to obtain shared features; the shared feature extraction layer is composed of The residual blocks are composed of a standard convolution layer and a hole convolution layer; Based on the shared features, an independent output layer is designed for each task objective, wherein the task objectives include predicting defect type, predicting defect occurrence time, and predicting defect severity; Based on the cross entropy loss function, the loss of each task target is calculated, and the total loss function is defined to calculate : Use the adaptive optimization algorithm Adam to optimize the total loss function and update the model parameters: in, Indicates the current parameters of the model; Represents the model update parameters; represents the learning rate; Represents the loss function Current parameters of the model gradient.

3. The intelligent all-in-one machine control method according to claim 1, characterized in that: The specific steps of constructing the intelligent decision-making model to generate the optimal production parameter adjustment plan include: Define a state space, an action space and a reward function; the state space includes current production parameters and potential defect conditions; the action space includes all adjustable production parameters; Creating a main network and a target network; the main network and the target network are two neural networks with the same structure but independent parameters; The intelligent decision model is pre-trained based on historical data. The training process includes: for each time step, obtaining the current state ,use Greedy strategy selects actions and executes them , get rewarded and the new state ; Create transfer sample and store it in the experience replay buffer; randomly sample a batch of transfer samples from the experience replay buffer, and calculate the target for each transfer sample value; according to the target Calculate the loss function and update the main network parameters using gradient descent ;Every Step 1 Update the target network parameters ; Apply the pre-trained intelligent decision-making model, and fine-tune the intelligent decision-making model through continuous learning in the application, calculate the optimal action, and convert the optimal action into the optimal production parameter adjustment plan.

4. An intelligent all-in-one control system, characterized in that: include: Multi-source data acquisition module, real-time acquisition of raw data during product production; A data preprocessing module preprocesses the raw data to obtain multi-source production parameters; A fusion data analysis module performs fusion analysis on the multi-source production parameters to identify the real-time operation mode; A fusion data construction module, which constructs fusion operation mode information according to the result of fusion analysis, and generates fusion time series data by combining the multi-source production parameters and the fusion operation mode information; A defect condition prediction module, which constructs a multi-objective defect prediction model based on historical data, and when the real-time operation mode is not the normal operation mode, inputs the fused time series data into the multi-objective defect prediction model to predict potential defect conditions; An intelligent decision-making module, which builds an intelligent decision-making model to generate an optimal production parameter adjustment plan based on the potential defect conditions; A decision-making execution module converts the optimal production parameter adjustment plan into specific control instructions, executes the control instructions, and adjusts the production equipment in real time; The fusion analysis of the multi-source production parameters to identify the real-time operation mode includes the following specific steps: defining an operation mode set ={Normal operation mode, Slight abnormality mode, Severe abnormality mode}; define each data source and run mode subset The distribution function between them is used to obtain the distribution probability of each data source to each of the operating mode subsets. ; Calculate the fusion assignment probability of all data sources to the subset of operating modes: in, Represents all data sources for the specified subset of operation modes The fusion allocation probability of Indicates The operating mode subset corresponding to the data source; express The intersection of the operating mode subsets corresponding to the data sources is ; Indicates the number of data sources; Indicates Data sources for the subset of operating modes The probability of the allocation; Indicates The weight of each data source; Represents the normalization factor, and the calculation formula is: in, express The intersection of the operating mode subsets corresponding to the data sources is an empty set; the fusion allocation probability of all data sources to the operating mode subsets {normal operating mode}, {slight abnormal mode} and {severe abnormal mode} is calculated respectively. , determine the real-time operation mode ,in, represents the parameter maximization function; The fused time series data, ;in, express The multi-source production parameters at the time; express The fusion operation mode information at the time; Indicates the time window size; the fusion operation mode information ;in, Indicates uncertainty; Indicates the degree of conflict; Indicates the real-time operating mode.

5. The intelligent all-in-one machine control system according to claim 4, characterized in that: The specific steps of constructing the multi-objective defect prediction model include: Construct a shared feature extraction layer to extract the features of the fused time series data to obtain shared features; the shared feature extraction layer is composed of The residual blocks are composed of a standard convolution layer and a hole convolution layer; Based on the shared features, an independent output layer is designed for each task objective, wherein the task objectives include predicting defect type, predicting defect occurrence time, and predicting defect severity; Based on the cross entropy loss function, the loss of each task target is calculated, and the total loss function is defined to calculate : Use the adaptive optimization algorithm Adam to optimize the total loss function and update the model parameters: in, Indicates the current parameters of the model; Represents the model update parameters; represents the learning rate; Represents the loss function Current parameters of the model gradient.

6. The intelligent all-in-one machine control system according to claim 4, characterized in that: The specific steps of constructing the intelligent decision-making model to generate the optimal production parameter adjustment plan include: Define a state space, an action space and a reward function; the state space includes current production parameters and potential defect conditions; the action space includes all adjustable production parameters; Creating a main network and a target network; the main network and the target network are two neural networks with the same structure but independent parameters; The intelligent decision model is pre-trained based on historical data. The training process includes: for each time step, obtaining the current state ,use Greedy strategy selects actions and executes them , get rewarded and the new state ; Create transfer sample and store it in the experience replay buffer; randomly sample a batch of transfer samples from the experience replay buffer, and calculate the target for each transfer sample value; according to the target Calculate the loss function and update the main network parameters using gradient descent ;Every Step 1 Update the target network parameters ; The pre-trained intelligent decision-making model is applied, and the intelligent decision-making model is fine-tuned through continuous learning in the application, the optimal action is calculated, and the optimal action is converted into the optimal production parameter adjustment plan.

Images