Sc. Park et al., EXPLANATION-BASED LEARNING FOR INTELLIGENT PROCESS PLANNING, IEEE transactions on systems, man, and cybernetics, 23(6), 1993, pp. 1597-1616
Citations number
60
Categorie Soggetti
Controlo Theory & Cybernetics","Computer Science Cybernetics","Engineering, Eletrical & Electronic
This paper explores the possibility of applying explanation-Based Lear
ning (EBL), a technique from machine learning, to intelligent process
planning. There are currently two major approaches to process planning
: the ''variant'' and the ''generative'' approaches. Each has advantag
es and deficiencies. Our hypothesis was that EBL could successfully un
ite these apparently disparate approaches. Weak method AI planning can
be viewed as subscribing to the generative notion of process planning
. Skeletal planning, on the other hand, has much in common with the va
riant approach. EBL can be employed to transition a traditional weak m
ethod planner into a strong method skeletal planner by acquiring stron
g method concepts which are generalized weak-method explanations of ob
served episodes. Thus, it would seem to be a natural vehicle to unite
variant and generative process planning. We implemented a learning pro
cess planner, called EXBLIPP, to test our hypothesis. We found that th
e system possesses many of the intended advantages. In particular, we
demonstrate that the EBL capability enables the process planning syste
m to learn new schemata which yield many of the advantages of variant
process planning. Unfortunately, the acquired concepts also manifest a
n inability to respond to unpredictable features of the environment. S
uch attributes are unavoidable in process planning applications which
require nontrivial scheduling decisions. The root of the problem can b
e traced to the fact that standard EBL forces all nonoperational decis
ions to be made a priori, thus leading to a brittleness that greatly l
imits the benefits of the acquired concepts. We were able to overcome
this deficiency through the use of a technique called contingent expla
nation-based learning. This was implemented as an extension to EXBLIPP
. By deferring certain planning decisions until execution time, the ex
tended EXBLIPP is able to adapt to the dynamic environment of a manufa
cturing system. We discuss the strengths and weaknesses of this approa
ch in the context of integrated planning and scheduling. We speculate
on how the techniques might be pushed to integrate control decision as
well.