@article {pub4310,
	title = {Real-World Anomaly Detection with Siamese Neural Networks and Digital Twin Systems},
	author = {Andrea Castellani AND Sebastian Schmitt AND Stefano Squartini},
	year = {2020},
	abstract = {Nowadays, with the continuously growing /mount of monitored
data present in Smart Company environment, the need for
Anomaly Detection technique has become more relevant in
order to identify some anomalous behavior from time-series
data generated by sensors. With the Digital Twin, a detailed
simulation of a complex physical system, it is possible to provide
more data to feed in a Machine Learning algorithm and thus
help in the anomaly detection task. This paper presents an
approach to the Anomaly Detection on real-world multivariate
sensor data, by the use of the Digital Twin to generate a synthetic
dataset. Those data are used to training some Artificial Neural
Networks (Autoencoders) in an unsupervised fashion, with the
aim of detecting anomalies in a real-world dataset. Then, a
novel method based on Siamese Autoencoder is presented. It
is aimed to improve the performance of such unsupervised
algorithms by only including a rather small subset of labeled
data in a weakly-supervised fashion. Furthermore, an end-to-
end Machine Learning model is presented, which is able to
work with raw data. De facto, it has the same structure of the
Siamese Autoencoder but uses Convolutional Neural Networks to
extract automatically meaningful features from the input data.
All the developed algorithms are compared with the state-of-the-
art approachfor Anomaly Detection and the overall results are
very promising. They show that the presented methods, during
testing, reach an ROC AUC measure and F2 score noticeably
higher the recent state-of-the-art Anomaly Detection techniques
chosen for comparison.},
	publisher = {IEEE},
	journal = {IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS}
}