A Knowledge-base Self-Learning System

G. Lowe and B. Shirinzadeh (Australia)


automation, robotics, self-learning, expert system, assem bly.


Expert systems in general require long development peri ods and are subject to knowledge obsolescence. This is particularly true in the case of assembly, where large sets of data and heuristics are inevitable. This paper employs a limited set of expert knowledge and develops a Knowledge Base Self-Learning (KBSL) system to generate new knowl edge. The system has two functions, firstly to evaluate se quences and prioritise for further evaluation by Reinforce ment Learning (RL), and secondly, to self generate new knowledge. This new knowledge is then available to com bine with existing knowledge for evaluating and prioritis ing product sequences. A key contribution of this self-learning technique, is establishing new knowledge, and adding this to the knowledge-base for application to new assembly sequence evaluations. One outcome of this approach is the require ment for complete knowledge is less onerous. However, a tradeoff between initial depth of knowledge coded and ex pert system performance must be considered. The method developed here was intended as a component in a hybrid system.

Important Links:

Go Back