Drone-based applications for tailings dam monitoring

Proactive management is necessary for a tailings storage facility (TSF) to operate successfully. Tailings are heterogeneous and anisotropic by nature, which presents numerous difficulties for their storage, design, operation and closure. The designs of monitoring systems are adapted to the distinctiveness of each TSF and anticipated operating performance, and they range from the most basic to the most sophisticated. Implementing a well-thought-out and thorough monitoring program is necessary to evaluate TSF performance. This program must produce high-quality data, collect and retrieve data quickly and reliably, and use the right methods to send this data to stakeholders for analysis and evaluation (Morrison, 2022). Monitoring instrumentation and supporting systems are subjected to exceedingly demanding conditions that require the cautious design of both the setting and long-term tailings characteristics. Monitoring systems generally fall into two categories of system: (1) terrestrialbased and (2) remote (SME, 2022). The focus of this paper is on the remote type of monitoring systems. Several hundred tailings dams have collapsed since 1910, according to the International Commission of Large Dams and the United Nations Environment Programme. One reason is the slope’s fragility. The instability of a slope might be caused by internal tailings movements (Cheng et al., 2021; Ozcan et al., 2013). External motions of embankment slopes, seismic activity, the weight of tailings and moisture content are further factors. Hence, these are the places that need to be monitored closely. With the help of drones and remote sensing technology, this study attempts to present the necessary research for a dependable and proactive tailings dam instability alert system. There are two types of remote sensing systems: active remote sensing and passive remote sensing (Morrison, 2022). The former assesses the surface feature by using energy, such as laser, radio waves or other techniques, and measuring the reflected condition. Passive remote-sensing techniques, on the other hand, utilize naturally existing energy that is radiated or reflected from the Earth’s surface and can be filtered depending on a distinct spectrum. Satellite-based systems are the most traditional and widespread method of remote sensing. The equipment used by small unmanned aerial vehicles (UAVs) for near-ground surface systems to measure or produce different types of energy for imaging has recently changed. For instance, RGB cameras, LiDAR, thermal imaging and hyperspectral sensors are getting smaller and lighter (SME, 2022). Using UAVs for remote sensing By measuring an area’s reflected energy (multispectral imaging) and emitted radiation (thermal imaging) from a distance, typically from a satellite or aircraft, remote sensing is the process of identifying and keeping track of its physical features (USGS, 2022). Satellites may cover wide areas of land but have certain downsides, such as irregular satellite visit times, limited picture quality, and a time lag between satellite flyover and sensor data reaching the enduser that can take many hours; hence, it cannot be utilized as a quick reaction resource. Moreover, cloud coverage is an essential factor for the availability of satellite data (Allison et al., 2016). Aircraft, on the other hand, can be deployed and are versatile, allowing them to target key areas and adjust as priorities shift. Aircraft frequently undertake lengthy missions and return to the target region, particularly when using unmanned platforms. UAVs, sometimes known as drones, fly at lower altitudes than traditional aircraft, which results in photos with excellent precision, flexibility and low operational costs (Maes and Steppe, 2019). Aerial remote sensing images taken by a UAV have greater resolution than those taken by satellites. Compared to UAV photos, satellite remote sensing images are more complex to comprehend and are greater in size (Lu et al., 2020a). However, the drones’ short range and low endurance present a challenge when using them. The system is not designed to support longer flights that would deliver continuous, uninterrupted data or lengthy mapping excursions. It is essential to mention that wind has a significant role when using drones. Strong winds will drain drone batteries quicker due to more energy exerted to combat winds. Similarly, flying an airplane requires more fuel against strong headwinds (Allison et al., 2016). The DJI Phantom 4 Pro (Fig. 1) is a professional 4k camera drone with a mechanical shutter that lets the user freeze high-speed objects, provides five directions of obstacle sensing and a 30-min maximum flight time (V1.0 user manual Phantom 4 Series, 2016). High-resolution photos can be captured using this drone. The Matrice 600 (Fig. 2), a six-rotor flying platform intended for professional aerial photography, is another illustration of a DJI drone. The Matrice 600’s retractable landing gear provides for clear 360-degree camera views. The aircraft has a maximum gross weight of 15.1 kg, can carry payloads up to 6 kg and employs six smart flight batteries to increase flying time (Matrice 600 user manual, 2017). Figure 1 Phantom 4 Pro (Phantom 4 Pro V2.0 – DJI, n.d.). Figure 2 Matrice 600 pro (Matrice 600 User Manual, 2017). Figure 3 Ebee X (EBee X, 2021). Due to their extended flying endurance and accuracy, fixed-wing drones are particularly used for surveying broad areas. Figure 3 shows an eBee X from senseFly. This drone’s features include a flight time of up to 90 min and exceptional accuracy down to 1.5 cm (EBee X, 2021). Drones’ application in mining During a study on employing UAVs and structure from motion (SfM) to remotely monitor tailing dams, 0.5 km2 of tailings were measured for potential subsidence (Rauhala et al., 2017). The application of SfM allowed for the creation of topographical models of the tailings surface, allowing for the monitoring of surface displacement over time. Tailings exhibit extremely complicated behavior and are easily impacted by process adjustments, sedimentation, consolidation and segregation. Thus, improving tailing monitoring is required. Given the size of tailing dams, contemporary remote sensing techniques can assist traditional geotechnical measurements, improving tailings management and providing in-depth understanding of tailings behavior. Drones are regularly used in the mining industry to keep a close watch on acid mine drainage (AMD). Fourie et al. (2006) used hyperspectral remote sensing (HSRS) to scan vegetation, seepage, and minerals to identify contaminated soils and groundwater in semi-arid settings, and to track the effectiveness of various tree species in reducing AMD. The AISA-ES uses a line of detectors oriented perpendicular to the direction of the spacecraft’s flight since it is a push-broom sensor. There are two scanners in it. Both the visible near-infrared and the short-wave infrared are covered. For vegetation mapping, the 980-nm band was used for mineral mapping: gypsum in a short-wave infrared I (SWIR I) range is related to the evaporation of AMD, while jarosite and copiapite minerals are associated with the weathering of AMD in the visible and near infrared (VNIR) range. Tailings frequently contain kaolinite, which has spectral characteristics in the SWIR II band. The water band at 851 nm was used to map seepage. The spectral mixing issue was resolved using the unmixing method, which produced the distinctive spectral signature per material within each pixel. A UAV was used to monitor AMD-affected areas, concentrating on mapping four prevalent AMD minerals: hematite, jarosite, schwertmannite and goethite (Jackisch et al., 2018). An Aibot hexacopter equipped with a Rikola hyperspectral camera with a wavelength range of 504 to 900 nm was used to collect the hyperspectral data at a flight altitude of 50 m (164 ft). A portable spectral evolution PSR 3500 spectrometer was also used during the four campaigns to examine 160 reference points. A digital elevation model (DEM) was created with orthophotos taken by a fixed-wing Senselfy eBee for topographic image corrections. Spectral Angle Mapper (SAM) was used for the classification, and the closer the two spectra were in radians, the more similar they were, leading to precise wavelengths for each mineral. Multispectral imaging To discover or describe a material’s characteristics, spectroscopy investigates how light interacts with those attributes. As a result, it analyzes the target’s light behavior and identifies materials based on their distinctive spectral signatures. The spectrum of the object can be used to identify these spectral fingerprints. A spectrum shows how much light is generated, scattered or transmitted from the target by describing the amount of light at various wavelengths. Typically, it is displayed as a graph with an intensity and wavelength scale (SpecimSpectral, 2019). Because multispectral images contain a wide range of spectral information, the user can identify specific materials, while multispectral imagery enables the user to collect more precise and detailed data (Shippert, 2003). The ability to categorize similar materials and gather subpixel information is conceded by the multispectral data. Reviewing some fundamental remote sensing ideas is necessary to comprehend the benefits of multispectral imaging. Initially, the energy level of each photon in light determines the wavelength of that photon. The wavelengths of light and many other types of electromagnetic radiation serve as indicators of their properties (Fig. 4). The principle of reflection is another important topic. Reflectance measures how much light a material reflects (Shippert, 2003) in opposition to being absorbed or transferred. The reflectance spectrum (Fig. 5) depicts the reflectance of various materials as measured over a spectrum of wavelengths. A certain wavelength of light will be reflected by some materials while being absorbed by others. Healthy vegetation, for instance, appears green to human eyes because it has a reflectance peak in the visible spectrum’s green region. Water bodies, on the other hand, absorb near-infrared wavelengths, which is why there is no reflectance in this region. Certain materials can be recognized via multispectral imaging by using the reflectance value of the materials. Figure 4 Electromagnetic spectrum (Gade and Moeslund, 2013). Figure 5 Reflectance spectrum (Siegmund and Menz, 2005). Figure 6 Multispectral data cube (Shaw and Manolakis, 2002). Multispectral cameras and technology A spectral imager is a type of hybrid camera that combines a standard camera with a spectrometer, which measures the wavelengths of light. It produces a 2D image, similar to what a digital camera would produce, but with the addition of a third dimension from the spectrometer. As a result, the data collected by a spectral sensor is a 3D data cube, with the spectrum data constituting the third dimension. Figure 6 provides an illustration of a hyperspectral data cube. There are numerous ways to create a hyperspectral image; the most popular techniques and cameras are discussed. Push-broom hyperspectral imaging. The scene is scanned one thin spatial line at a time, and this line is then divided into its different wavelengths. A row of pixels is scanned with a line of spatial information, and each pixel column is distributed with the spectral information for each point on that line. Figure 7 shows a representation of how a push-broom sensor captured the data. HySpex produces high-quality pushbroom hyperspectral cameras. The models are HySpex Classic, Mjolnir and Baldur. Only Mjolnir is able to be mounted on a drone. The cameras’ wavelength ranges vary from the visible light spectrum to 400–1000 nm for Mjolnir V-1240. Mjolnir S-620 covers the SWIR range from 970 to 2500 nm; Mjolnir VS-620 covers the VNIR and SWIR. Bragg tunable filter. Monochrome cameras are equipped with electronically tunable filters that generate an overlay of images at various wavelengths. It is demonstrated in Fig. 8 by showing images at the green, red and blue wavelengths. In order to create hyperspectral data cubes with high spatial and hyperspectral resolution for microscopy studies, the company Photon etc. manufactures hyperspectral imagers based on Bragg adjustable filters. Figure 7 Hyperspectral push-broom (Barnett, 2020). Figure 8 Tunable Bragg filter (Barnett, 2020). Figure 9 Concept of the Spectral Angle Mapper (Shippert, 2003). Spectral unmixing methods We can recognize spectrally related components and extract subpixel data using various hyperspectral image classification approaches. Hence, a more thorough image analysis is carried out. It is vital to note that using commercial software for spectrum analysis is commonplace. In addition, there are libraries of reflectance spectra that were collected from materials with defined compositions. These libraries act as a manual for categorization to specify targets. The most used image analysis and processing program is ENVI, which comes with 27 spectrum libraries. Spectral analysis techniques match the incoming image’s pixel spectrum to a reference spectrum (target). In addition to spectral libraries, this target can also be found in areas of interest or even specific pixels in a spectral image. The two types of classification algorithms are whole-pixel and subpixel techniques. By analyzing the similarity between the pixel and target spectra, whole-pixel approaches attempt to determine if one or more target materials are prevalent within each pixel. SAM, for instance, can be used to calculate the angle between a pixel and a target. For this reason, the lower the angle, the more comparable the materials are. Although the actual spectral angle computation is based on all of the bands in the image, Fig. 9 shows a 2D scenario for better understanding. Another method of matching target and pixel spectra is spectral feature fitting. The leastsquares method is used to compare the depth and shape of each pixel (the target and the feature), allowing the user to specify a variety of wavelengths over which a separate absorption attribute exists for the selected target. In order to determine how many target materials are contained in each pixel of a picture for the second category, subpixel algorithms are used, which can identify targets with amounts much lower than the actual pixel size. Linear spectral unmixing, or complete unmixing, states that each pixel’s reflectance spectrum derives from the linear combining of the spectra of all of its end members (spectrally unique surface materials). Each weight of the same end member in the linear combination corresponds to the area of the pixel containing that end member. The parts of each end member in the pixel are the variables at play in the solution of k (number of bands) separate linear equations for each pixel. This approach produces an abundance of images for each end member, each of which shows what portion of the pixel is made up of that end member. Another subpixel technique is matched filtering, also referred to as partial unmixing because equations are only partially solved as opposed to completely unmixing, which allows selected target spectra to maximize and suppress the response of the rest of the image. Figure 10 A thermal picture taken with MicaSense Altum. Figure 11 RGB image. Thermal imaging Since all matter emits temperaturedependent electromagnetic radiation at temperatures beyond absolute zero, thermal sensing concentrates on detecting emitted energy rather than reflected energy. The distribution maxima fluctuate inversely with temperature, and the radiance is not consistent throughout wavelengths. The peak occurs between 8 and 12 μm at ambient temperatures, and most of the radiant energy is found at wavelengths greater than 5 μm. As a result, the thermal imaging band refers to the wavelength range between 8 and 15 μm. The spectrum from 5 to 8 μm is unsuitable for remote imaging due to solid atmosphere absorption (Allison et al., 2016). Thermal cameras can be grouped into uncooled and cooled types based on the image detector type. Uncooled sensors function at ambient temperatures, while cooled sensors are packaged in a unit that keeps them at a shallow temperature due to cryogenics (Cooled vs. Uncooled Thermal Imagers, 2019). Greater details and the ability to detect little temperature variations are provided by cooled infrared thermal cameras. They use a lot of energy, are large and are expensive. Uncooled thermal cameras are less expensive, lighter and more suited for deployment on UAV-based platforms, but because of the poor contrast of the thermal picture, they significantly affect the creation of orthoimages (Zhou et al., 2021). Surface soil moisture (SSM) detection is one of the frequent applications for thermal sensors. Because moisture directly impacts the strength of the tailings, managing moisture is a key goal of managing these massive impoundments. This explains how moisture relates to tailings dam stability (Zwissler et al., 2017). Moreover, identifying early-stage seepage is crucial for the security of the dam (Mainali et al., 2016). Thermal infrared (TIR) sensors give us the ability to determine the soil surface temperature immediately or after processing, to determine the thermal characteristics of the soil surface. TIR techniques can be split into two categories (Petropoulos et al., 2015). The first is thermal inertia techniques. The rationale behind thermal inertia procedures is that SSM can improve thermal inertia and reduce diurnal temperature variance. As a result, SSM can be calculated by observing the diurnal temperature change. Nevertheless, this method typically requires supplementary information or prior knowledge of the soil characteristics, such as climatic variables or bulk soil density, and estimations are only allowed once per day. Contrarily, empirical techniques came up with relationships between SSM and remotely detected land surface temperature. These techniques frequently yield estimates with great spatial and temporal resolution, and are simple to apply. Although having some of the same limitations as empirical methods, they exhibit excellent efficacy under calibrated conditions (Tang and Esmaeili, 2021). Figure 10 shows a thermal image taken by the research and development team of the mining automation laboratory of the University of Nevada, Reno. The camera used was the Altum model from MicaSense, which allows a multispectral image with five waves: RGB or digital images, Red-edge, near infrared (NIR), long-wave infrared (LWIR) and thermal. Two gallons of water were poured on dry soil in Reno, NV. The red box shows where the water was poured and demonstrates that thermal cameras can easily detect moisture. Figure 12a Spectral filter applied in ImageJ. Figure 12b 3D surface plot. Bare soil moisture estimation In the work of Katarzyna et al. (2021), bare soil moisture content was evaluated using multispectral data from a UAV and hyperspectral data from a field spectrometer. By taking relative electric permeability readings, it verified the distant sensing data. Spectral reflectance analysis from 350 to 2500 nm is used in optical methods to determine the moisture content of bare soil. NIR and SWIR bands constitute the foundation for some applications. Three blue band normalized indices, with wavelengths spanning from 1800 to 2230 nm, produced a greater determination coefficient for this study (Yue et al., 2019). Based on the difference in water absorption between Sentinel-2 map photos and SWIR images, the estimated bare-soil moisture content. It is possible to lessen the impacts of the sun’s angle, the terrain and instrument mistakes by using the normalized transformation form of two or more spectral reflectances. Based on the SWIR1 and SWIR2 wavelengths, they suggest three normalized SWIR difference soil moisture indexes. Increased soil moisture causes a drop in Figure 12a Spectral filter applied in ImageJ. soil spectral reflectance, although the rate and extent of the decrease at various wavelengths varies because water has varied optical wavelength absorption capabilities. Hence, a greater difference in soil reflectance is caused by a greater amount of water in the soil. As described, moisture detection usually needs a specific type of hyperspectral sensor in the range of SWIR since the water absorption bands are more robust in this region: that is, 1400, 1900, 2400 nm. Thermal sensors can also be used for the same task. However, these sensors can sometimes be expensive and overload the UAV. But using a digital camera and processing the RGB images, soil moisture content can be determined. In Lu et al. (2020b), they may gauge the steppe’s soil moisture using UAV data in the visible band. The images’ brightness value was calculated using ImageJ software. The RGB image was first transformed into an 8-bit grayscale with 256 different gray brightness levels, ranging from 0 (the darkest) to 255 (the brightest). The grayscale image was given a spectral filter, which turned it into seven distinct colors: dark to bright red, orange, yellow, green, cyan, blue and purple. Using a 3D surface map to turn brightness into elevation and the idea of dark to brilliant into deep to shallow, a better interpretation was obtained. The findings showed that the brightness of the UAV’s photos might be used to estimate the surface soil wetness. Figure 11 shows the RGB image, Fig. 12a shows the spectral filter, and Fig. 12b is the result obtained by the research and development team of the mining automation laboratory of the University of Nevada, Reno, after using the method described in Lu et al. (2020). Conclusion Remote sensing technologies using drones allow mine sites to obtain accurate results when physical access is limited because of logistics, geometry, or hazardous conditions and concern or sensitivity to human safety. Drone-based monitoring is more effective than manual or satellite-based systems in terms of time and data accuracy. Small unmanned aerial systems are becoming more robust, airworthy, lightweight, maneuverable and versatile. The battery systems that control the vehicle lifting capacity are becoming progressively smaller and lighter, allowing for more extended aerial missions. Combined with high-resolution RGB cameras and thermal and multispectral sensors, it is possible to detect moisture content, one of the reasons for slope instability causing tailings storage facilities to fail. (References are available from the authors.)

Drone-based applications for tailings dam monitoring

Authored by:
Jose A. Gomez, Tebogo Letshwiti, Javad Sattarvand

To view the past issues of the UCA magazine you must be a member of SME/UCA.

Mining Engineering’s Weekly Rundown: Episode 73

Get your SME Foundation VIP Access Conference Pass to MINEXCHANGE 2027

“Check the box” and help support the SME Foundation

Mining Engineering’s Weekly Rundown: Episode 73

Get your SME Foundation VIP Access Conference Pass to MINEXCHANGE 2027

“Check the box” and help support the SME Foundation

To view the past issues of the UCA magazine you must be a member of SME/UCA.

Aquatech awarded contract for lithium refining at Thacker Pass

Keynote features mining experts on environmental, social and governance factors

Related Articles

Get your SME Foundation VIP Access Conference Pass to MINEXCHANGE 2027

“Check the box” and help support the SME Foundation

The Art and Science of the Flowsheet

UCA Division officer nominations

Mining Engineering’s Weekly Rundown: Episode 73

Get your SME Foundation VIP Access Conference Pass to MINEXCHANGE 2027

“Check the box” and help support the SME Foundation

Mining Engineering’s Weekly Rundown: Episode 73

Get your SME Foundation VIP Access Conference Pass to MINEXCHANGE 2027

“Check the box” and help support the SME Foundation

Use Single Sign On