Autonomous Load Profile Recognition in Industrial DC-Link Using an Audio Search Algorithm

Abstract

Industrial manufacturing plants, including machine tools, robots, and elevators, perform dynamic acceleration and braking processes. Recuperative braking results in an increased voltage in the machines' direct current (DC) links. In the case of a diode rectifier, a braking resistor turns the surplus of energy into lost heat. In contrast, active rectifiers can feed the braking energy back to the AC grid, though they are more expensive than diode rectifiers. DC link-coupled energy storage systems are one possible solution to downsize the supply infrastructure by peak shaving and to harvest braking energy. However, their control heavily depends on the applied load profiles that are not known in advance. Especially for retrofitted energy storage systems without connection to the machine control unit, load profile recognition imposes a major challenge. A self-tuning framework represents a suitable solution by covering system identification, proof of stability, control design, load profile recognition, and forecasting at the same time. This paper introduces autonomous load profile recognition in industrial DC links using an audio search algorithm. The method generates fingerprints for each measured load profile and saves them in a database. The control of the energy storage system then has to be adapted within a critical time range according to the identified load profile and constraints given by the energy storage system. Three different load profiles in four case studies validate the methodology.