JP4528962B2 - Design support method - Google Patents

Description

  The present invention relates to a method for supporting various designs.

  In recent years, the concept of Concurrent Engineering (CE) has been advocated, and it is necessary to process work in the product life cycle in parallel for the purpose of shortening lead time and improving quality such as safety. Sex is pointed out. This simultaneous and parallel does not simply mean that a plurality of processes are independently proceeding simultaneously, but means that each process advances its own processing while performing cooperative processing based on common data. At present, in the industry, a product development process that uses this technique to cope with diversification of products and high-mix low-volume production is becoming common.

  In the conventional design, regardless of whether it is CE or not, a design solution specified by a certain point value is obtained from the initial stage as shown in Fig. 1 (a), and whether the solution satisfies the required specifications and constraints is evaluated. However, if the evaluation result is inappropriate, a method is used in which correction is repeated until the solution reaches the design goal (this is referred to as point-based).

  However, in actual design, as the design progresses from conceptual design to detailed design, the obtained solution becomes inappropriate, and it is necessary to change the content that has already been determined due to the modification of the design solution and the effects of the modification. There is also. In the case of CE, there is a high possibility that other departments such as planning, development and production are working in parallel in addition to the design department, so a slight change may cause a large impact. As described above, the point-based method requires a lot of time because it is necessary to repeatedly correct the solution before the optimum solution is derived. In addition, the design specification itself may be changed due to a plan change or design change in the initial stage, or a request from the sales side of the product. In such a case, it is necessary to design again from the beginning, and the entire product development process may be prolonged.

  Recently, in order to improve this, as shown in Fig. 1 (b), a design solution is obtained as a wide set instead of a point value at the initial stage, and collaborative work is performed by each department based on the solution set. As the process progresses, a method called Set-Based Concurrent Engineering (SBCE) (see Non-Patent Documents 1-4 below) is proposed in which the design solution set is narrowed by gradually removing solution sets with poor feasibility. ing. Compared to a point-based approach, it takes a certain amount of time to solve a set of design solutions in the initial design stage, but in the later stages of design, the solution can be converged to the final manufacturing process in a shorter time. Overall, it's more efficient. Here, the set-based design is a method of obtaining a design solution as a wide set instead of a point value at an initial stage, and gradually narrowing the design solution set by excluding a solution set with poor feasibility.

  The characteristics and essence of each design method of non-CE type, CE type, and SBCE type will be described below. The conventional design process is a point-based non-CE type (Sequential Engineering, SE), and basically there is feedback from the downstream process after the design in the upstream process is completed, and the development period is long. Happens. In contrast, in CE, the contents of the downstream process are fed back from the initial design stage to the upstream process, and many requests are taken into consideration earlier, but essentially, as in the conventional design development flow, a single design solution is corrected. It will be repeated. For this reason, many corrections are always made in this method, aiming at the best design solution, and the development period may be further extended by applying CE. SBCE takes as wide a set of design solutions as possible at the initial stage of design, and collects a set of design solutions based on various information on the feasibility of design proposals through parallel and collaborative development between various departments. Gradually narrow. Although it seems that the efficiency is poor when looking at only individual departments, it can be said that it is a more efficient method because it does not invalidate the already determined contents as the whole system.

  In addition, the optimal solution at the detailed stage depends greatly on the selection of the design plan at the initial design stage. However, at the initial design stage, it is difficult to determine the exact required specifications, variations in manufacturing and assembly dimensions, changes in the usage environment, difficulty in selecting the best design proposal by the designer, and design solutions from other departments. There are various uncertainties such as the variability of. Therefore, in order to obtain an optimal and robust design solution from the viewpoint of performance, a design that considers various uncertainties is required. It can be said that SBCE is a method that can cope with such uncertainty.

  On the other hand, in the current design site, various analysis tools are often used for performance evaluation such as strength characteristics, vibration characteristics, collision characteristics, recyclability, drivability, and cost in order to confirm the validity of the design solution. I came. However, the human and time costs required for such an analysis are very high, and it is difficult to execute an analysis corresponding to all the required specifications and obtain an optimum solution satisfying the required specifications from the analysis results. To solve this problem, there are many metamodeling techniques that perform the minimum necessary analysis by design of experiments and derive an approximate calculation model representing the relationship between design variables and performance variables from the analysis results. It is used. Therefore, by using this approximate calculation model instead of actual analysis, it is possible to evaluate a plurality of performances quickly at a lower cost. In addition, the use of approximate calculation models is considered to be an effective method for realizing CE because of the advantage that a calculation model for multi-purpose / multi-domain design can be created by integrating individual analysis results.

Although SBCE is positioned as an effective method for realizing CE, a method that can be implemented on a computer has not been established. In addition, it supports uncertainties such as fuzzy set-based methods (the following non-patent documents 5-7), interval set-based methods (the following non-patent documents 8-10), probability-based methods (the following non-patent documents 11, 12), etc. Many researches on various set-based design methods and their computer support have been proposed, but there is still no method that can realize all the mechanisms of SBCE.
(1) Ward, A., JK Liker, JJ Cristiano, and DK Sobek II, "The Second Toyota Paradox: How Delaying Decisions Can Make Better Cars Faster", Sloan Management Review, 36 (3): 43-61, 1995. (2) Liker, JK, DK Sobek II, AC Ward, and JJ Cristiano, "Involving Suppliers in Product Development in the United States and Japan: Evidence for Set-Based Concurrent Engineering", IEEE Transactions on Engineering Management, 43 (2) : 165-178, 1996. (3) Sobek II, DK, AC Ward, "Principles from Toyota's Set-Based Concurrent Engineering", Proc. Of DETC'96, DETC96 / DTM-1510, Irvine, CA, August 18-22, 1996. (4) Sobek II, DK, AC Ward and JK Liker, "Toyota's Principles of Set-Based Concurrent Engineering", Sloan Management Review, 40 (2): 67-83, 1999. (5) Wood, KL, EK Antonsson, "Computations with Imprecise Parameters in Engineering Design: Background and Theory", ASME Journal of Mechanisms, Transmissions, and Automation in Design, 111 (4): 616-625, 1989. (6) Antonsson, EK, KN Otto, "Imprecision in Engineering Design", Transactions of the ASME Journal of Mechanical Design, 117 (B): 25-32, 1995. (7) Scott, MJ, EK Antonsson, "Preliminary Vehicle Structure Design: An Industrial Application of Imprecision in Engineering Design", Proc. Of DETC'98, DETC98 / DTM-5646, Atlanta, GA, September 13-16, 1998. (8) Ward, AC, T. Lozano-Perez and WP Seering, "Extending the Constraint Propagation of Intervals", Artificial Intelligence for Engineering Design, Analysis and Manufacturing, 4 (1): 47-54, 1990. (9) Finch, WW, AC Ward, "Quantified Relations: A Class of Predicate Logic Design Constraints among Sets of Manufacturing, Operating and Other Variables", Proc. Of DETC'96, DETC / DTM-1278, Irvine, CA, August 18-22, 1996. (10) Finch, WW, AC Ward, "A Set-Based System for Eliminating Infeasible Design in Engineering Problems Dominated by Uncertainty", Proc. Of DETC'97, DETC / DTM-3886, Sacramento, CA, September 14-17, 1997. (11) Chen, W., C. Yuan, "A Probabilistic-Based Design Model for Achieving Flexibility in Design", Transactions of the ASME Journal of Mechanical Design, 121 (1): 77-83, 1999. (12) Wallace, DR, MJ Jakiela and WC Flowers, "Design Search under Probabilistic Specifications using Genetic Algorithms", Computer-Aided Design, 28 (5): 405-421, 1996. (13) Zimmermann, H.-J., Fuzzy Set Theory and Its Applications, Kluwer Academic Publishers, 2001. (14) Luoh, L., W.-J.Wang, "A Modified Entropy for General Fuzzy Sets", International Journal of Fuzzy Systems, 2 (4): 300-304, 2000. (15) Terwiesch, C., A. De Meyer and CH Loch, "Exchanging Preliminary Information in Concurrent Engineering: Alternative Coordination Strategies", Organization Science, 13 (4): 402-419, 2002. (16) Scott, MJ, EK Antonsson, "Aggregation Functions for Engineering Design Trade-offs", Fuzzy Sets and Systems, 99 (3): 253-264, 1998. (17) Otto, KN, EK Antonsson, "Trade-Off Strategies in Engineering Design", Research in Engineering Design, 3 (2): 87-104, 1991. (18) Kusiak, A., J. Wang, "Dependency Analysis in Constraint Negotiation", IEEE Transactions on Systems, Man, and Cybernetics, 25 (9): 1301-1313, 1995.

  The present invention has been made in view of the above situation, and an object thereof is to provide a design support method that can be mounted on a computer and is effective for a set-based design method. Another object of the present invention is to provide a possibility distribution calculation method suitable for such a design support method.

The design support method according to the present invention includes the following steps:
(1) A step of setting a design variable range;
(2) dividing the design variable range;
(3) A step of calculating a preference level of required performance corresponding to the divided design variable range, possibility of performance value, and / or robustness. As will be described later in the embodiment, the required performance preference is, for example, p (x), the possibility of the performance value is, for example, q (x), and the robustness is, for example, PSI. However, it is not limited to these.

The design support method may further include the following steps;
(4) A step of excluding the corresponding divided design variable range from the object of design examination when any of the required performance preference, performance value possibility, or robustness is 0.

The calculation method of possibility distribution according to the present invention comprises the following steps:
(1) A step of calculating a calculation model indicating a probability distribution of performance values when the input preference is 0;
(2) A step of calculating a possibility distribution of performance values at an arbitrary input preference by interpolation using the calculation model. In these steps (1) and (2), the calculation model is, for example, a response surface, but is not limited thereto.

  According to the design support method of the present invention, it is possible to provide a design support method that can be mounted on a computer and is effective for a set-based design method. Further, according to the present invention, it is possible to provide a possibility distribution calculation method suitable for such a design support method.

  Hereinafter, an embodiment of the present invention will be described. In this description, first, basic items which are preconditions for explaining the present embodiment will be described. Thereafter, the present embodiment will be described using actual design examples (examples).

(Basic explanation to be assumed)
1. Set-Based Design Method Considering Preference 1.1 Design Solution Set for Representing Designer's Preference and Design Flexibility FIG. 2 shows a set-based design method in this embodiment. This is called PSD (Preference Set-Based Design) method. First, the designer prepares a calculation model that expresses the relationship between design variables and performance variables created by using some metamodeling technique (response phase method, neural network, inductive learning method, Kriging method, etc.). Next, in order to evaluate the target performance, the possibility range of the design variable and the performance variable is specified (step 2-1).

When considering a set of values to be taken as design variables, two types of sets are generally considered. Continuous and discrete. A set of continuous quantities has its upper and lower bounds defined as intervals (eg, noise is reduced to the range [15, 30] (dB)), or one of the upper and lower bounds is defined (eg, this component is at least requires a volume of 30 cm ^3). In the PSD method, a quantitative and intuitive expression method for expressing a designer's preference for a discrete quantity set using graph theory is possible, but in this embodiment, only a set-based design for a continuous quantity set is focused. To explain.

  As shown in FIG. 3, the preference number is not limited to a pure interval set, but can express a wide range of meanings by expressing information about the designer's preference. In addition, since the set-based concept obtains the final design solution as a set, it gives design flexibility in the sense that it can be arbitrarily changed within the interval of the design solution.

  The preference function is similar to a probability density function, a utility function, or a fuzzy membership function. However, the biggest difference is as follows. The probability density function shows certain types (distributions) of design variables due to statistical uncertainties. This is an uncertainty caused by the lack of accurate knowledge about a variable because it is a content or phenomenon that cannot be controlled or selected by the designer. Utility-functions must obey various assumptions and conditions. On the other hand, the mathematical meaning of the preference function indicates the designer's subjective degree related to the degree of desirability, and it is not necessary to follow the assumptions in utility theory. Moreover, utility evaluation is difficult in the initial design stage where there is not enough information to select the best or worst. On the other hand, fuzzy theory generally expresses inaccuracy among uncertainties and can handle this. Using the concept of fuzzy sets, for example, a fuzzy membership function of 0.5 indicates that the degree to which the design proposal belongs to the fuzzy set is 50%. However, it is not always clear how the designer should deal with the design proposal.

Furthermore, in the PSD method, the preference number is quantified by using a quantified symbol (generic symbol, existence symbol) according to Quantified Relations (QR's) (Non-Patent Documents 9 and 10). The quantified preference number (QPN) is expressed as follows.

  By using the quantified preference number instead of the preference number, it becomes possible to specify the possibilities for individual design variables. For example, there may be variables that the designer cannot control depending on the design variables, such as manufacturing errors and the possibility of being determined by other personnel, and all of the possibilities given for such design variables. A design solution that satisfies the requirements must be obtained. At this time, the preference number is quantified using a generic symbol. In addition, design variables that can be controlled by the designer can be quantified using existence symbols and adjusted in the designer's solution set to achieve good performance. Therefore, designers can incorporate preferences that structure the desirability of design when considering both the design space and the required performance space. In this way, the designer can set the design variable range (step 2-2 in FIG. 2).

1.2 Set Propagation Method Using Modified Discrete Fuzzy Operation Once the preference number as specified in Section 1.1 is specified for the design variable and the performance variable, it is necessary to determine the performance value that can be achieved by the design variable preference number. is there. Since the input value is given as a certain distribution, the output value also represents a certain distribution. Here, in order to obtain the QPN of the output performance, a method for propagating the QPN of the input design proposal with respect to the relationship between them is proposed (step 2-3 in FIG. 2). This method is similar to the method of decomposing fuzzy sets into level sets with α-cut. An arithmetic method for this can also be defined using set arithmetic. However, the method of the present embodiment can handle causal relationships between variables that are important in engineering design problems, and avoids overestimation of sets caused by standard fuzzy set operations by using extended interval operations. be able to. The latter feature can reduce the calculation time.

Each interval is given by an α cut at α-level p _j as follows:

  The subscripts L and U in Equation (3) represent the lower limit and upper limit boundary of the interval, respectively.

1.3 Experiment design method and set narrowing method using PRI 1.3.1 Creation of combination of design proposal subsets by experiment design method Initial design by using the set propagation method described in section 1.2 The probability distribution of performance achieved by the proposed QPN can be determined. If there is a region where the possibility distribution and the preference of the required performance overlap, it can be predicted that there is a feasible design solution in the initial design plan. However, there may be a region that is out of the required preference in the performance possibility distribution according to the initial design plan. In this way, the initial solution set is narrowed down to eliminate a subset that is not an effective solution in the initial design plan (here, step 2-6 in FIG. 2 will be described first). The PSD method uses experimental design (DoE) to create a combination of subsets of the initial design proposal, calculate the probability distribution of performance based on the combination, and the evaluation proposed in 1.3.2 below. A more robust design plan is selected by evaluating each performance using an index. The PSD method uses Taguchi's orthogonal array (Orthogonal Array) that allows simple factor analysis and creates the minimum number of combinations.

1.3.2 New evaluation index for evaluation of design preference and robustness
In the PSD method, as described in section 1.1, a preference function is used to express the degree of change in the preference for design variables and performance variables. In particular, the preference function p _X (x) for the performance variable X defines the relationship between the preference p and the level of the performance X. The preference function p _X (x) corresponds to an example of the required performance preference in the present invention. When the preference function for the design variable defines the performance, the performance obtained by the set propagation method described in section 1.2 is the probability distribution q _X (x) of the performance X depending on the preference function of the design variable. Become. The possibility distribution q _X (x) corresponds to an example of the possibility of the performance value in the present invention.

Here, an evaluation index for measuring the performance of the fluctuating performance corresponding to the fluctuation of the design variable is required. The PSD method employs a design preference index (DPI) (Non-Patent Document 11) in order to evaluate performance fluctuations that occur depending on the range of solutions. Mathematically, DPI is defined as the expected value from the design function preference function within the design range, and is expressed as the following equation (8).

  The information characteristics are information accuracy and information stability (Non-patent Document 15). Information accuracy is represented by the accuracy of one piece of information. The accuracy index for interval set information given as a design variable X (eg, X = [19, 21]) is defined as a 2 cm range divided by the median interval 20 cm, or a 10% range relative to the median. On the other hand, stability refers to a state that is no longer expected to be corrected further in the downstream process of the design. For example, Boolean information is completely stable. This is because the possibility of this information is only "yes" or "no" and there is a 100% probability that it will not change.

Therefore, the stability and accuracy of the design method using the PSD method can be interpreted as follows. The accuracy and stability of the probability distribution is similar to how much the curve shown by the distribution extends to its boundary values p = 0 and p = 1. In other words, PSI is a measure of how close the distribution is to 0 and 1. PSI was obtained by correcting Shannon's entropy index.

  Table 6 shows Shannon's entropy measure (S), γ-level measure (G), modified entropy measure (M), and the proposed PSI evaluation results for the six distributions shown in FIG.

  On the other hand, the basic idea of robust design is to improve product quality by minimizing the effects of fluctuations. In order to generate a design alternative as a combination of subsets of initial variables, the preference number is divided using an experimental design as described in 1.3.1. At this time, since the interval width of the initial variable is divided into equal sizes (in the embodiment described later, it is divided into two equal parts), it can be said that a design plan with a small spread of the possibility distribution of performance is a more robust design.

  In the PSD method, the product of DPI and PSI is used as a measure for simultaneously evaluating designers' preference and robustness. This is called PRI (Preference and Robustness Index).

At this time, the smaller PSI values represent better accuracy and stability, so the PSI should be normalized so that the smaller values contribute more to the PRI. Therefore, DPI and PSI are normalized by the maximum value of DPI and the minimum value of PSI, respectively, for relative evaluation of design alternatives. After all, PRI is calculated by the following equation.

  Here, NDPI (Normalized Design Preference Index) and NPSI (Normalized Precision and Stability Index) are normalized DPI and PSI, respectively.

  In general, in multi-purpose design problems, multiple performances are considered for design alternatives. Therefore, it is necessary to integrate PRI for each performance and evaluate all performances. This integrated PRI is called APRI (Aggregated Preference and Robustness Index).

The PSD method uses a weighted power average (Non-Patent Document 16) often used for multi-objective decision-making problems.

The following general average operator can be obtained by changing the variable S in the equation (13).

The average operator represented by Equation (14) shows different degrees of compromise (Equation (15)) from no compromise (MIN) to complete compromise (MAX) (Non-patent Document 17). For example, when MIN is used to determine APRI, the overall evaluation is determined by the lowest performance even if other performance is evaluated highly. In the case of MAX, overall evaluation is performed with the highest performance characteristics. The average operator between MIN and MAX calculates an average so as to compensate for a performance value that is highly evaluated in the order of HM, GM, AM, and QM.

2.4 Solution Set Correction Method Using PSI, Dependency, and Bisection Method
In the PSD, a probability distribution as a kind of output is calculated according to the propagation process of the set based on the preference number as an input. However, there are cases where the initial input preference number outputs a distribution different from the expected one and does not satisfy the performance requirement (step 2-4 in FIG. 2). For example, if p (x) × q (x) = 0, there is no possible set of solutions. Therefore, it is desirable to correct the input range (design variable range) (step 2-5 in FIG. 2). Here, the designer may not know how to modify the initial design proposal. Therefore, in order to perform design correction, it is necessary to determine the degree of influence of each input preference number. The method of correcting the set of PSD methods basically uses the concept of MoI (Non-Patent Document 5), but it is designed by adopting PSI as a measure of uncertainty and quantitative dependency (Non-Patent Document 18). The feature is that the priority of correction is determined in the draft and the amount of correction is simultaneously obtained.

  When evaluating the input preference number using a relational expression, some input values have little impact on output performance, and other inputs can have the most significant impact. Therefore, the relative relationship between the input and the output can be used to obtain a weight for the output of the input (Non-Patent Document 5). The PSI, a measure of uncertainty proposed in Section 1.3.2, can also be used as a qualitative measure that shows the relationship between input and output. This can provide the designer with guidelines for modifying certain inputs.

Here, y ₀ is the current value of the variable y, and y _new is a new value obtained by changing the value of the variable x by r%.

In order to obtain a quantitative dependency between the output distribution and the input preference level, all the input preference level values may be set to their highest value. The final output value corresponds to y _0. By changing each input preference number by r%, a new output value (y _new ) is obtained, and the quantitative dependency of each is obtained. However, since such quantitative dependence assumes linear approximation, an approximate value may be given outside the required value range when the relationship is nonlinear. Therefore, this correction method uses a two-step correction method. That is, as a first step, non-linearity is also dealt with by using the quantitative dependency as described above as a rough approximation and then performing more accurate correction using a bisection method.

2. Conclusion In the above, we have described the basic framework for set-based design for robust and flexible design based on simulation. The method consists of several computer-based models, including a set representation method, a set propagation method, a set narrowing method, and a set correction method.

  A set representation method that considers both the design solution space and the preference space while considering the range of possibilities of the initial design solution is important. A design solution set that satisfies the performance requirement preference can be obtained by a set propagation method based on a mapping between sets from the design space to the performance space. Further, the set is narrowed down by gradually excluding inappropriate design ranges from the initial design solution space (step 2-6 in FIG. 2). In that case, the evaluation method for evaluating the robustness of the design of the experiment and the design is used. Finally, a guideline is proposed to modify the initial design space by a set modification method. The above design method enables computer-aided implementation of set-based concurrent engineering (SBCE).

(Example)
Hereinafter, the present embodiment will be described in more detail by taking vehicle design, particularly door beam design as an example.

1． Introduction In the above, the theoretical aspects of the set-based design approach have been described. In this chapter and the next chapter, in order to verify the effectiveness, an example is given to a specific vehicle body structure, and the application to the benchmark design concerning the multipurpose design is explained.

  The set-based design method has the theoretical clarity as described above, and can overcome various difficulties in actual multipurpose design sites.

  Therefore, in the next chapter, we will first describe how the set-based design method responds to the actual situation of the design site (development process, multi-performance simultaneous satisfaction solution).

  Next, the concept of the set base design method and the actual approach will be described using a simple design example.

  Based on that, we will explain what was done as a benchmark. In the benchmark test, a specific structure is taken as an example, and a multipurpose satisfaction solution is obtained as a range of design variables, and a method for correcting to a better solution range is shown. We will also compare the design plans of multiple designers and explain the relationship between their design policy and design results.

2. Current status of design sites (difficulty in achieving multiple performances simultaneously)
2.1 Changes in the vehicle development process In the current vehicle development, we will promptly respond to user needs, improve safety (ratings in each country, swift response to laws and regulations), and respond to the environment (reducing weight, improving fuel efficiency). For this purpose, the development period has been greatly shortened. For this reason, many departments involved in development are pursuing Concurrent Engineering (CE), so that all the performance and layout requirements required for vehicles from the stage of body structure planning are satisfied within the given design. Are planning. However, at the initial stage of vehicle body structure planning, vehicle planning itself is fluid in order to flexibly respond to changes in social environmental requirements and market environment, as well as the requirements of given designs and units to be adopted. Many plan changes and design specification changes occur. Now that CE has advanced, the impact of changes in planning and design specifications is not limited to one department that is in charge of work at the time the change is decided, but it also affects multiple departments that use the same design information at the same time. In this case, it is necessary to change the design drastically, and if it is not possible to recover with the degree of correction, the design of each related performance is started from the beginning.

2.2 Many performance requirements of designers and problems caused by them The changes in vehicle planning and design specifications are not only due to environmental changes. Now that CE has advanced, designers must derive solutions that satisfy multiple design (design, production, planning, layout, and performance regulations) requirements at the same time, and solutions that achieve low targets for each individual performance. Even if is selected, design changes may be required later. In the initial design stage, the time given to the designer to obtain a solution that achieves multiple performances at the same time is often given only as much as in the era when conventional serial development has been carried out. At present, the design is changed to achieve the target.

  Each company applies various countermeasures to prevent changes in design specifications that cause such design changes. Especially when selecting solutions that satisfy a wide variety of performance requirements at the same time, design reviews involving vehicle development managers and functional specialists who are familiar with performance are often used, and decisions are made on the allocation of investment in conflicting performance. In many cases, the person who plays the central role in vehicle / body planning has to decide the solution to choose from past data / experience and planning concept, and the solution is aimed at satisfying multiple performance requirements as a vehicle. Even if we intend to do so, the achievement level for each performance requirement has not been verified, and modifications with design changes are necessary later. The impact on the overall vehicle development and the increase in the load on the individual can no longer be ignored.

2.3 Approach for obtaining solutions that satisfy various performance requirements at the same time In such an environment at the initial stage of design, the selection of solutions that simultaneously satisfy various performance requirements can include past data, experience, and CAE. Although prediction of performance using analysis or the like is effective, prediction within a range exceeding past experience is difficult, and CAE analysis itself has limitations such as securing time and analysis man-hours. Also, when the conditions given to select a solution that satisfies various performance requirements at the same time are given as a value with a certain range, or when the value itself may be changed later There is also. In vehicle development, completely independent performance items in which one performance does not affect the other are rare, and it is more common to interact with many related performances. For this reason, we thought that the following features would be required for a tool to obtain a solution that satisfies many performance requirements at the same time.
1) In order to obtain the necessary solution, grasp the range in which a given design specification can exist.
2) Check which of the multiple target performances has a large impact on other performances, and conversely what design specifications have low sensitivity.
3) If a solution that satisfies multiple target performances cannot be obtained, find out which target performance should be changed to which side and how much it will affect which performance.
4) If the relationship between these design specifications can be visually confirmed, it will be easier to grasp the relationship.

  These are items that have not been discussed much in the past or have not been realized in tools.

3. Outline of Set-Based Method This time, the inventors have taken the set-based method as a tool for obtaining a design solution that satisfies various performance requirements at the same time. It was verified by using a door beam which is a part of a vehicle body structural component in which various performance requirements influence each other. The “set-based method” that we have focused on has characteristics such as a to f.
a) The designer can specify the target performance setting within a certain range.
b) The designer's intention can be reflected within the specified target performance range.
c) The obtained solution can be obtained in a form that satisfies the given target performance and has a degree of achievement for the target.
d) When a solution that satisfies all the target performances cannot be obtained within a given range of design specifications, the design specification to be modified to obtain a solution and the amount of change in the design specifications are efficiently obtained. Can do.
e) The obtained solution can be confirmed quantitatively including the process of obtaining the solution.
f) Even if a design beginner designs, a solution similar to that of an experienced designer can be obtained, and at the same time, the designer's intention can be quantitatively factored in, so that the designer's personality is also demonstrated.

4). Example of application in door beam This is an example of actual door beam, and introduces the procedure for using the set base method with explanations of words.

  Here, the input preference and the required performance function will be introduced as an application example in the door beam 1 of the actual vehicle body model shown in FIG.

  First, there are three main performances required for the door beam: strength performance when the door 2 is opened and closed, vibration performance for preventing door resonance, and particularly side impact performance at the time of collision.

  In addition, here is the input preference for the design philosophy envisioned by the designer for the basic specifications of the plate thickness and diameter as a door beam (specifically, the plate thickness should be about this thickness considering various performances). Call it degrees.

4.1 Meaning of input preference
In the set-based method called Preference Set-Based Design (PSD), a design solution set is obtained by obtaining a design solution as a wide set instead of a point value at the initial stage of design, and gradually removing the less realistic solution set as the design progresses. To narrow. Therefore, PSD aims at robust design that can respond flexibly to changes due to various uncertainties, while reflecting preference, by introducing a preference function for design solutions and required performance.

  In PSD, a designer first defines a design solution ISS (Initial Solution Subspace ... corresponding to a design variable range) and a required performance PRS (Performance Requirement Subspace) for a design variable and a performance variable. Next, the designer calculates a performance value PPS (Performance Possibility Subspace) of the initial design solution ISS using an interval propagation algorithm (Preference Set Propagation Algorithm). At this time, since the initial design solution ISS is expressed as a continuous value preference function, the performance value PPS based on the initial design solution ISS represents a possibility distribution.

  Here, the diameter d and the wall thickness t of the door beam 1 shown in FIG. 4 are used as design variables, and the example shown in FIG. 5 where the maximum reaction force rf based on the FEM model assuming a side collision is used as a performance characteristic. Let's talk. An actual design example that satisfies multiple performances simultaneously will be described in detail later. In FIG. 5, reference numeral 3 indicates an impactor, and reference numeral 4 indicates a bracket for attaching the beam 1 to the door 2.

  For example, consider defining a design solution ISS with a preference function in the design solution space for the design variable diameter d and wall thickness t. At this time, PSD applies the input preference as shown in FIG. 6 to the design variable.

  This preference degree quantitatively handles the designer's subjective judgment in a form that includes ambiguity. For example, in FIG. 6 (a), the designer thinks that the diameter d is 23 mm to 26 mm is most desirable (preference is 1), and that the d is less than 23 mm or more than 28 mm is not desirable (preference is 0). ) For other diameter values corresponding to preference values from 0 to 1, the degree of desirability of the designer is expressed by the numerical value.

4.2 Meaning of Preference Function of Possibility Distribution and Required Performance Here, the side impact FEM analysis shown in FIG. 5 is performed on the initial design solution ISS having the input preference shown in FIG. 6 for the two design variables d and t. Obtain the performance value PPS of the maximum reaction force rf by. At this time, since the initial design solution ISS is expressed as a continuous value preference function, the performance value PPS based on the initial design solution ISS represents a possibility distribution.

First, the side collision FEM analysis shown in FIG. 5 is performed for various combinations of the diameter d and the wall thickness t, and the maximum reaction force rf is obtained. As a result, an approximate expression of the response surface of the maximum reaction force rf with the diameter d and the wall thickness t as variables is obtained in the form of Expression (17) and can be illustrated in the form of FIG.

rf = 17.09 - 1.281 × d + 2.617 × t + 0.02754 × d 2 - 0.001638 × d × t - 0.1292 × t 2 (17)

Next, the possibility distribution for the performance value PPS of the maximum reaction force rf is obtained according to the procedure of FIG. That is, in this procedure, first, the performance value when preference is 0 (that is, the maximum of appetite width), obtained using the calculation model (see Figure 7). The calculation model is, for example, a response curved surface.

Next, a design variable at an arbitrary preference (for example, 0.5) is extracted, and a corresponding performance value is calculated by interpolation using a calculation model (see FIG. 5). Thereby, in this embodiment, possibility distribution used for set base design can be calculated efficiently.

  Similar processing can be performed for performance values other than the maximum reaction force.

  With the above procedure, the probability distribution of the performance value PPS such as the reaction force rf is obtained for the performance space.

  By the way, for this performance space, the designer needs to give the required performance PRS of the designer in the form of a preference function similar to the input preference described above (FIG. 9).

  From this figure, there is a common area FPS (Feasible Performance Subspace) where the probability distribution of the performance value PPS derived from the initial design solution ISS defined by the designer giving input preference is overlapped with the preference function of the required performance PRS. You can see that it exists. Therefore, the design solution is considered to be within the designer's input range. As described above, the set base design is characterized in that the space for the feasible design solution can be reduced.

  The common area FPS of the performance value PPS and the required performance PRS needs to exist for all performance characteristics. There is a common area FPS for the performance characteristics of the maximum reaction force rf against a side collision, but if there is no common area FPS among the performance characteristics, all the initial design solutions ISS Since the required performance cannot be satisfied, the initial design solution ISS needs to be corrected again. The correction method will be described in detail later.

4.3 Introduction of actual examples Here, we introduce examples of the use of set bases in actual cases.

  First, it is necessary to define an input preference which is a subjective judgment on a design variable written by a designer.

  Specifically, in determining the design specifications of the door beam, the design concept that the designer envisions for the basic specifications of the plate thickness and diameter of the door beam (specifically, considering various performances, the plate thickness is I want to make it about this thickness), and presents the input preference for each designer for every thickness and diameter. This time, we compared the results of using the set base based on the difference in input preference between the two designers (Case A and Case B). Here, the contents indicated by the input preference will be described using a case B diameter example.

  As shown in FIG. 10, the horizontal axis is the diameter, and the vertical axis is the preference. The diameter B of case B is considered to be the most desirable design dimension (preference is 1), less than 24 mm and greater than 28 mm. The case is totally undesirable. (Preference is 0) Further, the larger the value is from 24 mm to 26 mm, the more desirable, and vice versa.

  As can be seen from the comparison with Case A, the difference in design philosophy between the two can be seen in the difference in preference.

  Similarly, regarding the thickness preference, it can be seen from the preference graph that there is a difference in design philosophy regarding the thickness of the two designers as shown in FIG.

  That is, the graph of design preference indicates the satisfaction plan of each designer for each design variable and the satisfaction of the designer in the design examination range. This point is not in the conventional way of thinking.

  Up to this point, the input preference indicating the degree of satisfaction of the designer in the design variables and the like has been determined, but it is also necessary to determine the satisfaction in various performances required by the designer as well as the input preference.

  The reason for this is that, in the conventional actual development, a great amount of man-hours are spent on the examination time to satisfy the required performance in the design examination range, and sometimes the required performance cannot be satisfied in the design examination range. The time required for the examination time may be wasted.

  Therefore, in order to reduce the examination time as described above and to eliminate the useless examination time for which the design satisfaction solution cannot be obtained, the fluctuation range of various required performances in the design examination range of the designer is obtained from the approximate expression, and the designer First, check whether the range of fluctuations is included in the range of required performance to be satisfied, and finally narrow down the design examination range (corresponding to the design variable range) where the degree of satisfaction with the required performance of the designer becomes high. As a result, design specifications (specifically, plate thickness, diameter, etc.) that increase the satisfaction of the designer can be determined efficiently.

  Here, the required preference function indicating the degree of satisfaction of the two designers with respect to the required performance, the above-mentioned reaction force natural frequency at the input preference with respect to the diameter and thickness of each designer's beam, the mounting portion Graphs overlaid with stress, cost, and mass probability distributions are shown in FIGS.

  The preference function of the required performance for the reaction force shown in FIG. 12 will be specifically described.

  The reaction force required for the door beam this time is smaller than 6KN, which indicates a problem in performance, and the larger the value is, the higher the performance is.

  Here, in the graph in which the reaction force possibility distribution and the required preference degree in FIG. 9 are overlapped, the hatched portion has a common area.

  The meaning of this part indicates that there is a design specification that achieves the performance required by the designer in the design examination range of the designer.

  However, for example, with respect to the reaction force and stress shown in FIG. 12 and FIG. 14, there is a possibility distribution region that deviates from the required preference. In other words, a part of the design study range that satisfies the designer's required performance indicates that the design study range cannot satisfy the designer's required performance.

  Therefore, in order to propose design specifications that satisfy the required performance efficiently, the design study range that satisfies the required performance of the designer can be narrowed (narrowed). )There is a need.

  The procedure for deleting the design examination range with poor design feasibility in case A of this example will be described.

  First, in each of the beam plate thickness and diameter input preferences, the beam variable thickness input preference value is appropriately set to 2 along the vertical axis as shown in FIG. To divide. Similarly, the diameter is divided into two as shown in FIG. Note that the number of divisions may be three or more.

Therefore, since it is necessary to consider all combinations, as shown in FIG. 19, four possible combinations of two factors and two levels are used as the probability distribution of each performance for all combinations using an L4 orthogonal table. The relationship of the preference function of the required performance was examined.
The results are shown in FIGS.

  Based on the results, in order to present a robust design proposal that satisfies the required performance preference higher, it is a product of DPI and PSI as a measure that simultaneously evaluates the designer's preference and robustness for each performance. A certain PRI (Preference and Robustness Index) has been proposed, but this measure is used for evaluation.

  In this case, we are studying design specifications that satisfy multiple performances such as stress, natural frequency, reaction force, etc. at the same time. APRI (Aggregated Preference and Robustness Index) integrated with PRI is used.

In addition, this time, APRI was obtained with the same weighting for each performance, and since there are multiple evaluation methods in APRI, evaluation was performed using several types of evaluation methods.
The results are shown in Table 2 and FIG.

  As can be seen from Table 2, in the combination of # A-1 and # A-2, since there is no common area for stress, DPI is 0 for stress, and PRI is the product of DPI and PSI. Is 0.

  Therefore, there is no effective solution that satisfies the required performance in # A-1 and # A-2. Therefore, in these cases, the divided design variable range (division range) can be excluded.

  Next, considering # A-3 and # A-4, in all APRI evaluation methods, the combination of the design proposals of # A-3 shows the highest APRI value, and # A- 3 can be selected as the best design proposal.

  Further, the design proposal is further narrowed down by repeating the above-described process for the design combination # A-3. FIG. 26 shows the possibility distribution of each performance and the required preference according to the design plan, and Table 2 and FIGS. As can be seen from the result, # A-3-4 was the best design proposal.

  As a result of performing the same for case B, the final design plan for cases A and B is as shown in FIG. In order to compare and examine the design proposals in the two cases, the preference as shown in FIG. 31 is normalized so that the maximum is 1, and the performance probability distribution by the design proposal having the normalized preference is obtained. Was obtained as shown in FIG. In particular, Table 4 below shows an interval set of the design proposal when the preference levels of cases A and B are 0 and 1. From this result, it can be seen that design proposals with different preferences produce different design solutions. Table 5 and FIG. 33 compare the final design proposals. As shown in FIG. 33, design plan B is better when technical performance such as reaction force, stress, and natural frequency is emphasized, and design plan A is better when economic performance such as cost and mass is emphasized. It can be said.

  Therefore, FIG. 31 shows a probability distribution of performance according to a design plan that normalizes the preference degree so that the maximum is 1 and has the normalized preference degree.

  It should be noted that the design support method and the possibility distribution calculation method of the present invention are not limited to the above-described embodiments, and it is needless to say that various changes can be made without departing from the scope of the present invention.

  For example, in the above-described embodiment, when PRI = 0 (that is, any of the required performance preference, the performance value possibility, or the robustness is 0), the divided design variable range (divided range) is excluded. is doing. However, when the PRI is not 0 and is smaller than the other divided ranges, the divided ranges can be excluded. In addition, it is theoretically possible to use DPI instead of PRI as a criterion for selecting a divided range, and use the preference of required performance, the possibility of performance value, or the robustness itself as a criterion for selection. is there.

It is explanatory drawing for comparing the concept of a point base design and a set base design. Figure (a) shows an iterative design approach in traditional point-based design. Figure (b) is a diagram showing a set-based design approach. It is a flowchart for demonstrating the procedure of a set base design. It is a graph which shows the relationship between a design variable range and a preference (PN). It is explanatory drawing of the door beam used as the design object in embodiment of this invention, Comprising: FIG. (A) is a figure of the state which attached the door beam to the door, FIG.5 (b) is sectional drawing of a door beam. It is a figure of the side collision FEM model (half model) with respect to a door beam. FIG. 5A is a graph showing a relationship between a design variable range and an input preference. FIG. 5A is an example when the design variable range is a diameter d, and FIG. . It is a graph which shows the response curved surface of the maximum reaction force rf. It is explanatory drawing for demonstrating the derivation | leading-out procedure of possibility distribution with respect to the performance value PPS of the maximum reaction force rf. It is a graph which shows the possibility distribution of performance value PPS, and the preference function of required performance PRS. It is a graph which shows the input preference with respect to a diameter (design variable range) in two designers (Case A, Case B). It is a graph which shows the input preference with respect to board thickness (design variable range) in two designers (Case A, Case B). It is a graph which shows possibility distribution of reaction force. In the figure, PN indicates the number of preferences. It is a graph which shows the possibility distribution of a natural frequency. In the figure, PN indicates the number of preferences. It is a graph which shows the possibility distribution of stress. In the figure, PN indicates the number of preferences. It is a graph which shows possibility distribution of cost. In the figure, PN indicates the number of preferences. It is a graph which shows the possibility distribution of mass. In the figure, PN indicates the number of preferences. It is explanatory drawing which shows the example of a division | segmentation of the design variable range in the input preference graph of plate | board thickness (in the case of Case A). It is explanatory drawing which shows the example of a division | segmentation of the design variable range in the input preference graph of a diameter (in the case of Case A). It is explanatory drawing which shows the combination of the input preference function in case A. It is a graph which shows possibility distribution of reaction force in each combination of Drawing 19. It is a graph which shows the possibility distribution of the natural frequency in each combination of FIG. It is a graph which shows the possibility distribution of the stress in each combination of FIG. It is a graph which shows the possibility distribution of mass in each combination of FIG. It is a graph which shows the possibility distribution of cost in each combination of FIG. It is a graph which shows the evaluation of APRI in each evaluation method. It is a figure for demonstrating narrowing down of the design proposal of # A-3. It is a graph which shows possibility distribution of reaction force in each combination of Drawing 26. It is a graph which shows the possibility distribution of the stress in each combination of FIG. It is a graph which shows the possibility distribution of cost in each combination of FIG. It is a graph which shows the possibility distribution of mass in each combination of FIG. It is a graph which shows the example of narrowing down of the design variable range in the example of cases A and B, Comprising: FIG. (B) is an example of narrowing down the plate thickness (# A-3-4 and # B-4-3). It is a graph which shows the possibility distribution obtained in the example of case A and B in the refined design variable range. Figure (a) is a graph showing the possibility distribution of reaction force. Figure (b) is a graph showing the probability distribution of the natural frequency. Figure (c) is a graph showing the potential distribution of stress. Figure (d) is a graph showing the probability distribution of costs. Figure (e) is a graph showing the probability distribution of mass. It is a graph for the comparison of the design solution set using the weight in a different design plan, Comprising: FIG. (A) is an example which placed weight on technical performance, and (b) is a non-technical requirement. This is an example in which weight is put on (for example, cost and mass).

Explanation of symbols

1 door beam 2 door 3 impactor 4 bracket

Claims (3)

A design support method comprising the following steps executed using a computer :
(1) and the design variables range, an input preference of the design variables, and required performance variable range, the computer instructions for setting the preference of the required performance is Ru receiving step;
(2) a step of receiving an instruction for dividing the design variable range by the computer ;
(3) the expected value and robustness of performance in divided design variables range to evaluate integrated manner, for each design variable range that is divided, based on the following equation, calculated in the computer Steps by which the means calculates PRI :
PRI = NDPI * NPSI
here,
NDPI: normalized DPI;
DPI: Possibility distribution (q (x)) of the required performance obtained from the preference (p (x)) of the required performance in the required performance variable range and the design variable range and the input preference of the design variable ) Expected performance value calculated from
NPSI: normalized PSI;
PSI: A measure indicating the accuracy and stability of the probability distribution (q (x)) of the required performance
It is . The design support method according to claim 1, further comprising the following steps executed by a computer ;
(4) When the PRI is 0, the calculation means in the computer excludes the corresponding divided design variable range from the design examination target. A method of calculating a probability distribution comprising the following steps executed by a computer to calculate the probability distribution according to claim 1 :
(1) a step in which the calculation means of the computer calculates a calculation model indicating a probability distribution of performance values for a design object when the input preference of the design variable is 0;
(2) A step in which the calculation means in the computer obtains the possibility distribution by calculating a performance value corresponding to a design variable at an arbitrary input preference by interpolation using the calculation model.