Paper Helicopter DOE musings

| No TrackBacks

As some of you already know, this semester I'm taking a graduate course in Design of Experiments(DOE). We recently performed the very popular paper helicopter experiment. Actually, I had done this experiment in 2002, as part of Six Sigma Green Belt training. After reviewing some literature, I discovered that most researchers did not emphasize the physics of the drag or air resistance. Although, I firmly believe that you must understand first principles or the fundamentals of physics, prior to undertaking a fluid dynamics problem. Otherwise, you're likely to misinterpret the results of your experiment. Statistical tools are fine, but common sense can keep you out of trouble.

Our instructor provided us with a challenge to improve the results of a 'canned' experiment, as the results were very poor. The objective was to design a copter with longest possible flight. The constraints were drop height and copter material.

The canned experiment failed miserably, as all of the various effects proofed to be significant (wing length, paper type,wing taper, body width, etc). So, we spent the next two weekends completing a screening experiment, and verification of the optimum helicopter design. The idea would be to improve the abysmal results of the canned experiment and also create a best practice methodology and helicopter DOE coursware kit for the instructor.

The preliminary screening essentially confirms which geometries and helicopter configurations would actually fly.
After timing the flights of the thirty-two paper helicopters from two different heights, we then fired up Minitab to determine which effects were most significant. We started with seven effects (wing length, body width, flaps, paper clips, taper, body length). As expected, we recognized that wing length and body shape were significant. Both of these effects interacted with each other. I won't go into a detailed analysis of the Minitab analysis. Our experiment required a two-level six factorial design. Doing the math that would be a 2^6 or 64 runs. Considering the time constraints, we ran a 1/2 fraction design. We did one replication to ascertain repeatability between first and second runs. Our designated timers and copter dropper were unchanged. The whole experiment required six heads.

  • Two stopwatch readers
  • Two runners (bring copters back for repeat drops)
  • Copter dropper
  • Data recorder

The following weekend, an optimization run was required. The data collected during the previous week provided keen insight and helped us develop a strategy for making the longest flying copter. We used a response surface design to provide the fine-tuning necessary to provide the desired results. As described previously, we had two-factor interactions that would be critical to our optimized design. The response surface helped us discern the sensitivity between these factors.

At the end of those two weeks, we were able to correct some of the numerous problems associated with the 'canned' experiment. The instructor was able to confirm our findings by having us run the experiment and demonstrate reproducibility.

Obviously, most of us were pretty tired of cutting out paper helicopters and running them up and down the steps for re-drops. Nonetheless, we did get some kudos from the instructor, and the results clearly demonstrated a well orchestrated DOE.

I will eventually post some pics in the usual places.

  • Making of a trebuchet
  • Is wider better? - Reply to James McGovern
  • Declaration and CBP musings
  • System architecting musings
  • No TrackBacks

    TrackBack URL:

    Monthly Archives


    OpenID accepted here Learn more about OpenID
    Powered by Movable Type 4.25

    About this Entry

    This page contains a single entry by AG published on April 22, 2006 6:41 AM.

    Friendster final chapter? was the previous entry in this blog.

    Understanding the OSS model is the next entry in this blog.

    Find recent content on the main index or look in the archives to find all content.