Chapter 5 Robust Performance Tailoring with Tuning - SSL - MIT

tion changes in the updated solution track analogous changes in the real system [57].
Cha and Pillis use the idea of configuration changes to perform model updating by
adding known masses to the physical structure to obtain a new set of measurements.
These new measurements are used in conjunction with previous data to correct the
mass and stiffness matrices of the model [25].
These ideas are readily applied to the tuning problem posed in this chapter with
a few modifications. The updating method of Cha and Pillis is developed to improve
the frequency and mode shape predictions of the model through manipulation of the
mass and stiffness matrices. In the tuning application it is only necessary to find
tuning parameters that successfully tune the model; it is not required that the model
be accurately updated. Also, in this application the data used for updating is a single
performance metric and not a set of modes and frequencies, and the uncertainty model
is parametric and assumed to be accurate. Therefore it is desirable to update the
uncertainty parameters and not operate directly on the mass and stiffness matrices.
However, the idea of tracking the hardware and model through configuration changes
is key in developing a tuning algorithm that is both successful and low cost.
The hybrid method, referred to as isoperformance tuning, exploits a design tool
called isoperformance to reduce the uncertainty space by considering the changes
in the performance data as the hardware is tuned. It is an iterative, cooperative
process in which tuning optimizations on the model guide the hardware tests, and
the resulting data further refines and guides the model tuning optimizations.
4.3.2 Isoperformance
Isoperformance is a methodology developed by deWeck for multi-objective design and
analysis of complex, high-performance systems [36]. Instead of fixing the design costs
or resources aprioriand optimizing for system performance within those constraints,
the isoperformance methodology constrains the performance and searches for a family
of designs that achieve this performance. The idea is that the “optimal” family of
designs can then be evaluated with respect to other considerations such as cost and
risk. In the thesis [36], deWeck develops and compares three methods of finding
137