CN-122021260-A - Assembly line concept design method and system based on Bayesian framework learning

CN122021260ACN 122021260 ACN122021260 ACN 122021260ACN-122021260-A

Abstract

The invention relates to the technical field of production line design, in particular to an assembly production line concept design method and system based on Bayesian framework learning. The method comprises the steps of S1, constructing a knowledge graph of product design, fine-tuning a large model by using the knowledge graph, generating a product design scheme by using the adjusted large model, S2, responding to the product design scheme, adopting a Bayesian framework to produce a plurality of production line schemes meeting production, S3, carrying out data verification on the plurality of production line schemes, judging whether the production line schemes meet preset conditions, if so, outputting the production line schemes meeting the preset conditions, and if not, re-executing the step S2. The technical problems in the conceptual design of the assembly production line, such as complex multi-objective optimization, uncertainty management and quick response of the change of the demand, are effectively solved, so that the design period is shortened, the production line performance is improved, and the development cost is reduced.

Inventors

ZHANG DING
TAN JIAHONG
PAN RUI
LIN ZEXIONG

Assignees

广东工业大学

Dates

Publication Date: 20260512
Application Date: 20251231

Claims (8)

1. The assembly production line conceptual design method based on Bayesian framework learning is characterized by comprising the following steps of: Step S1, constructing a knowledge graph of product design, performing fine adjustment on a large model by using knowledge graph data, and generating a product design scheme by using the adjusted large model; s2, inputting design requirements into the adjusted large model to generate a product design scheme; responding to the product design scheme, and adopting a Bayesian framework to produce a plurality of production line schemes meeting production; and step S3, carrying out data verification on the multiple production line schemes, judging whether the production line schemes meet the preset conditions, outputting the production line schemes meeting the preset conditions if the production line schemes exist, and if the production line schemes do not meet the preset conditions, executing the step S2 again.
2. The concept design method of assembly line based on bayesian framework learning according to claim 1, wherein the steps in the step S1 are as follows: Acquiring historical data of a process and equipment from a production line corresponding to a product, identifying association rules and frequent patterns among different data sources in the historical data of the process by a data mining technology, mapping heterogeneous data to a public semantic space by combining an ontology mapping method, and constructing a knowledge graph according to a knowledge graph construction method; injecting the knowledge graph into a large model; Inputting the design requirement into a question-answering system of a large-value model, and obtaining design parameters and a 3d modeling code; and determining whether the product design requirement is met based on the design parameters and the 3d modeling code, if so, outputting a product design scheme, and if not, inputting the design requirement into the large model again.
3. The assembly line conceptual design method based on bayesian framework learning according to claim 1, wherein the implementation step of the step S2 is as follows: step S21, collecting process flow data of a historical production line as a background set; S22, dividing all mobile phone schemes into a plurality of categories by adopting a UniForce single-mode forest method for design parameters in each historical production line design scheme in a background set; s23, carrying out cluster analysis on the design parameters and the samples in the background set by using a UniForce single-mode forest method, identifying the product category to which the design parameters belong, and obtaining a process skeleton corresponding to the design scheme; Step S24, finding out the corresponding type of equipment contained in each process in the process framework, and recording the total process quantity in the process framework as follows Finding out the corresponding category of the equipment contained in each process in the process skeleton, and respectively distributing the equipment in probability Sampling; step S25, in probability distribution for the total number of processes k Wherein the total number k of processes is greater than or equal to the process skeleton If not, re-executing the step to sample the total number k of processes; Step S26, inserting unnecessary processes into process gaps of the skeleton process, wherein the unnecessary processes of each process gap are required to meet probability distribution; step S27, if the process after sampling is repeated with the next process, the next gap is eliminated before sampling, and each time after the process gap sampling is finished, a test is carried out to judge whether the total number k of the processes is more than or equal to the process skeleton Unnecessary process Sum of all of Continuing to sample the unnecessary process until the next process gap, if so K remains unchanged and stops going to the next process gap.
4. The concept design method of assembly line based on bayesian framework learning according to claim 1, wherein the searching for the category to which the design parameter belongs in step S22 further comprises the steps of: S221, clustering the input design samples and the samples of the background set into a plurality of first sub-clusters according to k-mean cluster analysis; Deleting the first sub-clusters with the data quantity smaller than the quantity threshold value to obtain K second sub-clusters, and forming second sub-clusters; If the cluster where the input design sample is located is removed, classifying the design sample into the sub cluster closest to the design sample to obtain a new second sub cluster; Step S222, selecting any two second sub-clusters from the second sub-cluster sets, wherein one of the most data is marked as a big cluster Another marked as small cluster The big clusters are calculated respectively through Monte Carlo simulation for odd times And small clusters The method comprises the steps of (1) obtaining a perpendicular bisector hyperplane of a connecting line of the centers of two sub-clusters; step S223, calculating small clusters Symbol distances from all data points in the first distance clutch set to the vertical bisector hyperplane are obtained; step S224, initializing a voting list and setting the length as L; Step S225, from big cluster Medium random extraction and small clusters Calculating symbol distances between each sample and the vertical bisector hyperplane by using samples with the same data quantity to obtain a second distance clutch set; Step S226, combining the first distance clutch set and the second distance clutch set to obtain a third distance clutch set, executing dip-test operation on the third distance clutch set to obtain a first probability, judging whether the first probability is larger than a probability threshold, if the first probability is larger than the probability threshold, adding one to the number of votes, adding one to the number of loops, and if the first probability is smaller than the probability threshold, adding one to the number of votes, wherein the number of votes is zero; step S227, judging whether the cycle number is equal to L, if the broken cycle number is not equal to L, re-executing the steps S225-S226, and if the cycle number is equal to L, counting a first total value of the voting number; judging whether the first total value is larger than L/2, if the first total value is larger than L/2, clustering And small clusters Connecting, constructing into the same cluster, if the first total value is smaller than L/2, not forming a large cluster And small clusters Connecting; And step S228, repeating the steps S222-S227 until all the sub-clusters in the second sub-cluster are operated.
5. The bayesian framework learning based assembly line conceptual design method of claim 1, wherein the layout in the line plan has the following constraints: boundary constraint: ; ; ; ; Wherein% ) For the center coordinates of the device i, The width and length of the device i respectively, , The width and the length of the field are respectively, A safety distance coefficient for the device i; Anti-overlap constraint: ; ; ; ; ; ; Wherein the method comprises the steps of Indicating that device i is on the left side of device j, Indicating that device j is on the left side of device i, Indicating that device i is above device j, Indicating that device j is above device i; Size and orientation constraints: ; ; indicating the orientation of device i when When the value is 1, 0 degrees in the reference direction are indicated When the value is 1, 90 degrees in the reference direction is indicated, when When the value is 1, 180 degrees in the reference direction is indicated, when A value of 1 indicates 270 deg. in the reference direction, And The initial length and width of the device; ; ; Collinearly constrained transportation equipment: ; ; as a scale factor, the number of the elements is, Wherein , Central coordinates of the upstream device and the downstream device respectively; Orthogonal alignment constraints: ; If it is Then , ; If it is Then , ; Is a binary decision variable when When the value is 1, the upstream device u is aligned with the downstream device d in the horizontal direction, and when When the value is 0, the upstream device u is aligned with the downstream device d in the vertical direction, ; Distance calculation constraint: , , ; Wherein the method comprises the steps of , Representing the distance between device i and device j in width and length respectively, Representing the linear distance between device i and device j; Constraints on the layout in the production line scheme have the following objective functions: ; Wherein the method comprises the steps of For the distance deviation score to be a score, Score for time efficiency; ; , ; , ; , ; Wherein the method comprises the steps of Indicating the production rate of the upstream equipment, Representing upstream equipment transport speed and equipment transport capacity respectively, But not the distance between the upstream and downstream devices, Representing the ideal distance of the upstream device from the downstream device for the upstream device, Representing the ideal distance of the upstream device from the downstream device for the downstream device, 、 Respectively obtaining relative deviation of upstream equipment and downstream equipment through normalization; is a constant coefficient used for preventing denominator from being zero; ; ; ; Wherein the method comprises the steps of The bottleneck time of the system is that, Indicating the distance that the object carrying device is transported from the upstream device, Indicating the distance of transport of the object transport device from the downstream device, Indicating the distance the cargo has traveled from the upstream device to the downstream device, The transport speed of the material handling equipment, Indicating a reference time.
6. The concept design method of assembly line based on bayesian framework learning according to claim 4, wherein the step of extracting the process skeleton H corresponding to the corresponding category of the preliminary line is as follows: Finding out the set of processes in all production lines in the category, calculating the times of each process in each production line in the category, and marking the process as a process skeleton if the times of the processes are larger than a threshold value.
7. The concept design method of assembly line based on bayesian framework learning according to claim 3, wherein the step of performing data verification on the plurality of line schemes in step S3 is as follows: step S31, carrying out a quality cost analysis method on a plurality of production line schemes, obtaining the production line scheme meeting the cost condition as a first scheme, and if all the production line schemes are met, re-executing the step S2; Step S32, a generator function is greatly added to the first scheme to simulate a concurrency process, events are generated at discrete time points in the simulated concurrency process, simulation time is managed through environments, first time generated by products in the simulation time is obtained, the first scheme with the first time smaller than a time threshold is marked as a second scheme, and if the first time of all the first schemes is larger than the time threshold, the step S2 is executed again; Step S33, using the movement of the equipment in the second scheme of using the Liqun lie algebra to model, screening out the second scheme which can normally run, and if the second scheme can not normally run, re-executing the step S2; And step S34, inputting BYTwin codes of the third scheme to generate a large model, outputting executable codes, performing 3d operation simulation by utilizing BYTwin, obtaining the third scheme meeting the operation interference constraint in the 3d operation simulation, and if the third scheme does not meet the operation interference constraint as a final scheme, re-executing the step S2.
8. An assembly line concept design system based on Bayesian framework learning, which uses the assembly line concept design method based on Bayesian framework learning as claimed in any one of claims 1-7, and is characterized by comprising a scheme design module, a line design module and a screening module; the scheme design module is used for constructing a knowledge graph of product design, performing fine adjustment on the large model by using knowledge graph data, and generating a product design scheme by using the adjusted large model; the production line design module is used for inputting design requirements into the adjusted large model to generate a product design scheme; responding to the product design scheme, and adopting a Bayesian framework to produce a plurality of production line schemes meeting production; the screening module performs data verification on the multiple production line schemes, judges whether the production line schemes meet preset conditions, outputs the production line schemes meeting the preset conditions if the production line schemes exist, and re-invokes the production line design module if the production line schemes do not meet the preset conditions as a final scheme.

Description

Assembly line concept design method and system based on Bayesian framework learning Technical Field The invention relates to the technical field of production line design, in particular to an assembly production line concept design method and system based on Bayesian framework learning. Background The conceptual design process of the assembly production line is highly dependent on personal experience of engineers, when the experience of the engineers has field limitation, the traditional thinking framework is difficult to break through, and innovative design thought is lacking, so that the productivity of the production line is improved and the bottleneck is met. The system has a knowledge graph, covers a large amount of production line data, and can realize the advantage fusion of various fields during design. The existing design method excessively depends on experience of engineers, and many design defects appear due to experience deficiency when products in the brand new field are encountered. The BPL production line design is designed by learning the rules of the production line, and when the novel field products are faced, only a small number of historical cases are needed, so that the high-quality production line can be generated. The traditional design method is difficult to effectively cope with various uncertainty factors including equipment failure rate, equipment connection interference, material supply fluctuation and the like, and the influence of the factors on the overall efficiency of the production line often cannot be fully evaluated in the design stage. The system considers a plurality of constraints such as motion constraint and the like in design, and ensures the design quality. The traditional simulation flow has low efficiency, design parameters are required to be set manually according to a design scheme, simulation is operated and a result is analyzed, the whole process is time-consuming and labor-consuming, and the design iteration speed is seriously influenced. The system generates the software script by using the large model, thereby greatly reducing the simulation time. The simulation software function on the market at present has limitation, and most of the simulation software can only evaluate basic indexes such as throughput rate, cost and the like, lacks comprehensive simulation on the actual running condition of the production line, and cannot effectively predict various problems possibly occurring during running. Disclosure of Invention Aiming at the defects, the invention aims to provide an assembly production line concept design method and system based on Bayesian framework learning, which solve various problems possibly occurring when operation cannot be effectively predicted. The invention adopts the following technical scheme that the assembly line concept design method based on Bayesian framework learning comprises the following steps: Step S1, constructing a knowledge graph of product design, performing fine adjustment on a large model by using knowledge graph data, and generating a product design scheme by using the adjusted large model; s2, inputting design requirements into the adjusted large model to generate a product design scheme; responding to the product design scheme, and adopting a Bayesian framework to produce a plurality of production line schemes meeting production; and step S3, carrying out data verification on the multiple production line schemes, judging whether the production line schemes meet the preset conditions, outputting the production line schemes meeting the preset conditions if the production line schemes exist, and if the production line schemes do not meet the preset conditions, executing the step S2 again. Preferably, the step in the step S1 is as follows: Acquiring historical data of a process and equipment from a production line corresponding to a product, identifying association rules and frequent patterns among different data sources in the historical data of the process by a data mining technology, mapping heterogeneous data to a public semantic space by combining an ontology mapping method, and constructing a knowledge graph according to a knowledge graph construction method; injecting the knowledge graph into a large model; Inputting the design requirement into a question-answering system of a large-value model, and obtaining design parameters and a 3d modeling code; and determining whether the product design requirement is met based on the design parameters and the 3d modeling code, if so, outputting a product design scheme, and if not, inputting the design requirement into the large model again. Preferably, the implementation step of the step S2 is as follows: step S21, collecting process flow data of a historical production line as a background set; S22, dividing all mobile phone schemes into a plurality of categories by adopting a UniForce single-mode forest method for design parameters in each historical production line desi