melds optimal strategies from the dynamic generalized linear model is problematic—as the number of
statistics literature with flexible with a Reciprocal Inverse Gaussian stages increases, their detection
strategies from the active learn- (RIG) observational distribution. power decreases dramatically. To
ing literature. The merits of this The formulation motivates an achieve good detection power, they
approach are borne out in several extension of the Kalman filter to use a false discovery rate (FDR)
examples, including the motivat- this non-Gaussian scenario. This control approach. They develop
ing computational fluid dynamics results in a set of simple recursive two multistage process monitoring
simulation of a rocket booster. formulae where the current esti- and fault identification schemes,
Accelerated life tests (ALTs) mate of the parameter of interest an FDR-adjusted Shewhart chart,
are used in studying reliability to is updated as a weighted harmonic and an FDR-adjusted CUSUM
increase the number of failures at average of the previous estimate chart. To apply the FDR approach,
reasonable follow-up times. A fre- and the current observation. they derive the approximate distri-
quently asked question, near the Applying this Kalman-like filter to bution of the CUSUM statistics
end of an ALT program, is “What analyze traffic data collected by an using Markov chain theory and
do these test results say about ILD leads to a competitive alterna- Brownian motion with drift. The
field performance?” William Q. tive to estimate vehicular speed at detection and fault identification
Meeker, Luis A. Escobar, and Yili minimum cost. power of the new schemes are eval-
Hong show how to generate useful The next article, by Milan uated by the Monte Carlo method.
answers. They also will present this Žukovič and Dionissios T. The results indicate that the novel
work in the Technometrics session Hristopulos, concerns inference FDR-adjusted approaches are bet-
at the Fall Technical Conference for correlation parameters from ter at identifying the faulty stage
in October. spatial or temporal data. For than is the conventional type-I
ALTs are carefully controlled; many such problems, maximum error rate control approach, espe-
whereas, the field environment likelihood requires too heavy a cially when there are multiple out-
is highly variable. For example, computational burden and other of-control stages.
products in the field see different methods are needed. This article The issue concludes with
average use rates across the product develops and studies the meth- an article by Max D. Morris,
population. With good character- od of normalized correlations Brad Dilts, Stuart J. Birrell,
ization of field use conditions, it (MoNC). The authors compare and Philip M. Dixon titled
may be possible to use ALT results MoNC with maximum likelihood “Composite Response Surface
to predict the failure time distri- and weighted least squares using Designs for Factors with Jointly
bution in the field. When such simulated data and gridded and Symmetric Effects.” In most
information is not available, but scattered real data. For the syn- applications of response sur-
both life test data and field data thetic experiments, none of the face analysis, complete first- or
(e.g., from warranty returns) are, methods considered performed second-order polynomial mod-
it may be possible to find a model uniformly better than the oth- els are used, at least initially, to
to relate the two data sets. Under a ers. However, for regularly spaced represent the functional rela-
reasonable set of practical assump- data, MoNC is significantly faster tionships between independent
tions, this model can then be used and practically insensitive to the variables and expected responses.
to predict the failure time distri- sample size. For scattered data, However, there are situations in
bution for a future component or MoNC speed is reduced but still which knowledge of the physi-
product operating in the same use competitive. Potential applica- cal system under study can (and
environment. This paper describes tions of MoNC involve large data should) be used to modify the
a model and methods for such sets and fast pre-conditioning. model form. The work here is
situations. The methods are illus- The topic of statistical moni- motivated by a real experiment
trated by an example to predict toring of multistage processes has that is part of a research program
the failure time distribution of a attracted considerable attention in on the design of a control sys-
newly designed product with two recent years. An article by Yanting tem for an agricultural combine.
failure modes. Li and Fugee Tsung, titled Physical analysis of the combine
Management of road traffic “False Discovery Rate-Adjusted leads to the conclusion that some
can be greatly aided by real-time Charting Schemes for Multistage of the experimental factors have
data collection and analysis. Single Process Monitoring and Fault a kind of jointly symmetric effect
Inductance Loop Detectors (ILDs) Identification,” develops valuable on the responses of interest. The
provide online measurements of new methods. Li and Tsung for- authors show that these symme-
traffic volume, occupancy, and mulate the monitoring problem in tries imply a simplified response
vehicular speed. In an interesting terms of multiple hypothesis tests. surface model and consider how
article motivated by this problem, They show that use of multiple standard second-order composite
Baibing Li formulates the problem hypothesis testing methods that designs can be modified to take
of estimating vehicular speed as a seek to control the type-I error rate advantage of this structure. ■
24 AMSTAT NEWS MAY 2009
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48 |
Page 49 |
Page 50 |
Page 51 |
Page 52 |
Page 53 |
Page 54 |
Page 55 |
Page 56 |
Page 57 |
Page 58 |
Page 59 |
Page 60 |
Page 61 |
Page 62 |
Page 63 |
Page 64 |
Page 65 |
Page 66 |
Page 67 |
Page 68 |
Page 69 |
Page 70 |
Page 71 |
Page 72 |
Page 73 |
Page 74 |
Page 75 |
Page 76 |
Page 77 |
Page 78 |
Page 79 |
Page 80 |
Page 81 |
Page 82 |
Page 83 |
Page 84