Activity and Flow based construction site management

Project Team

M. Fischer, N. Garcia-Lopez

Overview

Field managers lack a method for analyzing the impact of workflow variability on activity execution and understanding how variability is propagated between activities. As a result, they cannot anticipate the impact of workflow variability and rely on their experience and intuition managing workflow variability during look-ahead planning. We propose to formalize the Activity-Flow Method (AFM), which uses a model representing construction activities and the flows connecting the activities, to analyze the in-project workflow variability and predict its impact on downstream activities. Using the AFM, field managers can identify targeted strategies to shield activities from workflow variability. The model for representing construction workflow, including the activities, flows between the activities, and the mechanisms causing workflow variability, will be developed from literature and validated with available variability tracking data from three projects. The AFM will be developed based on the application of the Last Planner System and input from field managers. The AFM will be validated by testing its implementation on three construction projects.

 Project Background

Research Motivation

Despite recent advances in the use of Building Information Modeling (BIM), lean approaches, and information technology systems, construction projects still continue to face problems managing on-site construction resulting in schedule and cost overruns (Jones and Bernstein 2014). In manufacturing, companies have achieved exceptional performance by systematically seeking to understand and manage variability (Womack and Jones 1996). Recently, construction researchers have revealed the negative impact that workflow variability has on construction performance such as: higher work in process, longer activity durations, and project completion delays (Arashpour and Arashpour 2015; Tommelein et al. 1999). However, there have been limited efforts developing methods that aid field managers to anticipate and manage workflow variability in the field.

This need motivated us to submit a Seed Proposal in 2014. We proposed to develop a simulation model to help field managers understand the impact of implementing different production management strategies on the production system metrics (time and cost). However, the findings from the initial case studies and a deeper exploration of the existing literature drove us to shift our focus. We found that field managers lacked a method that helped them understand the causes of workflow variability, and its impact on downstream activities. Hence, they were unable to anticipate the impact of workflow variability and implement targeted management actions to shield the activities. This motivated us to refocus the research and develop a method that uses the activity tracking data resulting from the Last Planner commitment tracking phase, namely the planned versus actual data for activity completion and the reasons for variance, together with the look-ahead schedule to predict the impact of workflow variability on downstream activities. Using these predictions, field managers can implement targeted measures to reduce the impact of variability on activity execution.

Observed Problem

Field managers lack methods to anticipate the impact of variability on downstream activities. As a result, they cannot formally manage workflow variability and rely on their intuition and past experience implementing measures to shield downstream activities from upstream variability. 

Research Objectives

1. To understand what variability factors affect what construction flows.

2. To understand how variability in the construction flows causes variability in the activities.

3. To understand how field managers can and should measure the variability factors, construction flows, and activity execution.

4. To develop a computational model that allows field managers to predict how variability is propagated to downstream activities. 

5. To understand how field managers can use the computational model to manage variability and its impact during look-ahead planning.

Proposed Construction Workflow Model

Construction workflow is defined as the movement of information, materials, and resources, through workspaces performing a sequence of activities on components (LCI 2015). Construction workflow is constituted by seven activity flows identified by (Koskela 1999): labor, equipment, materials and components, information, workspaces, predecessors, and external. Variability is defined as “a departure from uniformity” (Hopp and Spearman 2011). Using these two definitions, we define workflow variability as a departure from the baseline (or the plan) in the quality or quantity of the activity flows (Koskela 1999) necessary to perform the sequence of construction activities. Workflow variability causes variability in the execution of the activities, leading to variability in the activity start, duration, and finish. 

There are two main models that represent the construction workflow: the transformation view and the flow view. In the transformation view, the activities require a series of inputs that are transformed to create outputs. The activities are connected to each other via precedence constraints (Aalami 1998; Chapman 1997; Darwiche et al. 1988; Echeverry et al. 1991). The managerial focus is on optimizing the transformation. The transformation model does not explicitly represent the elements that make up the workflow (the activity flows). As a result, field managers cannot use it to anticipate the impact of workflow variability on downstream activities or understand how variability is transmitted through the activities.

On the other hand, the flow view represents production as a series of flows composed of transformation, inspection, moving and waiting times (Koskela 2000). This view focuses on both the optimization of the flows and of the transformation (Ohno 1988). The flow view has been partially formalized into some simulation models (Akbas 2003; Choo and Tommelein 1999; González et al. 2009; Tommelein 1998), but not at the production level of detail.

Figure 1: In the transformation view, dependencies between activities are represented by precedence constraints. In the flow view, dependencies between activities are represented by the flows that are shared between activities.

Workflow variability occurs because of two mechanisms: the occurrence of variability factors, and the untimely release of flows from upstream activities into downstream activities.

In the face of variability, field managers can implement buffers to shield activities from variability. There are three types of buffers: capacity, inventory, and time (González et al. 2004; Hopp and Spearman 2011). Since buffers are expensive, they should be adequately sized to match the level of variability. However, since current models do not explicitly represent and measure the flows, field managers are unable to collect data that could help them size the buffers adequately.

Figure 2 shows our proposed conceptual model of construction workflow integrating the concepts discussed above: the flows between activities, the activities, and the mechanisms causing workflow variability (variability factors and variability in the release of flows due to activity variability). 

Figure 2: Conceptual model of construction workflow. Note: PS: planned start, AS: actual start, PF: planned finish, AF: actual finish.

To formalize the proposed conceptual model of workflow variability into a computational representation, we need to overcome some theoretical and practical limitations. Firstly, we need to understand how variability in the activity flows leads to variability in the activity execution, and what variability factors affect what specific activity flows. Secondly, we need to develop measures and methods for tracking for the elements that make up the workflow model, i.e., the variability factors, the activity flows, and the activity execution. 

 

Activity-Flow Method

The Activity-Flow Method (AFM) will be a decision-support system that helps field managers to manage variability during look-ahead planning. The inputs to the method will be the look-ahead schedule, and the variability data collected up to that point in the project. The AFM will leverage the construction workflow model developed in the previous research task and the methods for measuring the components of the workflow model, to identify the downstream activity flows that are vulnerable to variability and the activities that depend on those flows. 


Using this information, the field managers can make decisions to shield the activities from variability by implementing buffers targeted at specific flows, or make changes to the look-ahead schedule based on the variability predictions. To support field planning, the AFM will be integrated into the look-ahead planning process by considering the steps outlined by the Last Planner System (Hamzeh et al. 2012) and the input from field managers. 

 

Research Tasks

 1.  Develop computational representation of workflow model

Status: Completed
Developed a web application to collect the data necessary for representing the proposed construction workflow model. The web application extends the Last Planner System by including information about the construction flows for each activity including: the quantity, status, and where they come from (previous activity or external). 


2. Validate the representation of the construction workflow model

Status: Completed
Collaborated with Graña y Montero collecting data at one of their projects during a 18 week period (8 weeks on site, 10 weeks tracking virtually). 
Findings: The data available to represent the model was readily available. The model represents a more complete view of the construction going on at the jobsite. A limitation is that it takes an extra amount of time to enter the data versus the traditional Last Planner System. However, the method allows reseachers to collect a large amount of data about the production system. On average, we collected the following data: 64 activities/week, 232 flows/week, 23,056 data points/week.

3. Develop the Activity Flow Method (AFM)

Status: 95% complete

We developed the AFM by leveraging the data collected using the construction workflow model and the activity variability data collected from the Last Planner System. We used data analysis techniques and machine learning to generate insights for field managers. We found that the AFM data structure supported the generation of better predictions compared to other construction models (Resource-loaded CPM, Line of balance, and the Construction Method Model).
we are currently developing methods for increasing the visibility of the activity flows by leveraging 4D modelling and developing dashboards to help field managers understand why variability is occurring.

4. Validate the AVM

Status: 90%
The AFM results will be validated in two phases:
Phase 1: Retrospective validation. Using the data collected from the project we will test the strength of the predictions.
Phase 2: Prospective validation at construction site. The AFM results will be introduced during the look-ahead meeting. We will measure the extent to which the field planners use the predictions to make changes to the look-ahead plan and shield the activities from variability. At the end of the validation period we will carry out interviews with field managers to undertand the advantages and limitations of the AFM.
 We have carried out the validation in three jobsites: 18-week validation in an office building in Peru (GyM), 4-week validation at a residential building in Colombia (Prodesa-SRC), and 4-week validation at a residential development in Denmark (MT Højgaard). 

Publications To Date:

  • Garcia-Lopez, N. and Fischer, M. (2016) A Construction Workflow Model for Analyzing the Impact of In-Project Variability. Construction Research Congress 2016: pp. 1998-2007. doi: 10.1061/9780784479827.199

Other Research Findings

Case Study Analysis:

We carried out two case studies to understand how field managers at construction sites manage workflow variability during look-ahead and commitment planning. We attended all scheduling and planning related meetings at the projects during a two month period. 

Findings:

  1. Field managers lack formal methods for managing variability and estimating its impact.
  2. Field managers rely on their intuition and past experience managing variability.

Analysis of activity variability data

We analyzed the activity tracking record of a building project that had implemented the Last Planner System. The data set consisted of 30,000 activity entries. We carried out a manual cleanup of the data, expanding the data set to include: activity classification (activity type and UNIFORMAT), and subcontractor type.

Findings:

  • Activity total variability is skewed to the right: median = 0, mean = 2.26 days. This means that it is more likely that an activity will finish late than early. 
     
  • Management activities have higher variability than other activities. MEP activities have lower variability than other activities. 
     
  • The relative importance of different reasons for non-completion varies by subcontractor group
     
  • We were unable to identify any clear clusters of activities according to their variability level by classifying them by Subcontractor type, UNIFORMAT classification, or activity type.
  • There is a strong relationship between the activity having a positive start variability (starting late) and the predecessor activity finishing late. 
     

Research Proposals

CIFE Seed Proposal 2016:

CIFE Seed Proposal 2015:

CIFE Seed Proposal 2014:

Last modified Tue, 15 Aug, 2017 at 9:06