This website uses cookies to store information on your computer. Some of these cookies are used for visitor analysis, others are essential to making our site function properly and improve the user experience. By using this site, you consent to the placement of these cookies. Click Accept to consent and dismiss this message or Deny to leave this website. Read our Privacy Statement for more.
FPE eXTRA Issue 51, March 2020

A Digitized Surveying Method Using Machine Vision to Collect Fuel Load Data in Buildings

By: Negar Elhami-Khorasani and Thomas Gernay

The fuel load density quantifies, per unit floor area of a compartment, the amount of energy available to fuel the fire. It is a crucial input for the definition of fire scenarios used in performance-based fire design. Yet in spite of the importance of this parameter for fire safety, fuel load density data are scarce because current surveying methods remain relatively impractical and/or inaccurate.

Statistics of data collected from fuel load density surveys are the basis of reported design values in codes and standards. For example, NFPA 557 [1] provides an Annex on previous fuel load surveyed data from 1957 to 2008, with only a handful surveys completed after year 2000. Surveying of fuel load density (defined in MJ/m²) entails quantifying the amount of combustible materials (in kg) and the nature of said materials (quantified by their calorific value in MJ/kg) in a compartment. The combustible materials in a compartment include movable and fixed contents such as furniture and interior finishes, such as floor finishing, and wall and ceiling linings.

The current preferred surveying approach, as recommended by NFPA 557, is a combination of inventory and weighing method, where direct weighing is applied to smaller items and the inventory method is used for heavier items. The inventory method obtains the mass of common furniture items based on established relationships between the visual characteristics of items and their mass from manufactorers’ catalogs. In a recent research project [2], a new method using machine vision was developed to increase the efficiency and accuracy of fuel load data collection.

Digitized methodology

Recent technological advancements provide an opportunity for more sophisticated yet easy-to-implement survey methodologies for collecting fuel load data in buildings. Specifically, fuel load surveys can be conducted using picture collection of building contents, virtual measurements, and image recognition techniques, as schematized in Figure 1. The four-step survey procedure consists of:

  1. Digital inventory: the surveyor collects on-site data through a digital application. Mostly, the surveyor takes snapshots and digital measurements of the room content using the camera and virtual measuring application in the smartphone or tablet running the application. The application also allows the input of general characteristics about the building and surveyed room, as well as uploading building drawings.
  2. Data organization: the developed application stores the snapshots and introduced data into a structured online database (SQL database in formatted tables).
  3. Item matching: off-site, the surveyor performs an image search on the snapshots to find matching furniture and corresponding specifications online. This search is based on image recognition and data mining techniques that are available in existing online inventory databases, to obtain mass and (if available) material composition of the item. At times, the surveyor needs to estimate the material composition of the item based on judgement or available information online.
  4. Fuel load calculation: the evaluation of fuel load is performed by the application by converting furniture specification (mass and material composition) into energy using the calorific values that are coded for the different materials.

Application

Figure 2 shows an application example of the survey methodology. A snapshot and virtual measurements of an office chair are taken using a smartphone (Figure 2a). The weight of this chair is measured using a regular scale as 10.4 kg. An image search is completed in an online database and returns a matching item (Figure 2b), with its dimensions and a listed weight of 10.4 kg. Comparison of the dimensions confirms the match. In general, the accuracy of the procedure depends on the availability of the item online (e.g. it may be harder to find a close match for very old furniture) and on the quality of the image (e.g. a clutter of various items on a bookshelf may pollute the image search on the bookshelf). For the chair, with assumptions on the material composition as 50% in mass non-combustible (metal frame, etc.), 30% in mass average plastic, and 20% in mass polyurethane, the fuel load calculated by the application is 165 MJ.

The survey method has been applied to three buildings in Buffalo, NY. A total office area of 1720 m2 was surveyed consisting of 34 closed offices and 161 cubicles within 12 large open plan office spaces. The measured fuel load density for movable content had a mean of 1115 MJ/m² with a standard deviation of 614 MJ/m². When including the fixed content, the measured total fuel load density had a mean of 1486 MJ/m² with a standard deviation of 726 MJ/m².

Moving forward

Advantages of the proposed digitized surveying method over previously applied inventory methods include: (1) application of image recognition to find matching items online, which speeds up the process to quantify item properties; (2) reduction of disruption for the room occupants (because the on-site procedure is faster and the digital inventory can be post-processed after the field visit); (3) interactive digitized form with possibility of standardization and data storage for an ever growing database of fuel load statistics.

Moving forward, the developed application can be used to survey buildings and populate the database over time with many furniture items. If the database grows large enough, the extra step on matching items, to find mass and input on material compostion can be skipped as AI technology can be used within the developed application to match the image of an item directly to the value of fuel load density in the database. This way, the post-processing could be fully automated.

Acknowledgements

The authors gratefully acknowledge the Fire Protection Research Foundation (FPRF) and the National Fire Protection Association (NFPA) for their generous support. The FPRF generously provided funding and guidance for this research project. In particular, the authors thank Amanda Kimball and the members of the Project Technical Panel for their support.

 

Negar Elhami-Khorasani

Thomas Gernay

Negar Elhami-Khorasani is with University at Buffalo, USA, and Thomas Gernay is with Johns Hopkins University, USA.

References

  1. NFPA. (2002). NFPA 557 Standard for determination of fire loads for use in structural fire protection, National Fire Protection Association, Quincy, MA.
  2. Elhami Khorasani et al. (2019). Digitized Fuel Load Survey Methodology Using Machine Vision, Fire Protection Research Foundation, Quincy, MA.

Corporate 100 Visionaries

About Us

© 2019 SFPE | All Rights Reserved
Privacy Policy