AECbytes "Building the Future"
Article (January 24, 2013)
Lean Construction: Discrete-Event Simulation for MEP Renovation
Jonghoon “Walter” Kim, Michael Marchione, Stephen Tolbert, and India Rose Hill
In the construction industry, renovation projects present a unique set of challenges that are different from ground-up building. Modifying, replacing, and removing elements within an existing structure, along with the typical challenges of scheduling and work sequencing, call for an innovative approach to spatial and temporal problem-solving. Implementing BIM (Building Information Modeling) technology on AEC projects has led to increased building efficiency, constructability, and quality. BIM can save time and money by anticipating problems before they occur, through generating, analyzing, and storing geometric and non-geometric data. Typically, a BIM model is a visualization tool for 3D clash detection and 4D simulation of the construction process.
This article describes a different case, in which the BIM model was used in a unique approach to project phasing for a large-scale MEP renovation of a medical research facility, Old Jordan Hall, at the University of Virginia. In this project, the model was employed as more than a strictly visual tool or geometry-based clash detection tool. By adding the component of time to the 3D model, at specific events during each phase and sub-phase of the basement mechanical room renovation, the model served as a true 4D coordination model, allowing the construction team to make informed anticipatory decisions. The 4D coordination model produced automated data for the team to perform discrete event simulation (DES), which evaluates specific, sequenced instances in a larger ordering system. The DES data from the 4D coordination model gave the team advance warning of potential problems and thus the ability to troubleshoot ahead of time.
Old Jordan Hall, part of the University of Virginia Health System, provides facilities to both the UVA School of Medicine and independently funded medical research laboratories. DPR Construction was hired for a comprehensive mechanical renovation to increase the capacity of the physical plant (Figure 1), incorporate new technology, and replace the majority of the original equipment. (This article only addresses the aspects of the project tied to the 4D model, although the scope of the renovation was much larger.) Built in 1972, the seven-story building housed eight original 20,000 CFM air handlers in the basement mechanical room, providing single pass conditioned air throughout the building by means of fifteen vertical supply air ducts. The internally lined supply ductwork and air handling units were moderately to severely deteriorated. The team demolished the air handlers two at a time, with two new temporary air handlers installed on the roof to back feed the laboratories while each pair was being taken offline. The old air handlers were demolished and removed piecemeal, and the new air handlers were built in place due to the physical restrictions of the building. The supply air ductwork was also demolished and replaced.
Figure 1. The mechanical system at Old Jordan Hall, University of Virginia.
The engineer of record generated the 3D model by combining the original 2D drawings from each trade with current field measurements of the existing field. This “Existing System” model (Figure 2) was given to the construction team at the start of the project. The team also created a “New System” model (Figure 3), to represent the new design to be installed. Both of these models were industry-standard, static (without a time component) models, using largely geometry-based data and populated with correctly sized and placed equipment using standard architectural naming conventions. The final basement mechanical room would not look exactly like either the “Existing System” or the “New System” models, but be a hybrid, because some existing equipment would remain throughout the renovation while other pieces were removed and/or replaced.
The models were generated at Level of Detail (LOD) 300, and used to generate construction documents. Both models can be thought of as “snapshots” in time; they represent two different states of the facility. By adding time, the model can provide a roadmap for how actual work should be performed in the field. In other words, if the Existing System and New System models are like “before and after” pictures, the 4D model is like a video tutorial explaining how to get from one to the other. With the 4D model, the team is able to easily simulate the construction process and identify potential work problems before they occur.
Figure 2. The model of the existing mechanical system.
Figure 3. The model of the new mechanical system.
When both static models are combined as singular instances, their data is useless, a spatial overlap with no temporal ordering (Figure 4). When time is introduced, the system becomes dynamic and actual clashes will be reported as they occur in the 4D coordination.
Figure 4. The new and existing systems superimposed.
The two static models were combined in Navisworks, and time was added for certain, specific events. In a 4D model, Navisworks checked for geometry-based clashes with the discrete introduction of time at specific events. The addition of time enables the calling-out of event occurrences, marking changes of states, and thus making the model dynamic. Here the two static models have ceased being snapshots, and, carrying the metaphor further, the combined systems, with a temporal function applied, can be thought of as frames on a film reel. The reel is the larger ordering system mentioned in the introduction (the span of the construction going on in this space), and the frames are the specific, discrete events in space-time that make up the larger ordering system. This may be a useful analogy later in the article as we unpack how Navisworks, combined with a phasing schedule generated in Microsoft Project, yields DES data from start to finish, with each removal of duct being a frame, or demolition of an air handler being a series of frames. Of course, the frames of concern are those parts in the film where a clash occurs. 4D coordination in Navisworks with a schedule linked to it surpasses the current practices of 3D clash detection and 4D simulation. The program will automatically flag each frame of concern before it can occur, preventing the film from turning into a construction horror movie.
Understanding how the space must change over time is particularly important for renovation projects. Three distinct activities are occurring simultaneously: objects are being removed from the space, objects are being brought into the space, and objects are remaining in the space. All of these activities have different workaround requirements, at different times. Furthermore, the renovation work was done during normal business hours in a fully occupied building. Some of the facilities’ uses were critical or sensitive, and could not be disturbed by any environmental consequences of construction; coordinating well in advance was critical. Notification was required no less than 5 days in advance of any utility shutdown, or activity that generated noise, dust, or vibration. It was imperative to complete activities within the approved time slot.
In addition to the schedule certainty constraints, the project faced several other challenges unique to renovation. For example, the size of the existing building meant that nothing larger than 10 feet could fit through the doorways, corridors, or elevators. This meant that anything larger had to be demolished in place and then removed, or delivered in pieces and then assembled in place. Also, with new and existing equipment and utilities in the same space, clashes were inevitable in the mechanical room.
The project team used the detailed 3D models for the existing and new systems to generate 2D construction documents. They checked space clashes before creating the 2D documents, and used separate sheets to represent a different phase of the renovation. But without the ability to see three-dimensional objects moving through time (i.e., a 4D model), it was unclear if the 2D drawings were clash-free or not.
Traditionally, 4D models are used to convey spatial and temporal aspects of construction activities together. They visually translate construction activities before/during construction and help project participants communicate what is expected to be accomplished, by who, in that window of time. (Ref: Bonsang Koo, “Feasibility Study of 4D CAD in Commercial Construction” in Journal of Construction Engineering and Management, 2000.) 4D models ensure smooth installation and avoid spatial interference between different systems. However, since the traditional review process is based on visual analysis, it is difficult to detect all the interferences and more likely to miss clashes that are not clearly shown in the animated model.
4D Clash Detection
In contrast to that visual analysis, this project afforded a practical opportunity to combine 3D clash detection and 4D sequencing into one process as “discrete-event simulation” 4D coordination. DES is a lean practice used in many industries, such as manufacturing, healthcare, and capital investing. In the AEC industry, DES has primarily been used for construction-related production planning and space-time conflict analysis, (Ref: Burcu Akinci, “Formalization and Automation of Time-Space Conflict Analysis” in Journal of Computing in Civil Engineering, 2002.) Current research speaks to the importance of BIM past the design and preconstruction phases to manage production and decrease project conflicts, though there is no body of work that specifically addresses DES (as anticipatory 4D coordination) in a renovation project.
DES is a process to break down a complex schedule into an ordered sequence of well-defined events (see http://whatis.techtarget.com/definition/discrete-event-simulation-DES). In DES, a system is ordered by chronological sequences of events that occur at specific (discrete) moments in time. The (discrete) event changes the state of the system, and that is why it is studied. The main purpose of discrete event simulation is to quantify the variables of interest and isolate them within the larger operations of the system. In this multi-phase infrastructure upgrade, 4D coordination simulated the demolition and installation of AHUs and ductwork (a sequence of events), and detected clashes for each event. A comparison of 4D Visualization and 4D Clash Detection is shown in Table 1.
Table 1. 4D Visualization and 4D Clash Detection Comparison
4D Model in Navisworks
To perform 4D clash detection, we modeled a series of events: demolition of AHUs along with associated ductwork and installation of corresponding new AHUs and ductwork, in a 4D model (Figure 5) using Autodesk Navisworks 2010. We linked the 3D model and schedule using the same unique tag in both the 3D model and the schedule. We used layer names in the 3D model as unique tags; each layer name had the type of system—“existing” or “new”—and sequence order in it. Figure 6 shows the layer-naming convention: for example, layer name “ME-PHASE A-DUCTS” is for the demolition of first AHUs and ductwork, where "ME" stands for “mechanical existing,” and "PHASE A" means the first phase of the whole process. Similarly, layer name “MN-PHASE A-DUCT” means “mechanical new—first phase—ductwork.”
Figure 5. The 4D Sequence model.
Figure 6. The Layer naming convention followed in the project.
To order demolition and installation, we used the sequence specified in the architectural and engineering documents. Figure 7 is an excerpt from the documents. It shows phasing notes on the right and highlights corresponding AHUs and ductwork in the drawing. We developed the sequence schedule using Microsoft Office. Level of activity detail in the schedule was the same as the 3D model. For the relationship between activities, we used only the "Finish-to-Start" relationship because each task always started after the activity in progress was completed. This fits well with the main principle of DES: one event at a time. The true duration of the activity was not as important than the sequence, because the purpose of the simulation was to detect clashes between events, not how long each activity would take. So, we simply used one day as the duration for every task instead of the true duration.
Figure 7. Demolition and installation sequence in the AE Document.
4D Clash Detection in Navisworks
We created batches to run clash detection. Figure 8 shows the clash batch setup for the first two new AHUs and ductwork against existing conditions. The right side of Figure 9 shows that all existing AHUs and ductwork were selected to run clash detection against the first air handler and ductwork. We did not know which existing system had already been removed, and what still remained. Linking to Timeliner allowed Navisworks to run clash detection for 10 different points in time, and eight of those instances specifically checked for clashes when the new AHU and associated ductwork had a relationship with existing equipment that either had to be demolished or remain. From the 4D clash detection, we discovered five major clashes (Figure 8). We present two instances of the clashes below. To resolve the clashes, we rearranged demolition and installation, and performed 4D clash detection repetitively until all the clashes were eliminated.
Figure 8. The clash batch setup for the new AHUs and ductwork against existing conditions.
Figure 9. File selection of all existing AHUs and ductwork to run clash detection against the first air handler and ductwork.
Clash Detection Results - #1
Clash detection results showed a clash between a new duct and existing duct on Aug. 10, 2010 (Figure 10). When the new duct installation started, this existing duct still remained, causing a clash. The “Sept 10, 2010” enlargement in Figure 10 shows what the new duct would be like once it is completed; no clash exists. In other words, the existing duct should have been removed before the new duct installation. This issue was resolved by simply ordering demolition of the existing duct ahead of installation of the new duct.
Figure 10. A clash between a new duct and an existing duct.
Clash Detection Results - #2
Figure 11 shows another clash between an existing AHU and a new AHU. The A/E’s sequence dictated replacing this AHU as the last phase, but when we ran 4D clash detection we found that its existing ductwork created three major clashes with other new AHUs and their associated ductwork that had to be replaced earlier in the phasing. We reordered from this information and replaced AHU 10 first (new associated ductwork was smaller than exiting ductwork), eliminating the three clashes. Had we not run a full 4D coordination, this situation would have caused a major schedule delay.
Figure 12. A clash between a new AHU and an existing AHU.
The successful 4D clash detection in this project made smoother sequencing possible during the mechanical upgrade, and consequently, the project saved time and money from the uninterrupted process. Simply using 2D plans, or even separate 3D models, would not have been enough to understand how each activity had to be sequenced in order to avoid clashing with others. But, applying the DES technique to the 4D model meant that we were able to analyze the intersection of specific activities and identify potential time and space conflicts.
About the Authors
Jonghoon “Walter” Kim is a senior BIM project manager for DPR Construction working from its Washington DC office. Since joining DPR, Jonghoon has been involved in BIM implementation and innovative construction technology adoption on various projects including health care, advance technology, higher education, and office buildings. He received a Ph.D. in Civil Engineering from Stanford University. At Stanford, he experienced a wide variety of R&D projects on Virtual Design and Construction, working with industry member companies and academic institutes to advance the research in Building Information Modeling. His works were published in academic publications. He also acted as an assistant consulting professor at Stanford to teach a class and advise research projects. Jonghoon is an active participant in the Digital Building Lab at Georgia Institute of Technology and teaches classes for the AGC Certificate of Management-Building Information Modeling (CM-BIM) training. He can be reached at firstname.lastname@example.org.
Michael Marchione is an MEP Coordinator/Superintendent at DPR Construction. He has 25 years experience with microelectronic, semiconductor, and pharmaceutical manufacturing, including 14 years managing the construction of High Purity Process Piping, MEP and Cleanroom Architectural Systems.
Stephen Tolbert is a Project Manager at DPR Construction. He has 30 years of construction experience serving highly complex mechanical and electrical projects in multiple capacities. Stephen led the UVA Old Jordan Hall Mechanical Renovation project, for which he was responsible for day-to-day management of the project team, and worked with all team members during procurement and construction to provide planning, schedule development and budget control.
India Rose Hill is a BIM Coordinator at The Boldt Company. She worked as an intern at DPR Construction on this project.
Note: AECbytes content should not be reproduced on any other website, blog, print publication, or newsletter without permission.
Have comments or feedback on this article?
Visit its AECbytes
blog posting to share them with other
readers or see what others have to say.
the Future > Discrete-Event Simulation for MEP Renovation > Printer-friendly