Preference-adjusted semi-automatic trade space search method, apparatus, and medium

Abstract

The application discloses a semi-automatic trade-off space searching method and device based on preference adjustment and a medium, and comprises the following steps: S1, acquiring all design schemes of an online visual detection system architecture and forming an initial trade-off space; S2, coloring the initial trade-off space to form a colored trade-off space; S3, coloring the Pareto front optimal solution or the Pareto front of a specific level in the colored trade-off space to form a colored Pareto front trade-off space; and S4, navigating the colored Pareto front trade-off space by using an utility constraint region and a cost constraint region and using a pole distance method to obtain an optimal design scheme. The application proposes a process of giving a system architecture suggestion solution set by simulation process data driving on the basis of a colored segmented Pareto front preference set, in combination with constraint conditions and the pole distance method for scale reduction and directional screening of the Pareto front scheme, and aims to strengthen the decision quality in the concept design stage of a high-quality online visual system.

Description

Technical Field

[0001] This invention relates to the field of strategy optimization technology, and more specifically, to a semi-automatic trade-off space search method, apparatus, and medium based on preference adjustment. Background Technology

[0002] In today's fast-paced technological landscape, rapid system iteration has become the norm, accelerating innovation while also bringing challenges. Complex system projects often suffer from high failure rates due to initial design flaws, with the effectiveness of conceptual design being particularly prominent in intelligent manufacturing systems. As industries move towards high-quality development, enterprise product manufacturing models are gradually shifting from large-scale mass production to small-scale batch production. Design flaws leading to frequent mass production failures often result in a large number of defective products and lengthy optimization times. In the product manufacturing process, online visual inspection systems serve as the last line of defense for quality control, playing a crucial role. For specific products, building an online visual inspection system that achieves good inspection results, stable operation, high accuracy, and low cost has become an urgent need for many enterprises. During the conceptual design phase, such systems need to consider multiple performance indicators, such as false negative rate and false positive rate. Optimizing these indicators across multiple objectives generates Pareto solutions. However, as the number of performance objectives considered increases, the size of the Pareto solution set expands dramatically. Simultaneously, physical verification resources, such as the number of hardware devices used for testing, testing time, manpower, and funding, have limitations, making the contradiction between these two factors extremely acute. To resolve this technical contradiction, during system construction, advanced algorithms can be used for optimization. For example, intelligent algorithms can be employed to weigh multiple objectives, focusing on solutions that better align with actual production constraints and needs when searching for Pareto solutions, thus reducing the generation of unnecessary solutions. On the other hand, simulation technology can be used to model the physical verification process, using higher-precision experimental verification data to pre-screen obviously unsuitable solutions, reducing reliance on real physical verification resources. This allows for the efficient configuration of a high-quality online visual inspection system architecture that meets specific product requirements within limited resource constraints. Our research has identified existing problems in Pareto front trade-off analysis and proposes a method based on colorimetric classification Pareto fronts. This method allows for real-time calculation based on parameter adjustments, redrawing the trade-off space during the trade-off process, and significantly reducing the number of solutions by combining extreme points and simulation data-driven approaches. This ensures that the system conceptual design aligns with the decision-maker's value proposition during the trade-off process, while also reducing the risk of project failure through human-machine intelligent bidirectional verification, preserving core technologies, and significantly reducing verification costs and improving verification efficiency before formally entering physical verification.

[0003] In multi-attribute decision-making problems, decision-makers typically need to weigh trade-offs across multiple dimensions, such as cost, utility, reliability, and environmental impact. These attributes often have different units, dimensions, and even properties. A multi-attribute trade-off space is precisely such a mathematical abstraction space, representing all possible solutions in vector form. Each vector component corresponds to a value of an evaluation attribute, and the goal of the decision is to find the optimal set of points that satisfy a certain optimization criterion. The competitive priorities of stakeholders are a key driver of these trade-offs, as stakeholders have very different perspectives on the trade-offs regarding task performance. While introducing diversity, this also brings challenges such as large-scale trade-offs and complex exploration processes.

[0004] Current research largely focuses on obtaining accurate Pareto fronts, while discussions on how to derive suggested solutions from Pareto fronts through trade-offs remain limited. Existing algorithms like NSGA-II may discard ideal solutions during crowding calculations, NSGA-III fails to address the issue of reasonable reduction of solutions in the trade-off space, and methods such as TOPSIS lack a matching and reasonable trade-off process. Furthermore, none of these methods consider incorporating decision-maker preferences, model design, and simulation data into the trade-off system. The decision-making and judgment process must always reflect the value of human intervention throughout.

[0005] Therefore, there is an urgent need for a semi-automatic trade-off space search method based on preference adjustment to address the problem that existing technologies generate too many suggested solutions for multi-objective optimization problems, which leads to significant time costs in deciding which designs to select for further testing or as the final design, and carries the risk of project failure, resulting in economic losses. Summary of the Invention

[0006] To address the aforementioned technical problems in related technologies, this invention proposes a semi-automatic tradeoff space search method based on preference adjustment, comprising the following steps:

[0007] S1, Obtain all design schemes of the online visual inspection system architecture, and use a multi-attribute utility function to measure all design schemes to form an initial trade-off space;

[0008] S2. Color the initial trade-off space to form a colored trade-off space; the coloring is to use different colors to classify and label the points in the initial trade-off space, with each color representing different attributes, design variables and simulation process data;

[0009] S3, in the coloring trade-off space, the optimal solution of the Pareto front or the Pareto front of a specific level is colored to form the coloring Pareto front trade-off space.

[0010] S4. Navigate the colored Pareto front trade-off space through the utility constraint region and the cost constraint region, and use the extreme distance method to obtain the optimal design scheme.

[0011] Specifically, the multi-attribute utility function is a cost-utility function.

[0012] Specifically, it also includes step S5, which further weighs the cost-effectiveness of different design schemes during the simulation process to generate a suggested design scheme.

[0013] Specifically, step S1 includes: obtaining multiple key capability attributes of the simulation scheme, and converting the multiple key capability attributes into a single utility evaluation according to the decision-maker's preferences; obtaining key design variables based on parameter space analysis in the simulation scheme, sampling the key design variables to form a design scheme, and then modeling the design scheme to obtain the cost and utility of each design scheme and obtain the initial trade-off space.

[0014] Specifically, step S2 involves coloring the design variables according to their different sampled values.

[0015] Specifically, step S3 involves: during the trade-off process, a trade-off is made based on the decision-maker's preferences regarding the Pareto frontier set.

[0016]

[0017] in For the m-th sampled value of the k-th design variable under the i-th Pareto front, The set of preferences is Pareto optimal by default (i = 1), where P indicates a clear preference in the trade-off process, NP indicates a negative preference, and N indicates no clear preference. The generalized preference set after weighing the options, i.e., the solution set that simultaneously contains P and N, is:

[0018]

[0019] PF i,p It is the set of generalized preference solutions after weighing the options.

[0020] Specifically, step S4 involves setting cost and utility constraints as a specific constraint region.

[0021]

[0022] Among them U min C is the lowest acceptable utility value. max and C minThese represent the highest and lowest acceptable cost values, respectively.

[0023]

[0024] Specifically, step S4 involves finding an optimal extreme point in the trade-off space that is optimal in terms of both cost and utility. in, and These represent the cost and utility values ​​of the poles, respectively; there is a global pole value z in the overall trade-off space. * = (0,1), the local pole value of its constrained region is in To constrain the cost value represented by the minimum boundary of the region cost, For schemes x that are in the constrained region and have the same Pareto front level i =(c i ,u i Its distance to the pole is:

[0025]

[0026] Find Afterwards, according to The optimal design scheme is determined by the distance.

[0027] Secondly, another embodiment of the present invention provides a semi-automatic trade-off space search device based on preference adjustment, comprising the following units:

[0028] The initial trade-off space acquisition unit is used to acquire all design schemes of the online visual inspection system architecture. The initial trade-off space is formed by measuring all design schemes using a multi-attribute utility function.

[0029] A coloring trade-off space acquisition unit is used to color the initial trade-off space to form a coloring trade-off space; the coloring is to use different colors to classify and label the points in the initial trade-off space, with each color representing different attributes, design variables and simulation process data;

[0030] The coloring Pareto front trade-off space acquisition unit is used to color the optimal solution of the Pareto front or the Pareto front of a specific level in the coloring trade-off space to form the coloring Pareto front trade-off space.

[0031] The optimal design scheme acquisition unit is used to navigate the colored Pareto front trade-off space through the utility constraint region and the cost constraint region, and using the extreme point distance method, in order to obtain the optimal design scheme.

[0032] Thirdly, another embodiment of the present invention provides a non-volatile storage medium storing instructions that, when executed, implement the aforementioned semi-automatic tradeoff space search method based on preference adjustment. Simultaneously, a display device displays the result of the instruction execution to the user. The user can use the displayed result to make a new adjustment instruction based on their own value judgment. The system recalculates based on the new adjustment instruction and feeds it back to the user. After multiple feedback loops, the user obtains a final satisfactory result.

[0033] This invention proposes a process that uses a colored segmented Pareto front preference set as a basis, combines constraints and the extreme point distance method to reduce the size and target the Pareto front schemes, and then provides a system architecture suggestion solution set through a data-driven approach during the simulation process. This aims to enhance the decision-making quality in the system conceptual design stage.

[0034] Furthermore, this application, based on the Pareto front for colorimetric classification and incorporating decision-makers' preferences, fully evaluates design data and simulation process data. By combining extreme points and simulation process data-driven approaches, the number of solutions is reduced to a smaller set of recommended solutions, saving time and costs. It can also provide a better technical solution more quickly and more effectively evaluate the ability of the online visual inspection system architecture to perform tasks, thereby reducing the risk of the task to some extent. Attached Figure Description

[0035] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0036] Figure 1 This is a flowchart of the semi-automatic trade-off space search method based on preference adjustment provided in an embodiment of the present invention;

[0037] Figure 2 This is a schematic diagram of the trade-off space exploration process provided in an embodiment of the present invention;

[0038] Figure 3 This is a schematic diagram of the initial trade-off space provided in an embodiment of the present invention;

[0039] Figure 4 This is a schematic diagram of the coloring trade-off space of the online machine vision inspection system architecture provided in this embodiment of the invention;

[0040] Figure 5 This is a schematic diagram of the color segmented Pareto front scheme set for the light source provided in the embodiments of the present invention;

[0041] Figure 6 This is a schematic diagram of the Pareto front region provided in an embodiment of the present invention;

[0042] Figure 7 This is a schematic diagram of the Pareto front scheme set within the constrained region provided in an embodiment of the present invention;

[0043] Figure 8 This is a schematic diagram of a semi-automatic trade-off space search device based on preference adjustment provided in an embodiment of the present invention. Detailed Implementation

[0044] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention are within the scope of protection of the present invention.

[0045] Example 1

[0046] refer to Figure 1 This embodiment discloses a semi-automatic tradeoff space search method based on preference adjustment, which includes the following steps.

[0047] S1, Obtain all design schemes of the online visual inspection system, and use a multi-attribute utility function to measure all design schemes to form an initial trade-off space;

[0048] refer to Figure 2 The overall framework for exploring the trade-off space is as follows: Figure 2 As shown, it is divided into four stages: extracting value propositions, generating concepts, system performance modeling, and exploring trade-off spaces. Analysts can integrate this structured process during the concept design stage.

[0049] Phase 1: Introducing the Value Proposition

[0050] Based on the mission and tasks, clarify and quantify the decision-maker's vision, goals, system boundaries, constraints, and system requirements under normal and critical states for the proposed system, define the decision-maker's value proposition, and initialize the model framework.

[0051] (1) State the mission and tasks: Through the human-computer dialogue window of text or visual model, understand the design requirements for executing the mission and complete the performance evaluation of the system architecture design based on capability attributes. For tasks in specific domains, other supporting tools or models can be used to complete the task.

[0052] (2) Identify the decision-maker: referred to as the stakeholder.

[0053] (3) Define a multi-attribute utility function: Based on understanding the preference ranking formula, and taking the decision-maker's value orientation as the basis, define the system's capability attributes. The capability attributes must meet the characteristics of completeness, operability, decomposability, non-redundancy, minimum and perceived independence. The constituent elements are the definition scope, measurement unit, weight factor and preference satisfaction, etc. Among them, preference satisfaction is presented as a 0-1 dimensionless monotonic piecewise preference utility function. All elements can be obtained through formal interviews with decision-makers or human-computer interaction.

[0054] In one embodiment, the multi-attribute utility function is a cost-utility function;

[0055] Phase 2: Concept Generation

[0056] In this phase, designers or engineers complete the conceptual design of the system by linking the system concept to the decision-maker's value proposition identified in Phase 1.

[0057] (1) Identify constraints: Constraints may come from physical laws, environment, or policies, and are conditions that the system must meet to operate. Some constraints that may change can significantly alter the "optimal" result for a particular problem.

[0058] (2) Propose Design Variables: When considering solutions for obtaining capability attributes, designers need to examine the capability attributes and propose various design variables, providing their relevant ranges and enumeration values. Design variables are quantitative parameters controlled by the designer, reflecting one aspect of a concept. These concepts combine to form a set that uniquely defines the system architecture. Each combination of design variables uniquely constitutes a specific design vector, also known as a design scheme. The set of all possible design schemes constitutes the design space.

[0059] (3) Mapping design variables to capability attributes: System conceptual design needs to reflect the decision-maker's value proposition, which is generally accomplished through experience. As the number of design variables increases, the design space grows exponentially. If computational resources are limited, design variables need to be pre-screened. This process can generally be done using the quantitative analysis tool Analytic Hierarchy Process (AHP).

[0060] (4) Determine the baseline design vector: The concept generation phase is completed after determining the range and step size of the design variables. For a given design variable, whether discrete or continuous, the choice of the step size can be divided into an enumeration phase and a sampling phase. In the enumeration phase, a "complete" range of values ​​needs to be given in order to drive the design variables within a large range. If computational resources are limited, a subset of values ​​can be selected from the enumeration range in the sampling phase for use in the simulation process and trade-off space analysis.

[0061] Phase 3: System Performance Modeling

[0062] To understand how the system meets the decision-maker's needs in the operating environment, designers need to model the design schemes in a physics-based parametric model and calculate the lifecycle cost and design utility of each design scheme.

[0063] (1) Develop simulation software architecture

[0064] Developing software architecture requires an N 2 The matrix describes how intermediate variables are used to map design variables to capability attributes, and this mapping needs to be performed with higher accuracy. The modules in the matrix allow the model to be decomposed to facilitate subsequent parallel development and validation.

[0065] (2) Link design vectors to capability attributes through simulation models.

[0066] Run the simulation model, input instantiated sample data into the model, and calculate the set of capability attribute values ​​for each design vector in the tradeoff space. Due to the geometrical growth of the tradeoff space, multidisciplinary optimization techniques may be needed to replace full factorial sampling.

[0067] (3) Convert design vectors into lifecycle costs

[0068] In addition to linking design variables to capability attributes, the model also needs to transform design variables into moderately accurate estimates of lifecycle costs. Developing cost models during the conceptual design phase of complex systems is a challenge, and parametric cost estimation methods are the best choice for conceptual design under time constraints when known physical, technical, and performance parameters can be correlated with costs.

[0069] (4) Applying multi-attribute utility functions

[0070] The utility of each option is calculated using a utility function, and the decision-maker's satisfaction with the design is comprehensively evaluated.

[0071] Phase 4: Balancing Space Exploration

[0072] 4.1 Coloring and Trade-offs of the Entire Trade-off Space

[0073] Trade-off space exploration differs from traditional cost-benefit-based trade-off methods. Its aim is to map decision-makers' preferences in the value domain to possible trade-off spaces in the technology domain. Designers should first spend time studying the entire trade-off space, mapping sampled values ​​of a design variable for each architecture by changing the color of each point, thus forming a colored trade-off space. Solutions of the same color in the trade-off space are clusters. The number of cost or utility solutions within a standard statistical interval can be used as a basic evaluation tool. Through cluster analysis of solution costs and utility, the study assesses how these values ​​form and influence the trade-off space, further understanding how each design variable drives the system architecture.

[0074] 4.2 Pareto Front Shading Segmentation and Trade-offs

[0075] In the tradeoff space, the optimal solution of the Pareto front or the Pareto front of a specific level is colored to form the colored Pareto front tradeoff space.

[0076] Let F(x)={U(x),C(x)}, x={x1,x2,…,x m Let $\frac{ ... A and x B If and only if the following conditions are met:

[0077]

[0078] Then the solution x is called the solution. A Dominate x B ,Right now If no solution in the current solution set dominates solution x, then x is called a Pareto optimal solution, also known as a non-dominated solution. All Pareto optimal solutions in the decision space constitute the Pareto optimal solution set. The Pareto front is a mapping of Pareto optimal solutions to the tradeoff space. The Pareto optimal solution set is a set of mutually non-dominated solutions, and no single solution is superior to all other solutions on all objective functions. Typically, decision-makers aim to determine the frontier of a Pareto optimal design or a solution "sufficiently close" to the Pareto front. These "sufficiently close" optimal solutions can be considered "suboptimal" solutions in the Pareto sense. A "suboptimal" solution is a new Pareto optimal frontier obtained by removing solutions from the Pareto optimal frontier. Pareto optimal solutions at different levels are distinguished by rank. Solutions with rank = 1 and rank = 2 are "suboptimal" solutions, except for those with rank = 1, which cannot dominate solutions with rank = 2. During the tradeoff process, the Pareto front set is weighed according to the decision-maker's preferences.

[0079]

[0080] in For the m-th sampled value of the k-th design variable under the i-th Pareto front, The set of preferences is Pareto optimal by default (i = 1), where P indicates a clear preference in the trade-off process, NP indicates a negative preference, and N indicates no clear preference. In the trade-off space, different colors are used to represent the components, and the shape is roughly segmented; therefore, it is called the colored segmented Pareto front. The set of generalized preferences after the trade-off, that is, the solution set that simultaneously contains P and N, is...

[0081]

[0082] PF i,p It is the set of generalized preference solutions after weighing the options.

[0083] 4.3 Pole Scheme Selection Based on Constraint Region

[0084] The colored Pareto front trade-off space may still contain many "optimal" or "near-optimal" solutions. To further narrow down the trade-offs, decision-makers can navigate the trade-off space by examining highly efficient and low-cost Pareto front designs to discover potentially valuable candidate designs. Cost and utility constraints can be set as specific constraint areas.

[0085]

[0086] Among them U min C is the lowest acceptable utility value. max and C min These represent the highest and lowest acceptable cost values, respectively.

[0087]

[0088] In the trade-off space, there exists a pole z that is optimal in terms of both cost and utility. * , in, and Let z be the cost and utility values ​​of the pole, respectively. There is a global pole value z in the overall trade-off space. * = (0,1), the local pole value of its constrained region is in To constrain the cost value represented by the minimum boundary of the region cost, For schemes x that are in the constrained region and have the same Pareto front level i =(c i ,u i Its distance to the pole is:

[0089]

[0090] Find Subsequently, it is generally believed that the closer the solution is to the poles of the constrained region, the better. It is intended only as a tool to narrow down the range of trade-off options and is not intended as the basis for the final ranking of options.

[0091] 4.4 Scheme Trade-offs Based on Simulation Process Data-Driven Approach

[0092] Pareto front coloring and piecewise trade-offs, as well as constraint-based region selection, are both trade-offs based on cost and utility weighted data, used to significantly reduce the range of options to be traded off. However, after reducing the range to a certain extent, it is insufficient to use only the final cost and utility results for trade-offs; it is necessary to further assist in the trade-off of options in the trade-off space by mining intermediate data from the model simulation process. Introducing intermediate data from the simulation model in this step is to avoid getting too bogged down in detailed data when there are large-scale options in the early stages of the trade-off. Taking the extraction of capability attributes and cost composition during the simulation process as an example, see Equation 9, U x and C x The values ​​represent the utility and cost of the proposed solution, respectively, and simulation process data is provided. and C j , C represents the contribution of the i-th capability attribute to the utility of the current solution. j This represents the j-th component of the cost of the proposed solution. Furthermore, the utility of capability attributes can be penetrated to different design variables through simulation models.

[0093]

[0094] By using simulation process data-driven analysis, the main components of the cost and utility differences between different schemes can be analyzed, not limited to formula (9), and after further weighing and screening, the final recommended solution set can be obtained.

[0095] S2, color the initial trade-off space to form a colored trade-off space; the coloring is to use different colors to classify and label the points in the initial trade-off space, with each color representing different attributes, design variables and simulation process data;

[0096] Specifically, in step S2, the design variables can be colored according to their different sampled values; each design variable can form a complete coloring classification trade-off space.

[0097] In one implementation, an example of a trade-off study of the architecture of an online machine vision inspection system is given.

[0098] The first stage involves identifying several key capability attributes of primary concern to the system. In one embodiment, this can be used in the architecture of an online machine vision inspection system. These key capability attributes include false negative rate, false positive rate, detection cycle time, equipment stability, deployment flexibility, and image processing latency. Based on the decision-maker's preferences, these capability attributes are transformed into single utility evaluation criteria, forming a mapping function from capability attributes to utility to quantify task benefits. The second stage involves exploring a wide range of possible parameter spaces and using a design value mapping matrix to evaluate the contribution of each design variable to the task. Through comprehensive analysis, six design variables with high overall impact are ultimately selected as the final design vectors for achieving the task objectives: optical resolution, light source configuration, algorithm configuration, number of parallel cameras, computing architecture, and communication protocol stack. Each design variable is then sampled to construct an architecture scheme set. In the third stage, a physical model is developed using Matlab to calculate the lifecycle cost and design utility of different architecture schemes, ultimately generating a trade-off space.

[0099] Figure 3 The diagram illustrates the trade-off space formed by all the proposed solutions, plotted with cost on the X-axis and utility on the Y-axis, with each point representing a specific solution. This allows for a visual observation and analysis of the distribution of all design options within the trade-off space.

[0100] The trade-off space is colored according to different sampled values ​​of the design variables. Each design variable can form a complete colored classification trade-off space. Figure 4 The study demonstrates the trade-off space for coloring using sampled light sources such as ring light, coaxial light, dome light, and mixed light. It can be found that the mixed light source scheme has the highest cost. The cost value of this scheme is only for illustration. The costs of other types are relatively low. It can be considered that the type of light source scheme is one of the main drivers of cost. The same trade-off can be made for other design variables.

[0101] S3, in the coloring trade-off space, the optimal solution of the Pareto front or the Pareto front of a specific level is colored to form the coloring Pareto front trade-off space.

[0102] Let F(x)={U(x),C(x)}, x={x1,x2,…,x m Let $\frac{ ... A and x B If and only if the following conditions are met:

[0103]

[0104] Then the solution x is called the solution.A Dominate x B ,Right now If no solution in the current solution set dominates solution x, then x is called a Pareto optimal solution, also known as a non-dominated solution. All Pareto optimal solutions in the decision space constitute the Pareto optimal solution set. The Pareto front is a mapping of Pareto optimal solutions to the tradeoff space. The Pareto optimal solution set is a set of mutually non-dominated solutions, and no single solution is superior to all other solutions on all objective functions. Typically, decision-makers aim to determine the frontier of a Pareto optimal design or a solution "sufficiently close" to the Pareto front. These "sufficiently close" optimal solutions can be considered "suboptimal" solutions in the Pareto sense. A "suboptimal" solution is a new Pareto optimal frontier obtained by removing solutions from the Pareto optimal frontier. Pareto optimal solutions at different levels are distinguished by rank. Solutions with rank = 1 and rank = 2 are "suboptimal" solutions, except for those with rank = 1, which cannot dominate solutions with rank = 2. During the tradeoff process, the Pareto front set is weighed according to the decision-maker's preferences.

[0105]

[0106] in For the m-th sampled value of the k-th design variable under the i-th Pareto front, The set of preferences is Pareto optimal by default (i = 1), where P indicates a clear preference in the trade-off process, NP indicates a negative preference, and N indicates no clear preference. In the trade-off space, different colors are used to represent the components, and the shape is roughly segmented; therefore, it is called the colored segmented Pareto front. The set of generalized preferences after the trade-off, that is, the solution set that simultaneously contains P and N, is...

[0107]

[0108] PF i,p It is the set of generalized preference solutions after weighing the options.

[0109] S4. Navigate the colored Pareto front trade-off space through the utility constraint region and the cost constraint region, and use poles to obtain the optimal design solution.

[0110] The colored Pareto front trade-off space may still contain many "optimal" or "near-optimal" solutions. To further narrow down the trade-offs, decision-makers can navigate the trade-off space by examining highly efficient and low-cost Pareto front designs to discover potentially valuable candidate designs. Cost and utility constraints can be set as specific constraint areas.

[0111]

[0112] Among them U min C is the lowest acceptable utility value. max and C min These represent the highest and lowest acceptable cost values, respectively.

[0113]

[0114] In the trade-off space, there exists a pole z that is optimal in terms of both cost and utility. * , in, and Let z be the cost and utility values ​​of the pole, respectively. There is a global pole value z in the overall trade-off space. * = (0,1), the local pole value of its constrained region is in To constrain the cost value represented by the minimum boundary of the region cost, For schemes x that are in the constrained region and have the same Pareto front level i =(c i ,u i Its distance to the pole is:

[0115]

[0116] Find Subsequently, it is generally believed that the closer the solution is to the poles of the constrained region, the better. It is intended only as a tool to narrow down the range of trade-off options and is not intended as the basis for the final ranking of options.

[0117] This embodiment addresses the problem of too many options in the system architecture trade-off space, leading to difficulties in making trade-offs. It proposes a method based on a colored segmented Pareto front preference set, combined with constraints and the extreme point distance method, to reduce the size of Pareto front options and perform targeted screening, aiming to enhance the decision-making quality in the system conceptual design stage.

[0118] Furthermore, this embodiment also includes step S5, which further weighs the cost-effectiveness of different design schemes during the simulation process to generate a suggested design scheme;

[0119] Pareto front coloring and piecewise trade-offs, as well as constraint-based region selection, are both trade-offs based on cost and utility weighted data, used to significantly reduce the range of options to be traded off. However, after reducing the range to a certain extent, it is insufficient to use only the final cost and utility results for trade-offs; it is necessary to further assist in the trade-off of options in the trade-off space by mining intermediate data from the model simulation process. Introducing intermediate data from the simulation model in this step is to avoid getting too bogged down in detailed data when there are large-scale options in the early stages of the trade-off. Taking the extraction of capability attributes and cost composition during the simulation process as an example, see Equation 9, U x and C x The values ​​represent the utility and cost of the proposed solution, respectively, and simulation process data is provided. and C j , C represents the contribution of the i-th capability attribute to the utility of the current solution. j This represents the j-th component of the cost of the proposed solution. Furthermore, the utility of capability attributes can be penetrated to different design variables through simulation models.

[0120]

[0121] By using simulation process data-driven analysis, the main components of the cost and utility differences between different schemes can be analyzed, not limited to formula (9), and after further weighing and screening, the final recommended solution set can be obtained.

[0122] By providing a system architecture suggestion solution through simulation process data-driven approach, the scope of solutions to be weighed can be further reduced, solving the technical problem that the final result data of cost and utility alone cannot be used to weigh and obtain the optimal design solution.

[0123] Example 2

[0124] This embodiment provides a semi-automatic trade-off space search device based on preference adjustment, including the following units:

[0125] The initial trade-off space acquisition unit is used to acquire all design schemes of the online visual inspection system architecture. The initial trade-off space is formed by measuring all design schemes using a multi-attribute utility function.

[0126] A coloring trade-off space acquisition unit is used to color the initial trade-off space to form a coloring trade-off space; the coloring is to use different colors to classify and label the points in the initial trade-off space, with each color representing different attributes, design variables and simulation process data;

[0127] The coloring Pareto front trade-off space acquisition unit is used to color the optimal solution of the Pareto front or the Pareto front of a specific level in the coloring trade-off space to form the coloring Pareto front trade-off space.

[0128] The optimal design scheme acquisition unit is used to navigate the colored Pareto front trade-off space through the utility constraint region and the cost constraint region, and using the extreme point distance method, in order to obtain the optimal design scheme.

[0129] Example 3

[0130] This embodiment presents a trade-off study example of an intelligent online visual inspection system architecture to evaluate the capabilities of an intelligent online visual inspection system architecture for inspection tasks, aiming to validate the analysis process and guide future research on intelligent vision system architecture.

[0131] In the first phase, six key capability attributes were identified, including false negative rate, false positive rate, detection cycle time, equipment stability, deployment flexibility, and image processing latency.

[0132] The Attr1 capability attribute, the false negative rate, is defined as the probability that a defect will not be identified. It is measured by standard defect template testing, and its utility direction is minimization. The engineering value range is [0.005%, 5.0%].

[0133] The Attr2 misclassification rate is the probability that a normal part is misclassified as a defect. It needs to be minimized and its value ranges from 0.1% to 12.0%.

[0134] The Attr3 capability attribute is the detection cycle time, which is the number of parts detected per minute. It is counted by the production line running at full capacity, with the goal of maximizing it. The engineering range is [40,400] FPM.

[0135] The Attr4 capability attribute measures device stability by mean time between failures (MTBF), which should be maximized, with a value of [1,500,50,000] hours.

[0136] The Attr5 capability attribute deployment elasticity is reflected in the production line changeover and reconfiguration time, which needs to be minimized, with a range of [3, 180] points;

[0137] The image processing latency, a capability attribute (Attr6), is the time consumed in analyzing a single frame of image, i.e., the time difference from acquisition to output. It needs to be minimized and has a value of [0.1, 500] seconds. The above capability attributes are transformed into a single utility evaluation criterion, forming a mapping function from capability attributes to utility, which is used to quantify the user's task benefits.

[0138] Assume the normalized weights of Attr1 to Attr6 are 0.11, 0.056, 0.11, 0.5, 0.167, and 0.056, respectively.

[0139] In the second phase, six design variables with significant overall impact were selected as the final design vectors for achieving the task objectives. These included optical resolution, light source configuration, algorithm configuration, number of parallel cameras, computing architecture, and communication protocol stack.

[0140] The DV1 optical resolution engineering value range is 5-50μm, and the sampling value is assumed to be [5,20,50].

[0141] The DV2 light source configuration options include ring light, coaxial light, dome light, and mixed light. Assuming the sampling value is [ring light, coaxial light, dome light, mixed light];

[0142] The configuration of the DV3 algorithm involves model architecture, confidence threshold and post-processing strategy. It is limited by computing power and memory boundaries. Assume that the sampled values ​​are [SSD, YOLOV11, Transformer, hybrid architecture];

[0143] The number of parallel cameras in DV4 ranges from 1 to 8, assuming the sampling value is [1, 2, 4, 8].

[0144] The DV5 computing architecture can be selected from X86, FPGA, GPU, and ASIC, assuming the sampled value is [X86, FPGA, GPU, ASIC].

[0145] The DV6 communication protocol includes EtherNet / IP, EtherCAT, etc., assuming the sampled value is [EtherNet / IP, EtherCAT].

[0146] In the third stage, physical models are developed using Matlab or other programming languages ​​such as Python and C++ to calculate the lifecycle cost and design utility of different architectural schemes, ultimately generating a trade-off space. The cost and utility calculations for a typical scheme are given below. Table 1 lists some design data for the trade-off space, and Table 2 lists some simulation process data.

[0147] Table 1

[0148]

[0149]

[0150] Table 2 Example of Simulation Process Data

[0151]

[0152] In this embodiment, Figure 3-4The diagram illustrates the trade-off space formed by all proposed solutions, plotted with lifecycle cost on the X-axis and utility on the Y-axis, with each point representing a specific solution. This allows for a visual observation and analysis of the distribution of all design solutions within the trade-off space. The costs in this example are illustrative and used only to compare the cost of different design solutions.

[0153] The trade-off space is colored according to different sampled values ​​of the design variables. Each design variable can form a complete colored classification trade-off space. Figure 4 The study demonstrates the trade-off space for coloring based on sampled values ​​of light source configuration schemes [ring light, coaxial light, dome light, mixed light]. The mixed light configuration type has the highest cost, while the other types have relatively lower costs. It can be considered that the light source configuration is one of the main cost drivers. The same trade-off can be made for other design variables.

[0154] After a preliminary assessment of the complete coloring trade-off space, Pareto fronts need to be generated at different levels and classified and colored according to the sampled values ​​of different design variables. Each design variable can form a complete Pareto front coloring map. Figure 5 Using the design variable optical resolution DV1 sampling values ​​of 5μm, 20μm, and 50μm as an example, we color-classify the trade-off schemes with Pareto front Rank=1. We can see that a track height of 50μm offers better utility and cost advantages. After the trade-off, we can assign the preference for the 50μm scheme set on the Pareto front as P, and the preferences for 20μm and 5μm as NP. The same approach can be used for Pareto front trade-offs for other design variables.

[0155] Within the Pareto front of the entire tradeoff space, there are 122 satisfactory Pareto front solutions (47 Rank 1 and 45 Rank 2), which is still a relatively large number. Further constraints can be added, such as setting utility above 0.6 and cost below 700,000 yuan as constraints. Within this constraint region, there are 21 Rank 1 solutions and 19 Rank 2 reference solutions, reducing the number of solutions by 82. Within the region formed by these constraints, distance calculations are performed using (0,1) coordinates within the tradeoff space as pole coordinates to further reduce the number of solutions. Figure 6 This illustrates the distribution of schemes within the Pareto front constraint region in the overall trade-off space, as well as the location of the poles. Figure 6 The ideal point is the pole.

[0156] Coloring is a way of presenting simulation process data, its purpose being to introduce process data. In this example, coloring the design parameters is one manifestation of this. "Designers should first spend time studying the entire trade-off space, mapping the sampled value of a design variable for each architecture by changing the color of each point, thus forming a colored trade-off space." The actual result solution corresponding to each process design parameter ultimately reflects its value; the horizontal axis represents cost, and the vertical axis represents utility. (Reference...) Figures 3 to 7 In this embodiment, on the one hand, coloring is used to allow humans to make judgments, with the aim of incorporating human preferences; on the other hand, the system automatically calculates and combines the two to form a semi-automatic trade-off.

[0157] Figure 7 The constrained region was magnified, and seven particularly noteworthy designs were identified and marked for further study. Among them, solutions #733, #736, #697, #1516, and #700 were selected as the top five solutions in the constrained region after eliminating unfavorable solutions, based on the extreme distance method. Solution #529, being the lowest-cost solution in the constrained region, was selected, and solution #703, being the most effective solution, deserves further attention. By comparing simulation data among the selected solutions, trade-offs can be clearly made between them. Decision-makers, based on their own value orientation, can consider solutions #700, #1516, #733, #736, and #529, whose differences are already very small, as a suggested solution set for higher-precision modeling to make a final decision.

[0158] As can be seen, this application, based on the colorized segmented Pareto front preference set, combines constraints and the extreme point distance method to reduce the size and target Pareto front solutions, and then provides a system architecture suggestion solution set through a data-driven approach during simulation. This application, based on the colorized classification Pareto front and combining extreme points and simulation data-driven methods, reduces the number of solutions to a smaller set of suggested solutions, saving time and costs, and providing a better technical solution more quickly. It can more effectively evaluate the ability of the online visual inspection system to perform tasks, and significantly reduce the risk of project failure.

[0159] Example 4

[0160] refer to Figure 8 , Figure 8This embodiment presents a schematic diagram of a semi-automatic trade-off space search and display device based on preference adjustment. The semi-automatic trade-off space search device 20 based on preference adjustment in this embodiment includes a processor 21, a memory 22, and a computer program stored in the memory 22 and executable on the processor 21. When the processor 21 executes the computer program, it implements the steps in the above method embodiments. Alternatively, when the processor 21 executes the computer program, it implements the functions of each module / unit in the above device embodiments.

[0161] For example, the computer program can be divided into one or more modules / units, which are stored in the memory 22 and executed by the processor 21 to complete the present invention. The one or more modules / units can be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in the preference-adjusted semi-automatic tradeoff space search device 20. For example, the computer program can be divided into the modules shown in Embodiment 2. The specific functions of each module are described in the working process of the device described in the above embodiments, and will not be repeated here.

[0162] The preference-adjusted semi-automatic trade-off space search device 20 may include, but is not limited to, a processor 21, a memory 22, a display device 23, and a user operation device 24. Those skilled in the art will understand that the schematic diagram is merely an example of the preference-adjusted semi-automatic trade-off space search device 20 and does not constitute a limitation on it. It may include more or fewer components than illustrated, or combine certain components, or use different components. For example, the preference-adjusted semi-automatic trade-off space search device 20 may also include input / output devices, network access devices, buses, etc.

[0163] The processor 21 can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. The processor 21 is the control center of the preference-adjustment-based semi-automatic tradeoff space search device 20, connecting all parts of the preference-adjustment-based semi-automatic tradeoff space search device 20 via various interfaces and lines.

[0164] The memory 22 can be used to store the computer programs and / or modules. The processor 21 implements various functions of the preference-adjusted semi-automatic trade-off space search device 20 by running or executing the computer programs and / or modules stored in the memory 22 and calling the data stored in the memory 22. The memory 22 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory 22 may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0165] The main purpose of the display device 23 is to display the calculation results under a specific preference.

[0166] The user operation device 24 can be a touch operation or a mouse click operation, and its main purpose is to provide feedback on the user's preference selection results.

[0167] The modules / units integrated into the preference-adjustment-based semi-automatic tradeoff space search device 20, if implemented as software functional units and sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by the processor 21, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content contained in the computer-readable medium may be appropriately added to or subtracted from the content as required by the legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium may not include electrical carrier signals and telecommunication signals.

[0168] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0169] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A semi-automatic tradeoff space search method based on preference adjustment, characterized in that: Includes the following steps: S1, Obtain all design schemes of the online visual inspection system architecture. The all design schemes are measured using a multi-attribute utility function to form an initial trade-off space. The multi-attribute utility function is a cost-utility function. Step S1 includes: obtaining multiple key capability attributes of the simulation scheme and converting the multiple key capability attributes into a single utility evaluation according to the decision-maker's preferences; obtaining key design variables based on parameter space analysis in the simulation scheme, sampling the key design variables to form a design scheme, and then modeling the design scheme to obtain the cost and utility of each design scheme and obtain the initial trade-off space. S2, coloring the initial trade-off space to form a colored trade-off space; the coloring involves using different colors to classify and label the points in the initial trade-off space, with each color representing different attributes, design variables, and simulation process data; step S2 specifically involves coloring according to different sampled values ​​of the design variables. S3, in the colored tradeoff space, the optimal solution of the Pareto front or the Pareto front of a specific level is colored to form a colored Pareto front tradeoff space; specifically, step S3 involves: during the tradeoff process, the Pareto front set is weighed according to the decision-maker's preferences: ， in In the first Under the Pareto front, the first The first design variable Sample values The set of preferences is Pareto optimal by default, i.e. , This indicates that there is a clear preference in the weighing process. Indicates a preference for negation. This indicates that there is no explicit preference; by default... The set of generalized preferences after weighing the options, that is, simultaneously containing... , The solution set is : ， ， It is the set of generalized preference solutions after weighing the options; S4, navigating the colored Pareto front trade-off space using the utility constraint region and cost constraint region, and employing the extreme point distance method, to obtain the optimal design scheme; step S4 specifically involves: setting cost and utility constraints as specific constraint regions: ， in The minimum acceptable utility value, and These are the highest and lowest acceptable cost values, respectively. ， In the trade-off space, there exists a pole that is optimal in terms of both cost and utility. , ,in, and These represent the cost and utility values ​​of the extreme points, respectively; there is a global extreme value within the overall trade-off space. The local pole values ​​of its constrained region are ,in To constrain the cost value represented by the minimum boundary of the region cost, , For schemes within the constrained region that are at the same Pareto front level... Its distance to the pole is: ， Find Afterwards, according to The optimal design scheme is determined by the distance.

2. The method according to claim 1, characterized in that: It also includes step S5, which further weighs the cost-effectiveness of different design schemes during the simulation process to generate a suggested design scheme.

3. A semi-automatic trade-off space search device based on preference adjustment, characterized in that: Includes the following units: The initial trade-off space acquisition unit is used to acquire all design schemes of the online visual inspection system architecture. The entire design scheme is measured using a multi-attribute utility function to form an initial trade-off space. The multi-attribute utility function is a cost-utility function. Step S1 includes: acquiring multiple key capability attributes of the simulation scheme and converting the multiple key capability attributes into a single utility evaluation according to the decision-maker's preferences; acquiring key design variables based on parameter space analysis in the simulation scheme, sampling the key design variables to form a design scheme, and then modeling the design scheme to obtain the cost and utility of each design scheme, thereby acquiring the initial trade-off space. A coloring trade-off space acquisition unit is used to color the initial trade-off space to form a coloring trade-off space; the coloring involves using different colors to classify and label the points in the initial trade-off space, with each color representing a different attribute; the coloring is performed based on different sampled values ​​of the design variables. The coloring Pareto front trade-off space acquisition unit is used to color the optimal solution of the Pareto front or Pareto fronts of a specific level in the coloring trade-off space, forming a coloring Pareto front trade-off space; during the trade-off process, the set of Pareto fronts is traded according to the decision-maker's preferences: ， in In the first Under the Pareto front, the first The first design variable Sample values The set of preferences is Pareto optimal by default, i.e. , This indicates that there is a clear preference in the weighing process. Indicates a preference for negation. This indicates that there is no explicit preference; by default... The set of generalized preferences after weighing the options, that is, simultaneously containing... , The solution set is : ， ， It is the set of generalized preference solutions after weighing the options; The optimal design solution acquisition unit is used to navigate the colored Pareto front trade-off space through the utility constraint region and the cost constraint region, and using the extreme point distance method, to obtain the optimal design solution by setting cost and utility constraints as specific constraint regions: ， in The minimum acceptable utility value, and These are the highest and lowest acceptable cost values, respectively. ， In the trade-off space, there exists a pole that is optimal in terms of both cost and utility. , ,in, and These represent the cost and utility values ​​of the extreme points, respectively; there is a global extreme value within the overall trade-off space. The local pole values ​​of its constrained region are ,in To constrain the cost value represented by the minimum boundary of the region cost, , For schemes within the constrained region that are at the same Pareto front level... Its distance to the pole is: ， Find Afterwards, according to The optimal design scheme is determined by the distance.

4. A non-volatile storage medium storing instructions that, when executed, implement the semi-automatic trade-off space search method based on preference adjustment as described in any one of claims 1-2.

Images