Design of a Database-Driven Soft-Sensor and its Application to a Control System

Abstract

In industries, it is difficult to measure physical quantities such as product quality. In order to estimate it, soft sensors have been proposed. In recent years, the Just-In-Time (JIT) method has attracted attention. According to this method, the output is estimated by extracting data from the database similar to the current situation. On the other hand, a database-driven control system has been proposed in which the JIT is equipped with a learning algorithm. This method calculates the control parameters using a database and has the scheme to learn the control parameters to minimize the control error. This paper proposes a new design method for soft sensors which uses a database-driven approach. Specifically, this paper proposes a method to improve the estimation accuracy without modifying the stored past data by introducing a new modifier c into the learning algorithm. Furthermore, a control system is designed in which the estimated value using the soft sensor is fed back.