Industrial robots play an increasingly important role in agrowing number of fields. Since the breakdown of a singlerobot may have a negative impact on the entire process,predictive maintenance systems have gained importanceas an essential component of robotics serviceofferings. The main shortcoming of such systems is thatfeatures extracted from a task typically differ significantlyfrom the learnt model of a different task, incurringfalse alarms. In this paper, we propose a novel solutionbased on transfer learning which addresses a well-knownchallenge in predictive maintenance algorithmsby passing the knowledge of the trained model from onetask to another in order to prevent the need for retrainingand to eliminate such false alarms. The deployment ofthe proposed algorithm on real-world datasets demonstratesthat the algorithm can not only distinguish betweentasks and mechanical condition change, it furtheryields a sharper deviation from the trained model in caseof a mechanical condition change and thus detects mechanicalissues with higher confidence.
Robust Predictive Maintenance for Robotics via Unsupervised Transfer Learning
2021-04-18
doi:10.32473/flairs.v34i1.128451
The International FLAIRS Conference Proceedings; Vol 34 (2021): Proceedings of FLAIRS-34 ; 2334-0762 ; 10.32473/flairs.v34i1
Article (Journal)
Electronic Resource
English
DDC: | 629 |
Robust Predictive Maintenance for Robotics via Unsupervised Transfer Learning
BASE | 2021
|Learning-supported and force feedback model predictive control in robotics
BASE | 2021
|Robotics in Highway Construction and Maintenance
Online Contents | 1995
|Aerial Robotics for Inspection and Maintenance
GWLB - Gottfried Wilhelm Leibniz Bibliothek | 2022
|