3.1. Findings Related to the Performance of DOS for Rock Bolt Monitoring
Part of the research effort was to further explore the capabilities and limitations of the use of the DOS as a tool for both in situ and laboratory investigations. As explored above, through tensile testing, the performance of the fiber for the application of capturing a full spectrum of loading was investigated, and results of the activation length was incorporated into the monitoring plan. This monitoring plan had a success rate of 75% for specimens under excess plastic loading (>160 kN) approaching ultimate state, and 43% survived shank failure. This allowed for observation of strain attenuations and insights into the active embedment length beyond bolt yielding. During this initial tensile testing, repeatability and performance at low level strains were also investigated to provide confidence in reported active embedment length observations. Even at loads below 1 kN (loads of 0.25 kN, 0.5 kN and 0.75 kN) with expected strain well below the accuracy of the sensor (expected strain is 4.6 με with the machine listed accuracy being +/− 25 με), the location of the start or end of the sensor was still clear. The largest positional difference between the actual location in which the sensor is no longer bonded and where it was perceived in the data was 13 mm. Repeatability was also inspected by looking at multiple scans (20 to 30 scans at 4 random times) under no load for two randomly selected 5 cm lengths. Range, average and standard deviation were inspected. The largest shift in average was 8.4 µε and 0.4 µε for each segment. The largest range from the mean observed was 24.8 µε and the largest standard deviation was 12.6 µε. These results provided further confidence in the ability to detect and discern between low and zero loads.
The use of fiber optics as a quality control instrument in a monitoring plan has been highlighted by research conducted in this group [
18]. The need for verification of ground support performance is required due to grouting issues such as gloving [
60]. This is often addressed through proof loading, in which the use of DOS can enhance insights provided to operators [
55]. Due to an issue with the grout pump used for the first round of specimen creation, the opportunity to highlight this enhanced insight was provided. In Specimen B-1-I there was a large air void roughly 20 cm in size located 40 cm into the embedment length. Because of this void, the same condition was retested with Specimen B-1-ii. When viewing the two displacement load profiles, it is not evident that one failure is due to grout shearing because of the reduced grout. However, when viewing the strain results from the DOS, the presence of the voids is apparent. This can be seen in
Figure 7. Due to the void’s location, it generally did not appear to have significant effects on the strain attenuation. This is likely because there was still some grout present, which was able to transfer load until its capacity was reached and the void become apparent. This exemplifies some of the additional insight gleaned from this cost-effective monitoring solution.
3.2. Effect of Ribs
Often, rock bolts utilize rebar, as the presence of their deformation improves performances via mechanical interlock [
61,
62,
63,
64]. To further explore this effect, 5 specimens were prepared and tested. Differences in strength and secant stiffness between the manufactured smooth bar were observed. Improved strength was seen with the presence of roughness and ribs, in line with other research. It was also noted that the rib specimens had lower secant stiffness. Although residual strength per unit length of the sanded smooth bars in the two annuli of B and A were not very large, there was a 72 kN difference in peak strength for grout annulus A observed. It should, however, be noted that these bars were smoothed by hand and could feature a slight difference in roughness, and that there existed a void in the grout for specimen B-6.
With the application of DOS, the effects of the presence of roughness and ribs can be explored further. Benefits in bond performance were clearly seen when comparing strain attenuation profiles. However, interestingly, this was not the case for loads below 30 kN. This can be seen in
Figure 8. Dilation effects were also observed where, at a higher load ribbed bars generated more dilation (at a load of 10 kN, the smooth bar generated more dilation than the rough bar). This high dilation is in line with the expected behaviour of the load transfer achieved through mechanical interlock. Viewing pre- and post-peak results of specimens B-1-I and B-6 again highlights the presence of dilation. Where the pipe is generally being compressed by forces transferred from the bar, however, dilation generates tension which then reduces the overall compression seen in the fiber positioned longitudinally on the outside of the pipe. In line with the expectation that mechanical interlock is responsible for a significant portion of strength, an evident change in response in the pipe pre- and post-peak for the ribbed bar when compared to the hand smoothed bar was observed. This can be seen in
Figure 9.
This investigation also allows an understanding of the limitation of the selected process for creating the rib spacings. The manufactured smooth bar had much lower strength when compared to rebar which was sanded smooth (145 kN vs. 12 kN). This is most likely due to either a change in friction between the two specimens or uneven sanding, leading to mechanical interlock. Due to this, it should be noted that specimens created through traditional manufacturing would likely have reduced bond performance when compared to the presented results, with the amount varying depending on end tolerances and coating. This is similar to the effect of rust seen on the strength capability of bolts [
63]. However, once cracking in the grout occurs, this effect would be minimal for mechanical interlock of ribbed bars.
3.3. Effects of Rib Spacing
During testing, three failure modes were observed. Specimens failed in the rebar shank, failed at the grout–rebar interface (pullout) and failed due to the pipe crushing (a result of pipe properties and grouting methodology). The majority of specimens reached rebar shank failure, as was the intention of the study. Specimen B-1-I and D-1 failed by pullout. However, for specimen B-1-I the large grout void was likely a factor in this. The failure of D-1 was also not indicative of the effects of the independent variables, as it experienced failure at the highest load of all specimens in grout annulus D (22.8 mm). This is likely due to a reduction in the area of the bolt during the rib removal process. As more ribs were removed, the reference of the original cross-section became further away and thus visually harder to match. The amount of diameter reduction to achieve the lowest value of the maximum applied load is approximately 2%; given the methodology employed, this is a reasonable explanation.
Another global parameter which is often investigated, as it is important to design and use, is the stiffness. In agreement with other conducted research on smaller embedment lengths, rib spacing was found to be a negative predictor of initial stiffness [
7]. Across all four selected grout annuli, it was seen that with increased rib spacing a decrease in stiffness was generally observed. An example from grout annulus A (7.7 mm) can be seen in
Figure 10. This effect was strongest in annulus B (9.9 mm). However, in the largest annuli, C and D (14 mm and 22.8 mm), this was less pronounced, except in specimen C-1 which was very stiff when compared to the others. This figure also highlights the general behaviours of the specimens, with most experiencing elastic loading, a brief plastic phase, strain hardening, then rupture. The testing and loading for specimens A1–A5 were ended prematurely due to the confining pipe yielding at the grout injection hole. Specimen A6 begin sliding prior to the yielding of the pipe. Specimens that began to slide generally exhibited a sinusoidal strain-softening phase which eventually reached a residual state. An example of a specimen reaching this residual state can be seen in
Figure 9. It is believed that this sinusoidal pattern is a function of the deformation-based control of the MTS. Specimen A7 went from elastic loading to grout failure and then sliding.
Multilinear regression was conducted to afford additional insights into the qualitative observation. No outliers were found, but a slight degree of non-linearity was observed. All assumptions of independence of observation, homoscedasticity, normality and additivity were found to be met, except homoscedasticity. To address this, a robust regression was run (bootstrapping, 5000 bootstraps and 95% confidence, JASP defaults) [
65]. The multiple regression model, Equation (7), statistically significantly predicts initial tangent stiffness F (2,17) = 33.002
p < 0.001, adj. R2 = 0.77. Coefficients can be found in
Table 6. This again showed that rib spacing negatively predicts initial tangent stiffness. Both the average unconfined compress strength (UCS) of the grout and rib spacing were statistically significant. Secant stiffness at peak loading did not have a clear trend in relation to rib spacing, but appeared to be related to the maximum applied load. However, this is likely a product of the rib removal process and the effects of reduced area.
Similar to the exploration of the effects of ribs, the strain attenuation patterns can be compared across the various grout annuli. Across all grout annuli, a reduction in bond performance was seen with increased rib spacing. This was observed by the strain being less effectively attenuated, and the attenuation rate appearing more linear, suggesting weaker bond performance. With increased rib spacing there also appeared to be more variance along the strain profile, which is logical as with increased rib spacing the locations of mechanical interlock are further apart. An example of this can be found in
Figure 11, which highlights results from annuli size A and B from loading at 90 kN. For the given annuli, it can be seen that the specimens with smaller rib spacing generally attenuated strain better. This is also true in annulus B, in which a large dichotomy exists between the altered and the unaltered bars. There are some exceptions to the general trend, where typically certain rib spacing may perform better than others in specific load ranges. For example, the 41 mm rib spacing in grout annulus A (7.7 mm) which attenuated stress the best for loads under 40 kN.
Utilizing LFRM, multiple linear regression confirmed this reduction in bond performance where rib spacing was seen as a negative predictor and the average grout strength was a positive predictor; see Equation (8). This relationship was seen as statistically significant where LRFM F (2,17) = 26.37,
p < 0.001, adj. R2 = 0.73). The regression coefficients and standard errors can be found in
Table 7. All required assumptions were found to be met; however, there was a degree of non-linearity. This non-linearity is logical as, if we examine either limit of rib spacing, it approaches a smooth bar. Therefore, caution should be used if the range of inputs are outside of the dataset used to form this model.
These strain data can also be explored through shear displacement profiles. When exploring these profiles, although the general patterns were similar, there seemed to be effects of increased rib spacing. Specifically with increased rib spacing, higher values of maximum shear stress developed, shear stress developed slower (higher levels of axial displacement required) and great variance was observed. An example of this can be seen below in
Figure 12. This shows the response of specimens D-1, D-3 and D-5, and was selected as it most clearly shows these observed patterns. This observation is in line with first principles and others’ findings, such that the delay in the development of shear stress is consistent with trends in initial stiffness and with a decreased number of ribs, the stress per rib would have to be higher while also driving the variance. The strain on the pipe was also explored under the assumption that with increased rib spacing greater levels of dilation would occur; however, no conclusive findings could be drawn from the collected results.
This overall negative bond performance was not in line with tests conducted on shorter embedment lengths [
7,
63,
64,
66]. However, differences in confining medium properties and grout material may offer alternative explanations for these differences in findings. In one study which featured comparable confinement stiffness [
7] they did find an increase in strength with increased rib spacing but it was not significant. This suggests a possible third variable to this relationship, which would require additional investigation.
3.4. Effects of Grout Annulus
The laboratory results were also cross analyzed to explore the effects of grout annulus. In a similar manner, global and component behaviour was examined. When comparing the strength and failure modes of the specimens, which can be seen below in
Table 8, it was observed that specimen D-1 failed in grout as opposed to shank. Visual inspection of the failure in the grout seemed to visually correspond to a reduction of confinement of the specimen, with cracking patterns resembling dilation slips failure; this can be seen in
Figure 13.
Figure 13 also highlights the observation of other cracking observed in specimens. Specimen B-1-i and C-1 both had large voids and thus should have resulted in lower strength because of the reduction of bond length, and specimen B-1-ii failed in shank at similar loads. This suggests that a reduction in bond strength can be associated with very large grout annuli. In line with the literature [
67,
68], it was expected that with increased grout annulus there would be a decrease in stiffness. However, no clear trends were observed across the data. Regression analysis could not be conducted as there were insufficient trends. When examining rib spacing 1, the axial deformation of the bar at 100 kN, B and C grout annuli (sizes) were stiffest.
Although there were no clear trends when examining the effect of grout annulus on bond performance across the data and multiple linear regression of LRFM could not be conducted, there was some support for a trend with the unaltered commercially available bars. When analyzing the strain profiles, it appeared that the unaltered rebar in B and C grout annulus performed better than the smaller and larger annuli. These can be seen visually in
Figure 11, comparing annulus A to B. For smaller loads, grout annulus A and D were more similar (this departure could potentially be caused by effects related to pipe deformation of the small pipe). Interestingly, beyond yield, it generally appeared that grout annulus C attenuated strain better than annulus B; this appeared to be caused by less rebar–grout debonding. However, when examining the strain at the free end, and the LRFM specimens in the grout, annulus B had better performance. However, the effects of the grout void could have impacted these results. When examining the shear profiles, A and D seemed to activate later and have lower values of shear stress even at higher displacement. Specimen B-1-I had the highest shear stress. This can be seen in
Figure 14.
In summary, there was some support for the theory that the bond performance of an unaltered rebar could be increased with increased grout annuli within a set range. Performance effects of increased stiffness, higher LRFM and better strain attenuation were observed with the unaltered bar. This range of grout annulus sized at 8.4× and 11.9× average rib height agrees with the finding of numerical modelling by Yokota, which suggested 9 times to 12 times rib height [
69] and other research which suggested the benefits of increased grout annuli [
5,
9,
70]. It should be noted that these trends were not seen globally across all the tested rib spacings. However, other selected research has shown that modifying the rebar–grout interface by wrapping the rebar with an additional rod does have an optimal hole diameter, suggesting that such a trend could exist for other alternative geometries [
70].
3.5. Model Results, Comparison and Discussion
3.5.1. Analytical Model Analysis
To compare the results of the physical testing program to analytical models, comparison plots at 10, 50 and 100 kN were prepared for all 24 specimens. An example of this can be seen below, in
Figure 15. Both selected models were generally exponential, matching what was largely seen across the data. It was seen that Ma’s model [
11] tended to provide a better fit to the data but was underpredicted in the attenuation rate resulting in higher strain compared to laboratory data at the same length along the specimens. Opposite to this was Li’s [
16] model, which was over-predictive in the attenuation rate of the strain. When examining the model’s fits to laboratory data across changes of the independent variables, Ma’s model appeared to have improved fit with increased rib spacing.
3.5.2. Numerical Modelling Results and Analysis
The overall response of the numerical model was consistent with the expected and observed behaviour of laboratory specimens. Specifically, it was observed that as the rebar is pulled axially, the bar elongates and slightly contracts. This displacement reduces as the distance from the loading position increased. This is because the ribs transfer the load to the grout and causes a slight dilation. This behaviour was congruent with their material properties and their interaction mechanics. It was also noted that the shorter models deformed more than the longer embedment length ones did. This was also seen in the laboratory specimens [
9]. It was found that the RS2 models created by our group for the longer length specimen such as 500 mm and longer were found to be over-predictive and did not match the general trends observed in physical testing. Despite a shift in strain attenuation pattern between the short and longer lengths not being observed in the RS2 model, it was successfully seen in the ANASYS contact model. During the first stage of the model, results were very close to physical testing. For the 100 mm and the 500 mm, the model performed very well, with the largest difference being ~15% (taken visually as original data were not available). However, generally the results were much closer. Although the 250 mm did display the correct pattern, it was over-predictive on the attenuation rate showing much greater strain departure when compared to the other models. This iteration of the model for the 750 mm standard rebar and altered bar (Stage 2.1 and Stage 3.1) did not perform as well as the smaller models (Stage 1 series models) and there was a greater difference between model and laboratory strain results. However, the global strain attenuation pattern matched much better when compared with the collected laboratory results, which can be seen below in
Figure 16. The results are most similar to the specimen 750CP and can produce very close results for 10 kN and to some extent 50 kN when the front is manually debonded. The possibilities of the differences could exist within the limitation of the model but also analysis of the strain data, there could be debonding of the section in this specimen as there is the appearance of a near-zero slope front section, which if present would lead to higher strain values along the embedment length. These initial findings suggest that a ANASYS provides a closer representation than the model using RS2. Additionally, further parametric analysis, calibration, and proper grout representation could be conducted to further improve the model ability to match the laboratory data.
It was seen that the model responded similarly to the physical results when rib spacing was increased. Increased rib spacing resulted in greater variance along the profile and a decrease in strain attenuation. The comparison between both 750 mm models (Models 2.1, 3.1) at 10 and 50 kN can be seen in
Figure 17. The model also showed a greater level of dilation, which was in agreement with the literature but was not observed in this laboratory investigation. In summary, it appears that this methodology shows promise to not only capture the general response of the system but also capture the effects of parametric changes without fitting.
3.5.3. Discussion of Numerical Model Design
Confirmation of sufficient meshing sizes was not possible with standard convergence studies due to the appearance of hot spots likely caused by contact singularity. Even with geometric refinement and the inclusion of 0.25 mm fillets, meshing studies did not converge. However, when looking at the overall stress distribution and strain response in the centre of the specimen, confidence in both the meshing and geometrical simplification was achieved. The comparison of stress between the two models can be seen in
Figure 18. Furthermore, as the main objective was the strain response, this was also compared between the two models; the largest difference between them was found to be 3%. To confirm this sufficient meshing with the larger models, an additional second verification was conducted. The 250 mm and 750 mm of the alternated bars were both investigated. For the 250 mm between the coarse and finer (117 k to 147 k nodes), there was a 0.16% difference in deflection, with the 750 mm having a similar level of consistency.
Another important factor for investigation is the contact normal stiffness factor. Generally, with increased normal stiffness the penetration will decrease, while contact pressure, stress and iterations to solve will increase. Values were selected by conducting convergence studies, with the first-stage models being 10 for the pipe grout contact and 3 for the grout rebar contact; however, this was increased to 15 for the pipe grout contact. Convergence models were generally in line with the stated expected response. However, in the first-stage model, maximum stress in other bodies (not the ones found at the contact interface) decreased with increased rebar–grout contact pressure, and the pipe grout seemed to peak at a normal stiffness factor of 10. These selected values are in line with expected contact behaviour where the pipe-grout is bonded and the grout rebar is frictional.
To ensure boundary and load conditions were properly represented, investigation into their response and effects on results was explored. The application of load was very closely represented to the laboratory condition, in the laboratory loading was achieved with a hydraulic grip ~20 cm from the front of the pipe and in the numerical model this was represented as the load was applied across the entire surface area of the front of the rebar in the numerical model. To keep the model efficient, the bearing plate used to hold the specimen in place was represented by a boundary condition that was applied across the top face of the pipe to constrain the X (plane of loading). The Y component was constrained by the plane of symmetry of the model. In the laboratory, specimens were restrained in the Z and Y by the friction between the pipe and the bearing plate. This could be represented as an additional friction contact surface. To simplify this, a plane of symmetry was created on the pipe and the Z was constrained on the edge which would interface with the pipe. The boundary conditions created slightly less confinement on the top of the pipe than the actual specimen, but was selected this way with consideration that the frictional force would be proportional to the applied load and with expected grout conical failure, very little force would be transferred in this region. When analyzing the load response of 20 kN of loading, the X boundary condition reaction forces were directly equal to the applied load. Reactionary forces in the other planes at the X and Z boundary condition were either small (~4% and ~2%) or zero. These results were in line with the expected response of the system. To understand the effects of making the model freer than the physical testing, another model was run but with the entire face of the pipe constrained in all directions. When comparing the models, a reduction of maximum stress of 3% and a reduction of the maximum total displacement of 8% was observed. Despite this large change in displacement, there was little effect on the strain profile in terms of shape and magnitude.
In the stage 1 models, the longitudinal ribs were left unaltered but in the subsequent stages the longitudinal rib was grooved to better match the laboratory specimen. These grooves were made slightly deeper than the actual specimen’s so that they were flush with the rest of the geometry to avoid thin geometries. To understand the effect of the inclusion or omission of the rib, model 2.1 was run both with and without the grooves. A large difference (~9%) around the front was noted but this difference converged further along with the rebar. This comparison confirmed the decision to include the grooves as it does affect the strain profile.
Materials were generally taken from either experimental collected values or literature values for the given materials. The only material modification was a reduction in grout stiffness. A ratio from the curve fitting efforts was utilized [
18,
71]. To understand the impact of this factor a model was run with both unmodified and adjusted values. In line with expected behaviour, strain attenuated quicker. This led to the model being even more over predictive when compared to the laboratory result. This directly leads to the largest limitation and source of error for this iteration of the model. The grout was modelled as linear elastic; however, it is clear from the laboratory results that cracking occurs and these should be included in subsequent models. This would result in greater displacement, with a plastic softening effect with increased loading. Future efforts could also include full plasticity representation, sensitivity analysis for friction values taken from literature as well as other material properties to understand their effects on results and the use of an axisymmetric model could be explored.