Data imputation of missing values from marine systems sensor data. Evaluation, visualisation, and sensor failure detection

Velasco-Gallego, C and Lazakis, I; (2021) Data imputation of missing values from marine systems sensor data. Evaluation, visualisation, and sensor failure detection. In: RINA Maritime Innovation and Emerging Technologies Online Conference 2021 Proceedings. Royal Institution of Naval Architects, London.

[thumbnail of Velasco-Gallego-Lazakis-RINA-2021-Data-imputation-of-missing-values-from-marine-systems-sensor-data]
Preview
Text (Velasco-Gallego-Lazakis-RINA-2021-Data-imputation-of-missing-values-from-marine-systems-sensor-data)
Velasco_Gallego_Lazakis_RINA_2021_Data_imputation_of_missing_values_from_marine_systems_sensor_data.pdf
Accepted Author Manuscript

Download (595kB)| Preview

    Abstract

    To enable Condition-Based maintenance, sensors need to be installed, and thus Internet of Ships (IoS) needs to be implemented. IoS presents several challenges, an example of which is the imputation of missing values. A data assessment imputation framework that is utilised to assess the accuracy of any imputation model is presented. To complement this study, a real-time imputation tool is proposed based on an open-source stack. A case study on a total of 4 machinery systems parameters obtained from sensors installed on a cargo vessel is presented to highlight the implementation of this framework. The multivariate imputation technique is performed by applying Kernel Ridge Regression (KRR). As the explanatory variables may also contain missing values, GA-ARIMA is utilised as the univariate imputation technique. The case study results demonstrate the applicability of the suggested framework in the case of marine machinery systems.

    ORCID iDs

    Velasco-Gallego, C and Lazakis, I ORCID logoORCID: https://orcid.org/0000-0002-6130-9410;