AI-based sensor signal monitoring at GfM

Ensemle Learning

Condition monitoring of mechanical drives generally involves measuring mechanical vibration and subsequently analysing the signal, looking for patterns that indicate irregularities. This works well and provides reliable and robust diagnostic information that can be used directly by the maintenance organisation. Operators of plant with mechanical drives derive a direct financial benefit from this.

As analysing vibration signals is not only a time-consuming process but, above all, one that becomes tedious over time, the founders of GfM began automating data analysis over 35 years ago. This automation is routinely utilised in GfM’s Peakanalyzer online condition monitoring system. Only in this way is continuous online spectral analysis of all measurement signals possible, which ultimately ensures maximum diagnostic reliability.

However, high data quality is a prerequisite for reliable condition monitoring. This applies not only to the selection of the correct sensors, their frequency response and, ultimately, the sampling rate, but also to ensuring that their functionality is maintained over the long term. Sensors themselves are generally not subject to any significant ageing process. So that is not the problem. However, sensors can become detached from the machine, sensor cables can be exposed to harmful mechanical stresses, and, finally, electrical problems can arise. If the analysis of sensor signals is carried out manually by a specialist, they will recognise when a signal is implausible. For fully automated machine diagnostics, an algorithm is required that performs this task automatically as well.

Years ago, GfM took on this challenge and set out to find a solution. As GfM has a large pool of annotated data sets at its disposal, it was obvious to utilise artificial intelligence tools. Ultimately, an AI-based plausibility check was developed, which is consistently applied to all sensor signals.

The scalable Peakanalyzer online condition monitoring system, with a maximum of 64 channels, comes equipped with this function as standard. This ensures that the already high diagnostic reliability of GfM systems is maintained even in adverse environments.