Data-driven soft sensors for electrical machines

Abstract

As electrical machines are widespread in industrial automation, operating them efficiently has significant potential to improve sustainability. Due to the complexity of electrical machines, obtaining direct measurement of energy consumption is challenging and cost intensive. Soft sensors are useful in inferring variables using available measurements in industrial processes. The data-driven approach to developing soft sensors requires a sufficiently large and diverse training dataset. Given the high cost to obtain voluminous sensor data, turning to simulation data as an additional data source is less expensive, although possibly inaccurate. With this motivation, we explore the need and benefit of combining measurement data from intelligent sensors with electrical machine simulation data for building soft sensors. We present an approach to leverage both, sensor measurements and simulation data to develop a soft sensor for energy efficiency. The soft sensor implementation results for an induction motor support the feasibility of the approach.