Predictive Performance Indicators for Capital Projects

This project investigates the applicability of multi-attribute predictive techniques such as neural networks or genetic algorithms to capital project performance assessment and prediction. A neural network is a learning tool wherein nodes are connected together to form a network. Within this network, algorithms operate, which alter the strength of network connections (nodal relationships) based on analyses of representative information from real-world environments such as completed capital project operations. These techniques seek to determine underlying network structures in order to understand, generalize, and predict future behaviors based on knowledge obtained from past events, occurrences and outcomes. The concept for this project is to “train” solution methodologies with respect to maximizing multiple dimensions of project performance using previously collected logistics service operations data. The historical data will be reviewed and organized to maximize fidelity to support neural network training. Typically, a portion of the available historical data (e.g., two-thirds) is used to train the model or proposed solution approach such that the remainder of the historical data (e.g., one-third) can be used to test the viability of the predictive method. In this way, potential performance can be simulated using historical data to assess the efficacy of the advanced predictive approach with respect to anticipating project performance in near real-time.

Benefits

This tool will be used to monitor a capital project’s life cycle and behavior to assess its “health” in near real-time to enable identification and timely corrective actions to mitigate schedule disruptions and cost overruns. The benefit realized by this tool is more effective project execution, which will translate to better project schedule adherence, risk management and lower overall project costs.

Opportunities/Status

This project was launched Q1, 2012 and is in the formation phase. To ensure completeness and viability, stakeholders and SMEs from any Materials Management Enterprise area are invited to participate and/or cosponsor this project projects.

Fiatech Staff Project Manager

Reg Hunter, This email address is being protected from spambots. You need JavaScript enabled to view it.

Project Leads

Dr. Scott J. Mason, Clemson University
John Fish, Ford, Bacon & Davis/S&B Engineers & Constructors
Ken Long, Panprojects
Dr. Bill O'Brien, The University of Texas at Austin

Additional SMEs needed to identify and develop optimal approach to this project including: refining success criteria and setting expectations; developing requirements, delineating tasks, refining deliverables; allocating and managing action items; setting and maintaining schedule; defining capital project outcome categories; evaluating and conditioning in-transit data, categories, attributes, and forcing functions; securing historical data to train and verify model; defining status dashboards; and performing viability and value assessments.

Participating Companies

Bechtel, Clemson University, HAL, Panprojects, S&B, and The University of Texas

Deliverables

The primary deliverable from this research project will include both the multi-attribute, predictive modeling tool described above and a user’s guide to allow Fiatech members to gain access to and use the tool. The guide will demonstrate the use of the tool with a pertinent example dataset and describe sources of potential data and lessons learned from the research process regarding effective data source identification, structuring and usage. The project will be conducted in phases with significant interaction among the stakeholders, subject matter experts and researchers. As such, the following list highlights the sequence that will be used on the project: research team identification and formation; dataset identification, selection and restructuring; desired performance measure identification; preliminary model concept specification inputs, required calculations and outputs; model trade studies, development and training using available historical data; alpha-version prototype model development and demonstration; user community assessment and feedback/recommendations; beta version predictive model for selected field testing and use by participating companies; final delivery of predictive modeling tool, user guide and lessons learned report; and presentations of research results at Fiatech member meeting, conferences and webinars.

Completion Date

Fall 2012