Digital Twin Model for Elevator Anomaly Detection: A LOF Approach

Abstract

Digital twins provide the capability to transfer real-time data into a virtual environment through
sensors, enabling the detection and prediction of abnormalities in production processes. A review of the
literature reveals that digital twins play a significant role in improving efficiency in production processes
and have become a crucial element in industrial competition. When anomalies are predicted and detected
in advance, our ability to intervene increases. This allows for the prevention of potential problems,
minimizing damage, and facilitating the implementation of predictive maintenance activities. Furthermore,
it reduces costs resulting from unexpected failures and contributes to the reliable operation of systems.
In this study, data from three sensors installed in an elevator were collected to attempt to create a digital
twin of the elevator. The aim was to detect anomalies in the collected data and improve effectiveness
through predictive maintenance. Real-time data analysis and anomaly detection were facilitated using the
Local Outlier Factor (LOF) algorithm, an anomaly detection algorithm. LOF evaluates the uniqueness of
each event based on its distance to its k-nearest neighbors. It is an unsupervised anomaly detection method
advantageous for cases where labeling large amounts of data is not feasible. In total, we collected 107,267
data points, of which 4,734 were identified as outliers, enabling us to comprehensively analyze the reasons
behind their outlier status.
The intended contribution of this study is to demonstrate that the creation of digital twins in systems leads
to the detection of anomalies in production processes, thereby increasing efficiency and reducing costs by
minimizing unplanned downtime.