|Supervisor:||Dr.-Ing. Roland Maas|
|Faculty:||Prof. Dr.-Ing. Walter Kellermann|
|Info:||A critical problem in the surveillance of complex industrial systems is the detection of malfunctions. A false alarm is highly undesirable since the costs of downtime may reach up to one million euros per day. On the other hand, a missed detection can be even more crucial because of subsequent permanent damages incurred to the equipment.|
Typical damages can, e.g., be caused by loose parts coming from the internal coatings or mechanisms. Their impact within the system generates structure-borne noise bursts, which propagate through its structure. The monitoring system then captures those acoustic waves by a set of sensors placed at different locations. However, apart from these loose elements, other non-critical events can cause similar effects in the sensor signals, e.g., due to processes specific to the normal runtime of the industrial system.
A major challenge is therefore the distinction of critical and non-critical events solely based on the sensor signals and some other observable system parameters.
In this thesis, the acoustic event detection problem of structure-borne noise bursts in complex industrial systems shall be investigated based on statistical classification. To this end, a feature analysis for obtaining an optimal feature set is to be carried out and an alternative statistical probabilistic or discriminant method of classification shall be explored and its performance be evaluated.