This is the first of three articles demonstrating the growing acceptance of High Performance Computing especially in new user communities and application areas. Major reasons for this trend are the ongoing improvements in user-friendly simulation software, the continuous performance improvement of HPC servers, and the easy and affordable access to HPC resources in the cloud. In this article series we present three use cases: we demonstrate how a US law firm applied multi-physics building simulation using HPC to prove building defects at a client’s property, how a class of ninth-graders in Norway designed and built a flying boat with a novel CAE software, and how India’s National Institute of Health was able for Schizophrenia to replace the current highly risky procedure of brain-invasive operation with a revolutionary non-invasive low-risk treatment based on HPC.
Use Case 1: Simulating Moisture Transfer in a Residential Condo Tower Helped PBBL Law Offices to Prove a Client’s Case
The Problem
One year ago, PBBL Law Offices in Las Vegas/Orlando approached UberCloud and Fraunhofer Institute for Building Physics asking for HPC support in a lawsuit dealing with a twin tower residential condominium. An extensive expert investigation established exterior plaster failure, water intrusion at improper window and roof installations, and high interior humidity levels with apparent biological growth (ABG) observed on interior walls, baseboards and between layers of interior gypsum board at unit partitions. Mechanical testing determined that condominium unit interiors were often under negative pressure, drawing in high-humidity, un-conditioned exterior air. A mechanical engineering evaluation found that the air conditioning units (serving each condominium) were improperly sized to adequately manage humidity.
Competing experts suggested that defective exterior plaster was a cause of the high humidity conditions. Exterior plaster failure included blistering of the coating system and saponification of the coating and substrate. Moreover, these experts contended that the existing mechanical system for each condominium unit was adequate to handle the dehumidification such that the ABG was caused by the moisture intrusion through the exterior wall system. Their opinion was that the negative air pressure was acceptable for 15/20 story towers.
Because of these competing opinions and the inability to field test the experts’ hypotheses, PBBL Law, for the first time in their history, chose HPC modelling to determine whether damage was caused by moisture transfer through the plaster coated exterior walls or that it was the result of negative pressure in the living units. Also, by advancing the modelling, HPC was used to determine the effect of the negative pressure and high humidity in the condominium living environment if left unmitigated.
The Project Team
The project team consisted of the end-users David Pursiano and Robert Simon from PBBL Law Offices, the software and expertise providers Florian Antretter and Matthias Pazold from Fraunhofer Institute for Building Physics, and the HPC Cloud expert Baris Inaloz from UberCloud.
The Software WUFI
To accomplish this the team used the WUFI Plus simulation environment, a hygrothermal building simulation software from the Fraunhofer Institute for Building Physics (IBP) in Germany, part of the hygrothermal simulation suite WUFI which is based on the calculation of the coupled heat and moisture transport across building components, like walls, roofs and floors. They simulate the temporal development of the heat and moisture profiles within a component and the heat and moisture exchange on the component surface. In addition, solar radiation through windows, inner heat and moisture sources or sinks, HVAC systems and ventilation are considered.
Parametric Study
With four different compute instances in the Microsoft Azure cloud, a benchmark was done to check the computation times and identify the ‘sweet spot’ system configuration. Tested were the 2-core DS11, 4-core DS12, 8-core DS13 and 16-core DS14 compute instances for best price/performance. The simulation period for the building model was set to one year. Table 1 shows the elapsed computation time on the different machines.
Table 1: Simulation time test results.
Simulations | DS11 (2core) | DS12 (4core) | DS13 (8core) | DS14 (16core) | ||
1 | 0:25:10 | 0:16:35 | 0:11:14 | 0:06:03 | ||
2 | simultaneous | 0:24:45 | 0:17:10 | 0:08:42 | elapsed time | |
0:12:22 | 0:08:35 | 0:04:21 | per simulation | |||
4 | simultaneous | (>>25min) | 0:34:17 | 0:16:00 | elapsed time | |
. | 0:08:34 | 0:04:00 | per simulation | |||
6 | simultaneous | 0:56:00 | 0:21:07 | elapsed time | ||
0:09:20 | 0:03:31 | per simulation | ||||
8 | simultaneous | 0:25:25 | elapsed time | |||
0:03:11 | per simulation | |||||
Time per simulation | 00:25:10 | 00:12:22 | 0:08:34 | 00:03:11 |
Interpretation of Computation Times
The DS11 took 25 minutes 10 seconds to simulate one year of the building, DS12 needed 16 minutes 35 seconds. During the simulation of one building model the CPU usage was lower than 100%. Due to this, the simulation was started twice at the same time (two simultaneous simulations), to get 100% CPU usage. Both simulations were finished in 24 minutes and 45 seconds. Regarding this and running two simulations at the same time on the DS12 machine it can be concluded that the simulation of one year lasts 12 minutes and 22 seconds on the 4-core machine. With the DS14 the test was done running the simulation up to 8 times in parallel. Those 8 simulations running parallel were finished in 25 minutes and 25 seconds. Concluding this, a simulation of one year takes 3 minutes and 11 seconds on the 16-core machine.
For the final parameter study, the DS14 with 16 cores was chosen, and set to run 8 simulations at the same time. Because parameter studies consist of independent simulations we’d be able to run 16 simulations in parallel on two DS14, or 32 simulations on four DS14, thus reducing the total simulation time from more than one month on the user’s workstation to less than two days in the Microsoft Azure Cloud!
Results
The parametric study resulted in 583 simulations investigating the building behavior depending on different conditions. With different count of simulated years per simulation (3, 5 or 8) in total 2790 years were simulated. Nearly 36 GB zipped data are stored, containing “the main” simulation results, like the air and material temperatures and humidity, and the material water content, for each hour in each simulated year.
Conclusions
For this specific project it was concluded, that the coating damages are not related to high indoor air relative humidity but to the vapor permeability of the coating and the driving rain leakage behind coating. Indoor climate humidity and mold issues are not related to coating permeability but to HVAC (heating, ventilation and air conditioning) equipment and infiltration of unconditioned and untreated outside air.
“The impact of multiple uncertain parameters on coating damages, elevated indoor humidity and mold issues can be assessed only with this kind of large-scale parametric simulations” said Florian Antretter from IBP. “The simulation models can be used to predict the potential future performance and related risk for renovation measures.”
“This study also demonstrates that, in general, multi-physics building simulation using HPC Cloud computing is accessible, affordable, and beneficial for our private clients”, added David Pursiano from PBBL Law. “It can be applied for initial design and repair to reduce downstream risk.”
HPC cloud computing enables the prediction of future performance of buildings by simulation for a broad range of input parameters in a reasonable time period due to the performance benefits with HPC cloud computing. Invasive / destructive building forensics can be reduced. Together with the ability to separate design, material, and workmanship deficiencies, the design process, potential forensic investigations, or litigation can draw huge benefits from utilizing HPC cloud computing with hygrothermal building simulation. Readers can download the detailed UberCloud Case Study #207.