Print Friendly, PDF & Email

Artificial Intelligence, Digital Twins, data-driven ship design, drones – this year’s conference on Computer Applications and Information Technology in the Maritime Industries, COMPIT 2018, brought together the forerunners of innovative thinking in the maritime sector, reports Hans Payer

More than 70 attendees from many European countries, USA, Canada and Asia got together in the Castello di Pavone in[ds_preview] the Piemont region, Italy, to discuss 45 papers on future developments in shipping and shipbuilding with today’s and future technological innovations. Artificial intelligence, artificial neural nets, knowledge-based systems, Digital Twins, Virtual and Augmented Reality, all the way to the autonomous ship were addressed.

Artificial Intelligence

Mankind has long been fascinated by Artificial Intelligence (AI). In the beginning it was Science Fiction films where man built smart robots which fulfilled various tasks autonomously. Several films developed stories of successful Artificial Intelligence as well as cases where things went wrong and mankind lost control. As we make rapid progress with the increasing digitalization of our world, the importance of ethical principles in dealing with Artificial Intelligence is becoming obvious. Several of the larger companies working on AI, such as Google or Amazon, take initiatives today to assure ethical conduct always keeping

the technology under control. Volker Bertram, organizer of the Conference, presented the paper Demystify Artificial Intelligence for maritime applications, explaining the key technologies used today.

The principles of machine learning are described going from simple examples to more complex tasks. Starting from data sets from concrete processes we frequently try to find regularities and patterns which can be used for extrapolation to new designs. The human brain is very good at trend spotting and pattern recognition. Machine learning tries to mimic this ability and take it further.

The most popular technique in machine learning and AI is Artificial Neural Networks (ANN) which allows mapping of multi-dimensional input/output data sets for an arbitrary number of input and output variables. An ANN structure may consist of several layers, with large numbers of nodes on each level, somewhat comparable to the brain. ANNs are convincing, nevertheless they do have limitations. They cannot predict the unpredictable, such as random events, or chaotic behavior, e.g. crash-stop maneuvers of ships.

Machine learning needs vast amounts of data points. Real-world problems in shipping for instance, depending on many factors, operational as well as ambient conditions, may easily lead to billions of data points required to train a neural network. This is a formidable job, even by today’s standards.

ANNs have been successfully used in system identification, deriving design formulas, to create response surfaces for interpolation of simulation results, automatic ship type identification, economic predictions etc.

The ambition is to extend automation on board through artificial intelligence further to a degree where the ship will eventually operate by itself, the autonomous ship. The operation of all components as well as the complete ship system are automatically controlled by mechanical and electronic devices that take the place of humans in observation, decision making and reaction. Technically we have come a long way. Nevertheless we cannot completely replace the human being. Jukka Merenluoto from DIMEC, Finland said: »The automation of ships will continue. We’ll see autonomous ships, but they will not be unmanned.«

The Digital Twin journey

With today’s possibilities we can create a complete and accurate digital representation of a ship, maintained during the construction period and throughout the operational life of the vessel. Such a Digital Twin is the platform of platforms. It should contain all relevant information about the vessel and should be contained within a single software platform, which is flexible enough to communicate with other related platforms.

Digital Twins have been in use successfully in several industries. Rolls-Royce for instance offers Digital Twins for airplane jet engines or thrusters for speed boats, usually maintaining the models in their data centers. Maintenance and replacement requirements are determined with the Digital Twin on the basis of operational data from the actual unit and prepared at suitable service centers worldwide. NASA confirms a Digital Twin for a spacecraft, manned or unmanned, as ultra-realistic. Christian Cabos, DNV GL, proposes to test a software update on the Digital Twin and only if it is successful to roll it out for the real object.

Setting up a comprehensive data model for a ship is quite elaborate as ships are very complex and mostly unique or from small series. Thus, Digital Twins are still rare for ships.

For Denis Morais from SSI, Victoria, Canada, the benefits of a Digital Twin in shipping are real, (see HANSA 5/2018). The creation of a Digital Twin of a ship will start from the geometric data and attributes coming from the engineering CAD tools, but include much more to finally represent the as-built condition of the vessel. The twin will always be up-to-date with the real ship, not only for structure and machinery, it must also have links to class approvals, simulations and calculations as well as paint information, to mention only a few. Ultimately, the system could include the use of »Internet of Things-sensors« to automatically keep the physical ship and its digital representation synchronous at all times.

The modern shipyard will benefit largely from having a Digital Twin available during construction of a ship and afterwards. But the main benefit will go to the ship owner/operator interested in higher safety and savings in operation over the lifecycle of his vessel. Several containership operators, for instance Maersk and Hapag-Lloyd, have signed up for DNV GL’s Digital Twin software to manage a virtual model of each of their vessels worldwide in the cloud.

Ships result from multi-tier collaborations between designers and manufacturers. With new »outcome-based« business models, shared responsibility for the integrity and performance of components and systems during operation is emerging. Therefore multi-tier Digital Twins will emerge in the industry. This means that condition data, asset representation, and behavioral models will be shared. Because of obvious interests for safeguarding intellectual property, access rights and change management will be important aspects of a system implementing shared Digital Twins.

Data-driven ship design

The computational power – speed and capacity – has been growing dramatically in recent years, bringing new possibilities for data collection and knowledge extraction from data never seen before. Good decision making comes from good data via DIKW, Data, Information, Knowledge and Wisdom hierarchy; KDD, Knowledge Discovery and Data mining; and DDD, Data-Driven Decision making. The expanded computational possibilities make new design methods feasible.

Henrique Gaspar from Brasil, professor at NTNU, Alesund, describes how efficient data-driven decision making is connected to a combination of intuition and analysis of data. Traditional methods of knowledge extraction and visualisation of information from data such as regression analysis and other fitting models are today augmented by methods such as clustering and text mining. In data-driven ship design data is both, studied top-down from regressions, previous designs and parametric studies based on existing solutions, as well as bottom-up, where it is connected to specific key elements and subsystems that directly affect the mission of the specific ship design. Data-driven ship design aims to integrate product and process content according to modern DevOps practices, such as versioning, tracking, monitoring, testing, tuning and feedback. An open and collaborative ship design library is briefly introduced as an initiative aiming to incorporate data-driven methods in ship design. A call for collaborative data-driven ship design using open standards closes the paper.

This year’s Compit Award went to Stefan Harries for the paper »Appification of Propeller Modeling and Design via CAESES«. The paper describes a web-based application (webApp) for geometric modeling and design of propellers. The app builds on CAESES, a flexible computer aided engineering environment which allows offering selected functionality sub-sets via a standard web browser. For the webApp the expertise of a propeller designer is combined with the design data of the Wageningen B-series. Building on CAESES’ parametric modeling techniques, a propeller, including blades, hub and fillets, is generated with just a handful of inputs. The App provides a watertight geometric model and determines the efficiency. In the final step, the geometry can be downloaded, e.g. as an STL file to be used in a numerical propulsion test or for 3d-printing.

Several papers proposed more use of the smartphone or tablet computer for design and inspection. Designers have easy access to design data and design software, surveyors have the calculation results or even the digital twin at their finger tips. Examples were shown. Herbert Koelman from SARC, The Netherlands, enquired: »why would you use the small screen of an iPhone rather than a bigger high resolution screen of a lap top?« The session chairman noted: The young generation sees the world differently from the established – older – generation.

The human element

Despite the ongoing dramatic developments in naval architecture the Human Element, HE, remains an important factor in risk and safety assessments in ship design and operation. Evaluations of ship accidents have shown that human error is the primary cause in many cases. IMO has reacted by setting up the International Safety Management Code in 1993, mandatory for sea-going vessels since 1998. It takes account of the human element, prescribing a. o. procedures and clear distribution of competence on board and ashore.

David Andrews from University College London points out that increased automation on board and employment of artificial intelligence in design and operation may reduce the risk of human error. As the tasks of the crew change and the complexity on board increases, however, a reevaluation of the human factor in the design phase as well as in normal operation is called for.

Augmented Reality

Stephan Procee and Delft University present their work using Augmented Reality to improve collision avoidance. Based on cognitive work analysis an ecological interface is designed and built for an Augmented Reality application, including the concept of velocity obstacles.

A ship domain, SD, is established to derive a realistic criterion for discriminating potentially dangerous targets from the rest. This is the basis to generate acceptable, effective alarms in the augmented reality interface. Studies show that the SD seems to be elliptical, about twice as long as it is wide, sometimes symmetrical around the own ship, sometimes shifted towards the bow. The basics of SD for use in the AR interface can be derived from these analyses.

This innovative way to visualize the problem space and solution space provides the navigator with real-time information about the possible combinations of course and speed that avoid intrusion into another ship’s protected zone. This results in better situation awareness, leading to less close encounters such as near misses, and fewer collisions.

Automated surveys

Replacing surveyor inspection by drone-based inspections is promising and has many advantages particularly for tankers and other large vessels. Two fields are important, drone-based inspection and automatic detection of structural defects such as cracks in the structure. Erik Stensrud from DNV GL describes results from an extensive R&D project into autonomous vessel inspections. Two crack detection models are presented using Convolutional Neural Networks, CNN.

The first model aims to classify cracks in static images by assigning a label to the image indicating the existence of the crack. Pre-trained classification CNNs were adapted to the model and their learned weights transferred by fine-tuning to the crack classification task. The second model aims at localizing the crack in a pixel-wise manner. It is a fully convolutional network. The two models were evaluated on a dataset which is limited but shows substantial variation in image quality. The study demonstrates the feasibility of machine-learning based crack detection but also shows the challenges of developing highly accurate and robust models for segmenting and classifying cracks.

Based on his experience with neural networks and artificial intelligence, Erik Stensrud observed: »The more we learn to know the limitations and dangers of Artificial Intelligence, the more we appreciate the marvel of the human mind.« The audience agreed.

Shipping becomes an early mover

In the past ten years or so the shipping industry was slow to pick up new trends and possibilities linked to digitalization. Compit 2018 conference strikes us, this is no longer so. The 45 conference papers show that maritime research institutes, shipping and shipbuilding are taking up the new possibilities and challenges created by rapid advancements. The conference was well attended to the end.

Denis Morais of SSI summarized his impression at the end of the conference: »This year seemed to be an explosion – in a good way – of hot topics that relate to the shipbuilding and shipping industry. Pretty much every type of technology we have been reading about was well represented at the conference.«

»We got the impression, we are at the cutting edge of developments,« as Johannes Hyrynen from VTT, Finland, stated. »In the future, the shipping industry will be radically different from what it is now.«

Only a few examples could be described in this synopsis of the conference. Those interested in more details should get hold of the proceedings online at www.compit.info.