Monitoring and early warning detection of collapse and subsidence sinkholes using an optical fibre seismic sensor

Abstract

We present and experimentally demonstrate a seismic ambient noise monitoring optical fibre sensor for early warning detection of sinkholes. The developed optical fibre sensor is designed for warning alert of subsidence and cover collapse sinkholes. The progressive process of sinkhole development causes structural change in the subterranean surface. The impact of this change and its influence on the subsurface acoustic modes was detected in the form of variations in the spectral content of the ambient noise signals monitored in the subsurface. Structural surface integrity was monitored through frequency response as the void increased. Vibrational states relating to unsteady structural conditions were identified. Significant instability events were captured giving timely warnings before collapse. The polarisation based single mode fibre sensor and monitoring method is proposed for implementation in a phase sensitive distributed acoustic sensor setup. Peak frequencies in the micro-seismic noise band of 0.1 Hz to 1.0 Hz were observed through cavity development and growth. Extended peak frequency shifts and bandwidth in the band >1Hz were recorded, indicating weakness and imminence of collapse. Early warning detection by the structural field model was achieved prior to the sudden subsurface failure which results in collapse sinkholes. By monitoring variations in the vibrating frequency modes when a subsurface cavity develops within the structure, trigger events and collapse precursor conditions are identified. We have successfully demonstrated an early response warning annunciator by using an algorithm to analyse combinational characteristics of the spectral components of the detected signals. The fibre sensor reduces the risk and socio-economic impact of infrastructural damage due to sudden collapse of sinkholes and has extended potential of monitoring earthquakes and landslides.

Keywords:

• Sinkhole monitoring
• optic fibre sensor
• fibre interferometer
• polarisation
• ambient noise

REVIEWING EDITOR:

• Zhongmin Jin
• Xian Jiao Tong University (China) and Leeds University (UK)

SUBJECTS:

• Applied Physics
• Environmental Physics
• Fibre Optics
• Vibration
• Acoustical Engineering
• Georisk & Hazards
• Instrumentation, Measurement & Testing

Jel Codes:

• c51
• c52
• c53
• c55
• c61
• c62
• c63
• c90
• c91

1. Introduction

Geo-hazards can cause catastrophic damage to infrastructure and the environment, and they can be fatal as well. The destruction of infrastructure due to sudden collapse of sinkholes without early warning has very high financial implications. Monitoring and early detection of sinkholes before collapse can either occur or reach extended locations, is critical for civil protection, environmental preservation and reduction of their adverse effects. A sinkhole forms when the ground material sinks downward into a void or the surface above collapses into a cavity formed by erosion or gradual removal of soluble bedrock (Dobecki & Upchurch, 2006; Ferentinou, 2020; Intrieri et al., 2015; Scotto di Santolo et al., 2018). Water has proved to be the most common agent and main factor in both natural and human induced sinkhole formation (Dobecki & Upchurch, 2006; Ferentinou, 2020; Heath & Oosthuizen, 2008; Intrieri et al., 2015, 2018; Kaufmann et al., 2018; Scotto di Santolo et al., 2018). Another important factor in sinkhole formation is the geology of the subsurface and specifically the rock type (Constantinou & Van Rooy, 2018; Ruth & Degner, 1984).

Although sinkholes are a localized phenomenon, they have adverse effects, causing costly property damage and loss of life in some instances (Heath & Oosthuizen, 2008; Oosthuizen & Richardson, 2011; Kaufmann et al., 2018). In South Africa, 2500 sinkhole and subsidence events had been recorded by 2011 and approximately 98% of them occurred in the Gauteng province (Heath & Oosthuizen, 2008). More events have been reported since then. Within the same year, the estimated cost of damage caused by sinkholes was in excess of USD 86.7 million while in a dolomite area west of Johannesburg, costs exceeding USD 600 million were projected to be incurred in relocating approximately 30 000 households to safer grounds (Buttrick et al., 2011; Ferentinou, 2020; Heath & Oosthuizen, 2008; Oosthuizen & Richardson, 2011). A total of USD 15.7 million was spent in the 2017/2018 and 2018/2019 financial years on preventive and remedial dolomite capital and maintenance projects within the Gauteng province according to Businesstech (2018). In Figure 1, subsidence and sinkholes developed in different ways are shown for dolomitic and non-dolomitic areas in South Africa.

Figure 1. Dolomite land in South Africa. >95% of recorded sinkholes are found in the Gauteng Province (within which Johannesburg and Pretoria are located). Inserts: Infrastructural damage due to sinkholes. TOP-Subsidence on a Highway in Johannesburg. MIDDLE-Collapse sinkhole in Durban due to subsurface erosion from a leaking storm water drainpipe (swallowed the road and part of a nearby house). BOTTOM-Sinkhole on road in Durban due to broken sewer pipe. (Map courtesy of CGS, and images taken by V. Nhlapho Times Live 2021, e-NCA 2021, L. Walford 2019).

Sinkholes have also caused extensive damages globally. In 1995, a 35-m-wide sinkhole in central Italy resulted in damage costs totalling USD 3.3 million (Buchignani et al., 2008; Intrieri et al., 2015), while in Florida, insurance claims totaling USD 1.4 billion were received inclusively between 2006 and 2010 (Intrieri et al., 2015; Kaufmann et al., 2018). Sinkhole prevalence in the past decade has by far superseded the figures from the preceding decades, due to the expansive growth of businesses and urbanization. The increased risk and cost have been a result of increased triggers from anthropogenic activities linked to residential, industrial and commercial development, as more infrastructure is being constructed on top of the active karstic landscape underlain by dolomite (Oosthuizen & Van Rooy, 2015). The catch in these areas is that they make up the best aquifers serving as major water supply sources also supporting boreholes and springs which yield a lot of good-quality groundwater (Businesstech, 2018). Drinking water is sourced from the dolomite aquifers to an amount of 40 million litres per day by the municipality of Tshwane in the Gauteng province of South Africa (Ferentinou, 2020).

Though sinkhole prone areas are mapped and known, the detection and assessment of a potential sinkholes by geologists or soil engineers depends on noticeable signs being reported. Monitoring and early collapse detection systems are essential to give warning alerts, especially in currently developed areas underlain by dolomite. Figure 1 shows areas of dolomitic land in South Africa (Heath & Oosthuizen, 2008). These zones are more susceptible to sinkholes and in most cases, human activities are the major triggering factors accelerating their formation. Recent image reports of infrastructural damages in Johannesburg and Durban are shown in the inserts in Figure 1. While cover subsidence in the top insert allows for remediation and avoidance of accidents, the cover collapse sinkholes in the other inserts rarely gives visible warning signs (Gutiérrez et al., 2019; Heath & Oosthuizen, 2008; Theron & Engelbrecht, 2018). Cover collapse sinkholes are the most dangerous compared to cover subsidence and solution type sinkholes. In naming and classifying sinkholes, the surface material and the nature of the formation processes involved are considered (Gutiérrez et al., 2019; Scotto di Santolo et al., 2018). For the purpose of this work, bedrock or caprock collapse and solution type sinkholes were not considered. In this study, surface depressions due to suffusion or sagging will be referred to as cover subsidence while cover collapse is implied for abrupt open surface expressions due to sudden subsurface failure (Ferentinou, 2020). The type and formation of sinkholes depends on the geology of the underlying surface (Constantinou & Van Rooy, 2018; Ruth & Degner, 1984). Triggering mechanisms responsible for the processes of cavity formation and growth are therefore determined by the nature and composition of the subsurface (Intrieri et al., 2015, 2018). Potential sinkholes are sometimes identified by warning signs of cracks on foundations and walls or outlining circular patterns on the ground, tilted trees and poles, gradual localized ground settlement and vegetation stress (Kukreja, n.d; Theron & Engelbrecht, 2018). These signs are usually confirmed with follow-up geological techniques like ground penetrating radar (GPR) and electric resistivity tomography (ERT) for detecting and mapping out features below the surface (Dobecki & Upchurch, 2006; Kaufmann et al., 2018; Oosthuizen & Van Rooy, 2015; Ruth & Degner, 1984). Interferometric Synthetic Aperture Radar (InSAR) is also used to identify, monitor, and locate potential sinkholes by scanning for depressions and ground subsidence (Ferentinou, 2020; Gutiérrez et al., 2019; Intrieri et al., 2015; Theron & Engelbrecht, 2018). The development leading to the formation of sinkholes is a typically slow process spanning years, even extending over decades (Heath & Oosthuizen, 2008). Cover collapse sinkholes can develop with little or no noticeable change on the surfaces upon which they suddenly occur with no warning indication of imminent danger (Gutiérrez et al., 2019; Heath & Oosthuizen, 2008; Theron & Engelbrecht, 2018). Such dramatic events are undesirable in urban development settings (Heath & Oosthuizen & Van Rooy, 2015). The optical fibre sensor devised in this study is an effort to complement existing methods, to curb this challenge and optimize detection to avoid missing subtle, invisible, and silent precursor events occurring before catastrophic collapse. We present an early warning annunciator system to allow preventive reinforcements or evacuation procedures thereby minimizing damage, cost, and loss. Such complementary detection and monitoring techniques and devices are critical for geologists, engineers and civil authorities in decision making, protection of communities and saving of properties and infrastructure.

The use of optical fibres as sensors has become commonplace over the past two decades (Hisham, 2018; Kapogianni et al., 2020; Matias et al., 2017; Sabri et al., 2015; Wei & Tjin, 2020; Yu & Wei, 2002). Due to their sensing mechanism and structural flexibility, optical fibre sensors have found significant applications across engineering and physical, life and earth science fields (Hisham, 2018; Matias et al., 2017; Sabri et al., 2015). They have several advantages compared to most other techniques used in sinkhole detection and monitoring. A most significant edge being that they can be distributed into series of concatenated virtual sensors along their entire length, and they have several degrees of freedom tailorable to sense different physical parameters. As a result, the fibre sensing technology has advanced immensely as it depends highly on developments in the telecommunication industry. Because of the different optical properties of light which can be manipulated, a wide variety of optical fibre sensors have been proposed, leading to a rapidly growing interest from different fields (Ferentinou, 2020; Hisham, 2018; Kapogianni et al., 2020; Matias et al., 2017; Sabri et al., 2015; Sladen et al., 2019; Yu & Wei, 2002; Wei & Tjin, 2020; Zhan et al., 2021,). In geological hazards detection and monitoring, optical fibre qualities that include environmental stability, durability, interference immunity and multipronged sensitivity have attributed to their significance in the area. Harnessing from the different optical signal properties and the optical fibre qualities, different designs and numerous techniques have led to repurposed wider applications, benefiting from the potential of optic fibres to be concurrently used for sensing alongside data transmission on the communication network (Sladen et al., 2019). Existing long-haul communication subsea optical fibre technologies have successfully detected oceanic seismicity and research is ongoing to fully achieve the goal of Scientific Monitoring And Reliable Telecommunications ‘SMART’ cables in these networks (Sladen et al., 2019; Zhan et al., 2021). Dark fibres which are already strategically deployed can also be utilized to realize further value in both unlit and unused decommissioned systems.

Optical fibre micro-seismic sensing complements existing sinkhole monitoring tools and it advances a step further by combining continuous monitoring and real time processing with collapse early warning alerts. With focus on addressing the challenge of untimely ground collapse which is characterized by subtle, invisible, and silent precursor events, the optical fibre sensor detects structural subsurface changes using relative variation in ambient noise vibrations before catastrophic collapse. A spectroscopic analysis algorithm achieves fast identification of spectral signatures typical of structural subsurface failure. It enables the early warning annunciator thereby allowing for remediation activities or implementation of evacuation procedures to prevent or minimize damage. This paper describes the application of an optical fibre-based polarization sensor in monitoring potential sinkhole development and for early warning detection of collapse. We outline below the unique capability and vital field characteristics of the optical fibre sensor, which existing technologies do not possess. A detailed description of the monitoring tests and miniature field detection experiments which were carried out to characterize the sensor signatures is accompanied by performance analysis of the fibre sensor.

1.1. Sinkhole formation

Sinkhole formation involves natural, and man induced processes which include erosion or gradual removal of slightly soluble bedrock (such as limestone, carbonate rock, salt beds, domes, gypsum etc.) by percolating water, the collapsing of a cave roof, and lowering of the water table (Ferentinou et al., 2020; Oosthuizen, 2008). Persistent dissolving of the rock salts causes enlargement of pores and cracks. These in turn carry even more acidic water creating huge cavities. Leaking storm water drains, burst pipes and extensive dewatering activities may leave voids by washing away weak unconsolidated material like crumbly volcanic rock, fine ash, sand, and other debris with grains small enough to be displaced. The growth of the created empty space is naturally a slow continual process spanning many years of dissolution and erosion. Anthropogenic activities cause subsurface changes which accelerate the processes involved in sinkhole generation (Dobecki & Upchurch, 2006; Intrieri et al., 2018; Scotto di Santolo et al., 2018).

Trigger factors like water ingress, dewatering, and seismic activity can induce sinkholes by either causing loss of buoyancy or reducing soil cohesion and pore pressure thereby breaching ground surface strength (Ferentinou et al., 2020; Gutiérrez et al., 2019; Heath & Oosthuizen, 2008; Ruth & Degner, 1984; Scotto di Santolo et al., 2018). The formation of sinkholes can be gradual or sudden without warning. Generally, the underlying event leading to sinkhole formation in karst areas, is the development and evolution of caverns beneath the surface which is a long dynamic process spanning years (Buttrick et al., 2011). For the dramatic cover collapse case where warning signs do not appear on the surface precursory to the sudden catastrophe event, monitoring the subtle, foregoing dynamic process can present useful information regarding preceding conditions pointing towards an impending collapse. As a result, safety and convenience can be achieved by accommodating timely decision making through warning alerts. Unlike the case with visual warning signs of subsiding surfaces, cracking walls, inclining poles and trees, and wilting vegetation, an alert system informs of concealed events and underlying danger. Though deep surface detailed surveys may be required after detecting potential sinkhole activity, further post processing and assessment will be unnecessary to predict collapse with the optical fibre sensor implementing real time detection.

1.2. Pre-collapse detection challenges

When the soil cover is more compact than sand, the sub surface structure may hold in position with no visible depression or subsidence until sudden failure and collapse. Warning signs or indicators pointing to underground activity may not be recognized as most monitoring and detection methods are driven by observable trigger events or ground displacement. As a result, a few monitoring instruments have been effective in cover collapse sinkhole detection, whereby subsidence or early warning signs are not significant enough to enable early alerts. Monitoring methods are classified according to the location of the measured disturbance and positioning of acquisition sensors for ground deformation data. The groups include subsurface methods, remote sensing methods and ground-based methods.

Subsurface methods can detect underground deformation for which there is no surface manifestation. These include micro-seismic monitoring using mini arrays of seismometers with geophones to detect weak micro-earthquake signals which are difficult to discriminate from noise leading to false positives. Their coverage is limited to monitoring spatially restricted sinkhole prone sites within 1 square km (Gutiérrez et al., 2019). Time domain reflectometry (TDR) devices are also subsurface monitoring tools therefore the Optical TDR (OTDR) falls in this category. When potential sinkholes are identified and detailed information is required, GPR can be used to enable measurement of depth and size of underlying cavities. Ground surface deformation data is collected from the air or space for remote sensing methods. The methods enable investigation over large areas, but they are limited by lack of displacement data as a result of vegetated surfaces which result in loss of signal coherence (Gutiérrez et al., 2019; Intrieri et al., 2015). Their other limitations are in detecting small sinkholes as sizes below 1.5 m are quite easily missed as with GB-InSAR and they have low temporal resolution which makes it less ideal for real time monitoring and early warning alert for the dramatic cover collapse sinkholes (Intrieri et al., 2015). Ground based methods in which the equipment is placed on the ground surface for data acquisition offer high spatial and temporal resolution as well as the possibility of real-time and warning alert. This is applicable to Differential Global Positioning System (Gutiérrez et al., 2019).

Some of the above sinkhole collapse detection challenges can be mitigated using of optical fibre sensor technology. The limit of wider spatial coverage can be addressed by using networking and multiplexing techniques to a distributed sensor setup. Effects of temperature, moisture changes, wind and vibration being the basis of ambient noise, are the driving factor for the proposed sensing technique. These environmental effects can be harnessed to monitor the structural strength and integrity of the subsurface through relative frequency response and spectral intensity variations. A unique type of fibre sensor with special reflectors known as fibre Bragg gratings (FBG) has successfully been demonstrated to be able to monitor sinkholes from strain changes induced by mass movement using a small-scale physical model (Ferentinou, 2020; Labuschagne et al., 2020). The sinkhole activity was detected by using correlated strain changes to monitor movement of soil. In FBGs different strains result in a change in the grating periodicity which reflect different corresponding wavelengths in the system. Simulations with numerical models were performed, in which the use of Brillouin OTDR (BOTDR) for sinkhole monitoring and collapse precursor alert was reported by two groups (Linker & Klar, 2015; Zhende et al., 2013). The first group used strain measurement to monitor sinkhole related movement and the latter group used wavelet decomposition to identify collapse precursor signals from other signals like temperature. A different approach was reported in which a small-scale physical model was used with a distributed sensing arrangement to detect sinkholes by monitoring strain (Cambridge CSIC, n.d). The high strain sensitivity in these sensors makes it possible to detect very small subsurface displacements and enable monitoring of collapse sinkholes. Optical fibre sensors have a potential solution to the challenge of early detection and alert of cover collapse sinkholes.

1.3. Phase and polarisation-based optical fibre detection

Optical fibre sensors are a fast-growing technology extending to numerous indispensable applications. In sinkhole monitoring, the field application challenge for optical fibre sensors is the quality of the contact at the sensor-ground interface and the signal coupling. However, as exhibited in the following experiments, the sensitivity of optical fibre sensors overcomes this shortcoming by using highly sensitive phase and polarization detection schemes (Jena et al., 2020). In addition, an interferometer can exceptionally enhance the dynamic range of the fibre sensor allowing for detection of very weak signals.

As a result, fibre sensors have the potential to operate effectively in different soil types. The proposed system developed in this study uses standard single mode fibre common to telecommunication networks and harnesses ambient noise to detect dominant vibrational frequencies and the associated intensity peaks, and spectral shifts within the subsurface.

1.4. Vibration monitoring techniques

Several techniques are used in structural health monitoring. These include acoustic emission, vibration analysis, strain sensing, ultrasound and thermography. Applicable for the use of unmodified, standard telecommunication fibre for monitoring sinkhole, are the acoustic emission and vibration analysis methods. The structural integrity of the surface is monitored through the spectral distribution of the vibrational response signatures detected within the structure. By analysing the amplitude, frequency and time history of the signals, the condition preceding collapse can be identified enabling early warning applications. Fluctuations due to these external modifications lead to reversible variations which can be distinguished from the irreversible changes in resonance frequency and seismic velocity. These changes include a drop in the magnitude of the resonance frequency which is an indication of incipient collapse (Colombero et al., 2021). The vibration analysis approach is a monitoring technique applicable to large structures for localised damage such as crack and delamination (Cawley, 2018; Shiwa & Kishi, 2005; Sarasini & Santulli, 2014). It is commonly used for structural health monitoring of rotating and moving parts, towers and foundations of mega structures (Guo & Infield, 2012; Lian et al., 2019; Tchakoua et al., 2014) and has therefore been adopted for subsurface monitoring. Use of fibre sensors to achieve vibration monitoring by Oh et al. (2015; Wang et al., 2017), shows that monitoring techniques, processing and analysis are unique for each application. Schemes including the frequency domain decomposition

(FDD) technique and the transfer matrix method have been used in vibration data analysis to identify model parameters of different structures (Kim et al., 2017; Wang & Wang, 2011). The designed approach is developed for sinkhole monitoring to ensure early identification of failure indicators before giving a warning alarm. The fibre sensor has potential for effective use in a variety of geographical locations characterised by different soil composition, texture, compactness and dryness. The effect of other factors may manifest through ambient noise variations which will be needed to drive signals through the structure with various frequency bands for spectral response analysis and structural health monitoring.

2. Materials and methods

2.1. Sinkhole monitoring and detection using optical fibre acoustic sensor

The optical fibre sensor was setup as shown by the schematic in Figure 2. A Michelson interferometer was used to improve the sensitivity while field induced change in the state of polarization (SOP) was monitored with a polarimeter. A detailed description of the development and characterization of the sensor is given in Jena et al. (2020).

Figure 2. Polarisation-based optical fibre acoustic sensor with an interferometric arrangement.

A continuous wave signal of linearly polarised light from the (Distributed Feedback) DFB laser is directed to the beam splitter through the directional coupler. The split signals are reflected from the end of the reference and sensing arms. The returning beams recombine and interfere giving a signal whose state of polarisation (SOP) is detected at the polarimeter. Without disturbance the SOP remains the same such that the associated Stokes vector components give horizontal traces. When the sensing fibre arm is perturbed, by a vibration for instance, the SOP changes causing corresponding shifts to the traces of the Stokes parameters. The signature of the traces matches the nature of the vibration in amplitude and frequency. Data is processed and analysed by the PC.

2.2. Sinkhole formation test model setup

The illustration in Figure 3 depicts the small-scale physical model used to simulate the sinkhole formation process and to experimentally observe the signals relating to the stages involved. A glass tank of volume 0.015 m³ (15 L) was constructed with a 15 mm pipe connecting through its bottom quarter. A deliberate break was introduced in the middle section across the pipe. Water was made to flow-in through one end after fine sand of 0.40 mm maximum grain size was added into the tank as shown in Figure 3. The flow of water creates a void as soil is washed away though the outlet pipe. Although the conditions and trigger factors in the field setting are seasonal and varying, a constant flow of water through the small-scale physical model was maintained in this simulation experiment. The analogue small-scale model was used in the experimental design and the analysis of the system is based on the similarity of process concept of Hooke’s approach (Green, 2014). By this approach, the dimensions, bulk material properties, composition and quantities used, are not moderated to scale. The quantitative data needed to establish collapse precursors is dependent on the properties of the compound structure which are determined by the characteristic soil properties. The desired metric qualitatively based on the integral structural defines the critical condition for early warning alert before dramatic collapse. Variations in the soil properties, constituent components, and quantity are manifest in the behaviour of the structure. The model is therefore, not limited in representation, to sinkholes of specific geology or formation but applies to different cover type sinkholes where a similar process occurs.

Figure 3. Small-scale field testbed for sinkhole monitoring with optical fibre acoustic sensor.

The fibre sensor in Figure 3 was used to monitor acoustic vibration signals before water was introduced into the system, during development and expansion of the void, and after the structural surface collapsed. Weakening of the subsurface layers due to loss of material underneath, results in ultimate collapse. This can either be gradual or sudden depending on dominant trigger factors, soil type and strength. As the cavity expands, the fibre sensor detects variations in the acoustic modes within the subsurface structure. The variations in the detected spectral components are analysed and monitored. Resonance, spectral shifts and temporal signatures relating to the presence of fractures in the structure are identified as precursor metrics for early warning detection. The detected spectral variations can be due to cavity growth and structural instability or changing ambient noise in the surroundings. To ensure structural mode related variations are monitored, controlled conditions were used, with the assumption that sources of ambient noise are constant with uniform signals in the system. Back reflections of the optical signal from the sensing fibre were monitored using a polarimeter. Data acquisition, processing and graphical analysis was done using the polarization shift signals as they correlate to the structural vibrations of the surface. Detected vibration signals were traced by the SOP variations and time series plots displayed alongside their corresponding spectral distributions.

3. Results and discussion

3.1. Strain and spectral monitoring

The sinkhole formation process was monitored using a three-pronged approach, namely visual, temporal and spectral. Video recording enabled visual image analysis. Observed events were correlated with time series signals to establish corresponding effects on amplitude, pitch and the trace signal datum level. Strain and soil mass movement were observed to cause a shift on the level of the trace signal datum line. Spectral distribution analysis facilitated monitoring of cavity growth, the state of the subsurface, failure indicator parameters and collapse events. The effectiveness of monitoring through visual inspection of indicative surface features is limited when the warning signs pointing towards sinkhole formation are insignificant to notice. Such warning signs can be detected via InSAR, allowing for necessary action to be taken. When there is no surface deformation similar tools are likely to fail. The subsurface monitoring optic fibre acoustic sensor uniquely addresses this challenge using vibration analysis for early warning alarm.

Using the optical fibre sensor, the timing of drastic collapse is achieved by monitoring variations in vibrating frequency and strain within the subsurface. Figures 4 and 5 plots show detected signals in the time and frequency domains. Figure 4(a) shows that peak frequencies from 0.1 Hz to 0.5 Hz were dominant vibration modes denoting the range of the fundamental and natural frequencies for the system setup. The ambient noise present in the experiment was observed in the respective time trace and constituted frequencies as high as 1.7 Hz. In Figure 4(b), water flow was introduced and a void began to form. Changes were seen in the temporal and spectral signatures. As the created empty space enlarged, more oscillations appeared on the time trace signals as shown in Figures 4(c) and 5(a). The peak frequency shifted to higher values like 3.3 Hz in Figure 5(c), while the bandwidth as well increased to spans reaching 3.8 Hz. The corresponding intensity peaks were also elevated. The collapse precursor conditions were evident at scalable period before the catastrophic collapse illustrated in Figure 5(b). These are characterised by twin high frequency peaks with high intensity and a large bandwidth. The Stokes parameters S1 (blue), S2 (green) and S3 (red) are the tri-axial polarization components which together define the state of polarization (SOP) of the optical signal. They are normalized to the total optical intensity, and they represent points within a Poincare sphere which is 3D space of unit radius. When the SOP changes due to external field variations the components fluctuate about the datum level determined by the average SOP of the signal (referred herein as the signal dc. level). When the average SOP changes the whole trace signal shifts as shown in Figure 5(b).

Figure 4. Optical fibre sensor temporal and spectral monitoring signals from sinkhole trigger events during cavity growth (a) steady state ambient noise signals (b) Void establishment (c) Cavity enlargement. Visual monitor [centre]: Time traces of change of State of Polarisation components in mW-S1 (blue), S2 (green), S3 (red) are the normalized Stokes Parameters.

Figure 5. Optical fibre sensor temporal and spectral signals from monitoring instability events (a) Collapse precursor, (b) Collapse, and (c) Post collapse. Visual monitor [centre]: Time traces of change of State of Polarisation components in mW-S1 (blue), S2 (green), S3 (red) are the normalized Stokes Parameters.

As the spectral pulse shifted to greater frequencies the strongest component peak was in the trailing edge. Meanwhile bandwidth increased as the cavity expanded. Prior to collapse the pulse propagates back towards zero, resembling a reflection as the higher peak trails in the lagging edge. This portrayed behaviour is typical of spectral energy emanating from very low frequencies near zero. The energy pulse then broadens as it slowly shifts towards a maximum point where it is back reflected to its initial state. Using ambient seismic noise spectral analysis to monitor a quartzite tower, two distinct frequency bands at 6 Hz and 8 Hz had respective vibration orientations that were controlled by two fracture sets which delimited the unstable volume (Colombero et al., 2021). The two peaks in Figure 5(a) are indication of fracture sets appearing ahead of structural surface failure. A decrease in the bandwidth and frequency is experienced before the system structure resonates and collapses. These are the critical warning metrics of the system in addition to auxiliary stress indicator parameters of surface movement observed as datum level shifts of the average SOP in the time series signals.

When vertical stress changes due to subsidence or loss of soil material by swallowing into the cavity a gradient is observed on the time signal. The slope steepness measures the rate of change of the quantity of soil in the subsurface indicating how fast the subsidence or sinking event occurs. The magnitude of the net shift in the dc-level of the time signal relates to the total volume of soil moved, whose mass is responsible for the strain on the fibre. Strain due to mass movement is clearly demonstrated by the temporal trace signal of Figure 5(b,c), whereby the structure was collapsing, and sediment slumping followed. Any strain in the system alters the dc-level of the signal average in addition to the rapid signal changes due to vibrations and noise influence on the optical polarization parameters of the sensor. As the void increases with more material wasting away, the structure attempts rearrangement of its soil particles to fill in the created spaces and support its own weight. In the process, the oscillations on the temporal trace signal are swamped and distorted. The respective spectrograms of Figure 5(b,c) indicate that the vibrations are dominated by the soil particle movement and resettlement.

The capillary action of the water through the soil fills up the pore spaces with moisture and the density of the medium changes. As finer particles soak and slip away the tensile strength and vertical stress of the soil are reduced. Applicable particularly to profiles that contain weathered altered dolomite (wad) which is rich in manganese oxide and has minimum strength when saturated (Montjane et al., 2020). Because the wad exhibits a sudden, dramatic drop in density from the bedrock it is easily mobilized leading to progressive flushing of substantial amounts of overlying material into voids creating sinkholes (Ali & Choi, 2019; Avutia and Kalumba, 2014). It is interesting to note that, likening to variations in the hydrogeological conditions, the water level in Figure 5(a) rises above its normal level determined by the pipe height and relative inlet-outlet flow rate. At this point the outlet had temporarily clogged and as the pipe unblocked the pressured flow agitated the water further while the level remained constant triggering further instability within the structure. The effect of changing underground water levels as an event trigger accelerating the formation of a sinkhole was demonstrated.

3.2. Performance analysis: event triggers and annunciator alarm

The dominant vibrational frequencies from the events in Figures 4 and 5 were recorded. They ranged from 0.1 Hz-3.3 Hz while the bandwidth varied between 0.1 Hz and 3.8 Hz. In addition, change in bandwidth and peak intensity values were measured and the readings plotted in Figure 6 and Figure 7, respectively, for each 10s log interval during sinkhole development. The alarm decision threshold was conditioned to trigger warning annunciators through an algorithm. Based on resonance, a combination of spectral parameters was used to identify instabilities within the structure. These are characterised by bandwidth and peak frequency above the steady state peak value of 0.5 Hz (Figure 4(a)). A corresponding increase in these quantities and in the magnitude of the bandwidth shift accompanied by a decreasing intensity from a value above the calibration threshold, together satisfy the instability condition. In addition, the presence of distinct peaks across the spectral profile is the ultimate indicator for reaching the irreversible state precursor to collapse. In the algorithm, warning alarms were categorised into 4 different alert levels for appropriate action. These are as follows; 1-alert-event detected, 2-high alert (multiple frequent alerts)-structure unstable, 3-warning-fault detected(evacuate), 4-dangerfailure/collapse (barricade). In Figure 6(a,b), when the structure was unstable, the warning alert indicators represented by shaded regions were obtained for the initial development of the cavity through stages of its growth and evolution respectively. Logs 1 and 6 on Figure 6(a) relatively correspond to monitor data of Figure 4(a,b).

Figure 6. Peak frequency and variation of spectral characteristics with time from optical fibre sensor response (a) initial cavity development and growth and (b) larger cavity and reduced growth rate.

Figure 7. Variations in spectral intensity, peak frequency and bandwidth with time, detected with optical fibre sensor (a) stable cavity and (b) collapse precursor resonance and sinkhole formation.

According to Figures 6(a,b) and 7(a), the frequency of warning alerts is greater during the void formation stage and decreases as the cavity growth slows. In Figure 6, the highest spectral peak intensities are associated with the 0.3 Hz frequency mode. Though the maximum bandwidth is close to 1.8 Hz the maximum peak frequency is lower than 0.8 Hz. The average intensity is above the calibration intensity threshold level determined by the steady state conditions.

From Figure 7(a), it appears that the structure regains stability as the cavity growth becomes less significant. The spectral peak intensity fluctuates more frequently about the threshold and a few additional vibrating frequency components become dominant. The average intensity is closer to the calibration threshold, which is now elevated, while greater frequencies as high as 1.5 Hz predominate the structure. As expected, if triggering factors like seismic and human activity are not present to destabilize the structure, the least alert alarms or none can be observed at this stage. Hydrological changes accelerate the collapse of subsurface layers into the developed cavity by reducing soil cohesion (Ferentinou et al., 2020; Ruth & Degner, 1984). A similar factor catalysed the collapse of the miniature structure in Figure 5(a,b) represented by logs 13 and 16 in Figure 7(b).

As the average spectral peak intensity begins to increase again Figure 7(b) shows a correlated instability in the structural surface after log 6. Highest spectral peaks appear at 0.2 Hz. More frequent alerts occur at the verge of structural failure and collapse. Peak frequency and bandwidth simultaneously increase to 3.3 Hz and 3.5 Hz respectively. This unique setting is the critical condition warning of imminent collapse where immediate and emergency response is called to action. The final alert just before collapse shows a unique behaviour demarcating the conditions which precede resonance and the associated structural surface failure. In Figure 7(b), a resonance frequency of 0.2 Hz shows the greatest maximum peak intensity while the bandwidth increases to its greatest span of 3.8 Hz. After the significant increase of both frequency and bandwidth they together decrease in sync, but bandwidth increases again as more frequency components get stronger in the structure as it weakens and rattles. During collapse, both intensity and bandwidth suddenly drop to the lowest magnitudes momentarily before shifting back towards their steady state averages. Frequency and bandwidth remain low after the sinkhole is formed as sediments resettle. Generally, when the cavity gets bigger, the spectral components broaden with a higher peak frequency and greater peak intensities. The unique precollapse condition from log 11 and log 12 is critical for warning alert as shown in Figure 7(b). In the spectral domain a rapid and synchronous change to values beyond the normal peak levels occurs, for the highest peak frequency, the bandwidth and the magnitude of the bandwidth shift.

4. Conclusion

In conclusion, we have successfully developed an optical fibre sensor for detection of sinkholes from ambient noise signals through monitoring subsurface response to vibrations. This enables early warning alerts and facilitates timely response to protect lives and save infrastructure from the usually unforeseen catastrophic damage by unanticipated sinkholes through sudden cover collapse. Subsurface structural health and integrity monitoring is essential where damage occurs without noticeable signs. The designed system is however, also applicable to subsidence sinkholes. Micro-seismic and ambient noise monitoring techniques might not identify the precise location, depth and development features of sinkholes but can still be used to detect growth and predict collapse. Because of the persistent existence of ambient noise signals, the continuous operation of the sensor in subsurface monitoring and sinkhole detection is enabled. The signal parameters may differ and change in relation to several factors linked to varying environmental sources. Consequently, any factor which may triggers and accelerate collapse are important to detect. A trigger event was identified which showed the impact of changes in the subterranean water level. Trigger factors from earthquake or landslide also induce events that accelerate the formation of sinkholes (Strakowski, 2018). The ability to monitor passive seismicity demonstrated by the optic fibre sensor means such events can be detected if present. The sensor can therefore find further use in monitoring earthquakes or landslides with separate dedicated setups or while monitoring the progression of sinkholes to give early warning alerts of critical failure conditions before collapse.

Acknowledgements

We are grateful for the support from African Laser Centre (ALC), Telkom, Dartcom, Ingoma, CISCO, National Laser Centre (NLC) South Africa, Department of Science and Technology (DST) South Africa, Council for Scientific and Industrial Research South Africa (CSIR), Technology and Human Resources for Industry Programme (THRIP) South Africa, and Square Kilometre Array (SKA).

Disclosure statement

The authors report there are no competing interests to declare. The funders had no role in the design of the study, in the collection, analysis, or interpretation of data, in the writing of the manuscript, or in the decision to publish the results.

Additional information

Funding

This research was funded by the National Research Foundation (NRFTWAS) South Africa, grant number 116090.