Optical properties of fire smoke are in the focal point of the fire science, highly relevant to both physical experiments and computer modelling of fires. The obscuration of light by smoke is usually presented in the form of a ‘
visibility in smoke’ parameter. In performance-based analyses for fire safety engineering of buildings, the visibility in smoke is often the first tenability criterion that exceeds its threshold value [
1], in consequence shaping the fire safety solutions in the building. The smoke obscuration also influences the visibility of evacuation signs, the response to the information on the signs, behaviour and the movement speed of people who are evacuating [
2]. Finally, low visibility in smoke will be one of the critical challenges that firefighters face when performing rescue operations.
Despite the profound relevance of visibility in smoke to the fire safety engineering, the existing theory of visibility in smoke was criticized as oversimplified, based on a limited range of experimental results and highly sensitive to user-dependent properties [
3]. Even though multiple studies were performed to improve the existing theory to include for more diverse visibility of various evacuation signs (e.g. [
4,
5]), the conventional approach is still the simplest model introduced by Jin [
6]. This model is based on a limited number of experiments (primarily from the 1960s and 1970s) on the obscuration of monochromatic light by homogenous cold smoke [
7,
8]. The development of this model was summarized in [
2,
9].
To better understand how light is obscured by the fire smoke and its implications on the performance of evacuation lighting and signage, it is essential to recognize the effects of polydisperse smoke (non-homogenous) on polychromatic light. This assumption is a significant improvement in the precision as compared to the traditional approach, but also resulting in higher complexity of the problem. To enable future research under these conditions, we have developed a novel construction of a smoke densitometer, that may be installed directly within a smoke reservoir and measure smoke extinction concurrently in five wavelengths. In this paper, we discuss similar developments in the past, the construction of the device with particular technical challenges that were faced in the development phase, calibration and exemplar use in an n-Heptane compartment fire experiment.
1.1 Obscuration of Light in Smoke Layers
The obscuration of light passing through a medium is often described with Bougher-Lamber-Beer law (Eq.
1). The ratio of light intensity measured at a target in obscured conditions (
I), to the light intensity measured with a clean optical path (
I0) is the transmittance (
τ). The light obscuration depends on three elements: specific light extinction coefficient (
σs), the mass density of smoke (
m) and the length of the optical path (
l). Another common way to present the light extinction of the smoke is the natural logarithm of
τ, which is called the obscuration density (OD) and is presented in [dB/m].
$$\tau =\frac{I}{{I}_{0}}={e}^{-{\sigma }_{s}ml}$$
(1)
It was recognised that this application of the Bougher-Lambert–Beer law has many limitations emerging from the preliminary assumptions of the law [
3], among which: (a) it should be used for monochromatic light, (b) it should be used for single or parallel light sources, (c) the light absorption is considered dominant over light scattering effects. Despite above mentioned profound limitations, this relation is still commonly used to determine the light extinction in fire smoke. Significant simplification in the practical use of this method was introduced by unifying the value of the specific light extinction coefficient (
σs) for flaming combustion, with a value of
σs = 8.7 m
2/g (± 1.1 m
2/g) [
10]. This value was determined based on seven studies involving 29 fuels, for light wavelength λ = 633 nm. A major implication of determination of a nearly universal value of
σs was enabling to determine the mass concentration of smoke based on light extinction coefficients. This universal value was widely adopted in engineering tools of fire science, among them the most popular CFD code FDS [
11]. Other values of
σs can be found in literature and engineering guidance – e.g. the value of 4.7 m
2/g used in the design of visibility conditions in road tunnels [
12] or other values of
σs reviewed by [
9].
In [
13], the value of
σs was discussed as a function of light wavelength. This was done by investigating a large batch of measurements of specific light extinction coefficient in wavelengths ranging from UV (385 nm) to far-IR (10 600 nm), with corresponding values of
σs from 12 m
2/g to 1 m
2/g respectively. The value of
σs decreased with increasing wavelength. A least-square fit to the experimental data was found (Eq.
2), with R = 0.98675. However, it must be emphasized that this fit applies to stoichiometric and overventilated combustion [
13].
$${\sigma }_{S}=4.8081{\lambda }^{-1.0088}$$
(2)
As the
σs asymptotically grows with the decreasing wavelength, the differences in
σs in the lower end of the visible spectrum are profound, which was confirmed in qualitative result analysis of some experiments on the visibility of evacuation signs, e.g. [
9,
14,
15] or research on the applicability of free space outdoor optic connectors in foggy weather [
16]. As the value of
σs can differ significantly within the visible spectrum, it is obvious that the light obscuration through the smoke at different wavelengths will take different values. In fact, past research was focused on exploiting this optical characteristic of smoke for remote analysis of characteristic smoke diameter, as summarized in Sect. 1.3.
1.2 Multi-Wavelength Smoke Densitometers in the Literature
Differences in the obscuration of light at different wavelengths were often connected to the distribution of particle sizes of the smoke [
15,
17,
18]. As the particle size distribution depends on the properties of burning material and the conditions in which the combustion occurs, this approach was considered for remote identification of burning fuel.
Jin [
15] has considered the impact of light obscuration at a different wavelength on the visibility of evacuation signs. He has recognized that the extinction coefficient in short wavelengths is larger than for long wavelengths and that the difference between them is also a function of time. The difference diminishes as an effect of coagulation of smoke particles. The particle size of 1 µm was proposed as the boundary above which the wavelengths stops being an important factor for visibility. He has calculated the ratios of the visibility in red light to the one of blue light, for smouldering and flaming polystyrene, flaming wood and flaming kerosene.
Dobbins and Jizmagian [
18] proposed two methods to determine the mean size of polydispersed dielectric spheres, first of which was the determination of a single spectral transmittance together with knowledge of particle concentrations. This approach is similar to how smoke obscuration effects are determined with Bougher’s law. The size of polystyrene particles used in the experiment (0.126–1.305 µm) was close to the size distribution of soot particles commonly found in fire smoke [
19]. For a polydispersion of known concentrations, they were able to compute the particle sizes with one transmittance measurement, following a revised version of Lambert–Beer law from their previous work [
20].
Cashdollar et al. [
17] proposed to use a white light source and a compact three-wavelength detector assembly to determine the average particle size and mass concentration of smoke. They have used the Mie theory and Dobbins revised Bougher’s transmission law [
18] to calculate sizes and concentrations. The beam separation was obtained with two cube beamsplitter filters, centred at wavelengths 450 nm, 630 nm and 1000 nm (nominal bandwidth 10 nm). The light intensity was measured with silicon photodiodes and the entire assembly was mounted in one small casing (8 cm × 10 cm × 5 cm). The smoke obscuration was measured at an optical path of 0.1 m, within a gas sampling test section connected to a fire-tunnel. For wood fire smoke log-transmission ratios were: ln
τ(1000 nm)/ln
τ(450 nm) = 0.25; ln
τ(1000 nm)/ ln
τ(630 nm) = 0.47 and ln
τ(630 nm)/ln
τ(450 nm) = 0.54. These ratios allow for determination for mean particle sizes. A reasonably good agreement was found between these measurements and two other sizing methods (transmission electron microscopy and ionization-type particulate detectors). Authors emphasized that this device may be used in field experiments.
Uthe [
21] did investigate the applicability of multi-wavelength LIDAR method to remotely determine extinction coefficient measurements and mean particle sizes of aerosols. The experimental study was performed with 14-wavelength transmissometer. Ten different wavelengths between 390 and 1640 nm, as well as 3900 nm, were produced with Zirconium arc-lamp; wavelength 3390 nm with a He–Ne laser and 10,600 nm with a CO
2 laser. The light intensity was measured by photodiodes (Silicone, Germanium, PbSe or HgCdTe) behind narrow-band optical filters. The facility consisted of a long (10 m) aerosol chamber with a cross-section of 0.50 × 0.50 m, to which compressed air with different dusts was supplied. Ratios of extinction between the transmittance measured at 1 045 nm and 514 nm were presented as a function of Saurer mean diameter for submicron particles, allowing for identification of the aerosol type. They have concluded that a single laser system (Nd:YAG 1060 nm) and its first harmonic (530 nm) would be useful to evaluate particles of sizes < 1 µm. For particles with sizes greater than 1 µm, the ratio of extinction coefficients at 1045 and 514 nm converges to value of 1.2 and is no longer sensitive to particle size. A similar conclusion was drawn in [
9] where it was stated that as the particle sizes increases, the integral optical property of the soot cloud approaches that of monodisperse particles and particle sizes are no longer important. This is somewhat similar to the observations of Jin [
15].
Limitations of optical transmission to the determination of the particle sizes were investigated by Swanson et al. [
22]. They have discussed the underlying theory and presented the derivation of the Bougher-Lambert–Beer law, as well as the theory of Mie and the radiative transport equation (RTE). The application of Bougher-Lambert–Beer law was verified in a scattering environment and the Hodkinson's findings related to the restriction of aperture related to scattered light were emphasized. A method of determination of the particle sizes with two measurements of transmittance was validated for monodisperse polystyrene particles with a size of 0–1 µm. Aspey et al. [
23] have further improved this method using High-Level Synthesis (HLS) approach and polychromatic LED. They have investigated the spectral change in transmitted polychromatic light, which was used to determine Mie scattering parameters in a wood smoke in the function of time. It was observed that for the wood smoke forward scattering of blue wavelength is approximately three times that for the green detector and the red detector measured negligible forward scattering.
The use of multi-wavelength light transmission was also proposed by Wilkens and van Hees [
24] to improve the research capabilities of a cone-calorimeter [
25]. This implementation would allow for determination of the mean particle diameter of the smoke produced in the calorimeter, through non-intrusive light transmission measurement. They have reviewed the methods of smoke identification used in the conventional standardized fire testing apparatus and performed theoretical considerations on the limitations of the proposed tool and also in-depth discussion of previous developments. An apparatus similar in the concept was built by Chaudhry and Moinuddin [
26]. A sealed box was used as a combustion chamber to burn different materials and the smoke was further transported into five separate light chambers. These chambers were equipped with transmittance measurements with specific wavelengths 275 nm, 365 nm, 405 nm, 620 nm and 960 nm. Both absorbed (measured at angle 0°) and scattered (at angle 90°) light were recorded. A Random Forest machine learning algorithm was trained based on the experimental results, in order to predict the burning material based on the measurements at these five wavelengths. The overall success rate of the algorithm was 96.6%–97.5% showing the high applicability of this method. For some materials such as cardboard and polystyrene, the success rate was lower, respectively 60.3% and 59%. The difficulties with the use of this approach were associated with obtaining meaningful results for both scattering (at a certain angle) and absorption.
Another recently developed approach was done with tomographic reconstruction of an image of LED, captured with a digital camera through a smoke layer. By comparing the predicted and observed shape and light intensity of the LED, the smoke layer opacity could be reverse engineered. This approach could be used for multi-wavelength observations, as typical LEDs consist of red, green and blue diodes [
27].
Summarizing previous efforts it was found, that the past use of multi-wavelength optical transmittance meters was limited to the determination of the mean diameter of smoke particle size and consequently, the type and source of the smoke. Most of the devices constructed were connected to bench-scale apparatus (boxes, tube burners, smoke tunnels). The Cashdollar’s apparatus [
17] was meant to be used in field experiments, although such use was not reported in the literature. The limitation of the proposed solutions was (1) requirement to know the mass density of smoke or (2) need for high accuracy of the transmittance measurements, which excludes the use of this type of apparatus in field experiments. The direct measurements in compartment scale fire experiments are needed as the optical properties of smoke generated in small and large scale may not necessarily be the same [
28]. It is possible to samples the gas from smoke layers, although as Wilkens [
24] stated, this would be considered an intrusive approach. Furthermore, following findings of Jin [
15] an Aspey [
23], the properties of smoke will change with time and that could distort the size of smoke particles and the light extinction measurements in consequence.
For our future research on the visibility in smoke in compartment fires, validation studies for computer models and smoke obscuration effects on evacuation lighting and signage we are interested in the differences of light transmittance through smoke-layers in full-scale experiments, rather than properties of smoke generated in bench-scale apparatus. To fit our purpose, the tool used to measure the light transmittance would have to (a) be mounted directly within the smoke layer, (b) be able to measure the transient change of the transmittance in real-time and (c) allow for result comparison with the existing body of measurements from EN 54–7 test chamber [
29], such as one used in [
30]. For this purpose, the most promising solution was to construct an optical densitometer based on laser diodes and photo-diodes, that could be mounted within a single casing and share the same optical path. The sources cannot interfere with each other and the interference from the fire source and other light sources (such as emergency lighting) should be limited by aperture, narrow-band filters and internal structure of the receiver. Following these limitations, we have constructed a densitometer that consists of five separate sets of laser—photo-diode couples within one casing, with their light extinction recorded simultaneously over an identical optical path. The sets differ by their specific wavelength and in our case these were: 450 nm (blue), 520 nm (green), 658 nm (red), 830 nm (IR) and 980 nm (IR).