Scoring of Machine-Learning Algorithms for Providing User Guidance in Special Purpose Machines

摘要

We propose a system to make complex production machines more user-friendly by giving the operator recommendations, such as "in the last 10 occurrences of this event the operators performed the following keystrokes". We describe algorithms to generate the recommendations based on data on former user-interaction and process values and to store them in a knowledge base. We also propose algorithms to retrieve recommendations suited to the current process state. We evaluate their performance on simulated data and data gathered from real production machines.