Description

Towards providing a holistic tool, PODIUM will be a designed as containerized micro-services, that will fulfil requirements related to data storage, latency and security. The functionalities of each component are:

1. Timeseries Database: A time-series database is a software system that is optimized for the handling of data organized by time. Time series are finite or infinite sequences of data items, where each item has an associated timestamp, and the sequence of timestamps is non-decreasing. Hence, this database will enable the real-time monitoring of the high frequency-sensor data while storing the historical data. The latter is mandatory to retrain/update the machine learning models for RUL estimation. To this end, we propose the RedisTimeSeries database as it is open source, also providing plugins for Grafana monitoring tool (see below) and most programming languages.
2. MySQL Database: A common SQL database, where generic information can be stored such as spare parts, or maintenance information (i.e., date, duration, technicians’ info).
3. Grafana Monitoring: Grafana is an open-source solution for running data analytics, pulling up metrics that make sense of the massive amount of data & to monitor our apps with the help of cool customizable dashboards. Grafana will be used to monitor real-time streaming data from sensors via the time-series database. Also, Grafana offers a built-in alert service in order to notify the users when an urgent abnormal behavior is present.
4. Custom RUL AI estimator: The model for RUL prediction is comprised of two sub-models. The first one leverages a Random Forest Classifier (RF) to predict whether the system in question is in normal or in degradation operation. In the case of normal operation, RUL estimation is not applicable. To this end, it could be applied an anomaly detection service based on the work done by our team and presented in 2020 IEEE 23rd International Conference on Intelligent Transportation Systems. Where, various approaches for anomaly detection were implemented on time-series data, utilizing machine learning on vessel’s sensor data, to predict the condition of specific parts of the vessel’s main engine and thus facilitate predictive/preventive maintenance. However, when the RF detects degradation in a system, the sensor data is used as input to the LSTM RUL Estimator which outputs the remaining useful life of the degraded component/system.

Economic sectors

Architecture diagram

Innovation

In forecasting, the degradation of a component or system is usually a non-linear function of many parameters. The degradation process accelerates over time until complete decomposition occurs. This makes fault prediction particularly difficult with conventional methods.

• The main contribution is a complete framework leveraging real time data for various maritime defect prediction. The framework consists of two major components, a classification mechanism classifying the conditions of the main engine of the ship according to the sensor data and a time-series forecasting method consuming the time-series when the main engine runs under abnormal behavior.
• The main difference of the proposed work compared to the existing ones, is the introduction of an innovative classification approach leveraging both supervised and anomaly detection methods as well as a probabilistic RNN model for timeseries forecasting. Furthermore, such AI techniques are not yet trusted and adopted to the maritime industry, thus our vision is to kame PODIUM a trusted and useful tool.

Innovation