Application of anomaly detection with autoencoder in railway maintenance using ultrasonic data
- This paper analyses the applicability of anomaly detection with Autoencoder on ultrasonic data. The data is based on ultrasonic measurements from regular maintenance inspections with standard maintenance trains. This means the varying conditions of rails at different speeds and different environmental conditions remain. This provides practical insights to complement laboratory findings from other research. The data combines signals from eight ultrasonic probes at different angles, merging them into a coherent data set. Autoencoders as a common method for anomaly detection have been used to identify defects and artifacts, based on the reconstruction error value. For that, we experimented with different architectures, network sizes, and node types, such as Fully connected Dense Layers, 2D-Convolutional Layers and LSTM layers. Dense Layer networks performed best, followed by Conv2D and LSTM Neural Networks.
Author: | Georg OlmGND, Thomas HeckelGND |
---|---|
URN: | urn:nbn:de:hbz:294-101170 |
DOI: | https://doi.org/10.13154/294-10117 |
Parent Title (English): | 34th Forum Bauinformatik / 34. Forum Bauinformatik (Bochum, 06. - 08.09.2023) |
Document Type: | Part of a Book |
Language: | English |
Date of Publication (online): | 2023/09/06 |
Date of first Publication: | 2023/09/06 |
Publishing Institution: | Ruhr-Universität Bochum, Universitätsbibliothek |
Tag: | AI; ML Autoencoder; Railway Maintenance; Ultrasonic |
First Page: | 374 |
Last Page: | 381 |
Institutes/Facilities: | Lehrstuhl für Informatik im Bauwesen |
Dewey Decimal Classification: | Technik, Medizin, angewandte Wissenschaften / Ingenieurbau, Umwelttechnik |
open_access (DINI-Set): | open_access |
faculties: | Fakultät für Bau- und Umweltingenieurwissenschaften |
Konferenz-/Sammelbände: | 34th Forum Bauinformatik / 34. Forum Bauinformatik (Bochum, 06. - 08.09.2023) |
Licence (German): | Creative Commons - CC BY 4.0 - Namensnennung 4.0 International |