Offshore Engineer - March 2017
Neural Networking By Design
Artificial neural networks are being used to help design floating production system hulls in the concept and frontend engineering phases. Elaine Maslin reports. Artificial Intelligence (AI) has long been an idea confined to science fiction films. But, the use of AI is creeping into our daily lives: think about Siri on your iPhone telling you how long it will take to get home. The power of processing – artificial neural networks, specifically – is also starting to make a mark in floating production vessel design for the offshore upstream industry. Artificial neural networks (ANN) are a type of information processing inspired by how biological neural systems, such as the brain, process information. ANN systems are composed of a large number of highly interconnected processing elements (i.e. neurons in the brain) working in unison to solve specific problems. What makes them useful is that they, like people, learn by example. They can be trained (values are given weight and treated accordingly) and can be configured for specific applications, such as pattern recognition or data classification. To oversimplify it, if you fed one (which has been purpose designed) a pile of floating production, storage and offloading (FPSO) vessel design data, it would learn from the data and be able to do the analysis that you need for initial sizing of new FPSOs. Libra and her possible smaller sisters generated using artifi cial neural network technology. This is Norway-based Inocean’s aim. Inocean, founded in 1996 (but, now part of TechnipFMC), is a naval architecture firm focused on floating production vessel design, with 30 FPSO newbuild and conversion projects under its belt. It also has bright young engineers who want to work with ANN techniques. “We have a lot of information and data [from] over the last 20 years that we want to use,” says Frode Kaafjeld, managing director, Inocean. “This new technology [ANN] opens up new ways of using this data.” Frode Kaafjeld By using existing data (former FPSO designs), you can train an ANN with an iterative algorithm to estimate the functional relationship between the input (main dimensions, draft, block coefficient etc.) and the output (hydrodynamic sectional loads). Normally, in the concept or front-end engineering and design phase, various iterations of an FPSO design will be evaluated, to find an optimal design within a set input criteria (storage capacity, risers, metocean conditions, etc.), which is a time consuming process because it includes calculating hydrodynamic sectional loads – for each iteration. This means using the likes of linear diffraction/radiation analysis and stochastic post-processing. Using an ANN estimator speeds up the entire process. “In order to speed up this process, we together with Technip are developing an automated iterative algorithm for sizing of FPSOs,” Kaafjeld says. “This algorithm works through numerous size combinations and automatically recalculates the effects on weight, stability, motions, etc., by integrated routines.” ANN is not new. It was first developed in the 1950s and 1960s, but the use of graphics processing units found in highend graphics cards has helped the creation of so-called Deep Nets, enabling computers to beat humans at quite complex games, such as the complex Chinese board game Go. For FPSO design, however, this level of system isn’t necessarily required, says Espen Engebretsen, one of the new generation Msc Naval Architects (born 1987). Inocean is drawing TechnipFMC’s FIDE (Floater Integrated Design Environment), a sizing tool that enables a trained user to perform sizing of spars, semisubmersibles and tension leg platforms (up to scantling level for class approval). Inocean and TechnipFMC have cooperated to include sizing of ship-shaped FPSOs into FIDE and are currently working on integrating the ANN for estimation of hydrodynamic sectional loads. “Within FIDE we have seen a great advantage of estimating hydrodynamic sectional loads without having to utilize the traditional diffraction/radiation analysis, which becomes very time consuming when a large number of size iterations are evaluated,” Engebretsen says. “By using [ANN], we have been able to create an estimator, which utilizes Inocean’s vessel database to instantly estimate hydrodynamic sectional loads.” While the tool’s development can take months (with the data it was fed carefully selected to provide an adequate representative database), and the algorithms built, the training is fast and once it is done, the results are also fast, Engebretsen says. “Work only took a few months,” he says. “ANN can be confi gured in multiple ways. What took time was optimizing it and avoiding some traps.” Inocean has trialed the system, performing initial hull sizing, but then verifying the results using traditional manual calculations, with positive results. As an example, the firm has produced indicative designs for sister vessels to the Libra FPSO (an Inocean design), using parts of the ANN algorithm it developed. The image shown depicts how the ANN algorithm works – how the designer can trim the design until a desired result is achieved. Of course, this is a conservative industry. Classification societies will not allow a design produced this way to be classified. What the tool can help with is getting through the initial iterative process faster, as the optimal hull size is obtained in a shorter amount of time, i.e. potentially helping to reduce costs, and before starting traditional calculations. The system is also being continuously developed and improved. “We still believe that we need to do the traditional hydrodynamic and structural hydrostatic calculations for the design, but as time goes by we will be able to integrate this analysis more into our tools,” Kaafjeld says. Espen Engebretsen A tender could be produced far faster, with options given, which can be quickly altered according to a changing specifi cation – additional risers, bigger tanks, etc. What’s interesting for Kaafjeld is that a new generation of engineers are pushing these techniques, but working alongside more experienced engineers. And hopefully, just as this technology is showing its capability in FPSO design, Kaafjeld thinks the FPSO market might be growing again. “The market for FPSOs in general seems to be growing again,” Kaafjeld says. “A lot of projects seem to be starting up again. After the drop in the market, we see clients that are more demanding and more conscious of what they want. They want more optimized solutions, they want cost-effectiveness and there is the ‘green’ dimension. All this is affecting the design of FPSOs.” Just as individual humans are not quite ready to put their lives in the hands of AI, floating production systems will not yet be fully designed using ANN. But, AI certainly has a part in making the process faster, reducing the time and effort required in that otherwise long iterative design process.
Published by AtComedia. View All Articles.
This page can be found at http://mydigitalpublication.com/article/Neural+Networking+By+Design/2721646/387465/article.html.