Using Multilinear Regression to Predict Key Engineering Quantities for Estimation in EPC Projects

Abstract

Abstract Objective The Engineering, Procurement and Construction (EPC) industry has been slow to adopt data-driven Artificial Intelligence (AI) and Machine Learning (ML) systems. Decentralized data storages make data engineering a challenging task, highlighting a recognizable shortage of harnessing data from these data storages, and using ML to optimize estimations for large EPC projects. Empowering proposal teams by using ML-based tools on the engineering disciplines data from previously executed projects can help them make better cost estimates and more informed decisions during project execution. Methods, Procedures, Process The McDermott AI team and domain Subject Matter Experts (SMEs) worked together to improve the process of using initial bills of quantities (BOQ) / Material take-off (MTO) for cost estimation during early stages of the project bidding. We combined the knowledge of EPC engineering experience and data into a real-life application - a ML-based multilinear regression application to predict MTO quantities for different engineering disciplines using MTO data from previously executed projects within the company. The database was developed by compiling MTO data from past projects and applying feature engineering on it. An example material quantity prediction is in Figure 1. Results and Conclusions As organizations look to modernize and optimize processes, ML is an increasingly powerful tool to drive automation. Unlike basic, rule- based automation—which is typically used for standardized, predictable processes—ML can handle more complex processes and learn over time, leading to greater improvements in accuracy and efficiency. The developed ML-based software can help proposal teams to improve initial cost estimates for engineering disciplines key quantities. In addition to this, in the paper we also share our experience and lessons learnt on how to prepare database from decentralized data banks, and how we can pre-process this data for our machine learning model. Novel/ Technical contributions The main aim of the developed ML solution is to harness the power of EPC project data and experience to improve work hour or cost estimation for engineering disciplines. The model has been developed using data from different types of EPC projects executed all around the world. This also presents a case on why and how companies can best utilize the data which is usually stored in decentralized data storage. The solution can assist proposal teams to make better estimates for scope of work based on the type of project. This product could be potentially applied to any EPC company working on oil & gas or petrochemical plants. It could also motivate other practitioners and researchers to replicate the experience to develop a similar tailor-made ML solution for different industries.