J-059

Interface Agent for Process Plant Operation; Toward the Next Generation Interface

Authors: Seiji Koide, Shingo Yamauchi
Affiliation: Ishikawajima-Harima Heavy Industries Co., Ltd.

Abstract
Agent technology has been illuminated these days in many fields including manufacturing. In this paper, we describe the notion of an interface agent for process plant operation starting from task-analysis on plant operation and generic characteristics of agency, then introduce an idea of case-based realization for agent$B!G(Js perception and memory. Sensation of anomaly was built on top of Memory Organization Package (MOP) that is developed by Schank et.al. Clustering of plant states by version space enabled to discriminate between a plant situation and another, and allowed operators to label them through teaching. The feasibility study using real data of a plant showed the competence of sensation for anomaly, and the demonstration on connecting to a plant simulator showed the usefulness of this cognitive interface agent in a malfunction scenario.

Seiji Koide
seiji_koide@ihi.co.jp