Log In   |   Sign up

New User Registration

Article / Abstract Submission
Register here
Register
Press Release Submission
Register here
Register
coolingZONE Supplier
Register here
Register

Existing User


            Forgot your password
December 2005
library  >  Application Notes  >  Steve Addison

Comparing Design vs. Analysis


introduction

far too often we hear the words "analysis" and "design" used as if they were synonyms. they are not! webster’s dictionary includes the following definitions:

 

design

a preliminary sketch or outline showing the main features of something to be executed . the arrangement of elements or details in a product or work of art . the creative art of executing aesthetic or functional designs.

analysis

the identification or separation of ingredients of a substance; a statement of the constituents of a mixture . an examination of a complex, its elements, and their relations . a statement of such an analysis.

 

 

it certainly makes it clear that they’re different but it’s not too easy to relate it back to engineering. far more useful are a couple of definitions from a classic german engineering design text (pahl & beitz, 1977):


"designing is the intellectual attempt to meet certain demands in the best possible way … in systematic respects, designing is the optimisation of given objectives within partly conflicting constraints. requirements change with time, so that a particular solution can only be optimized in a particular set of circumstances."


"an [analysis] is meant to determine the 'value', 'usefulness' or 'strength' of a solution with respect to a given objective."


although, in their translation from the german, pomerans and wallace choose to use the term "evaluation" rather than analysis. the key term in the definition of design is "optimization". design is an optimization process in which analysis is only one of the elements – albeit a critical element.

 

why should i care?


because, without an understanding of this fundamental distinction and awareness of the other stages in the process, it’s all to easy to spend thousands and thousands of dollars [or ecu’s or francs or yen or .] on analysis software, lab facilities and instrumentation when the basic problem with the design process lies elsewhere.

 

so, what is design

 

i like to view the design process in the following way:


the design process

 

there are five main stages:


1. preparation

before we even start the design process, we need to do some homework. we’ll need to define:

  • design parameters - those things which we can't change such as the power levels in the components, the operating environment .

     

  • design variables - those things that we can change such as number of fans, free-area-ratio of screens and grilles .

     

  • design constraints - characteristics which the final design must have such as a max weight, max acoustic noise .

     

  • the objective function - an encapsulation of what we are trying to optimize

     

we'll discuss the objective function - a critical and often misunderstood factor - in more detail later on.

2. synthesis

during the synthesis stage, we develop a design concept that we think will move us towards the ultimate goal. depending on the stage of the process, this might involve anything from a napkin sketch to building a full 3d solid model. in the context of thermal design, it is most likely to involve building a mathematical and/or physical model of the design for computational or experimental analysis respectively.

a smart organization may well have a "design catalog". this is a collection of previous designs (whether eventually committed to production or not, and even including design iterations that were unsuccessful) indexed by basic design parameters. in the case of thermal design, these parameters might be:

  • total system power
  • indoor or outdoor operation
  • rack width
you get the general idea! this can shorten the design synthesis stage dramatically through re-using similar design concepts as a jumping off point for new designs. however, the designer must always be careful that the existence of these designs doesn't precondition his/her thinking about the solution required for this particular set of design parameters.

3. analysis/evaluation

this is where analysis fits in! once we have a design, we evaluate its performance under the defined operational conditions.

analysis is, here, used in its very broadest sense. we can analyze thermal performance in a lab test, using cfd simulation, using correlations or any combination of thes techniques. we might also be looking at the emi, acoustic and stress aspects of the design. so the "analysis" might be a number of analyses in parallel.

4. scoring

once we have the predictions or measurements of the design's performance in the defined environment, we can calculate the objective function. we also check whether the design breaks any of the design constraints - for example, it might be too heavy, or the maximum temperature might be dangerously high for some of the materials.

5. modification

we have now evaluated the objective function and the performance of the design against the design constraints. but even if we have satisfied the design constraints, this is not necessarily an optimum solution. we need to modify the design by changing the design variables (more fans, less fins .) and restart the process. this is where the true skill in engineering lies. the designer blends together (where available):

  • previous experience (his/her own and that of others)
  • suggestions from formal optimization tools
  • suggestions from company design guidelines
  • supplier suggestions
  • their own creativity
to generate a new design option.

 

in the systematic approach to design, cfd and flow network modeling are analysis tools. cad or a solid modeling system are design synthesis tools. none of them is a design system in their own right. but all are critical elements of a complete design system.

 

you need to work on all 5 stages to develop a better design process!


about the objective function

 

the key to understanding optimization is that, although you can modify many design variables, you can only minimize or maximize one objective function at a time under a given set of circumstances. anyone who tells you otherwise does not understand the fundamentals of optimization theory.


let’s look at an example. i'll design a heatsink. in my ignorance, i start out to optimize the weight and cost and thermal performance.


  • minimizing the weight is easy … i simply make it out of the thinnest possible metal with no consideration of the thermal performance. sure, it might be hideously expensive to manufacture but .
  • the maximum thermal performance might involve filling practically all of the space envelope with a convoluted fin arrangement. too expensive to manufacture but .
  • the lowest manufacturing cost option might be a simple extruded fin design. not the lowest weight or the best thermal performance but .


so what is the "optimum" solution? it depends on how you weight the relative merits of cost, weight and thermal performance, and whether you have dictated any constraints on the outcome (weight < 4oz, cost < 12c). after you have struggled with this for a while, you will come up – either explicitly or implicitly – with a ranking of your priorities.


congratulations, you have just defined your single "objective function".


this, of course, is often a highly contentious issue and one which many designers prefer to leave in a gray area. after all, if you say that thermal performance is twice as important as cost, you’re simply creating a rod for your own back when you talk with the manufacturing guys. but failing to define a clear and explicit design goal means that any attempt to formalize the design (not analysis) process is ultimately doomed to failure. you can find lots of discussion on how to weight the analysis/evaluation results in p&b.


more about modification

 

the easiest way to visualize the effect of modification on the process is to think about a "design space". on the left, i've plotted a simple design space for a problem with just 2 design variables (it gets pretty difficult to plot anything in anything more than 3d!!). the vertical axis shows the value of the objective function as design variables a & b are varied.

 

fog_1_373

 


now imagine that you are looking down from above. what you will see as the design proceeds is the "trajectory" of the process. the more efficient your modification strategy is, the fewer jumps will be needed to find the minimum. some examples are shown below. so how can you become more efficient in this stage of the design process?


optimization algorithms

 
one way to do this is to employ mathematically derived optimization algorithms such as downhill simplex, conjugate gradient, simulated annealing . pretty high powered stuff. these approaches are aimed at the holy grail of fully automated optimization. they certainly work in well bounded problems but suffer from a number of limitations in problems as complex as electronics thermal design:


  • they are prone to finding local minima
  • they do not handle discrete value constraints very well
  • they precondition the answer and remove the designers creativity from the equation

 

genetic algorithms

 

an interesting technique used in some applications is based on evolutionary theory (darwin's survival of the fittest). here's the theory. develop a design. create n "mutations" (minor variants) and analyze them. kill all but the most successful. mutate the most successful . and continue until you're seeing little change. if you're interested in finding out more about this fascinating area, try the following link to [the genetic algorithms archive].

 

experience!

 

my favorite and, i believe, the most flexible and promising technique, is also one of the oldest - experience. there's still no computer in the world that can challenge the human brain for flexibility and creativity. but there are some things that can be done to help even in this area.


knowledge is the key here. knowledge acquired through training and textbooks, knowledge acquired from previous designs of a similar nature, knowledge encapsulated in design tools. if you have new engineers coming up to speed, you need to find ways to transfer that knowledge to them as quickly as possible. for more on that topic, see "investing in the human brain" in february's coolingzone magazine.

 

 

 

profiling the design process – a beginner’s guide


the first step to improving your current design process is to understand what you already have in place. a manufacturing manager wouldn’t dream of changing his/her plant without doing a detailed analysis of the capacity of each machine, the warehouse . a software developer will spend considerable time and effort analyzing how much time is spent in each part of a program. yet a design manager may invest considerable amount of $$$ and human capital in design tools without ever working through a profiling exercise. and it’s not that difficult. here’s how .

 



baseline profile for a typical design


let’s profile a typical design project for a new product. we’ll assume that:


  • the design is a clean sheet design – not a derivative design;
  • the company has invested heavily in cfd tools so this is the tool of choice for the analysis stages; and
  • we’ll need about 6 iterations to get to an acceptable design


here are the profiling spreadsheets for the baseline case. you'll see that design iteration #1 involves 56% of the man hours and about 50% of the elapsed time! why? principally because of the considerable time that it takes to gather the required data (20 manhours and 40 hours [1 week] elapsed time) and then convert it to an analysis model. how realistic is this? it depends on your situation but - if you have to try to extract useful thermal data from component manufacturers, it could take that time and much, much more! imagine how nice it would be if your vendors supplied you with ready to run thermal models!!


looking at the contributions of the various stages to the totals, one can immediately see that the synthesis stage occupies most of the engineering time, but that the analysis time is about 1/3 of the elapsed time.

 

reengineering the design process


once you’ve profiled the process, however roughly, you can begin to look at how you might best spend your meager budget – both of man time and $$$. there are a ot of things that you can try but not all of them will make sense for your particular organization. it’s here that you might get real value from an external consultant who can look at the process with a degree of detachment.


let’s look at a couple of possible options:


  • we’ll examine the effect of investing in an on-line database/intranet of data, and macros to create computational models from parametric input;
  • then, we’ll look at how deploying a second analysis technique – perhaps a correlation-based system – might affect the values.


option 1 - thermal library

 

if we assume that the company has spent money on implementing a central repository of thermal data, and has invested a little in developing macros to create models of commonly used parts, we might expect to see the following results:

 

 

we've (relatively conservatively) estimated that we can half the data gathering and model building time. other than that, everything is left the same. the new profile suggests that we could save about 30% of the man hours on each complete design, and cut the elapsed time by approximately 25% as well! if you're doing more than a couple of thermal designs per year, this adds up to significant savings!


option 2 - using preliminary correlations


now let's look at a slightly more complex example. we'll assume that we don't build a data library but that we develop an in-house system of correlations (maybe a spreadsheet system). furthermore, let's assume that we can run the first 3 design iterations using his new quick analysis system then do (maybe) 3 iterations using the cfd tool. this is what things might look like:

 

 

the savings aren't quite as clear-cut as the previous example. for instance, the number of manhours is roughly the same - basically because you now need to develop both a correlation based model and a cfd model. but there is a saving of a little under 20% in the elapsed time so it's still worthwhile. and one could also argue that we could do more correlation calcs up front to map out the design space and hence save more cfd time.


  manhours elapsed
baseline case 84 143
with thermal library 56.5 105.5
-33% -26%
using correlations 83.5 120.5
-1% -16%

 

how do you value the relative importance of manhour savings vs. savings in elapsed time? you'll have to decide on the relative weighting of these two aims for your particular organization. i learned a long time ago that different branches of engineering and industry value these two aspects very differently. for a consumer electronics company, time to market may be the ultimate driving force and product development times are measured in weeks. miss the christmas rush and you’ve missed half of your annual revenue.


for an aerospace company, product development times might be measured in years and the technical performance of the item is paramount.


but these are not necessarily in conflict. if i take fewer iterations to hit a desired performance goal, i can either bank the timesaving (reduced time-to-market) or reinvest the timesaving in further refining the design. either way, the key is to make each iteration more effective.


and in conclusion .

 

     

  • don’t confuse analysis with design! it's only one part of the puzzle.

     

  • don’t believe anyone who says they have "optimized" 2 (or more) things at once. dig deeper and find out what the implied objective function was.

     

  • understand the profile of your existing design process before spending money. you can get pretty big improvements from some relatively simple things.

 

want to find out more?

cover engineering design: a systematic approach
by g. pahl, w. beitz, k. wallace (translator), l. blessing (translator) 
english language translation of a classic german text. thorough and thought provoking, it acts as a valuable counterweight to analytical texts on design optimization.
cover principles of optimal design: modeling and computation
by panos y. papalambros, douglass j. wilde 
a comprehensive theoretical treatment of optimization. tremendous background material but remember that the real world isn't always as amenable to mathematical analysis!



 

 

 

 

 

 

 

about steve addison


from his early days as a project manager at british aerospace working on guided missile systems through phd studies in gas turbine engineering, post-doctoral work on aeroelasticity and vibration, consultancy in fields as diverse as the design of fans for vacuum cleaners, the development of diesel emissions sensors, analysis of power station heat exchanger performance.steve has worked across quite a spectrum of engineering disciplines.


from 1993 to 1999, steve worked for flomerics in the uk, california and massachusetts. amongst other things, he ran the san jose office, edited the user newsletter and developed web-based applications such as flopack and, more recently, flo/eda. he was also involved in the jedec jc15 committee on thermal standardization. in 2000, he worked for a time as the vp services for an e-learning consultancy headquartered in boston.


he is now living and working in seattle as an associate professor and program director for the oregon institute of technology and consultant for addison robson llc.


copyright © 2001 design-center.net inc. – all rights reserved.

Choose category and click GO to search for thermal solutions

 
 

Subscribe to Qpedia

a subscription to qpedia monthly thermal magazine from the media partner advanced thermal solutions, inc. (ats)  will give you the most comprehensive and up-to-date source of information about the thermal management of electronics

subscribe

Submit Article

if you have a technical article, and would like it to be published on coolingzone
please send your article in word format to [email protected] or upload it here

Subscribe to coolingZONE

Submit Press Release

if you have a press release and would like it to be published on coolingzone please upload your pr  here

Member Login

Supplier's Directory

Search coolingZONE's Supplier Directory
GO
become a coolingzone supplier

list your company in the coolingzone supplier directory

suppliers log in

Media Partner, Qpedia

qpedia_158_120






Heat Transfer Calculators