An assembly process monitoring and precision prediction method and system based on digital twinning

Abstract

The application discloses an assembly workshop monitoring and assembly precision prediction method based on digital twinning, and steps include: building a virtual assembly workshop; based on a nested fuzzy finite state machine model, the assembly process is subdivided into multiple states, and the switching conditions between states are a set; comprehensively utilize the analytic hierarchy process and the entropy weight method to determine the monitoring items; based on the measured data, a particle swarm-SG filtering data processing method is proposed to improve the reliability of the data; an assembly precision prediction model is established, the parameter optimization process of the support vector machine model is optimized, and based on the Monte Carlo simulation, the assembly precision is predicted; an assembly process optimization knowledge base is constructed, and when the prediction accuracy is not up to standard, the system automatically throws out the corresponding solution. Through the digital twinning technology and the machine learning model, the assembly workshop monitoring and assembly precision prediction are realized, not only guaranteeing the smooth completion of the assembly process, but also improving the assembly efficiency and the intelligent level.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of assembly workshop monitoring, in particular to a marine diesel engine cylinder head assembly workshop monitoring based on digital twinning. BACKGROUND

[0002] Assembly is to combine scattered parts and components into a product according to specified accuracy standards and technical requirements, using certain assembly process flow and various necessary assembly methods, which is one of the most critical links in the product life cycle. If problems occur during assembly, it will prolong the assembly time, reduce the production efficiency, and increase the production cost of the product, which will directly affect the production cycle and product quality of the entire product. For complex assemblies such as marine diesel engine cylinder heads, the assembly accuracy of each link is very strict, but because of the complexity of the constituent parts, it is difficult to detect problems during assembly, and the production quality and assembly accuracy are difficult to predict, which restricts the smooth completion of high-precision assembly work. Therefore, assembly process monitoring and assembly accuracy prediction have become the main content of the current research in the field of assembly of complex assemblies such as marine diesel engine cylinder heads.

[0003] Traditional marine diesel engine cylinder head assembly workshop has low production efficiency, opaque production process, scattered production data, and cannot timely detect problems, making it difficult to integrate management. This not only wastes time, but also increases the possibility of errors and repetitive work. In order to timely detect the production status of the cylinder head assembly workshop, realize centralized and unified management of workshop data, and make the production data of the workshop traceable, it is necessary to carry out research on cylinder head assembly workshop monitoring, establish overall control of workshop behavior, and ensure assembly accuracy. In assembly accuracy prediction, machine learning can analyze a large amount of historical assembly data, identify key factors affecting accuracy, and establish an accurate prediction model. This technology not only improves the accuracy of prediction, but also adapts to changes in workshop environment and assembly process by updating the model in real time, thereby providing more reliable support for the assembly process and improving overall product quality and production efficiency. When the assembly process is abnormal or the assembly accuracy is not up to standard, the constructed assembly process optimization knowledge base automatically pushes the solution strategy to assist workshop personnel in quickly handling abnormal situations, further improving assembly efficiency and reducing production cost. Therefore, based on digital twinning technology and machine learning algorithms, the monitoring of the assembly process and the prediction of assembly accuracy in the workshop can effectively solve the problems faced by the current assembly workshop.

[0004] Prior art CN112947294A - A kind of based on digital twin's automobile assembly workshop monitoring simulation system, the patent proposes a kind of based on digital twin's automobile assembly workshop monitoring simulation system, including assembly workshop static model establishment module, two-way passage data acquisition module, assembly workshop dynamic model establishment module, production operation synchronous correction module, assembly workshop entity control module.In the assembly workshop static model establishment module, the static model of preliminary establishment assembly workshop entity, utilizes two-way passage data acquisition module and passes through two-way communication channel to collect the real-time operation data of static model subitem, establishes the dynamic model of assembly workshop in combination with assembly workshop static model and the real-time operation data of static model subitem collected, synchronous correction module is used to monitor the dynamic model of assembly workshop, calculates data error and promptly corrects, synchronizes assembly workshop operation process and exports correction synchronization signal, finally in assembly workshop control module, the entity of assembly workshop is controlled by correction synchronization signal.But this method lacks effective processing of measured data, it is difficult to ensure the accuracy of data.

[0005] Prior art CN108171805A - An assembly precision prediction method, according to the corresponding assembly of each group of mating surface in the target model, respectively generate geometric surface conforming to the preset part design tolerance, carry out multiple simulation calculations on the generated geometric surface, obtain the relative position of each group of mating surface in the target model, obtain the assembly precision of at least two parts when assembling according to the results of multiple simulation calculations.But this method only considers the part surface topography and deformation influence, without considering other random errors in the assembly process, and the calculation process is complex, it is difficult to quickly predict the assembly precision.

[0006] The existing patent technology has the following problems:

[0007] (1) The traditional assembly workshop monitoring is mainly based on a simple model, which is difficult to comprehensively and meticulously describe the complex assembly process.

[0008] (2) The traditional scheme lacks effective processing of measured data, and the noise in the data affects the reliability of the assembly process data, which restricts the accuracy of the assembly precision prediction and the credibility of the monitoring data.

[0009] (3) The traditional assembly precision prediction method is mostly based on rigid body assumption or one-time deformation theory, ignoring the influence of multi-dimensional coupling error existing in the assembly process, the assembly precision prediction model is far from the real assembly process, which restricts the accuracy of the assembly precision prediction.

[0010] (4) The traditional workshop monitoring and precision prediction method only issues an alarm when an abnormal situation occurs, without providing a reference solution, which is not conducive to timely handling of abnormal situations. SUMMARY

[0011] Invention purposes: The application provides a complex assembly process monitoring and precision prediction method based on digital twinning, realizes real-time monitoring of the assembly process and accurate perception of the assembly precision, automatically pushes the solution strategy, and improves the assembly process; reduce production costs and improve the first assembly success rate.

[0012] Technical scheme: A kind of assembly process monitoring and precision prediction method based on digital twinning, comprising the following steps:

[0013] (1) according to the real scene of assembly workshop, build digital twinning assembly workshop;

[0014] (2) according to the requirement of assembly process file, establish assembly behavior model;The assembly behavior model adopts nested fuzzy finite state machine model;

[0015] (3) the project data needing monitoring in digital twinning assembly workshop is determined by using analytic hierarchy process-entropy weight method;

[0016] (4) the project data needing monitoring obtained in step (3) is optimized by using particle swarm-SG data processing method;The particle swarm-SG data processing method uses particle swarm algorithm to optimize the window size and polynomial order of SG filtering algorithm;

[0017] (5) construct assembly precision prediction model, contain a variety of error coupling superposition;

[0018] Improved support vector machine model is used for assembly deformation prediction;The improved support vector machine model uses hierarchical coarse and fine granularity grid search algorithm to optimize regularization parameter C and gamma parameter g;

[0019] When precision prediction, sample in error distribution interval, simulate random error in assembly process;

[0020] (6) the assembly process is simulated by using Monte Carlo method, enough simulation rounds are set, the assembly deformation prediction value in step (5) and the sampling value of random error are substituted into the assembly precision prediction model, and the assembly precision prediction value is obtained;

[0021] (7) for the precision that does not meet the requirements or abnormal situation appears in assembly workshop, establish assembly process optimization knowledge base in monitoring system.

[0022] Further, the assembly behavior model adopts nested fuzzy finite state machine model, the finite state machine model of subcomponent assembly process in equipment process is as a state of the finite state machine model of the equipment process, and the state switching condition is set as a set, when one condition in the state switching condition set is met, state switching can be realized.

[0023] Further, the analytic hierarchy process-entropy weight method first classifies the workshop data according to information categories, constructs a hierarchical structure model, and only retains the criterion layer; secondly, a judgment matrix is constructed, the analytic hierarchy process is used to solve the importance of each type of information in the criterion layer, sorting is performed, and unimportant information categories are removed; then, for various information under the important information categories retained, the entropy weight method is used to calculate the weight, and the final monitored project data is selected.

[0024] Further, the particle swarm-SG data processing method has the following specific steps:

[0025] The permission fluctuation range of data is set, and abnormal outliers are removed; after removing the abnormal values, the interpolation method is used to complete the missing values; the particle swarm algorithm is used to construct a model and design a fitness function, the window size and the polynomial order are used as the optimization dimensions, and the parameter combination that optimizes the performance of the SG algorithm is found through iterative search; after completing the parameter optimization, the best window size and the polynomial order are input into the SG algorithm, the SG algorithm fits the data according to the set polynomial order in the window range, and outputs the fitting value of the window center point as the processed data value; through the movement of the window, the entire signal data is processed step by step.

[0026] Further, the fitness function comprehensively considers the mean square error, smoothness and trend fitting degree of the filtered signal.

[0027] Further, the assembly precision prediction model:

[0028]

[0029] In the formula, ε is the assembly precision, δ 尺寸 is the size error, δ 变形 is the deformation error generated by the process parameters, δ 重复定位 is the repeated positioning error in the assembly process, δ 工装制造 is the tooling manufacturing error, δ 检测 is the detection error, represents the error coupling superposition. In step (5), the assembly deformation amount is δ 变形 , and the random error includes δ 重复定位 , δ 工装制造 , and δ 检测 .

[0030] Further, the improved support vector machine model uses a hierarchical coarse-fine grid search algorithm to optimize the regularization parameter C and the gamma parameter g, and the specific steps include:

[0031] 1) The search range of the parameters C and g is determined according to the maximum standard deviation of the input data, as shown in the following formula:

[0032]

[0033] where a is the base number for incrementing the parameter on a logarithmic scale, σ max is the maximum standard deviation of the input data, b is the step size, indicating the logarithmic value of each increment, determining the granularity of the parameter search, and l is the number of increments, lb is the increment of adjusting the search range;

[0034] 2) The parameter optimization process is divided into two stages of coarse-grained search and fine-grained search, and a set G of parameter combinations is established:

[0035] G={(C1,g1),…,(C1,g n );(C2,g1),…,(C2,g n );…,(C m ,g n )}

[0036] where (C m ,g n ) is a possible parameter combination, m and n are positive integers, in the coarse-grained search stage, a random sampling rate α is set, and a parameter combination is randomly selected to find the optimal parameter combination, as follows,

[0037] G′={G[i]|i∈randperm(N,k)}

[0038] where k=max(1,α·N) is the number of samples, N=m×n is the total number of combinations, G′ is the parameter combination after random sampling, and randperm(N,k) is a function of randomly selecting k indexes from 1 to N;

[0039] The fine-grained search interval is set around the optimal parameter combination obtained by coarse-grained search, and the fine-grained search interval is further divided into n search intervals for fine-grained search, and the values of C and g in the fine-grained search range are as follows:

[0040]

[0041] where C i and g i are the parameters of fine-grained search, C * and g * are the optimal C and g parameters obtained by coarse-grained search, δ is the parameter reduction ratio, ξ is the parameter growth ratio, n is the subdivision fraction, C * ×ξ-C * ×δ represents the length of the subdivision interval of parameter C, and g * ×ξ-g * ×δ represents the length of the subdivision interval of parameter g.

[0042] Further, the random error in the simulation assembly process is determined according to error statistical data, a distribution type and a distribution interval of the error are determined, and when precision prediction is performed, sampling is performed in the error distribution interval to simulate the random error in the assembly process.

[0043] Further, the establishment of the assembly process optimization knowledge base in the monitoring system numbers possible abnormal conditions in different states of different stations and stores corresponding solving strategies, when the monitoring system detects the corresponding abnormality, the solving strategy is automatically queried according to the abnormality number for reference of the user; the user can edit the knowledge base.

[0044] An assembly process monitoring and precision prediction system based on digital twinning includes a digital twinning assembly workshop, an assembly behavior model module, a data processing module, an assembly precision prediction model module, an assembly simulation module, and a knowledge base module.

[0045] The digital twinning assembly workshop is a digital twinning assembly workshop built according to a real scene of an assembly workshop.

[0046] The assembly behavior model module establishes an assembly behavior model according to requirements of an assembly process file; the assembly behavior model adopts a nested fuzzy finite state machine model.

[0047] The data processing module determines project data that needs to be monitored in the digital twinning assembly workshop by using an analytic hierarchy process-entropy weight method; the obtained project data that needs to be monitored is optimized by using a particle swarm-SG data processing method; the particle swarm-SG data processing method optimizes a window size and a polynomial order of an SG filtering algorithm by using a particle swarm algorithm.

[0048] The assembly precision prediction model module constructs an assembly precision prediction model that contains a variety of error coupling superposition; an improved support vector machine model is used for assembly deformation amount prediction; the improved support vector machine model optimizes a regularization parameter C and a gamma parameter g by using a hierarchical coarse-fine grid search algorithm; when precision prediction is performed, sampling is performed in an error distribution interval to simulate the random error in the assembly process.

[0049] The assembly simulation module simulates the assembly process by using a Monte Carlo method, sets a sufficient number of simulation rounds, substitutes assembly deformation prediction values and sampling values of the random error into the assembly precision prediction model, and obtains assembly precision prediction values.

[0050] The knowledge base module establishes an assembly process optimization knowledge base in the monitoring system for precision that does not meet requirements or abnormal conditions of the assembly workshop.

[0051] Beneficial effects:

[0052] 1. The nested fuzzy finite state machine model proposed in this invention can describe complex processes more comprehensively and in detail than traditional models, and is suitable for monitoring complex processes.

[0053] 2. By combining the Analytic Hierarchy Process (AHP) and the Entropy Weight Method, monitoring items are selected to avoid the problems of excessive subjectivity and missing or redundant monitoring items caused by traditional manual selection.

[0054] 3. Data processing can significantly improve the authenticity and reliability of data, further ensuring and enhancing the accuracy of monitoring and precision prediction.

[0055] 4. The support vector machine model has been improved to fully consider data features during training, thereby accelerating training speed and improving prediction accuracy.

[0056] 5. When predicting assembly accuracy, the influence of multidimensional coupling error is considered, making the established accuracy prediction model more consistent with reality. This overcomes the shortcomings of traditional accuracy prediction methods that only consider some factors and improves the accuracy of accuracy prediction.

[0057] 6. When the prediction accuracy is out of tolerance or other abnormal situations occur, the system automatically pushes out solutions based on the constructed assembly process optimization knowledge base, providing a reference for resolving the abnormalities. Attached Figure Description

[0058] Figure 1 This is a flowchart of a digital twin-based method for monitoring assembly workshops and predicting assembly accuracy.

[0059] Figure 2 A schematic diagram of the nested fuzzy finite state machine model established for this invention;

[0060] Figure 3 Select a flowchart for the monitoring project;

[0061] Figure 4 The flowchart is for the Particle Swarm Filter (SSG) algorithm.

[0062] Figure 5 This is a diagram showing the results of data processing.

[0063] Figure 6 This is a flowchart for predicting assembly accuracy. Detailed Implementation

[0064] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0065] like Figure 1 As shown, a method for monitoring assembly workshops and predicting assembly accuracy based on digital twins includes the following steps:

[0066] (1) Based on enterprise research, construct a product model library and build a virtual assembly scenario.

[0067] The product model library described in step (1) is used to store three-dimensional models of the marine diesel engine cylinder head assembly workshop, including workshop plant, equipment, production line, and assembled parts. These three-dimensional models are deployed in a virtual scene according to the layout of the real assembly workshop, and a virtual assembly workshop that highly restores the assembly scene is constructed, which is used to show the on-site state of the cylinder head assembly workshop. In addition, a friendly user interface is built, and the assembly process data is displayed in real time using charts to realize the visual display of the measured data, which facilitates users to timely perceive the assembly process data.

[0068] (2) Establishing an assembly behavior model

[0069] The assembly behavior model is established in accordance with the requirements of the assembly process file. Due to the complexity of the assembly process and the division into numerous stages such as sub-assembly and final assembly, the traditional simple model is not applicable, and therefore a nested fuzzy finite state machine model is proposed for the expression of the assembly workshop behavior. The specific steps are as follows:

[0070] (2.1) Determining the nesting level

[0071] According to the assembly process, the assembly behavior that is performed independently can be regarded as an independent state machine model. When the behavior is too complex, it can be appropriately divided and expressed using a multi-layer state machine. According to the assembly steps, each independent assembly process is divided into a finite number of assembly states.

[0072] (2.2) State machine nesting

[0073] According to the assembly process card requirements, the state machine model is nested. For example, if a component A and a part B need to be assembled in an assembly process C, the finite state machine model representing the assembly process of the component A is regarded as a state of the finite state machine model representing the assembly process C, so as to realize the nesting of the state machine model.

[0074] (2.3) Determining the state switching condition

[0075] The assembly process is continuous and uninterrupted, and the progress of the assembly process is essentially the continuous switching of multiple assembly states. In order to ensure that the switching between assembly states is orderly and accurately performed, the state switching condition needs to be set. The state switching condition of a complex assembly process is complex, and the switching condition of the traditional finite state machine model is single, which is difficult to meet the needs of a complex assembly process. Therefore, the traditional state machine model is improved, and the switching condition is set as a set. When one of the conditions in the switching condition set is met, the state switching is realized.

[0076] The application improves the traditional finite state machine model, nests multiple fuzzy state machine models, comprehensively and multi-level describes the assembling behavior, overcomes the shortcomings of poor adaptability and difficulty in describing complex assembling process of the traditional model, and lays a foundation for realizing workshop twin monitoring.

[0077] Figure 2 The model provided by the application is shown in the figure, wherein S i represents each state, S0 represents an initial state, F represents a final state, and S i，j represents the jth substate in the ith state, C m，n represents a transition condition from the state m to the state n, in order to improve the adaptability of the model to a complex production process, the application sets the state transition condition as a set, δ0 to δ k are elements in the set, and each is a transition condition, and when one of them is met, the state transition can be triggered.

[0078] As Figure 2 shown, based on the cylinder cover assembling process file, the assembling process is split, the states and the transition conditions are set according to the nested fuzzy finite state machine model provided by the application, the whole assembling process is comprehensively described through the nesting of multiple fuzzy state machines, and the basis for workshop monitoring is laid. When the state transition condition is met, the state transition is realized, the model is updated, and the virtual scene is synchronized with the real scene.

[0079] (3) Monitoring item selection

[0080] There are too much data in the assembling workshop, if all the data are collected and monitored, too many detection devices will be caused, the production cost is increased, and some of the data have little influence on the assembling process, and it is meaningless to collect these data. Therefore, it is necessary to optimize the monitoring items. The traditional monitoring items are determined by people, the subjectivity is strong, and the monitoring items are easily lost or redundant. In order to overcome the above-mentioned shortcomings, the application provides a method for determining the monitoring items by using the analytic hierarchy process-entropy weight method. The specific steps are as follows:

[0081] Firstly, the workshop data are classified according to the information categories, a hierarchical structure model is constructed, and only the criterion layer is reserved; secondly, a judgment matrix is constructed, the importance of each type of information in the criterion layer is solved by using the analytic hierarchy process, is sorted, and the unimportant information categories are removed; then, the weights of various information in the important information categories reserved are calculated by using the entropy weight method, and the final detection items are selected.

[0082] The method first screens key categories according to each information category by using the analytic hierarchy process, then calculates the entropy weight of each index under the key categories by using the entropy weight method, and finally determines the detection project, so that the serial combination mode effectively reduces the calculation amount of the model, avoids the result being too subjective, improves the efficiency on the basis of ensuring the scientific selection of the project.

[0083] As shown in Figure 3 The data of the cylinder head assembly workshop is classified according to information categories, and a hierarchical structure model is constructed, only the criterion layer is retained; secondly, a judgment matrix is constructed, the importance of each information category in the criterion layer is solved by using the analytic hierarchy process, and the unimportant information categories are removed; then, the weights of various information under the important information categories retained are calculated by using the entropy weight method, and the final detection information project is selected.

[0084] (4) Data processing

[0085] For the selected monitoring project, a data acquisition network is constructed, and data acquisition is performed. However, the data acquisition process may be affected by vibration and other factors, resulting in a large amount of noise in the measured data. In order to improve the reliability of the data, the present application proposes a particle swarm-SG data processing method, which uses the particle swarm algorithm to optimize the window size and polynomial order of the SG filtering algorithm, improves the performance of the SG filtering algorithm, and realizes data smoothing while realizing data filtering. The specific steps are as follows:

[0086] (4.1) Abnormal value elimination

[0087] Set the allowable fluctuation range of the data, and eliminate the abnormal outliers;

[0088] (4.2) Missing value completion

[0089] After eliminating the abnormal values, the missing values are completed by using the interpolation method;

[0090] (4.3) SG algorithm optimization and data processing

[0091] Firstly, an optimization model is constructed by particle swarm algorithm and a fitness function is designed, taking the window size and polynomial order as the optimization dimensions, to find the parameter combination that optimizes the performance of the SG algorithm through iterative search. The fitness function considers the mean square error, smoothness and trend fitting degree of the filtered signal to ensure the accuracy and stability of the filtering result. Then, after the parameter optimization is completed, the best window size and polynomial order are input into the SG algorithm. Within the window range, the SG algorithm fits the data according to the set polynomial order and outputs the fitting value of the window center point as the processed data value. Through the movement of the window, the entire signal data is processed step by step. The SG algorithm optimized based on PSO can significantly improve the data processing effect and achieve better smoothness and trend retention performance. The specific parameter optimization process is as follows:

[0092] 1) Design fitness function

[0093] To ensure the accuracy of the filtered data and retain the original characteristics of the collected data, improve the smoothness of the filtered signal, the present application considers three indicators, mean square error (MSE), trend error and signal smoothness, when designing the fitness function, and performs weighted synthesis to make the three indicators jointly determine the fitness function value. The fitness function is shown in formula (1):

[0094] Fitness=α·MSE+β·trend_diff+γ·std_filtered(1)

[0095] In the formula, Fitness is the fitness function value, MSE is the mean square error, trend_diff is the trend error, std_filtered is the signal smoothness, and α, β, γ are the weights of the three indicators of mean square error, trend error and signal smoothness, respectively.

[0096] The mean square error is used to measure the accuracy of the data processing result. Its calculation formula is as follows:

[0097]

[0098] In the formula, x i is the original signal, is the filtered signal, and N is the total number of samples.

[0099] The trend error trend_diff is used to evaluate the trend fitting degree, and its calculation formula is as shown in formula (3):

[0100]

[0101] In the formula, T i is the extracted trend signal, is the filtered signal, and N is the number of samples.

[0102] The standard deviation of the first-order difference of the filtered signal is used as the signal smoothness index, and the calculation formula is as formula (4):

[0103]

[0104] In the formula is the first-order difference value, is the average value of the difference value, and M is the number of difference samples.

[0105] 1) Establish an optimization model

[0106] Based on the above analysis results, the smaller the fitness function, the better the filtering effect. The optimization parameters of the application are the window size of the SG algorithm and the polynomial order. The constraint conditions of the model are set, and the optimization model is established as shown in formula (5):

[0107]

[0108] In the formula Fitness is the fitness function, w is the window size, w_min and w_max are the minimum and maximum values of its search range respectively, p is the polynomial order, p_min and p_max are the minimum and maximum values of its search range respectively, w must be an odd number to ensure that the filter has a unique center point, and L is the signal length.

[0109] The particle swarm optimization algorithm is used to solve the model, and the best parameters are assigned to the SG filtering algorithm for real-time filtering processing. After data processing, the influence of noise can be weakened, and the data reliability can be improved. Using the processed data for precision prediction can further improve the accuracy of precision prediction.

[0110] As Figure 4 shown, after outlier rejection and missing value completion, the mean square error, trend error and signal smoothness are considered comprehensively, and the optimization objective function is established by weighted synthesis. Combined with the variable range of the parameters, the optimization model is established, and the particle swarm algorithm is used to solve it. The best window size and polynomial order are obtained and assigned to the SG filtering algorithm. In the window size range, the data points are fitted according to the specified polynomial order, the filtered data is output, the window is shifted, and the process is repeated until all data processing is completed. The data processing effect is shown in Figure 5 . The processed data is finally obtained, the influence of noise is weakened, and the data reliability is improved. On the one hand, the data is displayed, and on the other hand, the system uses the processed data for assembly accuracy prediction.

[0111] (5) Assembly accuracy prediction overall model construction

[0112] The assembly process involves many aspects such as parts, equipment, assembly process and detection, so the assembly accuracy is often affected by multiple factors. The assembly parts are the objects of assembly, and their manufacturing errors and true sizes directly affect the assembly accuracy; the assembly equipment is the tool used in the assembly process, and its manufacturing errors and performance are also an important factor affecting the assembly accuracy; the force and torque in the assembly process will cause deformation of the assembly body, affecting the assembly accuracy and assembly quality, so whether the selection of assembly process parameters is reasonable will also affect the assembly accuracy; in the detection aspect, due to the errors of detection methods and tools, the measured assembly accuracy index deviates from the true value. In order to comprehensively consider the joint effects of these factors, the assembly accuracy prediction model is established as shown in equation (6):

[0113]

[0114] In the formula, ε is the assembly accuracy, δ 尺寸 is the size error, δ 变形 is the deformation error caused by process parameters, δ 重复定位 is the repeated positioning error in the assembly process, δ 工装制造 is the tool manufacturing error, δ 检测 is the detection error, indicates the coupling and superposition of errors. The assembly deformation amount in step (5) is δ 变形 , and the random error includes δ 重复定位 , δ 工装制造 , and δ 检测 .

[0115] The assembly process of a complex assembly body such as a marine diesel engine cylinder head involves the joint effects of multiple factors, and its assembly accuracy is difficult to guarantee, often requiring assembly accuracy prediction. The traditional accuracy prediction method does not consider the error sources adequately, and in order to improve the accuracy of assembly accuracy prediction, the assembly accuracy prediction process as shown in Figure 6 is proposed, and on the basis of sufficient analysis of error sources, the assembly accuracy prediction model as shown in equation (6) is constructed, which comprehensively considers the assembly deformation amount, part size error, repeated positioning error, tool manufacturing error, and measurement error.

[0116] (6) Multi-dimensional error value determination

[0117] (6.1) Assembly deformation amount prediction based on improved support vector machine

[0118] The deformation of parts in the assembly process is an important factor affecting the assembly accuracy. The deformation amount of parts is jointly affected by many factors such as material properties and assembly process parameters, and it is difficult to calculate using mathematical formulas, so the improved support vector machine model based on machine learning theory is used for assembly deformation amount prediction, and the specific improvement steps are as follows:

[0119] (6.1.1) Selection of adaptive parameter search range based on maximum standard deviation of input sample

[0120] The prediction performance of the support vector machine model depends on the selection of key parameters C and g, and the rationality of these parameters directly affects the training effect and prediction accuracy of the model. The traditional model parameter search range is specified by humans, ignoring the characteristics of the data itself, which easily produces unreasonable parameter combinations, reducing the model training speed and prediction accuracy. To fully consider the distribution characteristics of the data itself, the parameter search range of the present application is determined according to the maximum standard deviation of the input data, as shown in equation (7):

[0121]

[0122] where a is the base, used to increment the parameters on a logarithmic scale, σ max is the maximum standard deviation of the input data, b is the step size, representing the logarithmic value of each increase, determining the granularity of parameter search, l is the number of increments, lb is the increment of adjusting the search range, ensuring a wider coverage of possible values.

[0123] (6.1.2) Parameter optimization based on hierarchical coarse-fine grid search algorithm

[0124] To balance the parameter search efficiency and model performance, the present application improves the parameter optimization algorithm, using a hierarchical coarse-fine grid search algorithm for parameter optimization, dividing the parameter optimization process into two stages of coarse-grained search and fine-grained search, and establishing a set of parameter combinations as shown in equation (8):

[0125] G={(C1,g1),…,(C1,g n ),(C2,g1),…,(C2,g n ),…,(C m ,g n )}(8)

[0126] In equation (8), (C m ,g n ) is a possible parameter combination. In the coarse-grained search stage, a random sampling rate α is set, and a parameter combination is randomly selected to find the optimal parameter combination, as shown in equation (9).

[0127] G′={G[i]|i∈randperm(N,k)}(9)

[0128] where k = max(1, α·N) is the number of samples, N = m × n is the total number of combinations, α is the random sampling ratio, G' is the parameter combination after random sampling, and randperm(N, k) is a function that randomly selects k indexes from 1 to N. Through random search, the computational load can be significantly reduced, improving the efficiency of parameter optimization.

[0129] The fine granularity search range is further divided into n search ranges, and fine granularity search is performed, and the values of C and g in the fine granularity search range are shown in formula (10) and formula (11):

[0130]

[0131] In the formula, C i and g i are parameters of fine granularity search, C * and g * are optimal C and g parameters obtained by coarse granularity search, δ is a parameter reduction ratio, ξ is a parameter growth ratio, n is a subdivision fraction, C * × ξ - C * × δ represents the length of the subdivision interval of parameter C, g * × ξ - g * × δ represents the length of the subdivision interval of parameter g. By further searching around the optimal parameters obtained after coarse granularity search, a better parameter combination can be obtained.

[0132] The range of the neighborhood is determined by δ (the parameter reduction ratio) and ξ (the parameter growth ratio), and the specific value can be modified according to the prediction accuracy of the support vector machine model. The neighborhood here can be left and right intervals (δ < 1, ξ > 1), or only left interval (δ < ξ < 1), or only right interval (1 < δ < ξ). Here, the parameter search interval (the best fine granularity search interval) that makes the prediction accuracy of the support vector machine model highest can be determined through multiple tests. When this method is applied to other objects or problems, it may be different.

[0133] The present application can significantly improve the prediction performance of the model and reduce the model training time by reasonably determining the parameter search interval and improving the parameter optimization algorithm.

[0134] Assembly deformation is an important factor affecting assembly accuracy. It is difficult to calculate by mathematical formula because it is jointly affected by process parameters, part size, part material and other factors. Therefore, the improved support vector machine model is used for assembly deformation prediction. First, based on the maximum standard deviation of the input data, the search range of hyperparameters C and g is determined. In order to find the optimal parameter combination, coarse and fine granularity search is used for parameter combination optimization. First, all parameter combinations are placed in the set shown in formula (8), a sampling ratio is set, and k parameter combinations are randomly selected, and the MSE and R 2The optimal combination is selected, the parameter search interval is further subdivided according to formula (10) and formula (11), all parameter combinations in the interval are substituted into the prediction model to determine the optimal parameter combination, and the support vector machine model is substituted to perform assembly deformation amount prediction.

[0135] (6.2) Other error value determination

[0136] In addition to deformation errors, the assembly process of complex assemblies such as cylinder heads is also affected by other errors. However, these errors are often random values, such as measurement errors and repeated positioning errors, which are difficult to predict using specific formulas or models. To fully consider the impact of these errors on assembly accuracy, the present application determines the error distribution type and error distribution interval based on statistical data. When performing accuracy prediction, random sampling is performed on the error distribution interval to simulate random errors in the assembly process.

[0137] (7) Assembly accuracy prediction considering multi-dimensional coupling errors

[0138] To comprehensively consider the impact of multi-dimensional coupling errors on assembly accuracy, the present application uses the Monte Carlo method to simulate the assembly accuracy prediction model multiple times. A sufficient number of simulation rounds are set, the deformation prediction value and the sampling value of other random errors are substituted into the assembly accuracy prediction model established in step (5), and the process is repeated until the predetermined simulation number is reached. Finally, the average value is obtained to obtain the final assembly accuracy prediction value. This method can overcome the influence of chance.

[0139] (8) Abnormal processing

[0140] For assembly shop may appear the accuracy does not meet the requirements or other abnormal situation, the present application establishes assembly process optimization knowledge base, stores the abnormal situation and assembly accuracy requirement that may appear in the assembly process, and numbers each abnormal type. For the convenience of abnormal processing, the present application adds solving strategy in the assembly process optimization knowledge base, and allows the user to modify and improve, to adapt to different use demand. When the system finds that the assembly process is abnormal or the accuracy prediction result is out of tolerance, the system queries the corresponding fault number according to the assembly process and the abnormal type, and automatically retrieves the solving strategy according to the fault number, and feeds back the solving suggestion to the system interface in time for the user to refer. In addition, the user can edit the knowledge base to enrich the abnormal processing strategy and increase the abnormal processing capacity of the system.

[0141] The application discloses an assembly workshop monitoring and assembly precision prediction method based on digital twinning, and steps are as follows: based on the current situation of an enterprise assembly workshop, a product model library is constructed, and a virtual assembly workshop is built; on the basis of fully analyzing an assembly process, the assembly process is subdivided into multiple states based on a nested fuzzy finite state machine model, and switching conditions between the states are a set. Through nesting of multiple fuzzy finite state machines, the assembly process is comprehensively, clearly and completely represented; analytic hierarchy process and entropy weight method are comprehensively utilized to scientifically and reasonably determine monitoring items, and a data acquisition network is established to acquire real-time assembly data. Based on measured data, a particle swarm optimization (PSO)-SG (Savitzky-Golay) filtering data processing method is proposed, data smoothing is realized on the basis of eliminating noise of the measured data, and the reliability of the data is improved; an assembly precision prediction model considering multi-dimensional coupling errors is established, an improved grid search algorithm is used to optimize a parameter optimization process of a support vector machine (SVM) model, and based on Monte Carlo simulation, influences of accidental factors on assembly precision prediction are overcome, and the accuracy of assembly precision prediction is improved. An assembly process optimization knowledge base is constructed, when predicted precision does not meet a standard, the system automatically throws corresponding solutions, and reference is provided for users.

[0142] Based on the above process, assembly process monitoring and precision prediction based on digital twinning can be realized, users can timely perceive an assembly state, real-time process measured data, predict assembly precision, and timely process when an exception occurs, assembly precision of a marine diesel engine cylinder cover is improved, assembly cost is reduced, and a one-time assembly success rate is improved.

Claims

1. A digital-twin-based assembly process monitoring and precision prediction method, characterized in that, The method comprises the following steps: (1) a digital twin assembly workshop is built according to the real scene of an assembly workshop; (2) an assembly behavior model is established according to the requirements of an assembly process file; the assembly behavior model adopts a nested fuzzy finite state machine model; (3) the analytic hierarchy process-entropy weight method is used to determine the project data that needs to be monitored in the digital twin assembly workshop; (4) the project data that needs to be monitored obtained in step (3) is optimized by using a particle swarm-SG data processing method; the particle swarm-SG data processing method optimizes the window size and the polynomial order of the SG filtering algorithm by using a particle swarm algorithm; (5) an assembly precision prediction model is constructed, which contains a variety of error coupling superposition; an improved support vector machine model is used for assembly deformation prediction; the improved support vector machine model optimizes the regularization parameter C and the gamma parameter g by using a hierarchical coarse and fine granularity grid search algorithm; when performing precision prediction, sampling is performed in the error distribution interval to simulate random errors in the assembly process; (6) the assembly process is simulated by using the Monte Carlo method, a sufficient number of simulation rounds are set, the assembly deformation prediction value in step (5) and the sampling value of the random error are substituted into the assembly precision prediction model, and an assembly precision prediction value is obtained; (7) for the assembly precision that does not meet the requirements or abnormal conditions that occur in the assembly workshop, an assembly process optimization knowledge base is established in the monitoring system; the assembly behavior model adopts a nested fuzzy finite state machine model, the finite state machine model of the assembly process of a subcomponent in the assembly process is taken as a state of the finite state machine model of the assembly process, and the state switching condition is set as a set; when one condition in the state switching condition set is met, state switching can be realized; the particle swarm-SG data processing method comprises the following steps: the allowable fluctuation range of the data is set, and the abnormal outliers are removed; after removing the abnormal values, the interpolation method is used to complete the missing values; the particle swarm algorithm is used to construct a model and design a fitness function, the window size and the polynomial order are taken as the optimization dimensions, and the parameter combination that makes the performance of the SG algorithm optimal is found through step-by-step iteration; after completing the parameter optimization, the best window size and the polynomial order are input into the SG algorithm; within the window range, the SG algorithm fits the data according to the set polynomial order, and outputs the fitting value of the center point of the window as the processed data value; the whole signal data is processed step by step through the movement of the window; the improved support vector machine model optimizes the regularization parameter C and the gamma parameter g by using a hierarchical coarse and fine granularity grid search algorithm, and the specific steps comprise: 1) the search range of the parameters C and g is determined according to the maximum standard deviation of the input data, as shown in the following formula: where a is the base number for incrementing the parameters on a logarithmic scale, σ max is the maximum standard deviation of the input data, b is the step size, representing the logarithmic value of each increment, determining the granularity of the parameter search, and l is the number of increments, lb is the increment to adjust the search range; 2) the parameter optimization process is divided into two stages of coarse and fine granularity search, and a parameter combination set G is established: G = {(C1,g1),..., (C1,g n ) ; (C2,g1),..., (C2,g n ) ;..., (C m ,g n )} where (C m ,g n ) is a possible parameter combination, m and n are positive integers, a random sampling rate a is set in the coarse-grained search stage, a parameter combination is randomly selected, and an optimal parameter combination is searched for, as follows, G'={G[i]|i∈randperm(N,k)} wherein k=max(1,α·N) is the sampling number, N=m×n is the total number of combinations, G' is the parameter combination after random sampling, and randperm(N,k) is a function of randomly selecting k indexes from 1 to N. A fine-grained search range is set around the optimal parameter combination obtained by the coarse-grained search, the fine-grained search range is further divided into n search ranges, and fine-grained search is performed; the values of C and g in the fine-grained search range are as follows: where C i and g i are parameters of fine-grained search, C * and g * are optimal parameters of C and g obtained by coarse-grained search, δ is parameter reduction rate, ξ is parameter growth rate, n is subdivision number, C * × ξ - C * × δ indicates the length of the subdivision interval of parameter C, g * × ξ - g * × δ indicates the length of the subdivision interval of parameter g.

2. The assembly process monitoring and precision prediction method based on digital twinning according to claim 1, characterized in that, The analytic hierarchy process-entropy weight method first classifies workshop data according to information categories, constructs a hierarchical structure model, and only retains the criterion layer; secondly, a judgment matrix is constructed, the analytic hierarchy process is used to solve the importance of each type of information in the criterion layer, sorting is performed, and unimportant information categories are removed; then, for various information in the important information categories retained, the entropy weight method is used to calculate the weight, and the final monitored project data is selected.

3. The assembly process monitoring and precision prediction method based on digital twinning according to claim 1, characterized in that, The fitness function comprehensively considers the mean square error, smoothness and trend fitting degree of the filtered signal.

4. The assembly process monitoring and precision prediction method based on digital twinning according to claim 1, characterized in that, The assembly precision prediction model: where ε is the assembly accuracy, δ 尺寸 is the dimensional error, δ 变形 is the deformation error resulting from process parameters, δ 重复定位 is the repeated positioning error of the assembly process, δ 工装制造 is the tooling manufacturing error, δ 检测 is the inspection error, indicates the coupling of errors; the assembly deformation in step (5) is δ 变形 , the random error includes δ 重复定位 , δ 工装制造 , δ 检测 .

5. The digital-twin-based assembly process monitoring and precision prediction method of claim 1, wherein, The random error in the simulation assembly process is determined according to error statistical data, the distribution type and distribution range of the error are determined, and the random error in the simulation assembly process is sampled in the error distribution range during precision prediction.

6. The digital-twin-based assembly process monitoring and precision prediction method according to claim 1, characterized in that, The assembly process optimization knowledge base is established in the monitoring system, different abnormal conditions that may occur in different states of different stations are numbered, and the corresponding solution strategies are stored, when the monitoring system detects the corresponding abnormality, the solution strategy is automatically queried according to the abnormality number for reference by the user; The user can edit the knowledge base.

7. A digital-twin-based assembly process monitoring and precision prediction system utilizing the digital-twin-based assembly process monitoring and precision prediction method of any one of claims 1-6, wherein It comprises a digital twin assembly workshop, an assembly behavior model module, a data processing module, an assembly precision prediction model module, an assembly simulation module and a knowledge base module; The digital twin assembly workshop is a digital twin assembly workshop built according to the real scene of the assembly workshop; The assembly behavior model module establishes an assembly behavior model according to the requirements of the assembly process file; the assembly behavior model adopts a nested fuzzy finite state machine model; The data processing module determines the project data to be monitored in the digital twin assembly workshop by using the analytic hierarchy process-entropy weight method; the obtained project data to be monitored is optimized by using the particle swarm-SG data processing method; the particle swarm-SG data processing method optimizes the window size and polynomial order of the SG filtering algorithm by using the particle swarm algorithm; The assembly precision prediction model module constructs an assembly precision prediction model, which includes a variety of error coupling superposition; an improved support vector machine model is used for assembly deformation prediction; The improved support vector machine model optimizes the regularization parameter C and the gamma parameter g by using a hierarchical coarse-fine grid search algorithm; during precision prediction, the random error in the simulation assembly process is sampled in the error distribution range; The assembly simulation module simulates the assembly process by using the Monte Carlo method, sets a sufficient number of simulation rounds, substitutes the assembly deformation prediction value and the sampling value of the random error into the assembly precision prediction model, and obtains the assembly precision prediction value; The knowledge base module establishes an assembly process optimization knowledge base in the monitoring system for the precision that does not meet the requirements or abnormal conditions of the assembly workshop.

Images