Cell assemblies for diagnostic problem-solving

Abstract

We describe a neuronal model for diagnostic problem-solving. This model which is inspired by cell assemblies gives some hints on how diagnostic problem-solving might actually be performed by the human brain. The diagnostic process is described by a deduction system that performs an abductive inference. The abductive inference itself is described by the verbal category theory. A mapping of a diagnostic problem into a diagnostic system represented by an associative memory with feedback connections is presented. The associative memory with feedback connections offers a self-contained architecture for the administration and representation of manifestations and disorders. This can be implemented efficiently on a serial computer, requiring low memory space and low computational costs. Because of these advantages, this model was chosen for the implementation of a real embedded diagnostic system for a wire bonder machine. The knowledge base of this system is composed of 350 rules, which are stored in 11 modules. These modules model the error behaviour of the microcontroller based units of the machine and are arranged in a taxonomy which corresponds to the hierarchical chains that describe the relationship between disorders and manifestations.