The history of engineering shows many design solutions undergoing further improvements over time to optimize the system and eventually perfect it. However today, with a tendency for design specifications to be much stricter and with ever increasing commercial pressure to save time, materials and cost, there is a need to reach perfectible limits of an engineering system in the first step of the design process. It is therefore increasingly necessary to anticipate the design in a shorter time to maintain the schedule and therefore the competitiveness of the project. This emerging reality is most evident during the early phase of projects when the time is very short to prepare a robust budget in a competitive situation (either between solutions or providers). In this scenario, precise estimation of weight (which is strongly correlated to cost) and a prediction of its evolution during the course of all phases of an offshore project is critical when considering the overall economics, and hence the viability, of a development.
The objective of this paper is to provide a summary of a digital initiative, sponsored by the Onshore/Offshore Global Business Unit (GBU) of TechnipFMC, with the aim of using data science and data mining techniques to improve the operational efficiency of our activities. This digital initiative aims to better understand the evolution of the weight of offshore industrial projects that TechnipFMC designs and identify the impact of each influencing variable.
During the first phase, data mining algorithms are developed to automatically extract the influencing variable data from past project documentation including graphs from scanned documents using data science techniques such as transfer learning. In the second phase, the extracted data is automatically stored in a data base allowing analysis for specific causes of variance in weights. During the third phase, statistical models are used to highlight the influencing variables and their impact on the evolution of weight, giving trends which lead to an understanding of the increase or decrease of the weight. Finally, the developed algorithms can provide an estimation of the topside weight according to the design basis variables of a new project using machine learning techniques.