A digital twin workshop defect prediction method, system, storage medium and product
Through digital twin technology and machine learning decision tree algorithm, combined with physical data from textile workshops, defective product prediction is solved, and the problem of difficulty in discovering defective products in a timely manner in traditional workshops is improved, production efficiency and product quality are reduced, and production costs and risks are reduced.
Patent Information
- Application Number
- CN202411073703.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-07
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2044-08-07
AI Technical Summary
It is difficult for traditional textile workshops to discover defective products in a timely manner and deal with them during the production process, resulting in reduced production efficiency and unforeseen potential defects, increasing production losses.
Using digital twin technology and machine learning to build a digital twin model of the physical workshop, collect and preprocess production data in real time, predict the number of defective products through a decision tree algorithm, and build a visual platform to display production situation and prediction results.
It realizes the prediction of potential quality problems and faults, maintains and repairs in advance, improves the production efficiency and product quality of the workshop, and reduces production costs and risks.
Smart Images

Figure CN118586788B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of industrial manufacturing prediction, and in particular to a digital twin workshop defect prediction method, system, storage medium and product. Background Art
[0002] In traditional textile workshop production, defective products are usually inspected manually or automatically after production is completed, and defective products are repaired or scrapped. This method usually has two disadvantages: first, it cannot detect and deal with problems in time during the production process; second, it reduces production efficiency and cannot foresee potential defects to reduce production losses.
[0003] With the rapid development of intelligent manufacturing, the use of digital twin technology provides a new and effective method for predicting defective products in textile workshops. Digital twin technology combines physical workshops with virtual models and can collect various data on the production line in real time, such as production parameters and product quality. It is crucial to predict and visualize the number of defective products in the workshop through these data.
[0004] Machine learning is an important part of artificial intelligence. It has powerful data processing and pattern recognition capabilities. Combined with digital twin technology, it provides a newer solution for defect prediction in textile workshops. By training the model on processed data, it is possible to use the existing historical workshop data to predict possible problems in the future. Summary of the invention
[0005] The purpose of the present invention is to provide a digital twin workshop defect prediction method, system, storage medium and product that can predict defects in workshop production.
[0006] In order to achieve the above object, the present invention adopts such technical solution:
[0007] A method for predicting defective products in a digital twin workshop comprises the following steps:
[0008] S1: Build a digital twin model of the physical workshop, and bind and synchronize the digital twin model with the information of the physical workshop in real time;
[0009] S2: Real-time collection of production data from the physical workshop. The production data obtained under different production conditions is processed using corresponding preset processing methods to obtain the correct data set;
[0010] In step S2, the collected production data includes the finished fabric required for the fabric RQFF , calculated value of required warp yarn RBL , total output TP, number of gray fabrics RGF , Shrinkage SA , Fabric allowanceFA、 Total length of yarn received RCBL , number of meridians WARP、 Number of wefts WEFT、 Warp per inch EPI、 Picks per inch PPI and defect amount RE , a total of 12 feature data;
[0011] In step S2, for the production situation of yarn breakage and material change: the finished fabric required by the fabric is interpolated by random matrix method. RQFF、 Gray fabric number RGF、 Fabric allowance FA、 Calculation of required warp yarns RBL、 Shrinkage SA and defect amount RE The high-frequency random interpolation method is used to process the longitude number. WARP、 Number of wefts WEFT、 Warp per inch EPI、 Picks per inch PPI Processing, using the nearest principle to fill the total output TP and the total length of yarn received RCBL to process;
[0012] For production stoppages, the production data is processed using the longitudinal splicing method;
[0013] S3: construct a prediction model based on machine learning, train the prediction model by taking the feature data in the training data set as input and the number of defective products as output, and output the final prediction result;
[0014] S4: Build a visualization platform in combination with the digital twin model to present the production situation, real-time data and prediction results of the workshop on a visualization platform.
[0015] Preferably, in step S1, simulation software is used to establish a three-dimensional model and a digital twin model of a textile workshop. The digital twin model is used to monitor the overall condition of the textile workshop, equipment operation, shift efficiency, and order production parameter data in real time; the production status of the workshop is understood and corresponding adjustments are made through real-time data updates.
[0016] Preferably, step S2-1 is further included between step S2 and step S3: converting the numerical features and string features in the processed data set into categorical features, and performing encoding processing using one-hot encoding.
[0017] Preferably, among the 12 feature data collected, the correlation between feature data 1-feature data 7 and feature data 8-feature data 11 is negative, and decision trees are established using feature data 1-feature data 7 and feature data 8-feature data 11 respectively. When new textile data are used to predict defects, the feature data are input into two decision tree models to obtain predicted values respectively, and the predicted values obtained by the two decision trees are averaged as the final prediction result.
[0018] Preferably, the specific operations in step S4 are as follows: create a 3D scene in which 28 textile machines and 2 inspectors are placed, place a camera at a position where the entire 3D scene can be observed, and the lighting settings in the 3D scene include 1 spotlight and 3 parallel lights;
[0019] The 3D scene, camera and light are rendered on the screen through the browser, and streamer animation is added to make the scene more realistic. The data of the physical workshop is transmitted to the digital workshop and divided into four categories according to the data type: the overall situation of the workshop, the comparison of shift efficiency, the change of defective product output in the past 7 days, and the number of defective product output predicted by machine learning based on the parameters of the order. Different charts are drawn according to the characteristics of each data. All these data can be viewed in real time through the visualization function of the digital workshop.
[0020] A defect prediction system comprises a memory and a processor, wherein a computer program is stored in the memory, and the processor executes the computer program to implement the steps of any one of the above methods.
[0021] A computer-readable storage medium stores a computer program, wherein the computer program, when executed by a processor, implements the steps of any of the methods described above.
[0022] A computer product comprises a computer program, wherein when the computer program is executed by a processor, the method described in any one of the above is implemented.
[0023] By adopting the aforementioned design scheme, the beneficial effects of the present invention are as follows: the present invention adopts a combination of digital twin technology and decision tree algorithm in machine learning to predict the number of defective products in a textile workshop, builds a virtual model based on the physical textile workshop, obtains historical data and real-time data of the workshop, pre-processes the acquired data, and predicts the number of defective products through the decision tree algorithm in machine learning. In this way, potential quality problems and failures can be predicted, and maintenance and repair can be carried out in advance to avoid the generation of defective products; it can help enterprises better understand the production process and product quality of textile workshops, discover problems in time and take corresponding measures, improve workshop production efficiency and product quality, and reduce production costs and risks. BRIEF DESCRIPTION OF THE DRAWINGS
[0024] Figure 1 is a flow chart of the method of the present invention;
[0025] Figure 2 A method for constructing a digital twin model of the present invention;
[0026] Figure 3 It is a schematic diagram of the unique hot encoding of the present invention;
[0027] Figure 4 It is a schematic diagram of the prediction model structure of the present invention;
[0028] Figure 5 Schematic diagram of the digital twin platform of the present invention;
[0029] Figure 6 Schematic diagram of a single device of the digital twin platform of the present invention. DETAILED DESCRIPTION
[0030] In order to make the purpose, technical scheme and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings. Obviously, the described embodiments are only partial embodiments of the present invention, rather than all 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.
[0031] A digital twin workshop defect prediction method, such as Figure 1 As shown, the following steps are included:
[0032] S1: Building a digital twin of the physical workshop ( Digital Twin,DT ) model, bind the digital twin model with the information of the physical workshop and synchronize it in real time; the construction of the digital twin model is as follows Figure 2 shown.
[0033] In this embodiment, simulation software is used to establish a three-dimensional model and a digital twin model of a textile workshop to realize real-time monitoring of the workshop production situation. Through the digital twin model, the overall condition of the textile workshop, equipment operation status, shift efficiency, order production parameters and other data can be monitored in real time; through real-time data updates, the production status of the workshop can be quickly understood and corresponding adjustments can be made.
[0034] S2: Real-time collection of production data from the physical workshop. The production data obtained under different production conditions is processed using corresponding preset processing methods to obtain the correct data set;
[0035] In textile production, existing automation and sensor equipment are used to collect data.
[0036] In this embodiment, the collected production data includes the finished fabric required for the fabric ( RQFF), calculated value of required warp yarn ( RBL ), total production (TP), number of grey fabrics ( RGF )、Shrinkage( SA )、Defect Amount( RE )、Fabric excess( FA)、 Number of warps ( WARP)、 Number of latitudes ( WEFT)、 Warp per inch ( EPI)、 Picks per inch ( PPI ) and the total length of yarn received ( RCBL) , a total of 12 feature data.
[0037] According to the process mechanism, among the 12 characteristic data collected, RQFF, RGF, FA, RBL, SA These five features are strongly correlated, which is helpful for subsequent data processing. The following feature correlation formula is established:
[0038] ;
[0039] ;
[0040] During the data collection process, it was found that due to the complexity of the production environment, problems such as material change, yarn breakage and shutdown were often encountered. These three problems would affect the collection of production data, resulting in missing, repeated, erroneous or discontinuous data. Therefore, the collected production data needed to be processed according to these three problems.
[0041] Among them, material change refers to the situation in textile production where the machine stops due to the need to change yarn and other materials. This situation will affect the production progress and data collection, and needs to be handled in time to reduce downtime;
[0042] Yarn breakage refers to the phenomenon that yarn breaks during the production process due to factors such as yarn quality and tension;
[0043] Downtime refers to the situation where the production line is in a non-productive state due to various reasons such as equipment failure, human operating errors, and process problems.
[0044] In response to the above problems, the specific production data processing method is as follows:
[0045] Processing of production data for yarn breakage and material change: In this embodiment, the random matrix interpolation method is used to target 5 strongly correlated features among the 12 features. RQFF, RGF, FA, RBL, SA and 1 output feature RE. Yarn breakage will cause the collected production data to change from a numerical value to a null value, but the time data is continuous. Material change often requires downtime, the time data will be discontinuous, and the values of the production data before and after will be slightly different, but there will be no null values. It can be seen that material change and yarn breakage will cause data loss for a whole period of time. If 5 of the 12 strongly correlated features and output features are missing, the random matrix interpolation method and the above feature association are combined for preliminary complementation, and the interpolation polynomial for:
[0046] ;
[0047] in, For the original textile data RQFF, RGF, FA, RBL, SA The subsample data of these five features,
[0048] for The corresponding RE , is a 1×5 matrix,
[0049] The given data points are ,
[0050] Constructing the Difference Quotient ,in det Represents the matrix determinant.
[0051] In order to solve the missing values, 3000 sub-samples without missing values are randomly selected from the 22000 original textile data. The number of sub-samples selected each time is not repeatable. middle n There are 3000 rows in total, each with 5 feature data, so we can get an interpolation polynomial P n (x) Therefore, when finding missing values x_missing When the estimated value of P n (x) , solve the equation P n ( x_missing )= y_missing, in y_missing The missing data points correspond to RE The final value can be obtained and put into the data to interpolate the missing values. This can reduce the error in the interpolation process and improve the reliability of the results.
[0052] High frequency random interpolation method. For the other 4 features out of 12 features WARP, WEFT, EPI, PPI, the frequency of change in the production process is low, and there are usually a lot of repeated data. Therefore, in the case of missing values, it is suitable to use the high-frequency random interpolation method, arrange the remaining data in order from most to least, and take the first three data with the highest frequency: Features WARP The most common first three digits are 40, 50 and 30; Features WEFT The most common first three digits are 40, 50 and 30; Features EPI The most common first three digits are 110, 130 and 120; Features PPI The most common first three digits are 90, 70 and 80. Then one of the characteristic data is randomly selected for interpolation. This method takes into account the distribution characteristics of the original data, retains the overall characteristics of the data to a certain extent, and has a certain degree of randomness.
[0053] The nearest principle fills in the gap. Targeting 2 of the 12 features TP and RCBL The nearest sample data is selected to fill in the missing value. If there are adjacent sample data before and after, the previous data is given priority. This filling method is relatively simple and intuitive, and retains the overall trend and characteristics of the original data. Such data is more complete, ensuring the accuracy and comprehensiveness of data collection.
[0054] Use the vertical splicing method to process the production data of shutdown:
[0055] In the actual production process, there are occasional downtime due to unexpected factors such as power maintenance, mechanical failure, material jamming, stalling, yarn entanglement, etc., which cause a series of problems such as duplication, error, and discontinuity in the actual collected data. In view of this situation, we consider adding a data collection link. Generally, the whole day is divided into several shifts, and each shift is assigned a dedicated staff to monitor production. The production process can be monitored through digital twin technology, and the real-time data returned by the workshop is collected. We get a matrix for each group of collected data, for example A The matrix is m×n , B The matrix is k×n , the matrix is represented as follows:
[0056] ;
[0057] ;
[0058] in, m and k Indicates how many sets of data were collected. n =12, is the data in row m and column 12, The data is in row n and column 12.
[0059] In order to remove the missing data in these textile data and keep the complete data for splicing, the matrix B Add each row to the matrix A Below, a new (m + k)×n The matrix C , which can be expressed as:
[0060]
[0061] The matrices obtained from the data collected by other groups are spliced in the same way to obtain a new and complete data set. This method can effectively avoid inaccurate or missing data due to manual monitoring, and greatly improves the accuracy of the collected data.
[0062] The step S2-1 is also included between step S2 and step S3: converting the numerical features and string features in the processed data set into categorical features, and performing encoding processing using one-hot encoding.
[0063] When processing the collected data, the four feature values such as the number of warp and weft yarns are uniformly converted from numerical and string features to categorical features. Categorical features refer to features that represent a set of discrete categories or labels, without the concept of order or size. Unique hot encoding is then used for processing.
[0064] One-hot encoding ( One - Hot Encoding ) is also called one-bit effective encoding, which is done by using N Bit Status Register Pair N Each state has its own register bit, and only one of them is valid at any time. Directly representing the contained features with numerical values may cause the model to mistakenly believe that there is a certain size or order relationship between these values, which may cause the relationship learned by the model to be inaccurate. If one-hot encoding is used, each value will correspond to a dimension, the value of the corresponding value dimension is 1, and the values of other dimensions are 0. The encoding is as follows Figure 3 shown.
[0065] In this way, the features are not misled, and the model will not mistakenly believe that there is a size or order relationship between these feature values. Each feature value has a unique representation, which helps ensure that the model does not introduce wrong information when processing the data.
[0066] S3: Build a prediction model based on machine learning, use the feature data in the training data set as input and the number of defective products as output to train the prediction model and output the final prediction result.
[0067] The construction process of the prediction model in this embodiment is as follows:
[0068] A decision tree is a basic classification and regression method. It consists of nodes and directed edges. When using a decision tree for classification or regression tasks, it starts from the root node, tests a certain feature of the sample, and assigns the sample to its child nodes based on the test results; each child node corresponds to a value of the feature, and the samples are tested and assigned recursively in this way until a leaf node is reached.
[0069] like Figure 4 As shown, among the 12 collected feature data, the correlation between feature data 1 to feature data 7 and feature data 8 to feature data 11 is negative.
[0070] In this embodiment, feature data 1 to feature data 7 are RQFF, RBL, TP, RGF, FA, RCBL, SA ; Feature data 8-feature data 11 is WARP, WEFT, EPI, PPI This feature selection is limited to this patent. The decision tree used in other places needs to be set according to production conditions.
[0071] Therefore, by separating them into two decision trees, the optimal decision tree can be established independently for the fabric features, making full use of the information of each fabric feature, taking into account the interaction between fabric features with strong correlation, and reducing the computational complexity of a decision tree. When new textile data is used to predict defects, it is input into the two decision tree models to obtain the predicted values respectively, and the predicted values obtained by the two decision trees are averaged as the final prediction result.
[0072] Assumptions X and Y The input (11 features such as fabric required, total output, etc.) and output (number of defects) are given as data sets. A regression tree corresponds to a partition of the fabric input space and the output value on the partitioned unit.
[0073] When the division of the fabric input space is determined, it can be used R 2 coefficient To represent the prediction error of the decision tree for the training data, R 2 The optimal output value of each unit is solved based on the criterion that the coefficient tends to 1. R 2 We can see from the formula ESS The smaller, R 2 The bigger.
[0074] ESS is the residual sum of squares: ,
[0075] in, is the true value, represents the predicted value on the prediction curve,
[0076] However, there is no prediction curve that can be expressed in decision tree regression, so the regional mean approximation is used. The smaller the residual sum of squares, the stronger the generalization ability of the fitting curve is and the higher the fitting accuracy of the fitting curve is.
[0077] CART During regression, a heuristic method is used to divide the fabric input space and select the first i variables and its value j , as the segmentation variable and segmentation point, and define two regions:
[0078] ;
[0079] ;
[0080] Then find the optimal split variable i and the optimal split point j ;
[0081] Solution: Input variables for fixed fabric i The optimal split point can be found j .
[0082] in For R 1 The output value in the (i,j) region, For R 2 The output value in the (i,j) region, is the characteristic variable in this region.
[0083] Use the selected optimal split variable i and the optimal split point j Divide the area and determine the corresponding output value:
[0084] ;
[0085] ;
[0086] Traverse all the input variables of the fabric and find the optimal segmentation variable i ,constitute . The fabric input space is divided into two regions in turn. Then the above division process is repeated for each region until the stopping condition is met, thus generating a decision tree.
[0087] Assume that the fabric input space has been divided into N Units , and in each unit There is a fixed output value on , so the decision tree model can be expressed as:
[0088] ;
[0089] in, I is an indicator function, indicating a data x Is it R n , if it belongs to, the value is 1, if it does not belong to, the value is 0; is the feature space region represented by the nth leaf node.
[0090] The same is true for another regression tree model. After building a complete decision tree model, use the validation set to evaluate its performance. When the decision tree is divided too finely, it will overfit the data, so it is necessary to solve it through pruning to reduce the complexity of the decision tree to avoid overfitting. Pruning includes post-pruning, which is a top-down pruning method that uses the test set and validation set to determine whether the merging of these leaf nodes can reduce the test error and improve the model performance. If so, merge them.
[0091] Examples:
[0092] Table 1 shows some characteristics and data. x 1, x 2 and x 3, x 4 Two regression trees were built respectively.
[0093] Table 1 Part of the training data table
[0094]
[0095] The first regression tree follows x 1 Partition variables, where and For example, take j =31300.
[0096] at this time , , according to the calculation formula of mean and residual sum of squares, we can get Table 2:
[0097] Table 2 x 1 The mean and ESS value
[0098]
[0099] according to x 2 Divide the variables and take j =6At this time , , according to the calculation formula of mean and residual sum of squares, we get Table 3:
[0100] Table 3 x 2 The mean and ESS value
[0101]
[0102] Compare All ESS Value discovery, x 2 Characteristic j =8.1 ESS The smallest, at this time R 2 is closer to 1, so the first optimal segmentation variable is i = x , the optimal split point is j =8.1. , . Second division selection x 1 Minimize the feature ESS Find the optimal splitting variable and the optimal splitting point, set the maximum depth, number of leaf nodes, number of samples per node, etc., and repeat the above steps until a complete regression tree is obtained.
[0103] The second regression tree follows x 3 The results of the partitioning variables are shown in Table 4:
[0104] Table 4 x 3 The mean and ESS value
[0105]
[0106] according to x 4 The results of the partitioning variables are shown in Table 5:
[0107] Table 5 x 4 The mean and ESS value
[0108]
[0109] Obviously, press x 3 When divided into 0,1, ESS Minimum,R 2 is closer to 1. Therefore, the first optimal segmentation variable is i = x , the optimal split point is j =0.5. Second division selection x 4 Find the optimal split point based on the features. Set the maximum depth, number of leaf nodes, number of samples per node, etc. Repeat the above steps until a complete regression tree is obtained.
[0110] When new fabric sample data is input for defect prediction, the corresponding regression trees are trained according to the corresponding features, and the average of the obtained prediction values is the final defect prediction result.
[0111] S4: Build a visualization platform in combination with the digital twin model to present the production situation, real-time data and prediction results of the textile workshop on the visualization platform.
[0112] In order to more intuitively monitor the production situation of the textile workshop, the digital twin technology and machine learning algorithm are combined to build a digital twin three-dimensional visualization platform. First, create a 3D scene with 28 textile machinery and 2 inspection personnel, and place the camera at a position where the entire 3D scene can be observed. In this embodiment, the camera is placed at (-6.94, 361.56, -527.18) of the entire 3D scene to observe the entire model. The lighting settings in the 3D scene include a spotlight and three parallel lights. In this embodiment, the spotlight is placed at (500, 202.5, 1000) of the 3D scene, and the three parallel lights are placed at (-50, -150, 0), (80, 100, 50), (-200, 100, -200) of the 3D scene. Render the scene, camera, and light to the screen through the browser, add streamer animation operations to make the scene more realistic, and control the texture of the text (texture) in x and y Number of repetitions on the axis count :like count >20, texture.x and texture.y The value of is reset to 0; if count ≤20, texture.x The value of is increased by 0.02, texture.y The value of increases by 0.03.
[0113] The data from the physical workshop is transmitted to the digital workshop through sensors and other equipment, and is divided into three categories according to the data type: the overall situation of the workshop, shift efficiency comparison, and the change in defective product output in the past seven days. According to the characteristics of each data, the data of the overall situation of the workshop is plotted as a fan chart, of which 93.02% of the equipment is in operation and 4.65% of the equipment is abnormal; the shift efficiency is plotted in a tabular form, including the unit, personnel and efficiency, and the workshop efficiency is above 80%; the number of defective products produced in the past seven days is plotted as a bar chart, and the minimum output in a week is about 25,000. The number of defective products predicted by machine learning based on the parameters of the order is plotted as a line chart. The currently predicted number of defects per day does not exceed 300. All these data can be viewed in real time through the visualization function of the digital workshop. Textile digital twin platform such as Figure 5 As shown, a single device is displayed as Figure 6 shown.
[0114] The textile digital twin platform provides comprehensive data support and visual information for factory management, which can better monitor production conditions, improve production efficiency, and reduce production risks. Through digital means, it helps enterprises achieve intelligent production.
[0115] This embodiment also provides a defect prediction system for implementing the above method.
[0116] A defect prediction system includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of any one of the above methods.
[0117] This embodiment also provides a storage medium storing a computer program for implementing the above method.
[0118] A computer-readable storage medium stores a computer program, which implements the steps of any of the above methods when executed by a processor.
[0119] In this embodiment, a computer product for implementing the above method is also provided.
[0120] A computer product comprises a computer program, and when the computer program is executed by a processor, the method described in any one of the above items is implemented.
[0121] In summary, the present invention adopts a combination of digital twin technology and decision tree algorithm in machine learning to predict the number of defective products in textile workshops. A virtual model is built based on the physical textile workshop to obtain historical data and real-time data of the workshop, and the acquired data is preprocessed. The number of defective products is predicted by the decision tree algorithm in machine learning, which can predict potential quality problems and failures, perform maintenance and repair in advance, and avoid the generation of defective products. It can help enterprises better understand the production process and product quality of textile workshops, find problems in time and take corresponding measures, improve workshop production efficiency and product quality, and reduce production costs and risks.
[0122] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.
Claims
1. A digital twin workshop defect prediction method, It is characterized in that The following steps are involved: S1: Build a digital twin model of the physical workshop, and bind and synchronize the digital twin model with the information of the physical workshop in real time; S2: Real-time collection of production data from the physical workshop. The production data obtained under different production conditions is processed using corresponding preset processing methods to obtain the correct data set; In step S2, the collected production data includes the finished fabric required for the fabric QUR , calculated value of required warp yarn RBL , total output TP, number of gray fabrics RGF , Shrinkage SA , Fabric allowance FA, Total length of yarn received RCBL , number of meridians WARP, Number of wefts WEFT, Warp per inch EPI, Picks per inch PPI and defect amount RE , a total of 12 feature data; In step S2, for the production situation of yarn breakage and material change: the finished fabric required by the fabric is interpolated by random matrix method. RQFF, Gray fabric number RGF, Fabric allowance FA, Calculation of required warp yarns RBL, Shrinkage SA and defect amount RE The high-frequency random interpolation method is used to process the longitude number. WARP, Number of wefts WEFT, Warp per inch EPI, Picks per inch PPI Processing, using the nearest principle to fill the total output TP and the total length of yarn received RCBL to process; For production stoppages, the production data is processed using the longitudinal splicing method; S2-1: Convert the numerical features and string features in the processed data set into categorical features and use one-hot encoding for encoding; S3: construct a prediction model based on machine learning, train the prediction model by taking the feature data in the training data set as input and the number of defective products as output, and output the final prediction result; Among the 12 feature data collected, the correlation between feature data 1-feature data 7 and feature data 8-feature data 11 is negative. Feature data 1-feature data 7 and feature data 8-feature data 11 are used to establish decision trees respectively. When new textile data is used to predict defective products, the feature data are input into two decision tree models to obtain prediction values respectively. The prediction values obtained by the two decision trees are averaged as the final prediction result. S4: Build a visualization platform in combination with the digital twin model to present the production situation, real-time data and prediction results of the workshop on a visualization platform.
2. A digital twin workshop defect prediction method as claimed in claim 1, Features: In step S1, simulation software is used to establish a three-dimensional model and a digital twin model of the physical workshop. The digital twin model is used to monitor the overall condition of the textile workshop, equipment operation, shift efficiency, and order production parameter data in real time. The data is updated in real time to understand the production status of the workshop and make corresponding adjustments.
3. A digital twin workshop defect prediction method as claimed in claim 1, Features: The specific operations in step S4 are as follows: create a 3D scene with 28 textile machines and 2 inspectors, place the camera at a position where the entire 3D scene can be observed, and the lighting settings in the 3D scene include 1 spotlight and 3 parallel lights; The 3D scene, camera and light are rendered onto the screen through the browser, and streamer animation is added to make the scene more realistic. The data from the physical workshop is transmitted to the digital workshop and divided into four categories according to the data type: the overall situation of the workshop, shift efficiency comparison, the change in defective output in the past 7 days, and the number of defective output predicted by machine learning based on the order parameters. Different charts are drawn according to the characteristics of each data. All this data can be viewed in real time through the visualization function of the digital workshop.
4. A defect prediction system, comprising a memory and a processor, wherein the memory stores a computer program, Features: The processor executes the computer program to implement the steps of the method according to any one of claims 1 to 3.
5. A computer-readable storage medium having a computer program stored thereon, Features: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 3 are implemented.
6. A computer product comprising a computer program, Features: When the computer program is executed by a processor, the method according to any one of claims 1 to 3 is implemented.
