What is a floating roof tank

Floating roof tanks for petrol

Various tank shapes are used in refineries for the interim storage of petrol. This includes fixed-roof tanks and floating-roof tanks of various designs and retrofits to reduce the emission of volatile organic compounds. Floating roof tanks often represent an ecological and economic compromise between the highest possible emission reduction and comparatively low technical complexity. A fixed roof tank without upgrading is used as an emission reference for discussing the emission reduction achieved. An example floating roof tank is described below and its emission behavior is assessed. The main focus is on the discussion of key influencing variables and their uncertainties, which are included in the emission assessment.

PantherMedia / Gabriel777

Petrochemical products manufactured in refineries, such as fuels and raw materials for the chemical industry, are typically temporarily stored in storage tanks before they are sold, before they are decanted and transported to the customer. Fixed-roof or floating-roof tanks can be used for the storage of petrol, with floating-roof tanks often being used for petrol. These are characterized by a movable roof that follows the changing liquid level during emptying and filling processes and floats on top of the liquid. The moving parts are sealed by special sealing systems which, despite the mobility of the roof for volatile compounds such as petrol, can lead to lower emissions overall compared to fixed roof tanks. Furthermore, only a very small volume of air is available under the roof for the formation of saturated or partially saturated petrol vapors and a permanently active exhaust gas cleaning system or the use of a gas exchange between the tank and the transport container is not required. Although the direct emissions from appropriately equipped fixed-roof tanks are negligible, the operation of gas exchange (gas storage, recovery) results in higher technical costs and associated emissions due to the use of materials and energy. The ecologically relevant net pollution can therefore be more favorable for modern floating roof tanks.

Emissions from floating roof tanks result from escapes from sealing systems and from evaporation of lubricating films from the inside of the tank wall during removal from the tank. Since petrol, in particular, is a highly volatile and highly flammable mixture of substances, a non-zero probability of the formation of a dangerous explosive atmosphere must be assumed. For the evaluation of the probability of occurrence and characteristics of this dangerous explosive atmosphere, for questions about substance emissions into the environment, for substance dispersion analyzes and for emission reductions, emission estimates can be relevant. There are now recognized procedures that can determine the release of substances based on the tank design and depending on the stored substance. The rules API 2517 and 2519 [1] as well as their German implementation through the guideline VDI 3479 [2] should be mentioned here in particular. Using these guidelines, a representative floating roof tank should be estimated with regard to its emission behavior. The main focus is on the discussion of key influencing variables and their uncertainties.

The work was carried out as part of a DGMK research project [3], in which a floating roof tank for petrol was monitored with regard to its emission behavior as part of a long-term measurement. The aim of the project was the metrological assessment of whether the actual emission behavior corresponds to the expectations from the zone assignment of the hazardous areas or whether the zone assignment is over-conservative.

Structure and characteristics of the floating roof tank

Floating roof tanks are i. d. Usually standing cylindrical containers made of structural steel, which are positioned in a collecting space. With regard to the storage of petrol, specific requirements for the tank equipment are based on the Technical Rule for Hazardous Substances (TRGS) 509 and for emissions behavior from EC Directive 94/63 / EC (implemented in Germany by the 20th BImSchV). The top is closed by a floating roof that is sealed off from the atmosphere, so that there is no saturated or partially saturated gas phase above the liquid. The seal between the tank wall and roof (annular gap) and the roof ducts (e.g. roof supports, guide tube) is made using suitable sealing systems. In normal operation, the sealing elements represent more or less diffuse sources, or emissions occur through evaporation of the liquid, which stick to the steel shell when the liquid level drops. The source strength is strongly dependent on the type of annular gap seal (double seal, triple seal), the guide tube construction / seal and the degree of corrosion of the tank shell. Floating roof tanks with a primary and secondary seal are permitted for petrol (existing system) if these result in a minimum emission reduction (at least 97%) compared to a certain (not upgraded) comparable fixed roof tank. A schematic representation of a floating roof tank together with the dimensions used in this article is shown in Figure 1.

Fig. 1 Schematic representation of a floating roof tank together with the dimensions and naming of the assemblies relevant for the emission assessment, which are important for the emission assessment, taken and adapted from [2].

Photo: Otto von Guericke University Magdeburg.

Calculation method according to API 2517/19 and VDI 3479

The guideline VDI 3479 is an almost identical transfer of the API 2517/19 into German, so that there are no fundamental differences between the two calculation methods. During the transfer, however, conversions from Anglo-American to metric units were carried out and some negligible emission components were omitted. In this entry, designations and symbols according to VDI 3479 are used.

A distinction is made in these procedures: (i) Loss of standing due to imperfect seals, and (ii) Loss of extraction due to the evaporation of lubricating films when the roof is lowered. Emissions resulting from deviations from normal operation are not taken into account.

The total loss mass flow L.T (in kg a-1) is made up of the static loss mass flow L.S. (in kg a-1) and the traffic loss mass flow L.W. (in kg a-1) together:



The standing losses L.S. include the sum of all emissions arising from the edge seal between the tank roof and the tank shell, described by the mass flow L.R., from the built-in fittings, the mass flow L.F. and through the roof due to its special construction, described by the mass flow L.p (all sizes in kg a-1).

The loss mass flow L.p becomes in the present case to L.p = 0 selected because the tank roof is welded to look tight. So it follows:



The mass flow loss from the edge seal L.R. is a dimensionless function of ambient and vapor pressure p*, from a sealing loss factor KR. (in kmol m-1 a-1), from the tank diameter D. (in m) and of the molar mass M. (in kg mol-1) depending on the escaping substance:



The tank roof fittings loss L.F. from the various roof fittings is the roof fittings loss factor KFi (in kmol a-1) of the valve under consideration and the corresponding frequency NFi depending on this valve:




For the calculation of the traffic losses L.W. is the surface of the lubricating film, the density of the petrol W.L. (in kg m-3) and a conservative estimate of the thickness C. (in m) of the lubricating film is important. Here the surface is determined by means of the tank diameter D. and the annual withdrawal Q (in ma-1), which means for L.W. follows:




For the total loss L.T thus follows:






The loss factors KR. = KR.(u) and KFi = KFi(u) are given here as functions of the long-term mean value of the wind speed u and are also dependent on the specific construction of the valve. In both guidelines, the loss factors for four wind speed values ​​are u = 0; 2.2; 4.5 and 6.7 m s-1 specified. For intermediate values ​​of u, linear interpolation is used in VDI 3479, whereas in API 2517/19 fit functions of the form:



are specified. Since the loss factors are on convex curves as a function of u, the linear interpolation leads at least to conservative values.

Here the corresponding parameters a, b, and c for the interpolation functions in SI units were determined and used using the least squares method. Figures 2 and 3 show the specific (for the fittings and structural properties assumed here) loss factors depending on the wind speed and indicate the fit functions.

Figure 2 Representation of the loss factors KFi as a function of the wind speed u and the associated fit functions. Guide tube: KF (u) = 19.00 + (15.73 u) 1.36, breathing fitting, KF (u) = 3.55 + (0.40 u) 4.07, liquid level indicator: KF (u) = 6 , 42 + (3, 39 u) 1.115, sealing loss factor: KR (u) = 0.62 + (1.34 u) 0.99.

Photo: Otto von Guericke University Magdeburg.


Figure 3 Representation of the loss factors KFi as a function of the wind speed u and the associated fit functions. Entrance opening for people: KF (u) = 0.70, sounding and sampling pipe: KF (u) = 1.00, roof support: KF (u) = 0.90 + (0.23u) 0.78.

Photo: Otto von Guericke University Magdeburg.

It can be seen that the specified functions of type (7) fit the data points quite well.

The dimensionless function of ambient and vapor pressure p* is:





Here is pd the vapor pressure of the stored liquid and pa the ambient pressure.

Emissions estimates and fluctuations

The releases to be expected at floating roof tanks should be calculated using the methods described above. For this purpose, assumptions about the in Eq. (6) included sizes to be taken.

Input parameters and comments

Tank diameter D

Typical tank diameters are in the range from 20 to 60 m D. = 35 m assumed.

Withdrawal volume flow Q and QMax

The withdrawal volume flow Q is heavily dependent on the storage and retrieval (tank turnover per year) by the company and also on the dimensions of the tank. In order to be able to provide an orientation, data from a tank operator [3] were evaluated over a period of ten years. The data refer to a tank of comparable size.

When a tank is handled, it is often not the maximum volume that can be achieved in normal operation that is handled, but only a partial amount. The typical figure was 30 full tank turns per year. A deviation of  25% should be assumed here.

So it follows Q = (290,000  73,000) m3 a-1. In the guideline VDI 3479 and in the API 2517/19 no maximum value for the loss of traffic is calculated, but only an annual mean value. Peak values ​​can be higher because traffic losses are limited to the removal times as a first approximation.

If one also assumes that an exposed film of lubricant evaporates immediately, then peak emissions occur during the fastest extraction that is achieved in operation. Again underpinned by information from the operator [3], the maximum permissible volume flow during withdrawal QMax = 500 m3 H-1 = 0.139 m3 s-1 be accepted. If you put this in place of Q in Eq. (5), it follows L.Wmax (later given in kg s-1).

Molar mass M

Since the components escaping from the seals lead to the expectation of an increased proportion of volatile components, the conversion from kmol a-1 on kg a-1 in both directives with a molar mass of M. = 64 g mol-1 expected. Publications such as [4] show that the mean molar mass of the volatile components in the gas phase (rounded up slightly) is in the range of M. = 80 g mol-1 will lie.

The mean value was therefore used here M. = (72  8) g mol-1 used as a (conservative) compromise, since a higher molar mass mathematically results in a higher loss mass flow.

Vapor pressure function p *

For the estimation of p* become the vapor pressure of the petrol pd and the ambient pressure pa = 1.013 x 105 Pa needed. The vapor pressure of the petrol is strongly dependent on the temperature and also on the stored blend (explanation of terms below). In the guideline VDI 3479 for petrol for pd as an estimate pd = 0.4 x 105 Pa stated. This value is a compromise between the vapor pressures of summer OK (pd = 0.38 x 105 Pa) and winter OK (pd = 0.45 x 105 Pa), measured at a temperature of T = 293 K.

Here it was decided pd = (0.42  0.04) x 105 To use pa. The fluctuations take into account the distance to the summer or winter OK at a temperature of T = 293 K. This gives:



The uncertainty follows with the total error differential from the expected values ​​of the input variables and the specified fluctuations for the vapor pressure pd.

Liquid density WL.

The composition of petrol can vary considerably depending on the type of crude oil used in the refinery. There are also other additives for the end product. These different possible mixtures are usually called blends. An overview of this can be found in [5].

Furthermore, certain properties of the petrol are varied with the seasons or according to legal requirements. In summer fuel (summer OK, summer blend), fewer low-boiling alkanes are added in order to avoid the formation of vapor bubbles in the gasoline due to the typically higher temperatures, while in winter (winter OK, winter blend) the cold start properties are favorably influenced by increasing these proportions. This is also the reason for the previously discussed fluctuations in pd. The proportion of added ethanol (E5, E10) also has a significant influence on the composition. The exact composition of a blend may vary within certain legally regulated tolerances. In the standard DIN EN 228 [6], the minimum permissible density is WLmin = 720 kg m-3 and as the maximum permissible density of the blend W.Lmax = 775 kg m-3 (Values ​​at T = 288 K). It was therefore decided to use W.L. = (748  28) kgm-3 to be expected.

Wall wetting layer C

For the thickness of the wetting film in VDI 3479 a value of C. = 2.57 x 10-6 m used for a smooth tank wall (structural steel). For a heavily corroded tank wall, however, the value can also be five times higher. As the degree of corrosion and thus also the wall wetting layer C. is difficult to estimate without measurement, here for the film thickness (rounded up) C. = 1.3 x 10-5 m generally accepted.

Wind speed u

As mentioned earlier, is KR. or are the KFi depends on a (mean) wind speed that is neither specified in VDI 3479 nor in API 2517/19. However, it can be assumed that the speed is an average overflow speed of the assemblies on the tank or that the values ​​used can be traced back to emission tests on the assemblies at the specified wind speeds. Because the long-term means u = u the wind speed is strongly dependent on the tank location, this was considered a free parameter. As a mistake u 25% of the amount of u is used.

Edge sealing factor

The edge sealing factor F.R. is by means of KR. Calculated for a very well-fitting seal (designation according to VDI 3479) of the type mechanical slide shoe seal with edge-mounted secondary seal.

Roof fittings loss factor

To calculate the roof fittings loss factor, separate fitting type-dependent loss factors are used for all existing roof fittings KFi calculated that like KR. previously from the amount of wind speed (annual mean) u the overflowing wind and the built-in number are dependent. Here were included in the calculations (representative numbers):

  •  Slotted guide tube with scraper (1),
  •  Access opening for people, sealed (2),
  •  liquid level indicator (1),
  •  direction finder and sampling tube (1),
  •  ventilation valve (1),
  •  Tank roof supports, ring pontoon supports, not sealed (45).

It should be noted that high emissions are correlated with high wind speeds. At higher wind speeds, higher emissions, but also higher dispersion or higher transport, are to be expected.

General comments on the losses

Loss of standing must always be assumed if there is petrol in the tank. For the static loss mass flow (mass loss per year) it was assumed that the tank is used continuously. The annual stand losses determined can then be converted directly to a mass flow rate every second. For the traffic losses it must be noted that these with Eq. (5) can be calculated as mass loss per year. It is assumed that the film of lubricant that has formed has completely evaporated before the roof rises again.

Results of the emission estimation

The calculations were implemented using a specially designed C ++ program. The specified errors were calculated using the total error differential from the expected values ​​and the fluctuations in the input parameters. In picture 4 is the expected value of the annual losses L.i, i = S,L,T (in kg a-1) as a function of the wind speed u.

Figure 4 Representation of the annual loss mass flow mean values ​​Li, i = S, W, T as a function of the wind speed and the like.

Photo: Otto von Guericke University Magdeburg.

It can be seen that the annual mean values ​​of the stand losses L.S. compared to the mean traffic losses L.W. dominate. However, if one assumes conservative estimates of the withdrawal behavior as in Figure 5 - use of QMax = 500 m3 H-1 = 0.139 m3 s-1 instead of Q as W.L. = 775 kg m-3 without error tolerance in Eq. (5) - so follows L.Wmax = 1.6 x 10-4 kg s-1 and both types of emissions are of similar magnitude. For illustration, the maximum values ​​of the losses per second are shown in Figure 5 L.Smax = L.S. + L.S., L.Wmax and L.Tmax = L.Smax + L.Wmax shown as a function of u. These curves can be used for conservative immission calculations.

Figure 5 Representation of peak loss mass flows Limax, i = S, W, T as a function of u.

Photo: Otto von Guericke University Magdeburg.

Figure 6 shows a breakdown of the standing losses according to the individual assemblies. It can be stated that the highest emissions can be expected from the edge seal and the guide tube. Especially for higher wind speeds, losses from tank roof supports, sounding pipes, ventilation fittings and access openings for people can be neglected compared to the emissions from the slotted guide pipe. A graphical relationship between the static loss mass flow L.S. and the annual mean wind speed u is shown in Figure 7.

Fig. 6 Representation of the annual static loss mass flows broken down according to the individual assemblies LSi depending on u. The index i denotes the assembly (see legend).

Photo: Otto von Guericke University Magdeburg.


Figure 7 Representation of the static loss mass flow LS as a function of the annual mean of the wind speed u. The fluctuations shown are calculated for 25% of u and taking into account the error tolerances of the other input variables. The numerical values ​​entered are the static loss mass flows for the integer speeds in the units 1 x 103 kg a-1 (above) or in 1 x 10-4 kg s-1 (below).

Photo: Otto von Guericke University Magdeburg.

This shows how strong L.S. depends on the long-term mean value of the wind speed u. The corresponding values ​​for six comparative values ​​of u are given in Figure 7.

Concentration assessment and discussion

The expected values ​​for the emissions are in units of mol s-1 or in kg s-1 in front.

For the calculation of a concentration profile resulting from the emissions c(x, t) in kg m-3 or in ppm, no simple models of the dispersion of matter can be used, as there are many sources of different types (surface, line, point sources) which also emit diffusely. Furthermore, the material is dispersed with a turbulent wind field and the emission rate is dependent on the turbulent overflow itself. It is also important that the wind field depends on local conditions and cannot easily be taken from simple location weather data. A determination of c(x, t) is conceivable by means of complex CFD simulation (and numerous other assumptions that would have to be justified experimentally), while light gas dispersion models that are frequently used, such as e.g. B. Gaussian models (e.g. VDI 3783 Part 1 [7]) are excluded.

For a purely rough estimate, very abstract, but conservative assumptions should be made here:

1. The tank roof is at its maximum height, so that between the roof and the upper edge of the tank there is a remaining height of HR. = 1.5 m remains (typical values: 1 to 2 m).

2. The total emission of L.Tmax feed the remaining volume of VR. = 1 443 m3 (without air exchange).

With the ideal gas equation (x in ppm s-1) then roughly applies:




Using the results of L.Tmax = L.Tmax(u), the concentration increase x can be expressed in ppm s-1 to calculate. A representation of this relationship can be found in Figure 8 (calculated with T = 293 K).

If the lower explosion limit (LEL) of gasoline-air mixtures is assumed to be 6,000 ppm, the values ​​shown in Figure 8 can be used to determine the time until the LEL is reached. A corresponding curve is added to Figure 8 as an inset figure. However, it should be explicitly pointed out here that high emissions are correlated with high wind speeds. The estimates are therefore logically inconsistent (but very conservative), since the wind would carry away the evaporated substances. At lower wind speeds, the emissions are lower.

Fig. 8 Representation of the increase in concentration u in ppm s-1 with immission of LTmax into a closed volume as a function of u .. The inset shows the associated time until the LEL of 6,000 ppm is reached.

Photo: Otto von Guericke University Magdeburg.

If one assumes an average distance of 10 m to leave the tank roof as part of another estimate, the dwell time is tE. of an emitted petrol ensemble E. at a flow rate of u = 1 ms-1 (low wind, Beaufort scale 1) tE. = 10 s and at u = 0.1 ms-1 (Calm, Beaufort scale 0) tE. = 100 s. With dwell times of 100 to 10 s one calculates with Eq. (10) Concentrations in the one to two-digit ppm range. Higher wind speeds lead to lower concentrations here, since the dwell times are estimated to be shorter. Reference is made once again to DGMK research report 793 [3], in which these estimates could be confirmed on the order of magnitude by measurements with a photoionization detector.


It was shown how emissions of volatile gasoline components from floating roof tanks can be estimated. For this purpose, the guidelines API 2517 and 2519 as well as their German implementation through the guideline VDI 3479 were used. The emission estimates also reveal the influence of inherent uncertainties in the model parameters. The latter were included in the estimates as part of an error analysis. If you convert the annual emission values ​​into mass flows every second and roughly estimate the concentrations above the tank roof, you get one to two-digit ppm concentrations as conservative expected values. TS 615





[1] Manual of petroleum measurement standards - Evaporative loss from floating-roof tanks. API Publication 2517 and 2519. Washington: American Petroleum Institute 1997.

[2] VDI 3479: Emission Reduction - Mineral Oil Tank Depots Remote from Refineries. Berlin: Beuth Verlag 2010.

[3] Koehler, F .; Zinke, R .: Determination of potentially explosive areas in tank systems. DGMK research report 793. In press. Ed .: German

Scientific Society for Petroleum, Natural Gas and Coal V. Hamburg 2017.

[4] Henley, M. et al .: Health assessment of gasoline and fuel oxygenate vapors: Generation and characterization

of test materials. Regul. Toxicol. Pharmacol. 70 (2014), pp. S13-S17.

[5] Schmiedel, H.-P .: Composition of petrol from German refineries - winter goods 2001/2002. DGMK research report 502-1. Ed .: German Scientific Society for Petroleum, Natural Gas and Coal e. V. Hamburg, 2003.

[6] DIN EN 228: Fuels - Unleaded petrol - Requirements and test methods. Berlin: Beuth Verlag 2017.

[7] VDI 3783 Part 1: Spread of air pollution in the atmosphere; Spread of accidental releases; Security analysis. Berlin: Beuth Verlag 1987.


From Dr. rer. nat. Ronald Zinke, M. Sc. Florian Köhler and Prof. Dr.-Ing. Ulrich Krause

Dr. rer. nat. Ronald Zinke, M. Sc. Florian Köhler, Prof. Dr.-Ing. Ulrich Krause, Institute for Apparatus and Environmental Technology, Otto von Guericke University Magdeburg.