Anda di halaman 1dari 17

Materials and Design 25 (2004) 5167

Selection strategies for materials and processes


M.F. Ashbya, Y.J.M. Brechetb,*, D. Cebona, L. Salvoc
b a Engineering Department, University of Cambridge, Trumpington Street, Cambridge CB2 1PZ, UK L.T.P.C.M., Domaine Universitaire de Grenoble, BP75, 1130, rue de la Piscine, Saint Martin dHeres Cedex 38402, France c G.P.M.2., Domaine Universitaire de Grenoble, BP75, Saint Martin dHeres Cedex, France

Received 18 February 2003; accepted 21 July 2003

Abstract Engineering design draws on tens of thousands of materials and on many hundreds of processes to shape, join and finish them. One aspect of optimized design of a product or system is that of selecting, from this vast menu, the materials and processes that best meet the needs of the design, maximizing its performance and minimizing its cost. The problem, still incompletely solved, is that of matching material and process attributes to design requirements. Some of these attributes can be expressed as numbers, like density or thermal conductivity; some are Boolean, such as the ability to be recycled; some, like resistance to corrosion, can be expressed only as a ranking (poor, adequate, good, for instance); and some can only be captured in text and images. Achieving the match with design requirements involves four basic steps. (1) A method for translating design requirements into a specification for material and process. (2) A procedure for screening out those that cannot meet the specification, leaving a subset of the original menu. (3) A scheme for ranking the surviving materials and process, identifying those that have the greatest potential. (4) A way of searching for supporting information about the top-ranked candidates, giving as much background information about their strengths, weaknesses, history of use and future potential as possible. In this paper we review the strategies that have evolved to deal with this problem, the progress that has been made and the challenges that remain. 2003 Elsevier Ltd. All rights reserved.
Keywords: Selection strategies; Engineering design; Materials and process selection softwares; Expertise management

1. Introduction Life is full of difficult decisions and none is fuller than that of the designer. Among them are the decisions concerning the choice of materials and processes. There areit is estimatedbetween 40 000 and 80 000 materials and at least 1000 different ways to process them. The designer needs information about all of these if he is to optimize the choice. To put these numbers into perspective, compare them with the vocabulary of an ordinary European: approximately 5000 words. There is a self-evident need here for information-management systems. In recent years, the field of computer aided materials and process selection in mechanical design has evolved from a pedagogical tool into systems closer to the needs of design engineers. This evolution has been
*Corresponding author. Tel.: q33-47-682-6610; fax: q33-47-6826644. E-mail address: ybrechet@ltpcm.inpg.fr (Y.J.M. Brechet).

both in the development of structured databases for materials and processes and in the improvement of systematic methods to compare materials and processes for multi-criteria selection. As a consequence of this evolution, a number of new questions have arisen, suggesting new directions for research. The aim of this paper is to provide a review of the situation. Section 2 outlines the needs for materials and process selection at various steps of the design and the nature of information that these require. In Section 3 we present the possible selection strategies: the free search, the questionnaire based and the analogy. The choice of the most appropriate strategy for a given selection problem is a key issue in the development of efficient selection guides. The practical application of selection methods requires both the identification of a value function and the exploration of the space of possible solutions using optimization methods that depend on the nature of the set of possible solutions (discrete, semi-

0261-3069/04/$ - see front matter 2003 Elsevier Ltd. All rights reserved. doi:10.1016/S0261-3069(03)00159-6

52

M.F. Ashby et al. / Materials and Design 25 (2004) 5167

new methodological approaches and new optimization techniques. 2. Needs of the designer and nature of the required information Fig. 1 sets the scene. The backbone represents the design process, starting with a market need, proceeding through the stages of concept generation (the idea), embodiment development (the sketch) and detailed design (the working drawings), finally ending in production. New design starts at the top and proceeds downwards. Re-design starts at the bottom and loops upwards before descending. Materials and process information is necessary for every stage of Fig. 1approximate data for all materials and processes at the top; precise and detailed data for one or a few materials and processes at the bottom. The nature of this data is captured by the schematic of Fig. 2. At the top left sits a nugget of numeric data: values for precisely measurable properties like density, modulus, strength and thermal conductivity. Some properties are not so easily expressed by numbers: resistance to corrosion or to wear are examples. They can, instead, be described by a ranking: excellent, good, average, poor, awful, for instance, or simply A to E. Others are a matter of yesyno, Boolean, decisions: can the material be blow-moulded? Can the process shape aluminium alloys? Database technology is good at organ-

Fig. 1. The design flow-chart. Materials and process information is required at every stepbreadth at the top, detail at the bottom.

discrete or continuous). The problems of choosing a value function and selecting an exploration tool are detailed in Section 4. The methodology outlined here has led to number of generic and specific software tools. Section 5 will present examples of these, ranging from the most general to the most specialized. A diagnostic procedure for the development of specialized software is outlined. A number of open questions are identified in Section 6, suggesting the need for research in both

Fig. 2. Data for materials and processes takes a spectrum of forms, ranging from numeric data for standard properties (like density) to experience derived from past applications. One challenge in constructing selection tools is that of making the maximum use of all forms of data.

M.F. Ashby et al. / Materials and Design 25 (2004) 5167

53

izing, storing and manipulating information that can be expressed in any of these ways, which we will refer to as structured data. It is the stuff of materials handbooks and data sheets and is generally available, in standard formats, for almost all materials. However, think of all the stuff that is not in data sheets. There are suppliers guidelines for design with a material; there are case studies of its use in particular market sectors; there are analyses of failures caused by its ill-advised deployment; there is, in short, the accumulated experience of the used of the material (Fig. 2). Such experience may come from reference sources or may take the form of accumulated in-house experiencefailure reports, lab notebooks and the like. It is valuable information that we ignore at our perilto make a mistake once is a misfortune; to make it twice is gross carelessness. This sort of specific supporting information can be indexed by material name, but in any other regard it is not easily structured and exists only in patchy form, some bits for some materials, other bits for others. Also there is one more layer of information of a less specific kind: the supply-chain for the material, the codes, regulations and standards that determine its acceptability in a given application, the availability of specialized stress-analysis or optimization tools to enable its use in designwhat might be called the infrastructure needed to support its use. Design is by its nature an iterative process. The initial requirements lead to a first selection that suggests modifications to the requirements. If they are insufficiently restrictive too many solutions are obtained; if over-restrictive, none may remain, requiring a relaxation of constraints or the need to develop a completely new material or hybrid of two or more materials (see below). As a result, the designer must correct, modify and respecify his requirements and repeat the selection procedure. This puts strict requirements on selection tools: they must allow this iterative procedure and accommodate a range of databases appropriate to the various stages of the design. A simple architecture that meets these requirements has two main components. The first is a set of databases structured hierarchically (see Fig. 3), with data attached to records at each level of the hierarchy so that they can be used at the required level of precision and selection algorithms that allow ease of iteration. A crucial point is that these databases need to be comprehensive and complete, with data for every property of every material, requiring innovative methods for estimating properties where data are missing. The second component is an information system that can manage large amounts of supporting information in a range of data formats, which can be searched for background on a particular material by keywords or full-text searching. These are the basic

principle underlying the development of the software of the CES family (CES 4, 2002). 3. Selection strategies That, then, is the nature of the information. What are we going to do with it? Action is guided by strategy; strategy is shaped by the outcome that is sought and the means available to achieve it. A selection strategy has three components: a The formulation of constraints that must be satisfied if the material is to fill the desired function; b The formulation of a performance metric or value function to measure how well a material matches a set of requirements; and c. A search procedure for exploring solution-space, identifying materials that meet the constraints and ranking them by their ability to meet the requirements. Developments of selection strategies can be found in the texts by Dieter w1x, Charles et al. w2x, Farag w3x, Lewis w4x, Ashby w5,6x, Brechet et al w7x and ASM w8x. All, ultimately, seek ways to implement the process suggested by Fig. 4: it is to convert a set of inputsthe requirements of the designinto a set of outputsa selection of material and process. The role of the strategy is that of a transfer function. The second and third rows suggest two broad classes of transfer function: that of free searching using quantitative analysis, that of exploiting expertise capture in a questionnaire, and that of seeking previously-solved problems with features like that of the current problem (analogy). 3.1. Free searching, based on quantitative analysis Quantitative analysis, when it works, it is fast and efficient; it offers great freedom of application; and it has the ability to reveal solutions that are new and innovative. Howeverand here is the butit needs precisely detailed inputs in a form that can be analyzed by standard engineering methods. Find me a material for a beam of length L to support a force F at temperature T, and make it as light as possible can be analyzed by standard engineering methods. Find me a material to insulate my house cannot be approached without prompting about the type of house, the location in which it will be built and the building codes for that location. The steps in quantitative analysis are listed in Fig. 5. First establish the function of the component, the constraints it must meet and the objectives that are the targets for the design, and the variables that the designer is free to choose to meet the objectivesamong them, the choice of material. The function of the beam described above is to carry bending momentsthat is

54

M.F. Ashby et al. / Materials and Design 25 (2004) 5167

Fig. 3. The organization of structured data of the sort generally found in handbooks and data sheets. The material or process is indexed by name; the information listed under the name gives numeric values for material properties, a ranking of performance under standard conditions and yesyno categorization of materialprocess compatibility.

the definition of a beam. It must support the design loads without deflecting too much or failing and it must survive at temperature T (constraints). Further, it must be as light as possible (the objectivethat of minimizing weight). Selection proceeds in three steps: the steps of screening, ranking and supporting information illustrated in Fig. 6. The screening step has the function of eliminating materials and processes cannot meet the constraints. The ranking step orders the admissible solutions using performance metrics base on the objective. In the supporting information step, specific information is gathered about top-ranked candidates, some of it dependent on the local situation of the company.

Returning to the example of the beam, the constraint on the use-temperature T screens out potential candidates that cannot carry load at T, drawing on structured data for the maximum use temperature of materials to do so. The objective of minimizing weight whilst supporting the design load allows the survivors to be ranked again drawing on structured data. The uppermost of the ranked list become the prime candidates. Unstructured data for these is explored, seeking insight into their deeper personalities, enabling a final choice. Software to support decision-making of this kind takes the form illustrated by Fig. 7. Structured data for materials and processes are stored in a relational data-

M.F. Ashby et al. / Materials and Design 25 (2004) 5167

55

Fig. 4. Selection strategies act as transfer functions, converting design requirements into a listing of selected material and processes, with background information about each. Here we consider strategies, listed in the lower part of the figure.

base structure: materials information in one suite of records, process information in a second such suite. Records in the materials suite are linked to those processes that can be applied to them; consequently records in the process suite are linked to materials to which they can be applied. Records in both data tables are linked to text files and to web sites that contain unstructured supporting information. The method is most easily understood through an example. Fig. 8 shows the beam mentioned before. For simplicity, it has a square cross-section1 with an area Asb 2. It is required that the beam should be as light as possible while having a specified stiffness S and that it should operate without failure at 500 8C. The function, constraints, objective and free variables are listed in Fig. 9. We choose the mass m as the objective function; it is to be minimized. We write msALr (1)

constant that depends only on boundary conditions and the way in which the load F is distributed. From Eq. (2) AsC
B

12S E 2 3 F L2 D C 1E G

(3)

which, when substituted into Eq. (1) gives an expression for the mass m (the objective function): msC
D B 12S E 2
1

C1

F L2C
G

5B

r E F 1y2 DE G

(4)

The quantities S, L and C1 are specified by the design. The lightest beam is that made of the material with the greatest value of E 1y2 y r, which is called a material index. There are many such indices, each describing a combination of objective and constrainta few are indicated in Fig. 10.

where A is the cross sectional area, L is the length and r is the density of the material of the beam. The stiffness constraint sets a lower limit for A. The bending stiffness of a beam, S, is Ss C1EI C1EA2 s L3 12L3 (2)

where E is the modulus of the beam material, I is the second moment of area of its cross-section, and C1 is a
1 A similar treatment for a beam of arbitrary cross-section shape leads to a similar material index as that derived here containing a shape factor that accounts for the shape of the cross-sectionsee Ashby w5x for details.

Fig. 5. The starting point for rational analysis is the identification of function, objectives and constraints. The method is illustrated in Section 3.1.

56

M.F. Ashby et al. / Materials and Design 25 (2004) 5167

Fig. 6. The different stages for materials and process selection in design and the conceptual tools to carry them out.

Fig. 7. Typical contents and linkages in a relational database for selecting material and processes. The individual suites of records (circles) contain records for individual material and processes and their attributes, as suggested by Fig. 3.

We now have the information we need for screening and ranking. Initially, all materials are candidates. Using a database of material properties, software such as the

Fig. 8. A beam loaded in bending. In this configuration, C1s48.

CES4 w24x materials selection system allows the constraints to be applied (Fig. 11). Applying the constraint that the maximum working temperature of the material must exceed 500 8C (Tmax)773 K) eliminates all those in the gray box in the lower part of the figure. The same software can be used to generate the plot of E against r shown in Fig. 12. The scales are logarithmic, with the consequence that the criterion E 1y2 y rsC plots as a straight line with a slope of 2. The CES selector allows this line to be moved until only a few materials are left above it; these are the ones with the largest values of E 1y2 y r. The system combines the two stages (or more, if wished), listing their intersectionthat is,

M.F. Ashby et al. / Materials and Design 25 (2004) 5167

57

w8x), and software and data to support the activity are commercially availableCES4 w24x is an example. Its use does, however, place certain demands on the user, who must be able to formulate the design requirements precisely, to develop one or more performance metrics and express them in the form of indices (or know where to find them) and have the experience needed to find unstructured data and draw balanced conclusions from them. 3.2. The questionnaire strategy, based on expertisecapture
Fig. 9. The design requirements, expressed as in Fig. 5. The analysis given here is for the square section beam; a similar analysis for the more realistic shapes (I-beams, tubes, etc.) gives the same material index.

the materials that have Tmax)500 8C and exceptionally large values of E 1y2 y r. Supporting information can then be sought for these using handbooks, journals and the authoritative web-sources such as the ASM Handbooks (ASM 2002), retrieving past experience, design guidelines, commentaries on behavior in known environments, eco-hazards, toxicity and so forth. The user now has sufficient information to make an informed choice. The method is fast, systematic and has the potential for innovation: if the material database contains materials that have never before been used for light, stiff beams or if completely new materials are added to it, they will appear in the selection if they properly fill the design specification. The method is now well developed and has been applied successfully to material selection problems in mechanical and electro-mechanical design as well as process selection. The method is now well developed, techniques for meeting many constraints and more than one objective exist (Ashby w6x; Brechet et al.

Questionnaires guide the uninformed user through a more or less structured set of decisions, using built-in expertise to compensate for the lack of it in the user. Questionnaires to guide selection of materials and processes are constructed by documenting the ways in which experts do it. Howeverhere, too, there is a buteliciting this information is not easy (first capture your expert). The method, in principle, is to present the expert with a comprehensive set of specific questions and solicit from him or her the answers and the further questions that follow from those answers and the spectrum of answers to these, and the further questions until an unqualified answer is reached. Also therein lies the difficulty: there has to be a definitive answer to every question, and further questions following from every answer until the end is reached; nothing can be left hanging. The construction carries an enormous overhead in time and a still greater one in patienceabove all, that of the central character, the expert. Alsoa small additional considerationnot all experts agree. Therefore, it can be done, and when done well it achieves exactly what is wanted w10,11x. However, it has to be accurately targeted. Thus, a questionnaire to guide selection of aluminium alloys for die-castings is

Fig. 10. Material indices are the property groups that emerge from rational analysis of the performance of a component. This diagram shows those relevant to the design of a beam of minimum mass (the objective).

58

M.F. Ashby et al. / Materials and Design 25 (2004) 5167

Fig. 11. Applying simple constraints: here, that on modulus. Youngs modulus for a range of materials is plotted here. Those in the grey zone (E-10 GPa) are screened out. All polymers are eliminated. (Figure generated using the CES4, 2002 selection software.)

possible (and exists) w12x, as are others that map tightly limited domains. Such methods have their own special strengths. First, the novice can be trusted (or can trust himself) with its usethere is the pedigree of the expert behind it. Also by limiting its domain, a questionnaire can offer greater resolution. However, it does not innovatethe selections that emerge are those already known to the expert. Also a new material or process is lost herethe questionnaire will not recover it because it did not exist when the expert was consulted. A survey of current questionnaire-based selectors for materials and processes reveals a diversity of interfaces. We take the selection of processes for joining w13,14x as an example. At the diffuse end of the range is the locate yourself in these lists interface. The user is presented with columns of options like those in Fig. 13, each addressing a single question. What materials do you want to join? What is the geometry of the joint? How is it loaded? What additional features do you want? The questions can be answered in any order and any combinationthe boxes suggest one combination. Thus, far all we have is a constraint-based selector, something that can be done efficiently using the method

of Section 3.1. The expertise appears in the weightings the odds, to take a racing analogythe expert has attached to each of the choices. MIG-welding of low carbon steels? 5y4 on. Soldering of aluminum? 30y1 against. By weighting each choice and combining the weightings, a useful ranking is possible; material-process combinations with favorable odds win, those with unfavorable are dropped. This ability to rank materialprocess combinations is an example of the added resolution that the questionnaire approach allows: something that is harder to build into analysis-based methods. At the other end of the range is what might be called the narrow band-pass filter. Here all freedom of choice is gone; the user is led through a sequence of yesyno questions, each answer triggering a further compulsory choice, as illustrated by Fig. 14. No chance of a mistake herethe expert takes all responsibility; the user is forced to follow a pre-determined path, with no answer until the end of the chain of questions is reached. The simplicity and ease of use are obvious; the obvious difficulties lie in its creation and maintenance. However, the method does have one attraction: it offers still higher resolution. Many processes can butt-weld metal plates,

M.F. Ashby et al. / Materials and Design 25 (2004) 5167

59

Fig. 12. Ranking, using the material index E 1y2yr. The best choices are those that meet the constraint of Fig. 11 and have high values of the index. (Figure generated using the CES4, 2002 selection software.)

but the best choicemeaning the most economical depends both on the material and on the thickness of the plate. Experts know this from experience and if they can be induced to share the experience it can be incorporated into a domain-specific questionnaire. 3.3. Inductive reasoning and analogy Inductive reasoning has its foundations in previous experience. Here the inputs are design requirements expressed as a set of problem features; the transferfunction exploits knowledge of other solved problems that have one or more features in common with the new problem, allowing new, potential solutions (hypotheses) to be synthesized and tested for their ability to meet the design requirements. A central feature here is the library of previouslysolved problems or cases w15xa case is a problem, an analysis of its features, a solution and an assessment of the degree of success of this solution w16x. The challenge in assembling the library is that of appropriate indexingattaching to each case a set of index-words that capture its features. If the index-words are too specific the case is only retrieved if an exact match is found; if too abstract, they become meaningless to

anyone but the person who did the indexing. Consider, as an example w17x, the case of the redesign of an electrical plug to make it easy to grip, insert and pullout by an elderly person with weak hands. Indexing by electrical plug is specific; the case will be retrieved only if the plug is specified. Indexing under design for the elderly is more abstract and much more useful. Plugs are not the only thing elderly people find hard to use. Cutlery, taps, walking sticks and many other products are adapted for elderly people. Examining their shapes and materials and processes used to make them may suggest new solutions for the plug. The use of analogy-based selection is made clear by Fig. 15. The new problem is analyzed and its features identified. The library is searched for cases with features in common with those of the new problem. The retrieved cases document materials and processes used to create these features. The new problem is tackled and (with luck) solved by adapting and combining elements of the selected cases to meet the new need. 3.4. Which method, when? Each strategy has its strengths and weaknesses. There is room for all three, used singly or together. Here is an example.

60

M.F. Ashby et al. / Materials and Design 25 (2004) 5167

Fig. 13. The simplest sort of questionnaire asks the user to select from categories, as illustrated by the boxed selection show here.

The challenge: to choose materials for thermal insulation in two very different applications. The first: an X-ray telescope to be launched as part of the European Space Agency Program, and temperature-sensitive because it relies on diffraction from a single crystal silicon lattice. The second: insulation for the walls of a house in San Diego, California, where, at the time of writing, there is an embarrassing energy shortfall. The firstthe telescopeis a new problem seeking a novel solution. It must be light in weight, it will see extremes of temperature, and it is exposed to UV and other radiation. Here strategy 1quantitative analysis

works well. The requirement of maximum insulation at minimum weight can be expressed as an index (seek materials with high values of 1y rl where r is the density and l is the thermal conductivity). We apply the constraints that the insulating material must tolerate UV radiation and the extremes of temperature and rank those that do by the index 1y rl. We then seek supporting information about availability, shaping and joining of the top-ranked candidates, past experience of use in space and so on. However, analysis is not the only route. Strategy 3seeking cases with features in common with those of the X-ray telescopemight give

Fig. 14. The structured questionnaire. The user is led through an ordered series of questions. The system must have responses for every possible combination of choices if the user is not to be left hanging part way through, creating a heavy overhead in construction and maintenance.

M.F. Ashby et al. / Materials and Design 25 (2004) 5167

61

Fig. 15. Selection by inductive reasoning. A new solution is synthesized by combining elements of other solutions to problems with features that resemble that of the current problem.

useful results. A search on Design of Precision Instruments, which frequently have stringent requirements for thermal control and on other space applications in which thermal insulation is critical (liquid hydrogen lines, for instance) might provide helpful information. Now the secondthe house in San Diego. Here the domain is narrowthat of thermal insulation for private buildings. All countries have their building codes, and they differ; the codes dictate requirements that depend on the use of the building and the place in it for the insulation. The first question, then, is: which country? OK, what sort of building? And what part of the building? and more. Analysis is not of much help here; instead, strategy 2a structured questionnaire dedicated to the narrow domain of insulation for buildingsis a surer approach w18x. 4. Refinements: identifying value functions and exploring solution space 4.1. Identification of the value function The index method, illustrated in Section 3.1, measures the efficiency of a given material for a given elementary function (a beam, for instance) with a prescribed constraint and a single objective. Real problems are seldom that simple. Most involve many constraints and multiple, often conflicting, objectives (minimizing both mass and cost, for example). The selection must take account of all of these. When dealing with multiple constraints, it is necessary to identify the limiting constraint, requiring that each be modeled in the manner of Section 3.1. When there are multiple objectives, it is necessary to combine them into a single value function. This requires an exchange rate, a, between two objectives, measuring the value of one in units of the other: for instance when both mass m and cost C are to be minimized, it is necessary to estimate the acceptable cost to save 1 kg. The two

can then be combined into a single locally linear value function V: VsamqC (5)

for which a minimum is sought. The exchange constant can be obtained from a functional analysis of the object or from the analysis of existing solutions (Ashby w9x). Software tools to assist with these tasks have been developedthe CAMD and VCE packages w1921x. These two techniquesthe coupled equation method and value analysisallow the derivation of a value function that describes as objectively as possible the ability of a given material to meet a complex set of requirements. However, the extra information required to operate these methods are not always available. An alternative is to identify the relative importance of the various criteria. One way is to assign weighting factors to each objective, but they are heavily dependent on the judgement of the person assigning the weights. This can be partly overcome by the used of Fuzzy logic. The requirements in terms of properties or performance indices are expressed with some flexibility: for instance a minimum Young modulus is required, with a tolerance of 5% and a given strength with a tolerance of 20%. This allows for the possibility to trade off one property for another in a complex design. Another approach is to propose to the designer solutions that are at the margin of his requirements and to ask for a qualitative reaction to these solutions. The computer will then build an aggregation function (i.e. it will select a weighted average of the various criteria, arithmetic mean, geometric mean or more complex functions) which would give a reaction similar to that of the designer. This method, which has been implemented in the software Fuzzymat w22,23x, bypasses the difficulty of identifying a priori the weighting coefficient. However, it introduces in the procedure some subjectivity, while the coupled equationsyvalue analysis approach is totally the objective.

62

M.F. Ashby et al. / Materials and Design 25 (2004) 5167

4.2. Strategies to explore the solution space When a evaluation function has been defined from a quantitative analysis of the set of requirements, it is necessary to explore the solution space. Several techniques have been used; the most efficient one depends on the nature of the solution space. When the solution space is a finite list of discrete materials and processes (such as the data bases of CES), the most efficient way to explore this solution space is a simple screening algorithms: the evaluation function is calculated for each admissible solution of the database (i.e. all those that have passed the first screening stage) and the solutions are ranked in order of decreasing efficiency (CES 4, 2002) w24x. When the solution space is in principle infinite, but continuous (such as for instance in the case of optimizing the composition of a glass for a given set of requirements on its properties) standard optimization techniques such as the Simplex algorithm or steepest gradient methods perform efficiently the job of finding the optimum of the evaluation function w22x. When the solution space is infinite but non-continuous (such is the case for composite materials optimization or sandwich structure optimization, where the materials choice is discrete and the geometrical variables is continuous) genetic algorithms have proved to be effective w2325x. 5. Developing specialized software tools All the methodologies described in this paper have been applied at one time or another to the selection of materials and processes. Table 1 lists existing selection tools built with the different strategies. Free-search tools using strategy 1 are well suited for the early stages of design and have the ability to suggest innovative solutions. As the design becomes more specific, more focused tools are needed. Some couple search methods with micro-mechanical models to generate the space of possible solutions (for instance for optimizing composite materials or sandwich structures). Others use the questionnaire approach. Recently, the analogy strategy has been illustrated on a selector for joining processes w17x. The key issue is here to develop a tool to measure the proximity between cases. Cluster analysis and Multidimensional scaling (MDS) are possible strategies. Another possibility is to take advantage of existing knowledge to structure the cases by their positions on trees corresponding to a predefined questionnaire (like the material and process trees of Fig. 3) and to measure the distance between cases by the distance between their positions in the tree w26x. From these examples, a number of simple rules for developing specialized software have been identified. The databases for a free search strategy are best

constructed with a hierarchical structure like that of Fig. 3. The databases should not have any missing information, all the families of the database should be represented and the attributes should be identical for all the elements of the database, allowing comparability. The questionnaire approach requires that all the questions should be meaningful and referenced for all the elements of the database. A structure like that of Fig. 13 allows that all the questions are asked, but all are not necessarily answered; the discrimination increases with the number of answers. A structure like that of Fig. 14, by contrast, requires that all questions are answered, and in a prescribed order, giving high discriminationprovided all the answers area known. The variety of applications is representative both of the versatility of the procedures and of the variety of requirements encountered at various stages in the design procedure w36x. If the general methods can be applied in the first stages of the design procedure, the question of expertise is central to the further stages. Two questions have to be answered. What is the most efficient way of storing this expertise: a well-structured questionnaire or a rich database of cases with an appropriate exploring algorithm to guide the analogies? How can the stored expertise be a source of information for innovative design without being an encouragement to conservatism? Both these questions are still to be answered. 6. The future There are many unsolved problems in the field of optimal selection and many challenges in adapting the methods to meet specific requirements. Most require that a greater degree of modeling be incorporated into the selection procedures. The purpose of this last section is to outline pending questions requiring more research and to stimulate readers to contribute to this field. 6.1. Finite time design and expertise retrieval The selection methods described thus far do not include a consideration of component life. As a general rule, life is limited by creep, by fatigue, by corrosion or by wear. The first two depend only on the properties of the material and the way it is loaded; data for creep and fatigue behavior can be stored in data-structures like those already described. Corrosion and wear are more difficult because they depend not only on the properties of the component-material but also on those of the environment in which it is used (the corrosive medium) or the counter-face on which it rubs and the lubricant (if there is one) in between. Methods exist for safe design with a given material under conditions of creep, fatigue, corrosion or wear w37x. However, selection is more difficultit requires

M.F. Ashby et al. / Materials and Design 25 (2004) 5167


Table 1 Existing software for selection relying on the methodology of this paper Name of the software CES (Granta Design, 2002) w24x Objective of the software Materials and process selection, databases for materials and processes. A number f specialized database on materials and processes and possibility to create new ones Materials and process selection, same databases as CES, multicriteria selection using fuzzy logic algorithm Materials and process selection, same databases as CES, expert guide for developing the set of requirements implementation of coupled equation and value analysis Optimization of composite materials , matrix, fiber, and architecture. Expert compatibility database for process. Optimization of materials selection and dimensioning for structural Sandwiches Type of strategy Free searching

63

Evaluation method Screening of the database

Fuzzymat (Bassetti, 1997) w22,23x CAMD Landru, w19,27x

Free searching

Screening on the database

Free searching and questionnaire

Screening of the database by a recursive algorithm

Fuzzy composite Pechambert Duratti w23,28x

Free searching

Genetic algorithm and screening Micro-mechanics models to create possible solutions Genetic algorithm and screening Mechanics modes to create possible solutions Phenomenological modelling and screening method. Simplex algorithm coupled with fuzzy logic Screening

Sandwich selector Lemoine w28,29x

Free searching

Creep selector Lemoine w46x Fuzzyglass Bassetti, w22x

Selection of polymer in creep design Optimization of glass composition for properties and processability. Database of correlation coefficients. Optimization of aluminum extruded alloys selection, including extrudability and shape via expert rules Optimization of aluminum cast alloys selection, including hot tearing and mould filling via expert rules Selection of surface treatments according to the compatibility with the base material and the required function Identification of value coefficients in a design procedure from existing solutions Identification of possible applications for a material from a properties y performance profile Selection of optimal joining methods Selection of optimal joining methods from existing solutions Suggestion for industrial design from a database of objects Guide to failure analysis and possible solutions from a database of cases

Free searching

Free searching

Fuzzy extrude Heiberg, w30x

Questionnaire

Fuzzycast Bassetti, w22,12x STS Landru, w27,31x VCE Landru w27,32x MAPS Landru w27,33x Astek LeBacq w13,14x Astek expert Lae w17x CES aesthetics Johnson w34x Failure expert Bouget w35x

Questionnaire

Screening

Questionnaire

Screening Proximity on a tree

Analogy

Free searching

Screening

Questionnaire Analogy, case based reasoning Analogy, case based reasoning Analogy, case based reasoning

Screening Proximity on a tree

64

M.F. Ashby et al. / Materials and Design 25 (2004) 5167

that the response of all candidate materials are simultaneously evaluated and compared. The only practical way forward is to develop semi-empirical models (constitutive laws) for each mode of failure, fitting them to the incomplete data available in compilations such as those of the ASM (2002) Handbooks w38x or the NIMS (2002) data sheets w39x and others like them. Selection follows procedures that are parallel to those already described, with provision for the operator to enter the design conditionsthe temperature and design life in the case of creep or the R-value and fatigue life in the case of fatigue, allowing allowable stresses to be calculated. These allowables then replace the elastic limit in the indices for minimum mass and minimum cost design outlined in Section 3.1. Examples of the implementation of this method for creep and fatigue can be found in CES4 w24x. Corrosion and wear present greater challenges, as yet incompletely resolved except in narrow application domains such as the design of aircraft structures for which comprehensive data-sets can be assembled. Much information about both is available only as expertise, recorded as design guidelines about material compatibility and preferred geometry (avoid metal couples when aqueous corrosion is a possibility, sliding aluminum on steel can lead to scuffing, Titanium alloy X is susceptible to stress-corrosion cracking in salt water, liquid at rest in a container can lead to water-line corrosion). Such information is readily stored as text; the challenge is to capture, index and retrieve it accurately and at the relevant point in the design. 6.2. Process selection and modeling The process selection strategies described thus far rely on pre-stored attributes of the process, which include the materials it can treat, the shapes it can create and the characteristics of the finished productits surface finish, tolerance, etc. This enables screening to isolate those processes that that can meet the requirements, using the wide-ranging free-search method of Section 3.1 or higher resolution in a narrower domain using the questionnaire-based method of Section 3.2. However, neither of these captures the full complexity of the materialprocess interaction. Here, too, modeling is the way forwardthe modeling of interest here is that which is instrumental in selecting the process. Thus, modeling of the heat transfer from object to mould, in injection molding, allows prediction of the time before the mould can be opened and the part ejected. This time (and thus the rate of production) can be decisive in deciding whether injection molding is economically viable. Intelligent process selection could go further, predicting the process settings required for a given application. As an example: laser welding can be used to butt-weld

many metals, but the optimal laser power and tracking velocity depend on the metal and on the plate thickness and the condition of its surface. Modeling can capture this information andif based on physical understandingcan allow extrapolation to suggest optimal conditions even when no real data are available w40x. Thus, the need here is for coupling between process selection and process modeling. It is unlikely that a general format can be found in this direction, but welding technologies and surface treatments are promising fields of application for this approach. 6.3. Interfacing of the materials and process selection tools with geometric modeling and dimensioning tools Material and process selection are one part of a larger processthat of choosing the shape and dimensions of a component. Ideally, these activities should be coupled so that the geometric modeling correctly scales dimensions and creates features that make the best use of the material and the process and avoids those that do not. Many materials databases have the ability to export data in a format accepted by finite element packages. However, this passive coupling is only a first step; it does not deal efficiently with the coupling between the shape, material and process. Car wheels, for instance, can be made in cast aluminum and in stamped and welded steel, but they are so made in totally different shapes to match the capabilities of very different processes. There is a need for an active coupling between FE calculations, CAD tools and the materials and process selector. A requirement here is the ability of the CAD tool to recognize features, and an ability to index processes and materials with the features that they can create. In this way the designer can be warned of the restrictive nature of features that might then be changed and the materials selector can be warned of the restrictions that a commitment to a given process places on the design. 6.4. Multi-materials selection When no single material can meet all the design constraints or offer sufficiently high performance, a solution may be found by combining two or more materials to form a hybrid multi-material. Materials selection is incomplete if it fails to deal with the possibility of selecting woven composites, sandwiches or multi-layers. This requires the ability to optimize simultaneously materials choice and multi-materials geometry. Three questions need to be addressed: when is a multi-material approach necessary because of the incompatibilities of the set of requirements? What are the relevant type geometries worth investigating (cables, sandwiches and stiffened plates, filled tubes, etc.) and their advantges? What are the optimal choices for the materials and the dimensions of the chosen

M.F. Ashby et al. / Materials and Design 25 (2004) 5167

65

geometry? Partial solutions to these problems exist, but an overall strategy is still to be developed and implemented into software. The whole question is of critical importance when redesign is considered: it is important to start from the functional analysis and to evaluate the possible multi-materials solutions at the same level of generality with which the free search methods currently deal with a single material selection (see for instance Kromm et al. w41x). 6.5. Green design Design to minimize adverse impact of engineering products on the environment (green design) is assuming an increased importance in all branches of engineering. Eco-impact thus becomes an additional metric to be optimized, along with performance and cost. The problem is a complex one: eco-impact can be associated with the extraction and refinement of the material, with the manufacture of a product from it, with the use of that product and with its disposal (Wegst et al. w42x). Well-intentioned design to reduce the impact of one phase of life may have the effect of increasing it in another; it is a systems problem, not one of isolated events. Limited progress has been made, from which the obstacles can now be recognized. Briefly, they are these: what material attributes, not already available, are relevant for green design? How are these attributes to be measured and stored? And how should they be manipulated to minimize whole-life impact rather than just that of one part of the life cycle? 6.6. Aesthetics and industrial design The consumer in any developed country has, today, a wealth of choice. Any given product is available in many models, all with almost the same functionality and price. The choice the consumer makes is then determined by his or her perception of the product: its aesthetics, its associations, what it means to them. Much of this perceived value is created by the materials of which the product in made and the processes used to make and, particularly, finish it. Industrial designers express the wish for materials selection systems to assist in this choice. The challenge here is to identify the relevant attributes and devise ways of search solutionspace that can meet their need. Some progress has been made but the methods are in their infancy (see, for example, Ref. w34x). 6.7. Selecting functional materials The phrase functional materials needs clarification. Much of the time it is used to distinguish materials that are primarily used for non-structural purposesit is their thermal, electrical, magnetic and optical properties

that are exploited. We shall refer to these as passive functional materials to distinguish them from active materials that respond to a signal rather than simply transmitting one: materials with piezo-electic, magnetostrictive and other such behavior. Data for passive functional materials can be found in handbooks and databases; the methods for selecting them and the processes required to shape them are simple extensions of those described above. There is scope for exploring in greater depth the use of indices like those of Fig. 10 for the optimal selection of electrical conductors and insulators (which must often perform secondary mechanical or thermal functions at the same time), for hard and soft magnetic applications and for the optical properties of glasses. Active functional materials pose greater challenges. First, the coupling they provide between two different kinds of signal (mechanical and electoral, thermal and mechanical, thermal and electrical, magnetic and mechanical) is always associated with directionality, requiring a tensorial description or at least the determination of more than one interaction coefficient. Second, the response frequently depends on shape or can be magnified by shaping, so that selection cannot be decoupled from shape. Also third, the functionality they offer includes their ability to act as sensors and actuators, giving them a much richer set of attributes that include the frequency at which they can be driven, the power they can convert and the efficiency with which they do so w43x. 6.8. The challenge of miniaturization In the design of consumer electronics (mobile phones, laptop computers, PDAs, etc.) two of the dominant objectives, today, are the minimization of size and weight. Similar objectives influence the design of military equipment, driven by the need to pack ever more functionality and ever less space. Miniaturization imposes greater demands on materials: the intensity of mechanical, thermal and electrical loads all tend to increase as the size diminishes. This creates new opportunities for innovative selection, evident in the increased use of specialized materials for heat transfer (aluminum nitride heat sinks with metal foam heat exchangers) and of high strength, low density materials for chassis and casings (titanium and magnesium cases for lap tops and mobile phones, for instance). In these examples the scale, though reduced, remains in the range in which materials exhibit normal bulk properties. However, there is a growing interest in manufacture at an even smaller scale. Micro electromechanical systems (MEMS) aim at creating electromechanical (as opposed to opto-electronic) functionality at the millimeter-to-micron scale. As with more conventional miniaturization, this imposes greater loads on

66

M.F. Ashby et al. / Materials and Design 25 (2004) 5167

materials, but their properties, too, change with scale. The mechanical properties of thin films can differ greatly form those of the bulk material, and the ability, at the micron level, to create multi-layers and to utilize the processing methods of the micro-electronic industry to form complex multi-material combines, poses new challenges for the capture and storage of information about materials and processes and to model the process materialshape interaction w44x. 6.9. Identifying possible applications for new materials The inverse problem is an interesting one. Rather than seeking materials to fill a new application, how can we identify applications for a new material? Developing a new material is a slow, expensive process; to do so without knowing how it will be used (historically a common enough occurrence) may be a poor investment of time and money. Can a procedure be developed to take out some of the risk? Attempts have been made (w45x), largely based on a detailed comparison of the values of the properties and the indices (see Fig. 10) with those of existing materials, but there remains scope for the development of more refined methods and software to support them. 7. Conclusions In this paper we have given an overview of systematic methods to guide the selection of materials and processes. The emphasis has been on the strategies that have some demonstrated degree of success and generality. A number of teams are working on specific areas (corrosion, surface treatments, DFX methodologies, etc.) and further progress in the field requires closer collaboration between various sorts of engineering expertise. Dealing with the full complexity of problems of materials and process selection leads to challenging optimization issues because of the complex topology of the solution space. The work performed by two of the authors on the application of genetic algorithm stemmed from discussions with a specialist of spin glasses! w28x. It seems clear to us that some well-known techniques in statistical physics (neural networks, simulated annealing) are worth exploring as possible solutions to these difficult problems. The increasing need to store and retrieve expertise in engineering field will also require input from artificial intelligence experts, specialists of automatic translation, of sentence recognition, intelligent indexing and developers of web exploration engines. The expanding field of micro electro-mechanical systems (MEMS), and more generally the domain of functional materials is also a source of interesting problems for these methods, requiring a merging of the

skills of the physicist, the materials scientist and the engineer. The present paper is to be read as an incentive for the various fields of expertise to contribute to this new emerging field of selection for design. Acknowledgments Many people have contributed to the ideas outlined, only too briefly, here. We particularly wish to thank Dr H. Shercliff, Dr K. Johnson of the Engineering Department, Cambridge University, Dr D. Imbault from INPG, Dr V. Mandrillon from CEA Grenoble, for many helpful discussions, and to acknowledge the support of the Korber Foundation and of the U.K Engineering and Physical Science Research Council through a grant to the Engineering Design Centre at Cambridge. References
w1x Dieter GE. Engineering design, a materials and processing approach. 2nd ed. New York, USA: McGraw-Hill, 1991. w2x Charles JA, Crane FAA, Furness JAG. Selection and use of engineering materials. 3rd ed. UK: Butterworth Heinemann Oxford, 1997. w3x Farag MM. Selection of materials and manufacturing processes for engineering design. Englewood Cliffs, NJ, USA: PrenticeHall, 1989. w4x Lewis G. Selection of engineering materials. Englewood Cliffs, NJ, USA: Prentice-Hall, 1990. w5x Ashby MF. Materials selection in mechanical design. 2nd ed. Oxford, UK: Butterworth Heinemann, 1999. w6x Ashby MF, Cebon D. Case studies in materials selection. Cambridge, UK: Granta Design Ltd, 1999. w7x Brechet Y, Ashby MF, Salvo L. Selection des Materiaux et des Procedes de Mise en Oeuvre. Switzerland: Presses Polytechniques et Universitaires Romandes de Lausanne, 2001. (Volume 20 of the Series Traite des Materiaux). w8x ASM. ASM handbook. Materials selection and design. Metals Park, Ohio, USA: ASM International, 1997. w9x Ashby MF. Acta Mater 2000;48:359. w10x Lovatt AM, Shercliff HR. Mater Des 1998;19:205. w11x Lovatt AM, Shercliff HR. Mater Des 1998;19:217. w12x Lovatt AM, Bassetti D, Shercliff HR, Brechet Y. Int J Cast Met Res 1999;12(4):211. w13x Lebacq C, Brechet Y, Jeggy T, Salvo L. Materiaux et Tech 1998;5:39. w14x Lebacq C, Brechet Y, Shercliff HR, Jeggy T, Salvo L. Mater Des 2002;23:405. w15x Kolodner J. Case based reasoning. San Mateo, CA: Morgan Kaufmann Pub, 1993. w16x Ashby MF, Johnson, KW. Classification and Choice in Product Design, Cambridge University Engineering Department Report CUEDyC-EDCyTR108, 2001. w17x Lae L, Lebacq C, Brechet Y, Jeggy T, Salvo L. Adv Eng Mater 2002;4:403. w18x Guillanvarch Y, Gauthier T, Salvo L, Brechet Y, Quenard D. Adv Eng Mater 2002;4:407. w19x Landru D, Brechet Y. CAMD software. France: Institut Nation al Polytechnique de Grenoble, 1999. w20x Landru D. Ph.D. Thesis, France: Institut National Polytechnique de Grenoble, 2000.

M.F. Ashby et al. / Materials and Design 25 (2004) 5167 w21x Landru D, Brechet Y, Salvo L. Adv Eng Mater 2002;4:357. w22x Bassetti D. Ph.D. Thesis, France: Institut National Polytechnique de Grenoble. w23x Bassetti AS. Fuzzymat v.3.0 software. France: Institut National Polytechnique de Grenoble, 1997. w24x CES4, The Cambridge Engineering Selector, Granta Design, Rustat House, 62 Clifton Road, Cambridge CB1 7EG, UK (www.grantadesign.com), 2002. w25x Pechambert P, Bassetti D, Brechet Y, Salvo L. Proceedings of the 7th International Conference on Composite Materials (ICCM7), Institute of Materials, London, 1996:283. w26x Landru D, Brechet Y, Tassin C. Materiaux et Tech 2002;1 2:3. w27x Landru D. Ph.D. Thesis, France: Institut National Polytechnique de Grenoble, 2000. w28x Deocon J, Salvo L, Lemoine P, Landru D, Brechet Y, Leriche R. In: Banhardt J, Ashby MF, Fleck NA, editors. Metal foams and porous metals structures.MIT Verlag publishing, 1999. p. 325. w29x Bassetti D, Brechet Y, Heiberg G, Lingorski I, Pechambert P, Salvo L. In: Nicholson P, editor. Proceedings of the Conference on Composite Design for Performance, Lake Louise, 1998; 88. w30x Heiberg G, Brechet Y, Jensrud O, Roven HJ. Mater Des 2002;23:505. w31x Landru D, Brechet Y. Surface treatment selector, STS v.1.0 software. France: Institut National Polytechnique de Grenoble, 1999.

67

w32x Landru D, Brechet Y. Value coefficient estimator, VCE v.1.0 software. France: Institut National Polytechnique de Grenoble, 1999. w33x Landru D, Brechet Y. In: Yavari R, editor. Proceedings of the Colloque Franco-Espagnol, Institut National Polytechnique de Grenoble, 1996;41. w34x Ashby MF, Johnson KW. Materials selection, the art and science of product design. Oxford, UK: Butterworth Heinemann, 2002. w35x Bouget F. Diplome de recherche technologique. France: Univ ersite de Poitiers, 2002. w36x Brechet Y, Bassetti D, Landru D, Salvo L. Prog Mat Sci 2001;46(34):407. w37x Ashby MF, Brechet Y. Adv Eng Mater 2002;4:335. w38x ASM. ASM handbook series on line. Metals Park, Ohio, USA: ASM International, 2002. w39x NIMS. National Institute for Materials Science, Japan, Fatigue Data Sheets 190, 2002. w40x Shercliff H, Lovatt A. Prog Mat Sci 2001;46(3-4):429. w41x Kromm FX, Quenisset JM, Harry R, Lorriot T. Adv Eng Mater 2002;4:371. w42x Wegst U, Ashby MF. Adv Eng Mater 2002;4:378. w43x Shieh J, Huber JH, Fleck N, Ashby MF. Prog Mat Sci 2001;46(3-4):461. w44x Spearing MS. Acta Mater 2000;48(1):179. w45x Landru D, Brechet Y, Ashby MF. Adv Eng Mater 2002;4:343. w46x Lemoine P, Maldjian JM, Brechet Y, Ashby MF, Salvo L. Adv Eng Mater 2002;4(6):384.

Anda mungkin juga menyukai