The assembly of printed wiring boards (PWBs) typically involves the co
ordination of thousands of components and hundreds of part numbers in
a job shop environment with up to 50 different processes and workstati
ons. In this paper we present a greedy randomized adaptive search proc
edure or GRASP, for solving the daily scheduling problem that is found
in such environments. The advantages of the proposed methodology are
its ability to respond quickly to changing organizational goals, revis
ed customer requests, and a multitude of shop-floor contingencies. A f
lexible lot-sizing heuristic with user overrides allows the scheduling
algorithm to alter the production strategy in the face of random dist
urbances, such as machine failures, component stockouts, and demand pe
rturbations. The algorithms, embedded in a decision support system, ha
ve been implemented at Texas Instruments' (TI) Austin facility. This f
acility assembles boards for internal use and for a growing number of
external customers. Before implementation it was necessary to convince
management that the new approach would significantly outperform the s
cheduling methods that were currently in use. We were able to do this
by running a series of experiments using real data that compared TI's
rule set for scheduling starts and WIP with our methodology. The resul
ts indicated that over 10% increase in net revenues could be achieved
with the GRASP and that comparable improvements in cycle time, flowlin
e balance, and throughput could also be realized. Operational experien
ce over the last two years has borne this out.