Machine learning application : predictive modularization in EPC projects

Since the beginning of the new millennium, companies had to face an increasing number of challenges, such as the continuous reduction of both project delivery time and overall cost. This phenomenon has led firms to explore new digital innovations such as machine learning, to gain advantages using the huge amount of data, generated from projects already commissioned. The purpose of this thesis is to create and validate a methodology for machine learning application in EPC projects. A case study, in one of the global leader company in EPC sector, has been performed to validate the proposed methodology coming from literature review. The thesis starts with an analysis on data gathering in order to collect data in the best way for an easier extraction. Then, pre-processing techniques, needed to analyse data and prepare the dataset for the modelling, will be illustrated. Once the dataset is ready, a machine learning model, capable to learn from data and to predict new plant design parameters, will be created. The model will let the company design a new modular plant starting from a suggested design, avoiding redesigning it from scratch. It will give advantages in terms of: design time reduction, business process optimization, plant early estimation and cost reduction. Moreover, the methodology provided by this thesis helps companies to have different design alternatives with a predictive costs structure which helps them in a better decision-making process regarding project design. Therefore, any company working in EPC projects can use this approach to build up a machine learning model in order to obtain a competitive advantage.