Real-time monitoring

Machine Learning and A.I.

the ultimate decision-making support system

Predictive analysis
of future vessel performance

At METIS we use state-of-the-art Artificial Intelligence and Machine Learning technologies that enable us to predict the future behaviour of a vessel under variable operating conditions.

The predictions are made with respect to specific quantities and performance parameters (i.e. M/E or D/Gs Fuel Flow/consumption, M/E Power, T/C RPM, Scavenge Air Pressure, etc.). 

  • The first step of the process is to feed historical data to the machine learning algorithms in order to get trained. 
  • The next step is to validate the accuracy of the algorithms using a secondary set of past data.
  • The last step should be evident: a well-trained and verified model can be used to generate reference lines given real-world conditions. For example, propeller power versus speed or main engine fuel consumption versus propeller power and so on. Based on these reference lines METIS is able to produce predicted values for all the critical vessel quantities that define the behaviour of the vessel.

Machine Learning Predictive Models developed by METIS have the following characteristics :

  • Vessel-Specific
  • Dynamic, constantly adapting to new conditions
  • Self-trained, shelf-evaluated and self-improved in order to ensure the highest accuracy of the predictions provided

Combining machine learning with state-of-the-art mathematical models yields powerful processing chains, able to provide actionable intelligence to drive decisions in the real world.

As an example, our models can predict at significant accuracy,  the power of the vessel in various external conditions and provide a clear estimation of the effect of each external force (Wind, wave, swell, hull& propeller fouling) on the propulsion power of the vessel.

In another example, the goal is to “learn” the complex relation connecting shaft power to shaft RPM, turbocharger RPM, scavenge air pressure and fuel pump mark.

Dedicated Machine Learning models, provide predictions of all critical parameters and are made available to the user through multiple dashboards on the METIS web platform. Additionally, the user can make ad-hoc requests through the METIS chatbot or even schedule periodic delivery of a custom report through email.

METIS - Main Engine Monitoring
Machine Learning techniques to ensure data reliability

quality and accuracy assurance of
Sensor Measurements

Even the most advanced machine learning models have one weak point, this is the quality of the data to be used. The advantage that METIS provides is that is in full control of the data acquisition process ensuring the frequency and the quality of data gathered. However, on top of all the standard processes involved in the data acquisition workflow, METIS utilizes machine learning algorithms that monitor the data as an extra measure to ensure the quality and reliability of the analysis. Machine Learning models are assigned the task to identify irregular signal patterns with respect to the signal itself or even to other related signals. The results are well worth the effort and indeed quite impressive.

innovative, Machine-Learning-based user interface

Natural language processing

One of the most innovative aspects of the METIS platform is the availability of the first chatbot service specifically designed for the shipping industry. The METIS chatbot as a Virtual Personal Assistant is able to interact with the users in plain English in order to help them retrieve instantly, any kind of information, or to provide specialized analytics on operational as well as technical domains.

All this functionality behind the concept of Virtual Personal Assistant is implemented utilizing a combination of machine learning and big data analytics. The task for the system is to understand the requests provided by the users in plain human language (English) and provide the relevant content. 

Applying Natural Language Processing algorithms METIS is able to achieve the highest levels of understanding while at the same time the system is continuously trained to adapt to the specific characteristics of each user.