A hybrid methodology to estimate construction material quantities at an early project phase

Abstract

Preliminary project cost estimates are the first serious estimates made on a project. They play an important role during the decision-making process, and are the benchmark with which future estimates are expected to agree. This paper concentrates on the estimation of construction material quantities (CMQs) and presents a methodology to accurately estimate them during an early project phase. We make use of existing data and utilize regression analysis, neural networks and case-based reasoning to provide accurate results. It encompasses data collection, model development and evaluation, and the integration of different techniques. The use of the methodology is demonstrated by estimating CMQs of relevant structures. The accuracy of the methodology is investigated and compared with three state-of-practice approaches. The results obtained show a significant improvement over the state of the practice, and would improve the accuracy of preliminary project costs estimates. Through partial automation, it would likely reduce the time required to make estimates.

Keywords:

• resource estimates
• preliminary estimates
• hybrid estimation models
• neural networks
• backward elimination
• case-based reasoning
• construction material quantities

Acknowledgments

The authors would like to thank the capital expenditure (CAPEX) department at Holcim for financing and making their data available for this research. Special thanks are given to Mr. Rudy Blum and Mr. Roberto Nores.

Notes

1. With the intent to keep it at a high level so that it can easily be adapted to different project types and different industries.

2. In this paper, a project is related to the construction of a facility with multiple structures.

3. Including clinker production, cement mill and packaging.

4. This is done to avoid computational problems in the case that the input from the new structure being estimated is outside the range of the existing data by adjusting the range to accommodate the new value (as either a maximum or a minimum, whatever the case might be) and ensure that the scale between 0 and 1 is done properly.

5. Nevertheless, the proposed methodology allows for the adjustment of the similarity threshold by the estimator in case it is necessary.

6. Tall-frame structures were grouped according to their number of stages (levels) and their construction type, i.e. pure reinforced concrete (RC) and hybrid (HB) structures (a combination of reinforced concrete and structural steel, e.g. columns and slabs in reinforced concrete, floor beams and main beams in standard structural steel sections), yielding a total of 12 tall-frame structure subtypes.

7. This process was done in collaboration with the owners of the data and, when applicable, adjustments were made based on their knowledge about the data and expertise about the different projects where the data were originally collected.

8. For example, the design wind speed used was in accordance with the Eurocode 1, EN 1991 1-4 (2010), the spectral response acceleration used was in accordance with the 2009 IBC (based on the 2002 USGS National Seismic Hazard Maps), and the soil factor was used in accordance with the Eurocode 8, EN 1998 1-6 (2006).

9. Instead of ranking structures separately for each CMQ the individual percentages (P_i,a) can be combined to facilitate the identification of CMQ-relevant structures. This way the contribution of each structure subtype for each CMQ can be weighted using Equation (2).

10. The selected percentage (in this case 80%) is used as a threshold when determining the CMQ-relevant structures. Therefore, the higher the threshold value the more structures will be included and the lower the uncertainty built into the preliminary estimates for the entire project. However, this also means more work and effort required by the estimator since more data have to be collected and more models need to be developed.

11. Storage structure (STGS), types A–C, backward elimination technique (BET), concrete (CO), model number (4(4) ).

12. Storage structure (STGS), types A–C, backward elimination technique (BET), reinforcement (RE), model number (5(5) ).

13. AACEI (Christensen & Dysert 2011) indicated an accuracy outer range of -30% to +50% for Estimate Class 4; with average values of –23 to +35%.