Experimenting Smarter not Harder - PDF document

I zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA Experimenting Smarter not Harder zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA - Taguchi Methods Stella Q"vmd BAE SYSTEMS ISMOR 1 1 1 The Problem Developed zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA in Great Brirain zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA in zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA lxmLiw3 Design of Experiments the 1920's Developed extensively i n the academic enyMmeni 1

I zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA Taguchi Methods zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA Control Factors Force zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA RCS Characteristics >-- Mix Weather zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA Noise Factors - Threat zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA Threat &-Air Density zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA S A M Density 2

zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA I . Q m n d zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA Configuration Study 6 6 1 1 . m m n J - B Objectives zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA I 3

I zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA I zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA of Effectiveness - 1 zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA Measures First detections I Missiles launched I Pk Failures (Kills) Control Variables Noise Variables .Scenario .Ingress Route -Random seed 4

l3xz" zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA Taguchi Arrays zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA ;= zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA *Balanced Array *Independent Variables Experimental Design Matrix - LTaguchi Array m I canpumt)au ' a I 3 zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA 8- feplacsd with the Low. Medium or High Value of each tador zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA * Numban 1.2 zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA configuratimi out ol the tull fanorial sotol 12.754.y14 mntguratims -38 - 18 hour8 VS. 700 yeam Uertm,DIL.IOl.m-~m 5

- zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA 1 zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA I zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA Taguchi Control Factor Separation 1 L l ......... ... ... continued I L ktr . . . . . . zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA ..... I , . 6

rrrrrrrrmr zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA Influence of Control Factors on the three MoEs zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA 1 ........ I R 3 is not significant for zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA *Parameter 15 is most significant h i t s *Parameter Control Factors Ranked by Influence on Detections (cf. Hits &hunches) zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA *Parameter 6 is very imponant *Panmeter 2 is unimportant for Detections .Panmeter 16 is also relatively unimponant - I o D I , - m . ~ - o I I Control Factor8 Ranked by Influence on Launches *Parameter 6 is less imponant 'Paramem 2 is more imponant .Parameter 16 is very imponant 7

- zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA Control Factors Ranked by Influence on Hlts rEzzmr3 Advantages of using Taguchi *Reduced Experimentation 36 vs 12,754,584 experiments Multiple Analysts reach the same outcome zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA *Reduction in experimentation time 18 hrs vs over 700 years *Statistically Robust Outcome vs Pseudo Random Sample -Structured experimentation 8

c zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA i' 9