Hyperspectral imaging for asbestos identification in cement products

Hyperspectral imaging and chemometric methods are combined in an innovative approach for fast, in situ, and low-cost asbestos recognition.
08 September 2016
Giuseppe Bonifazi and Silvia Serranti

The unique technical properties of asbestos, especially those related to heat and corrosive chemicals resistance, mean that it was once used in many industrial and civil manufacturing/transformation applications.1 Although the term asbestos refers to two different fibrous mineral families2—serpentines (e.g., chrysotile) and amphiboles (including crocidolite, amosite, anthophyllite, actinolite, and tremolite)—chrysotile makes up more than 95% of the asbestos used in the construction sector (followed by crocidolite).3 Despite the useful properties of asbestos, in the 1980s it was recognized as hazardous to human health, and its use has since been banned in many industrialized countries. Among the asbestos-containing products manufactured in the past, the most potentially dangerous are thought to be those used for building and construction (i.e., where asbestos was used as a cement binder). Unsuitable techniques for the removal of asbestos-containing cement (and other products) during civil engineering activities, however, can cause uncontrolled exposure or liberation of asbestos and increased health risks. It is therefore important to implement fast, low-cost, and reliable strategies for identifying the presence of asbestos inside cement products that have a variety of characteristics.

Asbestos-containing materials are only dangerous when the asbestos is in a ‘friable state’ (e.g., when it has been reduced to a powder). This is because the mechanical manipulation (i.e., milling) of asbestos minerals produces finer, longer, and thinner particles than other minerals that are treated in a similar way.4 The smaller fibril diameter of the asbestos minerals means that they generate stable aerosols with very high inhalation potential. Asbestos particles can therefore easily penetrate the pulmonary architecture. Indeed, it has been shown that the state of the particles (i.e., the morphological and morphometric attributes of the fibril aggregates) affects their biological threat.5

Existing state-of-the-art techniques that are used for asbestos recognition and identification are based on physical sampling of the materials and laboratory analyses. The tests are usually conducted with polarized light microscopy (PLM),6 scanning electron microscopy (SEM),6 or transmission electron microscopy (TEM) to determine the elemental composition of the fibers. In addition, selected area electron diffraction (SAED) can be used to investigate the crystalline structure of the fibers.7 In all these cases, however, expert personnel are required to make the analyses. Moreover, the analyses are time-consuming and there is a lag time between the physical collection of the samples and obtaining the analytical results.

In our work,8 we have therefore been investigating ways to employ innovative sensing techniques for on-site characterization of potential asbestos-containing materials during construction/demolition work. Such techniques could play an important role in preventing uncontrolled diffusion of asbestos into the ambient environment. In particular, with our proposed approach, we combine the detection capabilities of hyperspectral-imaging-based sensing devices with the process capabilities of chemometric methods. We can thus perform direct detection of asbestos within a number of different mineral matrices, without the need for any physical sampling, de-localized laboratory equipment, or specialized expertise.

To demonstrate the effectiveness of our proposed technique, we performed experiments to try and identify the presence of chrysotile and crocidolite within manufactured cement products. For this study we used a SISUChema XL Chemical Imaging Workstation (from Specim, Finland) that we equipped with an ImSpector N25E imaging spectrograph. This spectrography works in the short-wave IR range (1000–2500nm). We also used a micro x-ray fluorescence instrument (BRUKER M4 Tornado) to validate our classification results.8

The measurement procedure that we thus developed consists of two main steps. The first step is the acquisition of spectra from different material classes as reference standards (i.e., chrysotile, crocidolite, and cement matrix), as shown in Figure 1. We collected, processed, and analyzed the spectral features of these standards (see Figure 2) so that we could build a reference spectral library. To eliminate unwanted effects (caused by lighting/background noise), we removed parts of the acquired spectra—see Figure 2(a)—at the high and low ends of the wavelength range. We then applied preprocessing algorithms (standard normal variate and mean centering) to highlight differences between the spectra of the different materials, as illustrated in Figure 2(b). Next, we performed a principal component analysis—see Figure 2(c)—to investigate correlations among the different variables and samples (i.e., clustering), to detect outliers and to quantify patterns. These features embed spectral attributes of the collected data set—see Figure 2(d)—that can then be used for the asbestos classification.

Figure 1. (Top) Red-green-blue (RGB), (middle) hyperspectral, and (bottom) prediction probability images of (a) crocidolite, (b) chrysotile, and (c) cement. These reference images of asbestos samples were used to build the spectral library for asbestos recognition in cement products.

Figure 2. (a) Representative short-wave IR spectra of crocidolite, chrysotile, and a cement matrix/paste. (b) Spectra after preprocessing algorithms (standard normal variate and mean centering) have been applied. These algorithms served to highlight differences between the spectra. (c) Plot of the first principal component (PC 1) versus the second principal component (PC 2) scores for a principal component analysis used to investigate correlations between the spectral variations. (d) Loading of crocidolite, chrysotile, and cement matrix for the reference samples (determined via the principal component analysis), as a function of wavelength.

The second step of our measurement procedure is the classification of actual manufactured cement products that contain crocidolite and chrysotile (see Figure 3). For this step, we follow the same strategies as in the first step (i.e., for the processing of the reference images). The red-green-blue and corresponding hyperspectral images we obtained for our test samples are shown in Figure 3(a) and (b), respectively. We obtained the resulting classification—see Figure 3(c)—through a partial least squares discriminant analysis. In our prediction maps, the pixels are assigned to one of three classes (crocidolite, chrysotile, or cement paste). With this approach we can assess the topology of the asbestos minerals, and we can analyze their morphological and morphometric characteristics (all of which are important in designing safe removal and handling strategies).

Figure 3. Imaging of crocidolite and chrysotile within cement products. The RGB (a) and hyperspectral (b) images are used in a partial least squares discriminant analysis to form the classification maps (c).

In summary, we have developed a novel method for fast, reliable, and low-cost identification/recognition of asbestos within cement products. In our approach, we combine the capabilities of hyperspectral imaging and chemometric methods, and we make use of a spectral reference library. We have also verified our technique by successfully identifying crocidolite and chrysotile components within cement products. Our approach is therefore suitable for rapid on-site material characterization during demolition activities, when the presence of asbestos-containing materials is suspected. In the next stages of our research we will develop a portable device for in situ recognition/identification of asbestos minerals and related manufactured products. This device will feature point-based scanning determination (i.e., asbestos detection) and sample surface mapping (i.e., asbestos identification and surface topological assessment) capabilities.

Giuseppe Bonifazi, Silvia Serranti
Department of Chemical Engineering, Materials, and Environment
Sapienza University of Rome
Rome, Italy

Giuseppe Bonifazi is a full professor of raw materials engineering. His work involves the characterization and processing of primary and secondary raw materials, in which he uses innovative optical sensing techniques (particularly hyperspectral imaging).

Silvia Serranti is an associate professor. Her research activity is focused on primary and secondary raw materials characterization and valorization. She has published more than 160 scientific papers.

1. F. Paglietti, S. Malinconico, B. Contestabile della Staffa, S. Bellagamba, P. De Simone, Classification and management of asbestos-containing waste: European legislation and the Italian experience, Waste Manage. 50, p. 130-150, 2016.
2. I. R. Lewis, N. C. Chaffin, M. E. Gunter, P. R. Griffiths, Vibrational spectroscopic studies of asbestos and comparison of suitability for remote analysis, Spectrochim. Acta Part A: Molec. Biomolec. Spectros. 52, p. 315-328, 1996.
3. C. Bassani, R. M. Cavalli, F. Cavalcante, V. Cuomo, A. Palombo, S. Pascucci, S. Pignatti, Deterioration status of asbestos-cement roofing sheets assessed by analyzing hyperspectral data, Remote Sens. Environ. 109, p. 361-378, 2007.
4. A. M. Langer, Identification and enumeration of asbestos fibers in the mining environment: mission and modification to the Federal Asbestos Standard, Regulat. Toxicol. Pharmacol. 52, p. 207-217, 2008.
5. A. M. Langer, R. P. Nolan, The properties of chrysotile asbestos as determinants of biological activity, Accomplishments in Oncology 2, p. 30-51, Lippincott, 1986.
6. W. C. McCrone, Asbestos Identification , p. 199, McCrone Research Institute, 1987.
7. L. De Stefano, F. De Luca, SEM quantitative determination of asbestos in bulk materials, Microscopy Anal. 16, p. 13-15, 2002.
8. G. Bonifazi, G. Capobianco, S. Serranti, A fast and reliable approach for asbestos recognition in complex matrices adopting a hyperspectral imaging based approach, 2016. Paper accepted at the 5th Int'l Conf. on Industrial and Hazardous Waste Management in Chania, Greece, 27–30 September 2016.
Recent News
Sign in to read the full article
Create a free SPIE account to get access to
premium articles and original research
Forgot your username?