DESIGNING EXPERIMENTS FOR CAUSAL NETWORKS

摘要

Causal networks are directed graphs thatrngeneralize Ishikawa diagrams to encompass multiplernresponses. Emphasizing tolerance design applications,rnthis work presents an optimal design algorithm whenrnthe variables are organized as a causal network. Therncausal network is first transformed into a causal map,rnwhich represents all factors and responses as points in arncommon D-dimensional metric space. The designrnapproach is algorithmic, optimizing Wynn’s entropyrncriterion. This criterion maximizes dispersion amongrnpredicted multivatiate responses, using a distance-inspacerncoefficients (DiSCo) model. A key constraint isrnblock self-containment—the blocks are analyzablernwithout reference to one another; these analyses are tornbe complemented by a unified all-block analysis. Alsornexplored is the benefit of skewing blocks by setting arnfew factors off-target.