Cost of Composites:a Qualitative Discussion 19 Figure 2.9 Co-cure of large complex parts(Courtesy Aurora Flight Sciences;see Plate 12 for the colour figure) any of the processes already mentioned.It may take the form of a building-block approach where fabrication of subcomponents of the full co-cured structure is done first to check different tooling concepts and verify part quality.Once any problems(resin-rich,resin-poor areas,locations with insufficient degree of cure or pressure during cure,voids,local anomalies such as'pinched' material,fiber misalignment),are resolved,more complex portions leading up to the full co- cured structure are fabricated to minimize risk and verify the design. Testing.During this phase,the specimens fabricated during the previous phase are tested. This includes the tests needed to verify analysis methods and provide missing information for various failure modes.This does not include testing needed for certification(see next item).If the design has opted for large co-cured structures to minimize recurring cost,the cost of testing can be very high since it,typically,involves testing of various subcomponents first and then testing the full co-cured component.Creating the right boundary conditions and applying the desired load combinations in complex components results in expensive tests. Certification.This is one of the most expensive nonrecurring cost items.Proving that the structure will perform as required,and providing relevant evidence to certifying agencies requires a combination of testing and analysis [8-10].The associated test program can be extremely broad (and expensive).For this reason,a building-block approach is usually followed where tests of increasing complexity,but reduced in numbers follow simpler more numerous tests,each time building on the previous level in terms of information gained, increased confidence in the design performance,and reduction of risk associated with the full- scale article.In a broad level description going from simplest to most complex:(a)material qualification where thousands of coupons with different layups are fabricated and tested under different applied loads and environmental conditions with and without damage to provide statistically meaningful values (see Sections 5.1.3-5.1.5)for strength and stiffness of the material and stacking sequences to be used;(b)element tests of specific structural details isolating failure modes or interactions;(c)subcomponent and component tests verifying how the elements come together and providing missing(or hard to otherwise accurately quantify) information on failure loads and modes;(d)full-scale test.Associated with each test level, analysis is used to reduce test data,bridge structural performance from one level to the next and
any of the processes already mentioned. It may take the form of a building-block approach where fabrication of subcomponents of the full co-cured structure is done first to check different tooling concepts and verify part quality. Once any problems (resin-rich, resin-poor areas, locations with insufficient degree of cure or pressure during cure, voids, local anomalies such as ’pinched’ material, fiber misalignment), are resolved, more complex portions leading up to the full cocured structure are fabricated to minimize risk and verify the design. Testing. During this phase, the specimens fabricated during the previous phase are tested. This includes the tests needed to verify analysis methods and provide missing information for various failure modes. This does not include testing needed for certification (see next item). If the design has opted for large co-cured structures to minimize recurring cost, the cost of testing can be very high since it, typically, involves testing of various subcomponents first and then testing the full co-cured component. Creating the right boundary conditions and applying the desired load combinations in complex components results in expensive tests. Certification. This is one of the most expensive nonrecurring cost items. Proving that the structure will perform as required, and providing relevant evidence to certifying agencies requires a combination of testing and analysis [8–10]. The associated test program can be extremely broad (and expensive). For this reason, a building-block approach is usually followed where tests of increasing complexity, but reduced in numbers follow simpler more numerous tests, each time building on the previous level in terms of information gained, increased confidence in the design performance, and reduction of risk associated with the fullscale article. In a broad level description going from simplest to most complex: (a) material qualification where thousands of coupons with different layups are fabricated and tested under different applied loads and environmental conditions with and without damage to provide statistically meaningful values (see Sections 5.1.3–5.1.5) for strength and stiffness of the material and stacking sequences to be used; (b) element tests of specific structural details isolating failure modes or interactions; (c) subcomponent and component tests verifying how the elements come together and providing missing (or hard to otherwise accurately quantify) information on failure loads and modes; (d) full-scale test. Associated with each test level, analysis is used to reduce test data, bridge structural performance from one level to the next and Figure 2.9 Co-cure of large complex parts (Courtesy Aurora Flight Sciences; see Plate 12 for the colour figure) Cost of Composites: a Qualitative Discussion 19
20 Design and Analysis of Composite Structures justify the reduction of specimens at the next level of higher complexity.The tests include static and fatigue tests leading to the flight test program that is also part of the certification effort. When new fabrication methods are used,it is necessary to prove that they will generate parts of consistently high quality.This,sometimes,along with the investment in equipment purchasing and training,acts as a deterrent in switching from a proven method (e.g.hand layup)with high comfort level to a new method some aspects of which may not be well known(e.g.automated fiber placement). The relative cost of each of the different phases described above is a strong function of the application,the fabrication process(es)selected and the size of the production run.It is, therefore,hard to create a generic pie chart that would show how the cost associated with each compares.In general,it can be said that certification tends to be the most costly followed by tooling,nonrecurring fabrication and testing. 2.3 Technology Selection The discussion in the two previous sections shows that there is a wide variety of fabrication processes,each with its own advantages and disadvantages.Trading these and calculating the recurring and nonrecurring cost associated with each selection is paramount in coming up with the best choice.The problem becomes very complex when one considers large components such as the fuselage or the wing or entire aircraft.At this stage is it useful to define the term 'technology'as referring to any combination of material,fabrication process and design concept.For example,graphite/epoxy skins using fiber placement would be one technology. Similarly,sandwich skins with a mixture of glass/epoxy and graphite/epoxy plies made using hand layup would be another technology. In a large-scale application such as an entire aircraft,it is extremely important to determine the optimum technology mix,i.e.the combination of technologies that will minimize weight and cost.This can be quite complicated since different technologies are more efficient for different types of part.For example,fiber-placed skins might give the lowest weight and recurring cost,but assembling the stringers as a separate step (bonding or fastening)might make the resulting skin/stiffened structure less cost competitive.On the other hand,using resin transfer molding to co-cure skin and stringers in one step might have lower overall recurring cost at a slight increase in weight(due to reduced strength and stiffness)and a significant increase in nonrecurring cost due to increased tooling cost.At the same time,fiber placement may require significant capital outlays to purchase automated fiber/tow placement machines. These expenditures require justification accounting for the size of the production run, availability of capital,and the extent to which capital investments already made on the factory floor for other fabrication methods have been amortized or not. These tradeoffs and final selection of optimum technology mix for the entire structure of an aircraft are done early in the design process and'lock in'most of the cost of an entire program. For this reason it is imperative that the designer be able to perform these trades in order to come up with the'best alternatives'.As will be shown in this section these 'best alternatives'are a function of the amount of risk one is willing to take,the amount of investment available,and the relative importance of recurring,nonrecurring cost and weight [11-14]. In order to make the discussion more tractable,the airframe (load-bearing structure of an aircraft)is divided in part families.These are families of parts that perform the same function, have approximately the same shapes,are made with the same material(s)and can be fabricated
justify the reduction of specimens at the next level of higher complexity. The tests include static and fatigue tests leading to the flight test program that is also part of the certification effort. When new fabrication methods are used, it is necessary to prove that they will generate parts of consistently high quality. This, sometimes, along with the investment in equipment purchasing and training, acts as a deterrent in switching from a proven method (e.g. hand layup) with high comfort level to a new method some aspects of which may not be well known (e.g. automated fiber placement). The relative cost of each of the different phases described above is a strong function of the application, the fabrication process(es) selected and the size of the production run. It is, therefore, hard to create a generic pie chart that would show how the cost associated with each compares. In general, it can be said that certification tends to be the most costly followed by tooling, nonrecurring fabrication and testing. 2.3 Technology Selection The discussion in the two previous sections shows that there is a wide variety of fabrication processes, each with its own advantages and disadvantages. Trading these and calculating the recurring and nonrecurring cost associated with each selection is paramount in coming up with the best choice. The problem becomes very complex when one considers large components such as the fuselage or the wing or entire aircraft. At this stage is it useful to define the term ’technology’ as referring to any combination of material, fabrication process and design concept. For example, graphite/epoxy skins using fiber placement would be one technology. Similarly, sandwich skins with a mixture of glass/epoxy and graphite/epoxy plies made using hand layup would be another technology. In a large-scale application such as an entire aircraft, it is extremely important to determine the optimum technology mix, i.e. the combination of technologies that will minimize weight and cost. This can be quite complicated since different technologies are more efficient for different types of part. For example, fiber-placed skins might give the lowest weight and recurring cost, but assembling the stringers as a separate step (bonding or fastening) might make the resulting skin/stiffened structure less cost competitive. On the other hand, using resin transfer molding to co-cure skin and stringers in one step might have lower overall recurring cost at a slight increase in weight (due to reduced strength and stiffness) and a significant increase in nonrecurring cost due to increased tooling cost. At the same time, fiber placement may require significant capital outlays to purchase automated fiber/tow placement machines. These expenditures require justification accounting for the size of the production run, availability of capital, and the extent to which capital investments already made on the factory floor for other fabrication methods have been amortized or not. These tradeoffs and final selection of optimum technology mix for the entire structure of an aircraft are done early in the design process and ’lock in’ most of the cost of an entire program. For this reason it is imperative that the designer be able to perform these trades in order to come up with the ’best alternatives’. As will be shown in this section these ’best alternatives’ are a function of the amount of risk one is willing to take, the amount of investment available, and the relative importance of recurring, nonrecurring cost and weight [11–14]. In order to make the discussion more tractable, the airframe (load-bearing structure of an aircraft) is divided in part families. These are families of parts that perform the same function, have approximately the same shapes, are made with the same material(s) and can be fabricated 20 Design and Analysis of Composite Structures
Cost of Composites:a Qualitative Discussion 21 Table 2.1 Part families of an airframe Part family Description Skins and covers Two-dimensional parts with single curvature Frames,bulkheads,beams,ribs,intercostals Two-dimensional flat parts Stringers,stiffeners,breakers One-dimensional (long)parts Fittings Three-dimensional small parts connecting other parts Decks and floors Mostly flat parts Doors and fairings Parts with compound curvature Miscellaneous Seals,etc. by the same manufacturing process.The simplest division into part families is shown in Table 2.1.In what follows the discussion will include metals for comparison purposes. The technologies that can be used for each part family are then determined.This includes the material(metal or composite,and,if composite,the type of composite),fabrication process (built-up sheet metal,automated fiber placement,resin transfer molding,etc)and design concept(e.g.stiffened skin versus sandwich).In addition,the applicability of each technology to each part family is determined.This means determining what portion in the part family can be made by the technology in question.Usually,as the complexity of the parts in a part family increases,a certain technology becomes less applicable.For example,small skins with large changes in thickness across their length and width cannot be made by fiber placement and have low cost.Or pultrusion cannot be used (efficiently)to make tapering beams.A typical breakdown by part family and applicability by technology is shown in Table 2.2.For convenience,the following shorthand designations are used:SMT=(built-up)sheet metal, HSM=high-speed-machined aluminum,HLP=hand layup,AFP=automated fiber place- ment,RTM=resin transfer molding,ALP=automated(tape)layup,PLT=pultrusion.The numbers in Table 2.2 denote the percentage of the parts in the part family that can be made by the selected process and have acceptable (i.e.competitive)cost. It is immediately obvious from Table 2.2 that no single technology can be used to make an entire airframe in the most cost-effective fashion.There are some portions of certain part families that are more efficiently made by another technology.While the numbers in Table 2.2 are subjective,they reflect what is perceived to be the reality of today and they can be modified according to specific preferences or expected improvements in specific technologies Given the applicabilities of Table 2.2,recurring and nonrecurring cost data are obtained or estimated by part family.This is done by calculating or estimating the average cost for a part of Table 2.2 Applicability of fabrication processes by part family Part family SMT HLP HSM AFP RTM PLT ALP Skins and covers 100 100 15 80 100 0 50 Frames,etc. 100 100 65 100 10 30 Stringers etc 100 100 5 0 100 90 0 Fittings 100 85 U 0 100 0 0 Decks and floors 90 100 35 40 90 10 20 Doors and fairings 80 100 5 35 90 5 10
by the same manufacturing process. The simplest division into part families is shown in Table 2.1. In what follows the discussion will include metals for comparison purposes. The technologies that can be used for each part family are then determined. This includes the material (metal or composite, and, if composite, the type of composite), fabrication process (built-up sheet metal, automated fiber placement, resin transfer molding, etc) and design concept (e.g. stiffened skin versus sandwich). In addition, the applicability of each technology to each part family is determined. This means determining what portion in the part family can be made by the technology in question. Usually, as the complexity of the parts in a part family increases, a certain technology becomes less applicable. For example, small skins with large changes in thickness across their length and width cannot be made by fiber placement and have low cost. Or pultrusion cannot be used (efficiently) to make tapering beams. A typical breakdown by part family and applicability by technology is shown in Table 2.2. For convenience, the following shorthand designations are used: SMT ¼ (built-up) sheet metal, HSM ¼ high-speed-machined aluminum, HLP ¼ hand layup, AFP ¼ automated fiber placement, RTM ¼ resin transfer molding, ALP ¼ automated (tape) layup, PLT ¼ pultrusion. The numbers in Table 2.2 denote the percentage of the parts in the part family that can be made by the selected process and have acceptable (i.e. competitive) cost. It is immediately obvious from Table 2.2 that no single technology can be used to make an entire airframe in the most cost-effective fashion. There are some portions of certain part families that are more efficiently made by another technology. While the numbers in Table 2.2 are subjective, they reflect what is perceived to be the reality of today and they can be modified according to specific preferences or expected improvements in specific technologies. Given the applicabilities of Table 2.2, recurring and nonrecurring cost data are obtained or estimated by part family. This is done by calculating or estimating the average cost for a part of Table 2.1 Part families of an airframe Part family Description Skins and covers Two-dimensional parts with single curvature Frames, bulkheads, beams, ribs, intercostals Two-dimensional flat parts Stringers, stiffeners, breakers One-dimensional (long) parts Fittings Three-dimensional small parts connecting other parts Decks and floors Mostly flat parts Doors and fairings Parts with compound curvature Miscellaneous Seals, etc. Table 2.2 Applicability of fabrication processes by part family Part family SMT HLP HSM AFP RTM PLT ALP Skins and covers 100 100 15 80 100 0 50 Frames, etc. 100 100 65 55 100 10 30 Stringers etc. 100 100 5 0 100 90 0 Fittings 100 85 5 0 100 0 0 Decks and floors 90 100 35 40 90 10 20 Doors and fairings 80 100 5 35 90 5 10 Cost of Composites: a Qualitative Discussion 21
22 Design and Analysis of Composite Structures 5 mean cost≈l4hrs/kg standard deviation 4 of cost≈11 hrs/kg approx normal ■actual distribution data 2 2 4 681012141618202224262830 Cost(hrs/kg】 Figure 2.10 Distribution of recurring cost of HLP skins medium complexity in the specific part family made by a selected process,and determining the standard deviation associated with the distribution of cost around that average as the part complexity ranges from simple to complex parts.This can be done using existing data as is shown in Figure 2.10,for technologies already implemented such as HLP,or by extrapolating and approximating limited data from producibility evaluations,vendor information,and anticipated improvements for new technologies or technologies with which a particular factory has not had enough experience. In the case of the data shown in Figure 2.10,data over 34 different skin parts made with hand layup shows an average (or mean)cost of 14 hr/kg of finished product and a standard deviation around that mean of about 11 hr/kg (the horizontal arrows in Figure 2.10 cover approximately two standard deviations).This scatter around the mean cost is mostly due to variations in complexity.A simple skin(flat,constant thickness,no cutouts)can cost as little as I hr/kg while a complex skin(curved,with ply dropoffs,with cutouts)can cost as high as 30 hr/kg.In addition to part complexity,there is a contribution to the standard deviation due to uncertainty.This uncertainty results mainly from two sources [12]:(a)not having enough experience with the process,and applying it to types of part to which it has not been applied before;this is referred to as production-readiness;and(b)operator orequipment variability.Determining the portion of the standard deviation caused by uncertainty is necessary in order to proceed with the selection of the best technology for an application.One way to separate uncertainty from complexity is to use a reliable cost model to predict the cost of parts of different complexity for which actual data are available.The difference between the predictions and the actual data is attributed to uncertainty.By normalizing the prediction by the actual cost for all parts available,a distribution is obtained the standard deviation of which is a measure of the uncertainty associated with the process in question.This standard deviation (or its square,the variance)is an important parameter because it can be associated with the risk.If the predicted cost divided by actual cost data were all in a narrow band around the mean,the risk in using this technology(e.g.HLP)for this part family (e.g.skins)would be very low since the expected cost range would be narrow. Since narrow distributions have low variances,the lower the variance the lower the risk. It is more convenient,instead of using absolute cost numbers to use cost savings numbers obtained by comparing each technology of interest with a baseline technology.In what follows, SMT is used as the baseline technology.Positive cost savings numbers denote cost reduction below SMT cost and negative cost savings numbers denote cost increase above SMT costs
medium complexity in the specific part family made by a selected process, and determining the standard deviation associated with the distribution of cost around that average as the part complexity ranges from simple to complex parts. This can be done using existing data as is shown in Figure 2.10, for technologies already implemented such as HLP, or by extrapolating and approximating limited data from producibility evaluations, vendor information, and anticipated improvements for new technologies or technologies with which a particular factory has not had enough experience. In the case of the data shown in Figure 2.10, data over 34 different skin parts made with hand layup shows an average (or mean) cost of 14 hr/kg of finished product and a standard deviation around that mean of about 11 hr/kg (the horizontal arrows in Figure 2.10 cover approximately two standard deviations). This scatter around the mean cost is mostly due to variations in complexity. A simple skin (flat, constant thickness, no cutouts) can cost as little as 1 hr/kg while a complex skin (curved, with ply dropoffs, with cutouts) can cost as high as 30 hr/kg. In addition to part complexity, there is a contribution to the standard deviation due to uncertainty. This uncertainty results mainly from two sources [12]: (a) not having enough experience with the process, and applying it to types of part to which it has not been applied before; this is referred to as production-readiness; and (b) operator or equipment variability. Determiningthe portion of the standard deviation caused by uncertainty is necessary in order to proceed with the selection of the best technology for an application. One way to separate uncertainty from complexity is to use a reliable cost model to predict the cost of parts of different complexity for which actual data are available. The difference between the predictions and the actual data is attributed to uncertainty. By normalizing the prediction by the actual cost for all parts available, a distribution is obtained the standard deviation of which is a measure of the uncertainty associated with the process in question. This standard deviation (or its square, the variance) is an important parameter because it can be associated with the risk. If the predicted cost divided by actual cost data were all in a narrow band around the mean, the risk in using this technology (e.g. HLP) for this part family (e.g. skins) would be very low since the expected cost range would be narrow. Since narrow distributions have low variances, the lower the variance the lower the risk. It is more convenient, instead of using absolute cost numbers to use cost savings numbers obtained by comparing each technology of interest with a baseline technology. In what follows, SMT is used as the baseline technology. Positive cost savings numbers denote cost reduction below SMT cost and negative cost savings numbers denote cost increase above SMT costs. 0 1 2 3 4 5 Cost (hrs/kg) Number of parts 24681012141618202224 262830 standarddeviation ofcost≈11hrs/kg meancost≈14hrs/kg actual data approxnormal distribution Figure 2.10 Distribution of recurring cost of HLP skins 22 Design and Analysis of Composite Structures
Cost of Composites:a Qualitative Discussion 23 Also,generalizing the results from Figure 2.10,it will be assumed that the cost savings for a certain technology applied to a certain part family is normally distributed.Other statistical distributions can be used and,in some cases,will be more accurate.For the purposes of this discussion,the simplicity afforded by assuming a normal distribution is sufficient to show the basic trends and draw the most important conclusions. By examining data published in the open literature,inferring numbers from trend lines,and using experience,the mean cost savings and variances associated with the technologies given in Table 2.2 can be compiled.The results are shown in Table 2.3.Note that these results reflect a specific instant in time and they comprise the best estimate of current costs for a given technology.This means that some learning curve effects are already included in the numbers. For example,HLP and RTM parts have been used fairly widely in industry and factories have come down their respective learning curves.Other technologies such as AFP have not been used as extensively and the numbers quoted are fairly high up in the respective learning curves. For each technology/part family combination in Table 2.3,two numbers are given.The first is the cost savings as a fraction (i.e.0.17 implies 17%cost reduction compared to SMT)and the second is the variance (square of standard deviation)of the cost savings population. Negative cost savings numbers imply increase in cost over SMT.They are included here because the weight savings may justify use of the technology even if,on average,the cost is higher.For SMTand some HLP cases,the variance is set to a very low number,0.0001 to reflect the fact that the cost for these technologies and part families is well understood and there is little uncertainty associated with it.This means the technology has already been in use for that part family for some time.Some of the data in Table 2.3 are highlighted to show some of the implications:(a)HLP skins have 17%lower cost than SMT skin mostly due to co-curing large pieces and eliminating or minimizing assembly;(b)ALP has the lowest cost numbers,but limited applicability (see Table 2.2);(c)the variance in some cases such as ALP decks and floors or AFP doors and fairings is high because for many parts in these families additional nonautomated steps are necessary to complete fabrication.This is typical of parts containing core where core processing involves manual labor and increases the cost.Manual labor increases the uncertainty due to the operator variability already mentioned. Table 2.3 Typical cost data by technology by part family [14] Part family SMT HLP HSM AFP RTM PLT ALP Skins and covers 0.0 0.17 0.2 0.25 0.08 0.08 0.32 0.0001 0.0061 0.02 0.009 0.003 0.06 0.01 Frames,etc. 0.0 0.1 0.28 0.1 0.18 0.40 0.0001 0.0001 0.006 0.06 0.008 0.08 Stringers,etc. 0.0 -0.05 (in skins) 0.05 0.40 0.35 0.0001 0.0001 0.002 0.001 0.09 Fittings 0.0 0.2 -0.10 0.0001 0.005 0.015 Decks and floors 0.0 -0.01 0.15 -0.15 0.20 0.0001 0.0001 0.01 0.008 0.02 Doors and fairings 0.0 0.1 0.25 -0.10 0.35 0.0001 0.0021 0.026 0.01 0.05
Also, generalizing the results from Figure 2.10, it will be assumed that the cost savings for a certain technology applied to a certain part family is normally distributed. Other statistical distributions can be used and, in some cases, will be more accurate. For the purposes of this discussion, the simplicity afforded by assuming a normal distribution is sufficient to show the basic trends and draw the most important conclusions. By examining data published in the open literature, inferring numbers from trend lines, and using experience, the mean cost savings and variances associated with the technologies given in Table 2.2 can be compiled. The results are shown in Table 2.3. Note that these results reflect a specific instant in time and they comprise the best estimate of current costs for a given technology. This means that some learning curve effects are already included in the numbers. For example, HLP and RTM parts have been used fairly widely in industry and factories have come down their respective learning curves. Other technologies such as AFP have not been used as extensively and the numbers quoted are fairly high up in the respective learning curves. For each technology/part family combination in Table 2.3, two numbers are given. The first is the cost savings as a fraction (i.e. 0.17 implies 17% cost reduction compared to SMT) and the second is the variance (square of standard deviation) of the cost savings population. Negative cost savings numbers imply increase in cost over SMT. They are included here because the weight savings may justify use of the technology even if, on average, the cost is higher. For SMTand some HLP cases, the variance is set to a very low number, 0.0001 to reflect the fact that the cost for these technologies and part families is well understood and there is little uncertainty associated with it. This means the technology has already been in use for that part family for some time. Some of the data in Table 2.3 are highlighted to show some of the implications: (a) HLP skins have 17% lower cost than SMT skin mostly due to co-curing large pieces and eliminating or minimizing assembly; (b) ALP has the lowest cost numbers, but limited applicability (see Table 2.2); (c) the variance in some cases such as ALP decks and floors or AFP doors and fairings is high because for many parts in these families additional nonautomated steps are necessary to complete fabrication . This is typical of parts containing core where core processing involves manual labor and increases the cost. Manual labor increases the uncertainty due to the operator variability already mentioned. Table 2.3 Typical cost data by technology by part family [14] Part family SMT HLP HSM AFP RTM PLT ALP Skins and covers 0.0 0.17 0.2 0.25 0.08 0.08 0.32 0.0001 0.0061 0.02 0.009 0.003 0.06 0.01 Frames, etc. 0.0 0.1 0.28 0.1 0.18 0.40 0.0001 0.0001 0.006 0.06 0.008 0.08 Stringers, etc. 0.0 0.05 (in skins) 0.05 0.40 0.35 0.0001 0.0001 0.002 0.001 0.09 Fittings 0.0 0.2 0.10 0.0001 0.005 0.015 Decks and floors 0.0 0.01 0.15 0.15 0.20 0.0001 0.0001 0.01 0.008 0.02 Doors and fairings 0.0 0.1 0.25 0.10 0.35 0.0001 0.0021 0.026 0.01 0.05 Cost of Composites: a Qualitative Discussion 23