The other approach was data-based, using an algorithm to learn the system’s behaviour and the influences of parameters such as velocity, acceleration, torque, position and current consumption. The real values are then compared with the learned description so as to define deviations.
In both cases, any potential problems – such as increased friction on a spindle, or wear on a belt drive – could be detected via changes in measured current and torque values, either through increases in the values, or by detecting anomalies through frequency analyses. In either case, an alarm would be raised and the causes shown on a dashboard.
The two approaches to condition monitoring differ not only in terms of their principles but also how the data is evaluated. The model-based evaluation will usually be performed by a control system because it does not need significant computing power. The data-based approach needs machine learning and artificial intelligence evaluations which are normally implemented as cloud applications.
At SPS, Lenze demonstrated two approaches to condition monitoring based on interpreting existing information, without needing extra sensors. Instead, devices in the machine are used as sensors.
Lenze has developed algorithms for various applications and says it can help engineers to turn their knowledge of machines into condition-monitoring models that will improve machine performance.
