A production line construction method and system based on comprehensive performance evaluation

Abstract

The application belongs to the technical field of production line production, and particularly discloses a production line construction method and system based on comprehensive performance evaluation, which comprises the following steps: S1, determining evaluation indexes of the production line, and determining comprehensive performance evaluation indexes of the production line by weighted fusion of the evaluation indexes; S2, obtaining evaluation values of the comprehensive performance evaluation indexes corresponding to different production line parameters through modeling simulation, and then obtaining a sample set; S3, training a prediction model through the sample set, and obtaining the trained prediction model; and S4, optimizing parameters of a production line to be constructed based on the prediction model, so that the comprehensive performance evaluation indexes meet the requirements, and thus the production line construction is completed. Meanwhile, the genetic algorithm and the prediction model are combined to realize global optimization of the production line configuration parameters. The method is the basis for realizing predictable production, and is of great significance for design of a to-be-constructed line and improvement of the production line.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of production line production, and more particularly relates to a production line construction method and system based on comprehensive performance evaluation. BACKGROUND

[0002] A production line is a highly efficient production organization mode in which professional production equipment is determined and arranged according to the process path of a product. Compared with the traditional production organization mode, the production line has high production efficiency. The current research on the performance of the production line involves various dimensions, including capacity, availability, robustness and balance; and when the production line is constructed, each performance index needs to be quickly and accurately evaluated to select the optimal production line parameters and improve the production efficiency.

[0003] However, the current research mostly only targets a certain dimension of the performance of the production line, and the reflection of the performance of the production line has certain one-sidedness. Meanwhile, it is time-consuming to obtain the performance index corresponding to different production line parameters by using simulation software, especially in the design stage of the production line, because many configuration parameters are not determined, and various design schemes can only be changed one by one to change the simulation model parameters, and multiple independent repeated simulations are performed again. A large amount of time and computing resources are required to complete this process. SUMMARY

[0004] In view of the above defects or improvement needs of the prior art, the application provides a production line construction method and system based on comprehensive performance evaluation, which aims to comprehensively consider multiple production line performance evaluation indexes and improve the adjustment and optimization efficiency of the production line parameters, and quickly construct a production line meeting the comprehensive performance requirements.

[0005] To achieve the above-mentioned purpose, according to one aspect of the application, a production line construction method based on comprehensive performance evaluation is provided, including a model training stage and a model using stage.

[0006] The model training stage includes:

[0007] S1, determining the evaluation indexes of the production line, and determining the comprehensive performance evaluation indexes of the production line by weighted fusion of the evaluation indexes;

[0008] S2, obtaining the evaluation values of the comprehensive performance evaluation indexes corresponding to different production line parameters by modeling simulation, and then obtaining a sample set;

[0009] S3, training the prediction model by using the sample set, and obtaining the trained prediction model;

[0010] The model using stage includes:

[0011] S4, based on the prediction model, optimizing the to-be-constructed production line parameters so that the comprehensive performance evaluation index meets the demand, thereby completing the production line construction.

[0012] As a further optimization, when adjusting the existing production line: in the model use stage, according to the sensitivity of the production line parameters to the evaluation index, some production line parameters are input into the trained prediction model, and the production line parameters with higher sensitivity are adjusted preferentially, so that the comprehensive performance evaluation index predicted by the prediction model meets the demand, thereby completing the adjustment and construction of the production line.

[0013] As a further optimization, the calculation method of the sensitivity of the production line parameters to the evaluation index is: first, use the Morris method to qualitatively screen some production line parameters with higher sensitivity; then, use the Sobol method to quantitatively analyze the sensitivity of the screened production line parameters, and obtain the sensitivity of each production line parameter.

[0014] As a further optimization, when reconstructing the production line: in the model use stage, the XGBoost prediction model is combined with the genetic algorithm to optimize the production line parameters, specifically taking the production line parameters as population individuals, taking the comprehensive performance evaluation index calculated by the XGBoost prediction model as the fitness, and determining the optimal production line parameters through iterative optimization by the genetic algorithm, thereby completing the reconstruction of the production line.

[0015] As a further optimization, the evaluation index of the production line includes production line capacity, availability, robustness and balance rate, and the comprehensive performance evaluation index of the production line is determined by weighting and fusing each evaluation index by the analytic hierarchy process.

[0016] As a further optimization, in step S2, the production line comprehensive performance evaluation index is obtained by Plant Simulation simulation.

[0017] As a further optimization, the simulation warm-up time is determined by the Welch moving average method, and after the simulation is completed, the evaluation values obtained during the simulation warm-up time are deleted, and the remaining evaluation values are used to construct the sample set.

[0018] As a further optimization, the construction of the sample set from the remaining evaluation values is specifically: constructing a data set with the production line parameters and the evaluation values of the corresponding comprehensive performance evaluation index, selecting samples in the data set by Latin hypercube sampling, thereby obtaining the sample set.

[0019] As a further optimization, the production line parameters include the processing cycle of each part in different processing units, the processing unit fault maintenance time, the processing unit fault interval time and the processing unit buffer size.

[0020] According to another aspect of the present application, there is provided a production line construction system based on comprehensive performance evaluation, comprising a processor configured to execute the above-mentioned production line construction method based on comprehensive performance evaluation.

[0021] Overall, compared with the prior art, the above technical solutions conceived by the present application mainly have the following technical advantages:

[0022] 1. The present application assigns weights to each performance indicator to obtain a comprehensive performance evaluation index through weighted fusion, and based on the establishment of the comprehensive performance evaluation index, obtains the required data for evaluation through a simulation model; then uses a prediction model to replace the simulation model to avoid the tedious and time-consuming process of obtaining results from the simulation model, and optimizes the parameters based on the prediction model to quickly construct a production line that meets the comprehensive performance requirements. The present application method is the basis for realizing predictable production, and is of great significance for the design of the proposed line and the improvement of the production line.

[0023] 2. For different application scenarios, the present application designs different parameter optimization methods:

[0024] For the improvement of existing production lines, the change of each production line configuration parameter will affect the evaluation results, but each parameter has different effects on the evaluation results, some parameters have greater effects on the evaluation results, and some parameters have smaller effects on the evaluation results; the present application determines the key parameters that have greater effects on the evaluation results of the production line through Morris-Sobol sensitivity analysis, and can use the analysis results to reduce the dimensionality of the input space, reduce the calculation amount during modeling, realize local optimization of parameters, and improve the parameter optimization efficiency.

[0025] For the construction of a new production line, the XGBoost prediction model is combined with the genetic algorithm to optimize the production line parameters, which can avoid falling into local optimization, realize global optimization of parameters, and make the comprehensive performance evaluation index optimal.

[0026] 3. Through the design of the prediction model, the present application can have faster prediction speed, consume less computing resources, and has the following advantages: it can predict the comprehensive performance evaluation index results of the production line scheme with different parameter settings; it can eliminate the influence of randomness in simulation and smooth the simulation results; application of the prediction model can more conveniently explore the influence degree of each configuration parameter on the comprehensive performance evaluation index results of the production line; application of the prediction model can more quickly and conveniently optimize the parameters of the production line.

[0027] 4. The present application selects representative indicators from the four dimensions reflecting the performance of the production line, comprehensively considers the performance of each dimension of the production line, and can comprehensively evaluate the production line, so that the evaluation value can better represent the comprehensive overall performance of a production line, so as to better grasp the overall state of the production line and better design and optimize it.

[0028] 5. In the simulation stage, the discrete-time simulation has an initial transient problem, which leads to the phenomenon that the output results of the model are different from the results in the steady state, a simulation warm-up time is determined by using the Welch moving average method, after the operation of the simulation warm-up time, the observation value of the model reaches the performance in the steady state, the observation value in the simulation warm-up time can be deleted, only the remaining observation value is used for statistical solution, and the accuracy of the simulation result is improved.

[0029] 6. All sample data sets are obtained by using the discrete-time simulation model to perform computer simulation experiments, and the samples are randomly sampled in layers, specifically, Latin hypercube sampling is used for sample selection, so that the sample distribution is more uniform, and the randomness is also higher. BRIEF DESCRIPTION OF DRAWINGS

[0030] Figure 1 A production line construction method flow chart based on comprehensive performance evaluation of the embodiment of the application is shown in the figure;

[0031] Figure 2 An analytic hierarchy process construction comprehensive performance evaluation index structure diagram of the embodiment of the application is shown in the figure;

[0032] Figure 3 A simulation flow chart for obtaining the comprehensive performance index of the production line of the embodiment of the application is shown in the figure;

[0033] Figure 4 A production line comprehensive performance index prediction model and parameter optimization diagram of the embodiment of the application is shown in the figure;

[0034] Figure 5 A commutator production line structure model of the embodiment of the application is shown in the figure. DETAILED DESCRIPTION

[0035] In order to make the purpose, technical scheme and advantages of the application clearer, the application is further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the application, and are not used to limit the application. In addition, the technical features involved in each embodiment of the application described below can be combined with each other as long as they do not conflict with each other.

[0036] The production line construction method based on comprehensive performance evaluation provided by the embodiment of the application is described by taking a commutator production line as an example, as shown in the figure, Figure 1 The method comprises the following steps:

[0037] S1, selecting representative evaluation indexes for the production line, and determining suitable weights for each index by using the analytic hierarchy process, and obtaining the comprehensive performance evaluation index of the production line by weighting and fusing each index.

[0038] Further, the evaluation indexes include: a production line capacity index reflecting the production capacity of the production line; a production line availability index reflecting the production line's ability to maintain normal production capacity; a production line robustness index reflecting the production line's ability to respond to environmental changes; and a production line balance index reflecting the continuity and rationality of the production line.

[0039] Specifically, the capacity refers to the number of products that the production line can produce within a certain time or the time required to produce a certain number of products under given organizational and technical conditions, and is used to measure the production capacity of the production line, which is the most important dimension of the performance of the production line. For a multi-variety production line, it is more convenient to use piecework to calculate the total capacity by product method, and the formula is as formula (1):

[0040]

[0041] Wherein: PN represents the total number of products produced; T total represents the total production time.

[0042] The production line availability refers to the ability of the production line to perform the specified functions under the specified conditions and within the specified time, excluding external factors such as supply, personnel, finished product storage, etc. The availability is defined as the degree of the system being in a workable state, i.e. the ratio of the time the system can work to the sum of the time it can work and the time it cannot work, which is the most important index for evaluating the availability of the system, so the availability is selected as the availability index of the commutator production line. The production time of the commutator production line is the working time of the production line minus the whole line downtime. The formula is as formula (2):

[0043]

[0044] Wherein: A represents the availability of the production line; WT represents the working time of the production line; PRDT represents the production time of the production line; OLST represents the whole line downtime of the production line, OLST = WT-PRDT.

[0045] The production line robustness refers to the ability of the production line to resist disturbances during production. The commutator production line can effectively alleviate the propagation of such disturbances by adding buffer zones between production units. The downtime conversion rate can be selected as the robustness evaluation index of the commutator production line. The downtime conversion rate of the commutator production line is defined as the probability of converting equipment failure into whole line downtime, i.e. the ratio of the sum of the failure time of each production equipment to the sum of the production line downtime and the production line equipment failure time. The formula is as formula (3):

[0046]

[0047] Wherein, DCR represents the production line downtime conversion rate; The sum of the failure time of all devices in the production line; ttr ij The duration of the jth failure of the ith device; n represents the total number of devices in the production line; m i The number of failures of the ith device; OLST represents the production line downtime, which is the production line uptime minus the production time.

[0048] Production line balancing refers to balancing the load of each production unit by reasonably allocating the work of each production unit to balance the processing time of each production unit. The production line balancing rate is the most important indicator for measuring the balancing of the production line. Therefore, the balancing rate is used as the indicator for balancing the commutator production line, and the formula is as shown in equation (4):

[0049]

[0050] Where LB represents the balancing rate, n represents the number of production units in the production line, t i The processing time of the ith production unit, The total processing time of all production units, P represents the processing time of the bottleneck process, i.e., the processing time of the process with the longest processing time.

[0051] Further, the four evaluation indicators selected, i.e., the production line capacity, availability, robustness, and balancing rate, respectively reflect a certain dimension of the performance of the commutator production line, and each has a certain one-sidedness. To comprehensively evaluate the commutator production line, the evaluation results of each indicator need to be integrated. The analytic hierarchy process can be applied to assign appropriate weights to the four indicators, and then the weighted sum of each indicator is calculated according to the weights, so as to obtain a comprehensive performance evaluation indicator that can comprehensively reflect the performance of each dimension of the commutator production line. The process of weighting each indicator using the analytic hierarchy process is to construct the hierarchical relationship between the production line setting parameters, the four selected evaluation indicators, and the comprehensive performance evaluation indicator, determine the relative importance relationship between each indicator, construct a suitable judgment matrix, and perform consistency check to determine whether the relative importance ranking of the four evaluation indicators relative to the comprehensive performance evaluation indicator is correct, and finally give the weight of each evaluation indicator in the form of the eigenvector of the judgment matrix. According to the weight values of the four indicators obtained, the four indicators can be normalized and combined to construct the comprehensive performance evaluation indicator of the commutator production line.

[0052] Specifically, the comprehensive performance evaluation indicator of the production line is constructed based on the analytic hierarchy process, as shown in equation (5), which includes: Figure 2

[0053] ​S11, establish hierarchy model. The comprehensive performance evaluation index, the four selected evaluation indexes and the production line parameter settings are divided into target layer, criterion layer and scheme layer respectively. The meaning of each layer is as follows: the comprehensive performance evaluation index of the target layer represents the result to be evaluated; the four evaluation indexes of the criterion layer represent the factors considered by the target layer; and the production line parameter settings of the scheme layer represent different schemes participating in the evaluation, i.e. the production line under different parameter configurations.

[0054] S12, construct judgment matrix. In order to avoid using the results of simple qualitative judgment in determining the weight of each evaluation index, the judgment matrix can be constructed by using the consistency matrix method. The judgment matrix is determined by comparing the relative importance of the four selected evaluation indexes of the criterion layer with respect to the comprehensive performance evaluation index of the target layer. The element a ij in the judgment matrix represents the importance of the i-th factor with respect to the j-th factor.

[0055] S13, hierarchy single sorting and consistency check. The hierarchy single sorting is the sorting of the relative importance of the evaluation indexes of the criterion layer with respect to the comprehensive performance evaluation index of the decision layer. The consistency check is the process of determining whether the single sorting is correct. The steps of the consistency check are as follows:

[0056] 1) calculate the maximum eigenvalue of the judgment matrix;

[0057] 2) calculate the consistency index: where CI represents the consistency index; λ max represents the maximum eigenvalue of the judgment matrix; and n represents the number of factors, i.e. the number of rows or columns of the judgment matrix.

[0058] 3) determine the consistency index RI;

[0059] 4) calculate the consistency ratio: where CR represents the consistency ratio; CI represents the consistency index; and RI represents the average random consistency index.

[0060] 5) judgment. If CR<0.1, it indicates that the judgment matrix has satisfactory consistency, otherwise the judgment matrix needs to be modified.

[0061] 6) eigenvalue method to obtain the weight of each factor. If the consistency check of the judgment matrix (a ij ) n×n is passed, the maximum eigenvalue λ ij of (a n×n ) max is the characteristic vector corresponding to each evaluation index. Normalizing the characteristic vector, the sorting weight vector The elements are the order weights of the relative importance of each evaluation index of the criterion layer relative to the comprehensive performance evaluation index of the target layer.

[0062] S14, the four indicators are combined to construct a comprehensive performance index. The values of the three indicators of production line availability, downtime conversion rate and production line balance rate are between 0 and 1, and the value of the production capacity exceeds this range. If direct weighted summation is performed, the proportion of the production capacity in the result is too large, and the weight value loses its meaning. Therefore, the production capacity of the production line needs to be normalized before effective weighting.

[0063] S2, a discrete-time simulation model of the production line is constructed, the evaluation value of the comprehensive performance evaluation index of the production line is obtained based on the simulation model, so as to obtain data (sample set) for calculation, and the simulation repetition number and simulation warm-up time are determined to make the model more accurate.

[0064] Further, the production line is modeled by discrete events, the Plant Simulation software is used as the modeling software, and the comprehensive performance evaluation index of the production line is recorded and solved; at the same time, in order to reduce the influence of randomness in the simulation process, the number of independent repeated simulation experiments is determined, and the Welch moving average method is used to explore the initial transient problem in the discrete event simulation.

[0065] For a new production line still in the design stage, the cost of establishing a new production line and putting it into production is high, and a good design scheme is of great significance to the construction of a high-performance production line. By establishing a simulation model, the feasibility of different design schemes can be verified, and the design scheme with better performance can be selected. When adjusting and optimizing the production line, the cost is very high if the actual production line is directly adjusted and experimented. Through the computer simulation model of the production line, the performance of the production line under different conditions can be explored, so that the method of improving the production line can be explored without affecting the actual production. In the simulation process, the influence of each input on the output can be explored by changing a certain input parameter and observing the change of the output. For a complex production line, it is difficult to obtain an accurate analytical solution by a pure mathematical model, but a more detailed simulation model can obtain a more accurate solution.

[0066] Specifically, as shown in Figure 3 , step S2 includes:

[0067] S21, production line modeling assumption. In order to facilitate subsequent modeling of the production line within a reasonable range and reasonable index evaluation.

[0068] S22, production line simulation modeling. The model of the commutator production line should have the following characteristics: the modeling parameters of the production line are adjustable; the comprehensive performance evaluation index can be solved and recorded; the production line with different parameter settings can be independently simulated repeatedly. The discrete event simulation model established by using Plant Simulation software realizes the calculation of the comprehensive performance evaluation index of the commutator production line.

[0069] S23, determination of simulation times and simulation warm-up time. The number of simulation experiments can be determined by using the Experiment object. The Experiment object can count the output results of multiple independent repeated simulations and calculate the confidence interval of the output results under a certain confidence level, so that the appropriate simulation times can be determined by observing the change range of the output results and the interval size of the confidence interval under different simulation times. In the simulation stage, there is an initial transient problem in the discrete time simulation, which leads to the phenomenon that the output results of the model are different from the results in the steady state. The Welch moving average method is used to determine a simulation warm-up time, and after running for the simulation warm-up time, the observation value of the model reaches the performance in the steady state, so that the observation value in the simulation warm-up time can be deleted, and only the remaining observation value is counted and solved.

[0070] S24, sample set acquisition. In the production line system, X can be used to represent the factor vector affecting the behavior of the system, x1, x2, …, x R represent the factors of X, that is, the input parameters of the system; Y represents the response vector of the production line system, that is, the comprehensive performance evaluation index of the production line. Therefore, the response relationship of the source system can be defined as: Y=G(X)=G(x1,x2,…,x R ). All sample data sets are obtained by using the simulation model to perform computer simulation experiments, and the final sample set is obtained by stratified random sampling of the samples in the sample data set; specifically, Latin hypercube sampling is used for sample selection, so that the sample distribution is more uniform and the randomness is also higher.

[0071] S3, a prediction model is constructed based on a machine learning algorithm to replace the simulation model, and the prediction model is trained through the sample set to improve the efficiency of obtaining the comprehensive performance evaluation index of the production line.

[0072] Further, the XGBoost model is used as the final prediction model. In order to prove the effect of the XGBoost model, the present application constructs a prediction model of the comprehensive performance evaluation index and other evaluation indexes of the commutator production line based on the GBDT and XGBoost models, and analyzes the current commutator production line using the two prediction models.

[0073] Specifically, the Gradient Boosting Decision Tree (GBDT) algorithm and the eXtreme Gradient Boosting (XGBoost) algorithm are both ensemble learning models, the base learner is a regression tree, and a strong learner is composed of multiple base learners in series, each regression tree corrects the result of the previous one, and the model is continuously optimized. For parameter setting of the GBDT model, the main parameters to be determined include boosting framework parameters and base learner parameters. The boosting framework parameters mainly include the maximum iteration number n_estimators of the weak learner, the weight reduction coefficient learning_rate of the weak learner, the subsampling subsample, and the loss function loss. The base learner parameters mainly include the maximum depth max_depth of the decision tree and the minimum sample size min_samples_split required for internal node subdivision, etc. The XGBoost model has more settings for the regularization weight term alpha, the L2 regularization weight term lambda, etc. After training, the validation set and the test set are tested respectively, and the effects of the two are compared. The comparison results show that the prediction effect based on the XGBoost model is better.

[0074] S4. According to different application scenarios, the production line parameters are optimized based on the prediction model, so that the comprehensive performance evaluation index meets the demand, and the production line construction is completed.

[0075] (1) When adjusting the existing production line: according to the sensitivity of the production line parameters to the evaluation index, some production line parameters are input into the trained prediction model, and the production line parameters with higher sensitivity are adjusted preferentially, and the result is predicted again, so that the comprehensive performance evaluation index predicted by the prediction model meets the demand, thereby completing the adjustment and construction of the production line.

[0076] Further, since the number of input parameters of the production line prediction model is large, for multi-dimensional input parameters, the Morris method and the Sobol method are combined to analyze the parameter sensitivity of the index; the Morris method is a qualitative sensitivity analysis method, and the Sobol method is a quantitative sensitivity analysis method. Specifically, the Morris method with lower calculation cost is used for preliminary screening, and several parameters with higher sensitivity are qualitatively screened out to reduce the model dimension, and then the Sobol method is used for quantitative sensitivity analysis of the several key parameters screened out to obtain the specific sensitivity value of each parameter, and the configuration parameters with large contribution to the output result are preferentially adjusted during optimization.

[0077] (2) When the production line is restructured: in the model use stage, the XGBoost prediction model is combined with the genetic algorithm to optimize the production line parameters; the production line parameters are taken as the population individuals, the comprehensive performance evaluation index calculated by the XGBoost prediction model is taken as the fitness, the genetic algorithm is iteratively optimized, the optimal production line parameters are determined, the production line is restructured, and the optimization process is as shown in Figure 4 .

[0078] Specifically, the fitness function in the genetic algorithm is used to evaluate the good and bad of individuals and populations, and the higher the fitness, the better the individual or population. By encoding the individual, that is, the input production line parameters, subsequent selection, crossover and mutation operations are performed, and the comprehensive evaluation index value, that is, the fitness, is obtained by using the XGBoost prediction model. After a certain number of generations of evolution, the fitness of the individuals in the population is improved, the individual with the highest fitness in the population is selected, and the parameters of the individual obtained by decoding are the final optimization results.

[0079] To test the effect of the XGBoost genetic algorithm, the XGBoost genetic algorithm and the GA tool of Plant Simulation are used for experiments, and the results are compared. Specifically, the parameter settings of the genetic algorithm part of the present application are as follows: the population size is 20, the evolution generation number is 500, the crossover probability is 0.4, and the mutation probability is 0.2. In addition, the genetic algorithm (GA) tool GAWizard for parameter optimization in the Plant Simulation software is used, and important parameters of the genetic algorithm such as population size, evolution generation number, fitness function and input and output are defined in the GAWizard; at the same time, the number of repeated observations for each individual is set, and the statistical reliability of each individual is ensured by repeated observations. The comparison results show that the XGBoost genetic algorithm has less calculation time under the condition that the optimization effect is similar to that of the GA tool.

[0080] The method of the present application is described below through a commutator production line specific example:

[0081] The structure model of the commutator production line is as shown in Figure 5 , wherein B 11 ~B 52 indicates the buffer area, B i1 (i=1, 2, 3, 4, 5) indicates the buffer area for storing parts to be processed, B i2 (i=1, 2, 3, 4, 5) indicates the buffer area for storing processed parts, and the capacities are K i1 and K i2 (i=1, 2, 3, 4, 5), and K i1 =K i2(i = 1, 2, 3, 4, 5). When the AGV vehicle transports the parts through each processing unit, the transported parts are first transferred to the buffer area for storing the parts to be processed, and then the parts stored in the buffer area for storing the processed parts are transferred to the AGV vehicle.

[0082] T1~T5 respectively represent the processing periods of the processing units S1~S5, wherein the third processing unit S3 is composed of two same processing devices S 31 and S 31 in parallel, and thus its processing period is: T3 = (T 31 + T 32 ) / 2.

[0083] wherein the processing unit preparation time refers to the preparation time for replacing different processing parts and replacing the clamp and the like. The tbf and ttr of the processing unit refer to the failure interval time and the failure maintenance time of each processing unit, which are obtained by recording the failure and maintenance records of the processing unit in the production process. Generally, tbf and ttr are random variables subject to certain statistical distribution, and the distribution form in the commutator production line is Weibull distribution. There are three parts in total, Part 1 is Part 1, Part 2 is Part 2, and Part 3 is Part 3.

[0084] The above parameters are important parameters of the production line, and the input parameters for subsequent prediction are selected from these factors. The considered factors (configuration parameters) and their change ranges are shown in Tables 1-3.

[0085] Table 1 Change range of processing period (s)

[0086]

[0087] Table 2 Change range of ttr distribution (h) and tbf distribution (min)

[0088]

[0089] Table 3 Change range of buffer size

[0090]

[0091] That is, there are 26 production line parameters: T 11 , T 21 , T 12 , T 32 , T 13 , T 23 , T 33 , T 14 , T 24 , T 25 , T 35ttr1, ttr2, ttr3, ttr4, ttr5, tbf1, tbf2, tbf3, tbf4, tbf5, K1, K2, K3, K4, K5. Wherein, T ij represents the processing cycle of the i-th part in the j-th processing unit; ttr k represents the k-th processing unit fault maintenance time; tbf n represents the n-th processing unit fault interval time; K m represents the m-th processing unit buffer size.

[0092] The weight values of the four indexes of capacity, availability, robustness and balance obtained by using the analytic hierarchy process are: 0.4146, 0.0982, 0.2436, 0.2436, and the four index values differ greatly, so normalization is carried out before weighting. Then, the XGBoost network is used, and 26-dimensional parameters are input to predict the four-dimensional indexes and the comprehensive performance index, and the prediction model is obtained.

[0093] Further, the Morris-Sobol method is used for parameter sensitivity analysis of the index, the Morris method is used for preliminary screening, and the several input parameters that have greater influence on the comprehensive performance evaluation index are T31, ttr1, ttr3, ttr4, K1, K2, K4 and K5. Further sensitivity analysis is carried out by using the Sobol method, and the two parameters with the highest sensitivity are K1 and ttr3. Therefore, when improving the comprehensive performance evaluation index of the commutator production line and adjusting the configuration parameters, the two parameters can be preferentially optimized to improve the efficiency of the optimized production line.

[0094] In order to verify the optimization result, the parameters are input into the Plant Simulation simulation model for simulation experiment, and the absolute error and relative error of the parameter optimization result and the simulation experiment verification result are very small, which verifies the correctness of the optimization result. At the same time, in the case that the optimization effect is almost the same as the simulation software, the efficiency is improved by about 650 times.

[0095] In summary, the prediction model is used instead of the simulation model in the present application, which avoids the tedious and time-consuming process of obtaining results by the simulation model, and the parameter sensitivity analysis and parameter optimization of the production line comprehensive performance evaluation index model. The method of the present application can be used in the design stage and the running stage. In the design stage, the application of the evaluation method of the present application can quickly predict the comprehensive performance evaluation index result of the new production line scheme with different parameter settings before the establishment of the production line and reduce the calculation cost. In the running stage of the production line, the evaluation method of the present application is used for the production line, and through sensitivity analysis and parameter optimization, the existing production line can be improved and optimized, avoiding direct adjustment of the actual production line, so as to not affect the direct production.

[0096] It is to be understood that the above description is intended to be illustrative and not restrictive. Many other embodiments will be apparent to those of skill in the art upon reading and understanding the above description. The scope of the application should, therefore, be determined with reference to the appended claims, along with the full scope of equivalents to which such claims are entitled.

Claims

1. A production line construction method based on comprehensive performance evaluation, characterized by, The model training stage and the model using stage are included; The model training stage includes: S1, determining evaluation indexes of the production line, and determining a comprehensive performance evaluation index of the production line by weighted fusion of the evaluation indexes; the evaluation indexes include: a production line capacity index reflecting production capacity of the production line; a production line availability index reflecting the production line's ability to maintain normal production capacity; a production line robustness index reflecting the production line's ability to respond to environmental changes; and a production line balance index reflecting continuity and rationality of the production line; The comprehensive performance evaluation index of the production line is constructed based on the analytic hierarchy process, including: S11, establishing a hierarchical structure model: the comprehensive performance evaluation index, the four selected evaluation indexes and the production line parameter settings are respectively divided into a target layer, a criterion layer and a scheme layer; S12, constructing judgment matrix: when determining the weight of each evaluation index, a consistency matrix method is used to construct a judgment matrix; the judgment matrix is determined by comparing the relative importance of the four selected evaluation indexes of the criterion layer to the comprehensive performance evaluation index of the target layer; the elements in the judgment matrix represent the importance of the first factor relative to the second factor ​​​ S13, hierarchical single ordering and consistency check: hierarchical single ordering is the ordering of the relative importance of the evaluation indexes in the criterion layer to the comprehensive performance evaluation index in the decision layer; and the consistency check is to determine whether the single ordering is correct, and the steps of the consistency check are as follows: 1) calculating the maximum eigenvalue of the judgment matrix; 2) Calculate the consistency index: where CI denotes the consistency index, denotes the largest eigenvalue of the judgment matrix, denotes the number of factors, i.e. the number of rows or columns of the judgment matrix; 3) determining the consistency index RI; 4) Calculate the consistency ratio ; 5) judgment: if then it indicates that the judgment matrix has satisfactory consistency, otherwise the judgment matrix needs to be modified; 6) Eigenvalue method to get the weight of each factor: if the consistency test of the judgment matrix is passed, the maximum eigenvalue of the judgment matrix is obtained, and the eigenvector corresponding to the maximum eigenvalue is the weight corresponding to each evaluation index; then the eigenvector is normalized to obtain the ranking weight vector , , and the elements of the ranking weight vector are the ranking weights of each evaluation index of the criterion layer relative to the overall performance evaluation index of the target layer. S14, constructing the comprehensive performance evaluation index of the production line in combination with the four evaluation indexes: normalizing the production line capacity, and then determining the comprehensive performance evaluation index of the production line by weighted fusion of the four evaluation indexes; S2, obtaining evaluation values of the comprehensive performance evaluation index corresponding to different production line parameters through modeling simulation, and then obtaining a sample set; S3, training the prediction model through the sample set to obtain the trained prediction model; The model using stage includes: S4, based on the prediction model, optimizing the parameters of the production line to be constructed so that the comprehensive performance evaluation index meets the demand, thereby completing the construction of the production line.

2. The production line construction method based on comprehensive performance evaluation according to claim 1, characterized by, When adjusting an existing production line: in the model using stage, according to the sensitivity of the production line parameters to the evaluation indexes, some production line parameters are input into the trained prediction model, and the production line parameters with higher sensitivity are adjusted preferentially, so that the comprehensive performance evaluation index predicted by the prediction model meets the demand, thereby completing the adjustment and construction of the production line.

3. The production line construction method based on comprehensive performance evaluation according to claim 2, characterized by, The calculation method of the sensitivity of the production line parameters to the evaluation indexes is: first, qualitatively screening some production line parameters with higher sensitivity by using the Morris method; and then, quantitatively analyzing the sensitivity of the screened production line parameters by using the Sobol method to obtain the sensitivity of each production line parameter.

4. The production line construction method based on comprehensive performance evaluation according to claim 1, characterized by, When reconstructing a production line: in the model using stage, the XGBoost prediction model is combined with the genetic algorithm to optimize the production line parameters, specifically taking the production line parameters as population individuals, taking the comprehensive performance evaluation index calculated by the XGBoost prediction model as the fitness, and determining the optimal production line parameters through iterative optimization of the genetic algorithm, thereby completing the reconstruction of the production line.

5. The production line building method based on comprehensive performance evaluation according to claim 1, wherein, In step S2, the comprehensive performance evaluation index of the production line is obtained through Plant Simulation simulation.

6. The production line construction method based on comprehensive performance evaluation according to claim 5, wherein, The simulation warm-up time is determined by the Welch moving average method, and the evaluation values obtained during the simulation warm-up time are deleted after the simulation ends, so as to construct a sample set with the remaining evaluation values.

7. The production line construction method based on comprehensive performance evaluation according to claim 6, wherein, The sample set is constructed by using the rest evaluation values, specifically, a data set is constructed by using the production line parameters and the corresponding comprehensive performance evaluation indexes, sample selection is performed in the data set by Latin hypercube sampling, and thus the sample set is obtained.

8. The production line building method based on comprehensive performance evaluation according to any one of claims 1 to 7, characterized by, The production line parameters include processing periods of each part in different processing units, processing unit fault maintenance time, processing unit fault interval time, and processing unit buffer size.

9. A production line construction system based on comprehensive performance evaluation, characterized by, The application further provides a processor for performing the production line construction method based on comprehensive performance evaluation according to any one of claims 1-8.

Images