The reliability of Heavy-Duty Vehicles (HDVs) is critical for continuous operations in sectors like transportation and logistics. However, the complexity of these vehicles' subsystems, including the Air Pressure System (APS), poses significant challenges where failures lead to costly downtimes and safety risks. This paper introduces a novel semi-supervised anomaly detection approach based on a Long Short-Term Memory Autoencoder (LSTM-AE) model to identify APS failures in HDVs. Leveraging 30 days of operational time-series data from 140 vehicles, of which 30 experienced APS failures, our study presents a semi-supervised formulation of the problem bypassing the limitations of supervised classification and addressing the scarcity of labeled data in the real-world scenarios. After applying several essential preprocessing steps, the proposed model was rigorously trained and validated to ensure robustness. It achieved an F1 score of 0.75 with a corresponding accuracy of 91.4%. The proposed framework in this research promotes enhanced vehicle uptime and improved safety standards, providing practical implications for both HDV manufacturers and operators.


    Zugriff

    Zugriff prüfen

    Verfügbarkeit in meiner Bibliothek prüfen

    Bestellung bei Subito €


    Exportieren, teilen und zitieren



    Titel :

    Air Pressure System Failures Detection Using LSTM-Autoencoder


    Beteiligte:
    Mumcuoglu, Mehmet Emin (Autor:in) / Farea, Shawqi Mohammed (Autor:in) / Unel, Mustafa (Autor:in) / Mise, Serdar (Autor:in) / Unsal, Simge (Autor:in) / Cevik, Enes (Autor:in) / Yilmaz, Metin (Autor:in) / Koprubasi, Kerem (Autor:in)


    Erscheinungsdatum :

    26.06.2024


    Format / Umfang :

    1734606 byte




    Medientyp :

    Aufsatz (Konferenz)


    Format :

    Elektronische Ressource


    Sprache :

    Englisch