Abstract:
In this paper, the data-driven safety level assessment problem for dynamic systems with only normal scenario data is studied, i.e. no abnormal scenarios data are availabl...Show MoreMetadata
Abstract:
In this paper, the data-driven safety level assessment problem for dynamic systems with only normal scenario data is studied, i.e. no abnormal scenarios data are available. An expert-augmented data-driven safety level assessment framework is proposed, which combines expert knowledge and data-driven algorithms to tackle the problem. The user-defined high-level description of the system is introduced firstly, which consists of the attribute that denotes specific subsystem is abnormal or not. In this way, expert knowledge can be represented as the attributed description prototype matrix with corresponding safety levels. Considering the data is non-stationary and usually comes in the stream, the fast incremental support vector data description (FISVDD) algorithm, which can be trained incrementally, is used to detect the attribute description based on the online measured variables. At last, the nearest neighbor search is used to determine the safety level by calculating and sorting similarities between the attribute description and prototypes. A practical case study based on JiaoLong deep-sea manned submersible demonstrates the effectiveness of the proposed scheme.
Published in: 2021 CAA Symposium on Fault Detection, Supervision, and Safety for Technical Processes (SAFEPROCESS)
Date of Conference: 17-18 December 2021
Date Added to IEEE Xplore: 01 February 2022
ISBN Information: