1 Introduction

Fuel load density is one of the primary parameters in determining the temperature demands when evaluating safety or designing a structure for fire. The prescribed values of fuel load in codes and standards are based on data collected from fuel load density surveys. Major surveys conducted in the US go back to 1970’s, where Culver and Kushner adapted an inventory method to quantify the weight of furniture using manufacturer’s catalogues [1, 2]. The results of a series of comprehensive building surveys in 1970’s showed that the magnitude of fuel load correlated with the room use and room area [3, 4]. In 1996, Caro and Milke conducted surveys of two office buildings but used direct weighing method, which is more accurate than the inventory method [5]. Several surveys have been conducted in Canada between 2005 and 2011, in both residential and office buildings, using different methodologies, ranging from less accurate web-based questionnaire or real-estate website methods to more accurate direct weighing and inventory methods [6,7,8,9]. A series of surveys were completed in Europe in early 2000s in industrial, commercial, and public buildings across different countries [10, 11]. The surveying method varied from direct measurement of volume for furniture with simple geometry to inventory method for more complex ones. Surveys in other countries, such as India [12] or Australia [13] are also reported in the literature. The results of surveys from different countries showed a relatively high scatter in fuel load density, as well as the challenge in conducting efficient surveys while maintaining accuracy in the collection of data.

Elhami-Khorasani et al. [14] established an innovative fuel load survey methodology based on picture collection of building contents, virtual measurements, and image recognition techniques. The primary objective was to develop a methodology to increase the efficiency and accuracy in conducting fuel load density surveys using automated procedures. This paper applies the developed methodology to survey offices in three buildings. This serves to demonstrate the feasibility of the newly developed procedure, as well as to collect new fuel load data for office buildings. The surveys cover closed and open space offices with different compartment areas. The collected data is analyzed and compared with results of existing surveys. The paper also includes a sensitivity analysis of the collected data to a number of factors, such as office size, office use, and material composition.

The decision on recommended fuel load design value should be based on probabilistic calculations and a target safety level for the design [15]. Higher safety requirements lead to selection of a higher percentile of fuel load for calculation of design fire, in turn leading to increased thickness of insulation on members to achieve satisfactory performance under the calculated design fire. The target safety should be acceptable to the public, and at the same time as low as reasonably practicable (commonly referred to as ALARP) [16]. In determining the probability of failure of a structure under fire, fuel load density is an input from the demand side, to quantify the temperature–time evolution within the structure [17, 18]. This paper discusses the distribution of fuel load data based on the surveyed buildings, but the ultimate design recommendation for fuel load density in regulatory codes should be based on a comprehensive reliability-based analysis with a target safety level. Meanwhile, report of fuel load distribution as part of this paper can advance performance-based fire engineering where the target performance objective can be determined by the stakeholder.

2 Development of a Prototype Fuel Load Survey Methodology

2.1 Brief Description of the Methodology

Details of the proposed methodology adopted in this research is provided in [14]. A brief overview of the approach is discussed here. The proposed procedure includes the following four steps:

  • Digital inventory: collect general information on the building (address, age, number of floors, etc.) and room (floor, room use, dimensions, etc.) that is being surveyed, and take images of furniture and digitized measurements using a smartphone or tablet.

  • Data organization: store the collected data in an organized database. This step is automatically completed as part of the developed methodology.

  • Item matching: process the collected data by conducting an inverse image search of furniture photos to find matching items online, and document mass and material composition of the items. This step relies on existing image recognition and data mining techniques from the available online inventory databases.

  • Fuel load estimation: convert mass (kg) to energy (MJ) using calorific values corresponding to material composition of the item. This step is completed automatically within the developed methodology.

The procedure has been programmed into an electronic application. The application contains the surveying form and enables automatic completion of the method through the four steps cited above.

2.2 Demonstration of the Methodology on One Office Room

2.2.1 Step-by-Step Description

The developed methodology is first applied to one sample office room to illustrate the process. The office plan dimensions are 3.8 m by 2.9 m. The office is located on the campus of the University at Buffalo, NY. The process and results are reported in this section.

Based on the methodology summarized in Sect. 2.1, the surveyor used the digitized survey application to complete the following steps when surveying the room:

  1. 1.

    Take measurements of each item in the room. For instance, measurements for a bookshelf include height, width, and depth;

  2. 2.

    Take a few images of each item for image search after the survey;

  3. 3.

    If applicable, record percent of volume filled (e.g., percentage of volume of combustible content in an open bookshelf), and

  4. 4.

    If applicable, record whether an item is enclosed or not (e.g., drawers or bookshelves).

The surveyor collected dimensions and images of 14 different furniture items in the office. These include two different chair types, a desk, two bookshelves and their content, a phone, a monitor, a printer, two different diploma frame types, a tea kettle, two trash bins, and miscellaneous paper piles.

As stated earlier, the measurements are obtained using the virtual measuring applications on a cellphone or a tablet. The advantage of digitized measurements compared to the standard tape measure is twofold: (a) One person can complete the survey and take all measurements without much effort, compared to the possibly difficult task of single-handedly using a physical tape measurer for some items such as tall windows, items above shelves and out of reach, etc., (b) Images of measurements are recorded and stored by simple click of a button on the tablet/smart phone, while fixing a standard measuring tape in place and taking a photo might require more than one person. Different virtual measurement applications were tested to identify their accuracy and ease of use. The process showed that the surveyor (or the user) of the application should make a few trial attempts to understand how the application works and how to select the points on a surface to obtain an accurate measure of distance between the two points. The Tape Measure and Measure applications on iPad provided reasonable accuracy and were relatively user-friendly. In addition, a small sensitivity analysis was conducted by having two different and independent users complete virtual measurements of different furniture items, and comparing the virtual measurements with values from a physical tape measurement. The discrepancy between measurements is quantified in Table 1. The maximum error is about 10%, while keeping in mind that exact measurements are not necessarily needed as these measurements are used to guide the post-processing procedure to find a matching item. The error is generally higher when the surface is not perfectly flat, such as height or width of a chair.

Table 1 Quantification of error for virtual measurements
Full size table

The office under study includes two openings (one door and one window). Virtual measurements from the window are shown in Fig. 1. The door dimensions are 1.85 × 0.85 m.

Figure 1
figure 1

Comparison of actual and virtual measurements for the window in the example study

Full size image

The office floor is covered with carpet, which is included in the computation of the fixed fuel load density. The four boundary walls are made of concrete blocks, which is considered as a non-combustible material, and therefore not included in the total fuel load. The ceiling tiles are assumed to be combustible and included as part of the fixed fuel load, to be conservative. The type and material composition of ceiling tiles are not accessible; therefore, the result will clearly specify the contribution of these ceiling tiles to fuel load density for further clarification. The mounting structure for the ceiling is made of aluminum and is not considered as a combustible material.

Figure 2 shows a sample of images with measurements obtained for one frame and one chair. The measurements shown for the frame are obtained using the “Measure” app while the ones for the chair are obtained using the “AR Measure” app (the Tape Measure app, as mentioned earlier, was more user friendly, and was selected as the main application after the initial trials with different applications). The error in measurement on the height of the frame is about 6% and the error for chair measurements are on average 7%.

Figure 2
figure 2

Comparison of actual and virtual measurements for a frame and a chair

Full size image

The next step is to process the images and find the mass of the surveyed items. The Google Images search, and websites with inventories such as Wayfair.com, were used for the two chair types, the desk, bookshelf, and frames. Items that are easy to locate within online inventories, given their model, such as phone, monitor, and printer were directly searched and exact masses were obtained. Figure 3 shows the selected matching chair with the one shown in Fig. 2 when searching with image recognition techniques. The chair of Fig. 3 was automatically extracted from an online database based on the uploaded picture of Fig. 2. Items such as chairs and desks are perhaps the hardest to find the exact match. It should be noted that the traditionally applied inventory method provided large standard deviations in estimated mass of such items as part of this pre-listed inventory options from manufacturer catalogues. In this case, the matching chair from the image recognition search appears visually similar and has matching dimensions with the surveyed chair. The reported mass of the chair from the online inventory is 10.4 kg (23 lbs). Following this exercise, the surveyor weighed the chair shown in Fig. 2 with a digital scale as a validation of the process. The measured mass was 10.4 kg (23 lbs), exactly matching what was found during the image search. Such validation is not possible for the bookshelves and the desk, and it is not expected that the implemented procedure provides such level of accuracy in general, but this example suggests that the error in obtaining mass can be reduced using image recognition when compared to traditional inventory method.

Figure 3
figure 3

Selected matching chair using image recognition

Full size image

Mass of books, binders, and piles of papers in bookshelves are quantified by converting the volume to mass using an average density. A typical pack of letter size paper (8.5 × 11 inches paper size with a pack thickness of 2 inches) weighs 2.3 kg. However, the typical paper pack is a relatively dense pile of paper when compared to binders and different book sizes on a bookshelf. A number of measurements, using a digital scale, for piles of books and binders were conducted as part of this study. Based on these measurements, an average density of 500 kg/m3 was derived for books, binders, and piles of papers typically found in bookshelves and drawers. This average density of 500 kg/m3 is applied as reference during the surveys presented in this report. However, the digitized form is setup such that the user can modify the input density for the paper if desired.

2.2.2 Results

Table 2 lists the furniture items and fixed contents (carpet, ceiling panel tiles, and door) surveyed in the office. For each item, the table provides the following information:

Table 2 Summary of surveyed items in a sample office
Full size table

  • Assigned mass. This mass is based on the image processing procedure for most items, and direct match to the model for electronic equipment;

  • Assumed material composition. This composition is either indicated by the available information online or based on the judgement of the surveyors;

  • Fuel load. The fuel load is obtained by converting the mass to energy taking into account the material composition and the net calorific values.

It should be noted that the application includes a drop-down list of materials with default calorific values. The list of materials and the corresponding calorific values are included in a companion paper that describes the methodology (Elhami-Khorasani et al. [14]). The user has also the option of creating a new material by adding the material name and calorific value. The assigned calorific values within the application are independent of furniture geometry, as there is not enough information available to differentiate heat release characteristics at such level.

The total mass density for the movable and fixed content is calculated as 67 kg/m2 (13.7 psf) and 13.2 kg/m2 (2.7 psf), respectively. The measured mass density of furniture items in this office is larger than the reported mean value by Culver’s survey [3] (introduced in Sect. 1) but it is comparable with reported values from Caro and Milke’s study [5]. The fuel load density for the movable and fixed content is calculated as 1327 MJ/m2 and 308 MJ/m2 (ceiling tiles contribute to 96 MJ/m2 of the fixed fuel load density). This office contains two relatively large bookshelves (one completely open where derating factor would not be applicable and the other half open with two drawers).

Initial application of the methodology showed that the surveyors became familiar with the digitized form and post-processing procedure reasonably fast. They practiced using the digitized form and post-processing items for 2 days (repeating some tasks) to become proficient. Surveying duration in an office depends on the number of furniture and the amount of miscellaneous items that are present in the room. Survey of a typical office with a number of furniture items and various piles of papers would take approximately 30 min. A team of two surveyors work best, where one surveyor takes digitized measurements of items, while the second surveyor records the rest of the information. The two surveyors can also enter information into the database simultaneously. Post-processing of standard items such as phones, computers, printers, as well as papers, books, etc. are fast (one material composition is assumed). Chairs, desks, and bookshelves need more careful processing. Only in rare cases, with old and/or odd furniture items, a good matching item through image search was not found.

3 Application of the Methodology to Three Office Buildings

3.1 General Description of the Surveyed Buildings

Following the development of the methodology, three office buildings were surveyed using the proposed methodology. The three buildings are located on the campus of the University at Buffalo. Two of the buildings included closed faculty offices primarily, with a few closed clerical offices and open plan student offices. The third building, with administrative offices, primarily included open plan cubicle work stations with only a few closed offices. The selection of buildings covers different office types in terms of size, use, how long they have been occupied, and floor plan layout. Construction of buildings 1 and 2 was completed in 1980’s while building 3 was completed in 1970’s. The surveys were performed in Summer and early Fall of 2019.

Table 3 summarizes the number of surveyed offices and their occupancy types. A total of 34 closed offices, including 20 faculty/researcher offices, 9 administrative or clerical offices, and 5 common areas, were surveyed. The common areas refer to meeting rooms, computer labs, etc. In addition, 161 cubicles or work stations as part of 12 large open plan offices were surveyed. The range of areas for closed offices were 8 m2 to 87 m2 (86 ft2 to 936 ft2) with an average of 17 m2 (183 ft2), while the open plan offices had areas in the range of 24 m2 to 345 m2 (258 ft2 to 3713 ft2) with an average of 96 m2 (1033 ft2). Overall, a total office area of 1720 m2 (18,514 ft2) was surveyed as part of this study.

Table 3 Summary of surveyed offices in three buildings
Full size table

The fuel load density survey collected information on both moveable and fixed contents. Fixed content mainly covered doors, carpets, and ceiling tiles/panels (panels in two offices included wood ornaments). Two of the surveyed buildings had interior and exterior concrete block walls, the third building had concrete block walls in the exterior and drywalls at a number of interior locations. All three buildings had carpet flooring, except for very few offices. The majority of offices in the two buildings built in 1980’s had ¾ hour fire rated steel doors while the third building had ¾ hour fire rated commercial wood doors. A few cases, such as the main reception or a department office, had glass doors. The primary movable contents found in the offices included chairs, desks, drawers, bookshelves, drawer and bookshelf content, paper, phone, computers and electronics. Some occasionally observed contents included window blinds, white boards, and plants. In case of open plan offices, partitions between cubicles were also included as part of the movable content.

3.2 Fuel Load Data

Table 4 provides results of surveyed fuel load data for the three buildings. The fuel load density is given for movable content, fixed content, and total (i.e. the sum of movable and fixed contents). Listed results include the minimum, maximum, mean, and standard deviation of the fuel load. Besides, results are grouped separately for the different use category of the office. The last rows provide the aggregated results for all surveyed rooms.

Table 4 Breakdown of surveyed fuel load based on office use
Full size table

Overall, a mean value of 1486 MJ/m2 has been measured for the 34 closed offices and 12 large open plan offices (161 cubicles) surveyed in the three buildings. This includes 1115 MJ/m2 of movable content and 371 MJ/m2 of fixed content. The standard deviation is 726 MJ/m2. The value of fixed fuel load density is calculated conservatively, assuming combustible ceiling tiles in cases that the material composition is not clearly known. Excluding the contribution of those ceiling tiles, the mean and standard deviation of the total fuel load density becomes 1401 MJ/m2 and 729 MJ/m2, and the mean and standard deviation of the fixed fuel load density become 284 MJ/m2 and 537 MJ/m2. The results are further discussed in the next sections.

3.3 Data Analysis

3.3.1 Influence of Parameters on Fuel Load Density

From Table 4, an influence of the use category on the fuel load can be observed. Specifically, it is found that closed offices of faculty and researchers have the largest fuel load density, followed closely by open offices for administrative staff. Some of these offices had very large quantities of accumulated paper and books over a number of years. Although age is not considered as a factor, it was observed that offices that were occupied more than 10 years had a larger amount of fuel, when compared to those occupied under 5 years. At the other end of the spectrum, open offices for students (cubicles) and closed common areas (e.g., meeting rooms) have smaller fuel load density. Based on this classification, the results suggest that the use category within the office occupancy (e.g. common area vs. administrative office) has a greater influence on the fuel load density than the distinction between closed offices and open-space cubicles. However, the statistical database in this study is too limited to draw definitive conclusions.

Figure 4 shows the surveyed data from the three buildings in this study as a function of office area. Offices with larger areas tend to have smaller fuel load density, given that the total fuel load is normalized by the larger area to yield the fuel load density. The scatter in data between the three buildings is relatively consistent, while the three buildings cover different office types (administrative versus faculty or researcher offices). Based on observations during the survey, the majority of offices had a relatively heavy load of paper.

Figure 4
figure 4

Surveyed fuel load density as a function of office area

Full size image

Table 5 provides a breakdown of the movable fuel load density in the surveyed offices. The breakdown is presented both in terms of types of items contributing to the movable fuel load (furniture content, electronics, paper, other), and in terms of materials contributing to the movable fuel load (wood, plastic, paper, other). As a comparison with previous studies, of the total movable fuel in the survey conducted by Caro and Milke [5], papers and books contributed 54% of the total load, furniture contributed 25.1%, partitions contributed 11.4%, computer equipment contributed 6.2%, and miscellaneous items contributed 3.3%. The results of this study are relatively consistent with Caro and Milke’s survey outcomes, with 54% paper, 36% furniture, 8% electronics, and 2% other items. It should be noted that in the study by Caro and Milke, the fuel is reported as mass density (psf) and mass values are not multiplied by the calorific values, which complicates the direct comparison. Table 5 also reports the breakdown of movable fuel load density in terms of material composition, with 30% of all content made from wood, 16% plastic, and 54% paper.

Table 5 Breakdown of movable fuel load density in terms of item type and material composition
Full size table

Based on the above discussion, paper contributes to more than 50% of fuel load density in surveyed offices. It must be noted that closed parts of bookshelves and drawers are assumed to be filled with paper. As stated previously, no derating factor is assumed for enclosed spaces. Figure 5 shows a sample of observed paper load in different offices. Figure 5a and b compares accumulation of paper in two faculty offices that were occupied for more than 10 years and 5 years, respectively. Figure 5c shows an open plan administrative office with series of filing cabinets and archives.

Figure 5
figure 5

a closed faculty office, occupied for more than 10 years, with open bookshelf and accumulation of paper over years; b closed faculty office, occupied for more than 5 years, with large open stacks of paper aside from enclosed drawers; c large open administrative office with full height shelves for file archives and a series of enclosed filing cabinets

Full size image

Table 6 shows the breakdown of fuel load density based on the office type (closed versus open). Open offices have slightly smaller total fuel load density, but the difference is not as significant as could be expected (mean total of 1400 MJ/m2 vs. 1519 MJ/m2). Open offices have slightly smaller movable fuel load density, but more importantly, a smaller standard deviation for movable load as well. It can be seen that maximum value of fixed fuel load density is much larger than the mean for both open and closed offices. The large values of the maximum fixed fuel load density (2513 and 2953 MJ/m2) come from two special cases (a meeting room and a reception office) with special ceilings composed of decorative wood finishes. The mean values of fixed fuel load density for open and closed offices change to 354 MJ/m2 and 259 MJ/m2, respectively, if the conservative assumption on fuel contribution from a set of ceiling tiles is removed from the calculations.

Table 6 Breakdown of surveyed fuel load based on office type (open vs. closed)
Full size table

3.3.2 Probabilistic Distribution

Figures 6 and 7 show the cumulative distributions of the movable content and total fuel load densities of all data, respectively. A Generalized Extreme Value distribution (Type I) fit is used. The Generalized Extreme Value distribution is equivalent to Gumbel distribution, which is the prescribed distribution for fuel load density in the Eurocode [19]. The fit shows good agreement with the data. The mean of the fitted distribution for movable content is 1116 MJ/m2 with a standard deviation of 604 MJ/m2, which are within 2% of calculated values for the data as reported in Table 4. Similarly, the mean of the fitted distribution for total fuel load density is 1485 MJ/m2 with a standard deviation of 719 MJ/m2, which are within 1% of calculated values for the data. Table 7 shows the parameters for the fitted distributions for both movable and total fuel load density.

Figure 6
figure 6

Generalized extreme value distribution fit to the movable fuel load density for the surveyed offices

Full size image
Figure 7
figure 7

Generalized extreme value distribution fit to the total fuel load density for the surveyed offices

Full size image
Table 7 Parameters of fitted generalized extreme value distribution to fuel load density data
Full size table

3.3.3 Sensitivity Analysis

In obtaining the above results, one main assumption was in regards to the content of closed drawers and closed sections of bookshelves. It was assumed that on average, 50% of the volume of the closed drawer or bookshelf is filled with paper (for determination of calorific value). During the survey, a few of the occupants provided input in terms of the content in their closed spaces and permitted the surveyors to open the drawers. The actual volume and type of content was similar to the assumed approach in this survey. Nevertheless, a sensitivity analysis of the results to the assumed volume of the available content in closed drawers and bookshelves is completed to investigate the influence of this assumption.

Table 8 shows the change in the amount of movable fuel load density if the percentage of the drawer and closed bookshelf contents change to 40% or 30% of volume from the original 50% assumption. The maximum value of movable fuel load density drops by 336 MJ/m2 when lowering the content from 50% to 30% volume. Under the same assumption, the mean value of movable fuel load density drops by 131 MJ/m2. The latter represents an 11.7% decrease in the mean movable fuel load density, and an 8.8% decrease in the total fuel load density (reducing it from 1486 MJ/m2 to 1355 MJ/m2). This sensitivity study is also similar to application of a derating factor to the enclosed content. As was shown in Fig. 5, a number of offices in this study had large stacks of papers and books in open bookshelves, therefore, the change in assumed percentage of content in enclosed shelves or drawers does not significantly lower the amount of fuel load density.

Table 8 Sensitivity of results to assumed volume of paper in enclosed drawers and bookshelves
Full size table

The sensitivity of results was also investigated with respect to two different users post-processing the items. Five rooms were randomly selected, and the movable contents were post-processed by two members of the team independently. As stated earlier, the post-processing includes finding the weight of an item through image search and including material composition. The objective was to investigate how the user (surveyor) affected the results through different choices in post-processing. The calculated values for movable fuel load density reported by User 1 and User 2 showed some level of discrepancies. When summing the absolute differences of all reported values, the average absolute difference amounted to 135 MJ/m2. However, no systematic under- or over-estimation was consistently made by a user compared to the other; in other words, differences in reported values tended to compensate, and the mean movable fuel load densities reported by User 1 and User 2 differed by only 16 MJ/m2.

4 Discussion and Recommendations

4.1 Comparison with Previous Surveys

The obtained results are compared with previously conducted survey by Culver in 1970’s [3]. Table 9 shows the statistics of fuel load density data for government and private office building, based on Culver’s survey. Culver reports fuel load density in terms of mass density or lbs/ft2, while the reported values in Table 9 are in MJ/m2. The majority of fuel in 1970’s was assumed to be made of wood (cellulosic); therefore, a caloric value of 18 MJ/kg was conservatively assumed to covert mass density to fuel load density (taking into account the possibility of different material compositions). NFPA 557 [20] also makes a reference to Culver’s data but applies a calorific value of 15 MJ/kg for the conversion. Culver used the inventory method for surveying offices, where the mass density of an item was calculated using a series of transfer functions based on an inventory of furniture collected from manufacturers’ catalogues. While the maximum fuel load density from Culver’s survey is comparable to the range of data in this study, the average value from Culver’s survey (615 MJ/m2 and 659 MJ/m2) is considerably smaller than that of the current study (1486 MJ/m2 for all offices).

Table 9 Results of Culver’s survey of fuel load density from 1970’s [3]
Full size table

A Canadian survey from 2011 [9] reported results in MJ/m2 and compared fuel load data obtained from direct weighing, inventory, and combined surveying techniques. The inventory method resulted in a mean of 852 MJ/m2 and standard deviation of 484 MJ/m2. The survey had few data points below 500 MJ/m2. Besides, the Canadian study used derating factors for enclosed items, which reduces the fuel load due to papers.

To make further direct comparison with other studies, the collected data in this survey is also quantified in terms of mass density (not taking into account the material composition and calorific value). The mean mass density of all offices in this study is 62.2 kg/m2 (12.7 psf). As a comparison, Caro and Milke [5] reported mean mass density for movable items in the order of 71 kg/m2. Therefore, the amount of movable content within surveyed offices in both studies is comparable.

4.2 Comparison with Design Codes

Eurocode 1 [19] provides a recommendation for the mean and 80% fractile of a Gumbel distribution for movable fuel load density. The Eurocode 80% fractile fuel load densities for office, library, and dwelling are 511, 1824, and 948 MJ/m2. The 80% fractile fuel load density from the fitted distribution in this study is 1560 MJ/m2. Figure 8 shows a comparison of Eurocode fuel load distributions for office, library, and dwelling with the obtained distribution in this study. It can be seen that the current survey yields much greater values of fuel loads than the Eurocode distribution for office occupancies. Yet, data from the survey fall within the bounds of the Eurocode data when considering the different types of occupancies. From this observation, it can be questioned whether the office classification is adequate for the type of rooms surveyed in this study.

Figure 8
figure 8

Comparison of current survey result with fuel load distributions from Eurocode [19]

Full size image

NFPA 557 [20] recommends an average content fuel load density of 600 MJ/m2 (with standard deviation of 500 MJ/m2) for office buildings, mainly based on Culver’s survey [3]. The mean fuel load density for moveable content in this study is 1115 MJ/m2. In a previous study, Elhami Khorasani et al. [21] analyzed Culver’s surveyed data, considering the correlation of fuel load with the room type and area of the room, and compared Culver’s data with the 95% fractile design fuel load density from the background Eurocode data [19] for office occupancy. The data were organized into two main categories: (1) compartments with lower fuel load (general, clerical, lobby and conference) and (2) compartments with higher fuel load (file and storage). It was concluded that the Eurocode value was conservative for light weight categories but the majority of data points in the heavy-weight category exceeded Eurocode’s design value. In this study, it is clear that the surveyed rooms fell within the ‘higher fuel load’ category given the large amount of paper stored in these academic offices.

Based on comparison of fuel load for different occupancies in the Eurocode (office versus library), the influence of room use on value of fuel load density, and given that 54% of the moveable fuel load in the surveyed offices comes from paper, one consideration for future studies could relate to sub-categories of occupancy types for determination of fuel load density. For example, an office occupancy could include different sub-categories of general and clerical offices, educational institutions, businesses, etc.

Another observation made during the surveys, from the design perspective, is that the fuel load density across large open floor plan offices (with the number of cubicles/workstations in the order of 20 or more) is distributed uniformly across the floor. The workstations are relatively similar in terms of furniture (one desk, one or two chairs, drawer, computer, monitor, keyboard, mouse, phone, trash can, and paper). The workstations are generally arranged in a grid across the floor. There is typically a common room for a photocopier/printer, fax, etc. that serves one or few units in the office.

4.3 Summary and Recommendations

The following observations can be formulated from the application of the digitized fuel load surveying method to three office buildings.

In terms of the fuel load data collection, a total office area of 1720 m2 was surveyed consisting in 34 closed offices and 161 cubicles within 12 large open plan offices. The measured fuel load for movable content ranged between 109 MJ/m2 and 3143 MJ/m2, with a mean of 1115 MJ/m2 and a standard deviation of 614 MJ/m2. When including the fixed content, the measured total fuel load ranged between 368 MJ/m2 and 3451 MJ/m2, with a mean of 1486 MJ/m2 and a standard deviation of 726 MJ/m2. The data is well captured by a Generalized Extreme Value distribution.

The obtained fuel load results based on the given sample of offices are considerably larger than recommended values in standards for office buildings (e.g. Eurocode [19], NFPA 557 [20]). While the findings of this report are based on a relatively small sample and office buildings that are located on a university campus with accumulated piles of books and papers, the extent of the difference is noticeable. A comparison to previous surveys is provided. It is noteworthy that older studies report fuel load density in terms of psf, hence not taking material composition into account, which makes a direct comparison more difficult. The mean mass density for movable items reported by Caro and Milke [5] was larger than that obtained in this survey. In the present study, movable fuel load was composed at 54% by paper, and no derating factor was applied to enclosed content. Observations of the surveyed offices across the three buildings (including administrative and academic offices) showed a large amount of accumulated paper. Even when modifying the assumption on volume of paper in enclosed drawers and bookshelves, the obtained fuel load remained larger than prescribed design values in current codes and guidelines. This pleads for future fuel load data collection efforts to either confirm or update the existing codes and guidelines for fuel load density in office buildings. The results also suggest that a new definition of occupancy sub-categories might be needed for determination of fuel load density, at least to distinguish between office occupancies with lower versus higher fuel loads.

In terms of the methodology, technological advances over the last decades in mobile technologies, computer vision, artificial intelligence, and cloud storage, provide opportunities for breakthrough in surveying methods for fuel load data. This proof of concept for a novel methodology harnessing (some of) these advances has demonstrated the feasibility and interest of a digitized inventory procedure using machine vision.

However, a number of directions can be identified for improvement, and require further works. In general, updated databases of calorific values for common furnishing, not just component materials, are needed to decrease uncertainty in fuel load calculations. In terms of the methodology, future developments could include the ability to automatically recognize the material composition of complex items (e.g. electronics); the ability to automatically evaluate the fuel load associated with fixed contents (e.g. detecting materials and thickness of room lining); the ability to extract information based on a single 360° picture of a room rather than individual pictures of all items; and the possibility to fully automatize the post-processing by populating the fuel load data based on the matching pictures and measurements. The first improvement could rely on targeted studies to create a database of material composition and fuel load of electronics and other equipment commonly found in offices. The second and third improvements will be linked to advances in machine vision, and it seems reasonable to expect that the technology will be available in a not-too-distant future. The fourth improvement could naturally derive from wide adoption of the methodology that would result over time in an extensive database of digitized fuel load surveys.

5 Conclusions

This paper builds on a newly established digitized fuel load survey [14]. This methodology takes advantage of recent technological advances, notably in machine vision, to modernize the inventory method with a digitized approach. The paper applies the methodology to three office buildings and provides results of the surveyed fuel load density. The application shows the feasibility and practicality of the methodology, notably in terms of matching results obtained automatically through image search for the movable items, in terms of speed, and in terms of consistency and robustness of the approach to particular choices made by the surveyor, compared to existing surveying methodologies. The application allowed collecting new data on fuel load in office buildings. The surveyed compartments had a mean fuel load of 1486 MJ/m2 with a standard deviation of 726 MJ/m2. These relatively large values are thought to be due to the nature of the offices surveyed, the large amount of paper, and the fact that no derating factor was applied to enclosed content. Yet, considering these values and the outcomes of other recent surveys, it seems critical to collect additional information and conduct more surveys on fuel load density of different occupancy types to provide recommendations for design.

The proof of concept for the new methodology illustrated that breakthrough of survey methods can be realized by taking advantage of recent technological advances. However, this methodology is the first-generation of its kind and directions for improvement are identified in the paper. The long-term objective of this study is to develop the foundation of a database with an inventory of images and the corresponding fuel load density values. Such database could then dramatically speed up the surveying process by directly matching (through computer vision) surveyed items with a fuel load value, i.e. bypassing the current step of identification of the weight and material composition. From the survey presented herein, the built database includes information on 135 chairs, 94 desks, and 59 bookshelves. The work has established the foundation toward a fully automatized method, by defining the methodology, setting up the electronic form, and a structured database of recorded information. A wide adoption of this method could provide more benefits by progressively populating such digital fuel load database. The results can then be used to provide design guidelines for fuel load density in codes and standards, for application in performance-based design.