Laboratory Experiments on Rock Friction Focused on Understanding Earthquake Mechanics
Grant Number: 99-HQ-GR-0068
Terry E. Tullis
Tele: (401) 863-3829
Program Element II - Understanding Earthquake Processes
Key Words - Laboratory studies, Fault dynamics, Source characteristics
We have continued our series of experiments at intermediate slip velocities to evaluate the effect that dynamic slip is likely to have on stresses during earthquakes. Although we are unable to conduct experiments at speeds of 1 m/s typical of dynamic earthquake slip, we can slide for large distances at velocities of up to 5 mm/s, about three orders of magnitude faster than in most laboratory friction experiments. We can therefore attain high enough temperatures generated by frictional heating to cause interesting effects. In future work we hope to attain higher speeds where the behavior may be even more directly relevant to seismic slip resistance.
Our experimental suite has recently been expanded to include experiments conducted under unconfined conditions. We have also made more detailed calculations of both the average surface temperature and the transient local so-called ‘flash temperature’ at frictional contacts. Thermal dyes on the fault surface in several experiments were also employed to further constrain the temperature.
Confined experiments. Results from our confined
experiments on quartzite are summarized in Fig. 1. Samples were slid at an
initial run-in velocity of 1 m m/s for the first few
millimeters of slip, then at close to the highest velocity presently possible in
our high pressure apparatus, 3.2 mm/s, for the remainder of the experiment.
Plot of friction coefficient vs. displacement data for confined samples slid at normal
stresses of 25-112 MPa. One experiment at 112 MPa normal stress (the lowermost trace
in the figure) showed extraordinary weakening in only 1.5 m of displacement. Microprobe
analyses indicated the presence of significant fluorine in this sample, which may have
induced melting (see text for discussion), resulting in the observed low strength.
As shown in Fig. 1, we observe a dramatic decrease in strength for samples deformed at a normal stress of 28 MPa. The friction coefficient decreases from an initial low speed run-in value of 0.7 to 0.85, to a final value <0.4, over displacements approaching 3 m. One sample slid at a normal stress of 112 MPa exhibited extraordinary weakening at smaller displacements, with friction decreasing to a value of 0.14 in a displacement of 1.7 m, as shown in Fig. 1. It was primarily on the basis of this experiment that we reported last year that the degree of weakening we observed in rapid sliding experiments appeared to depend markedly on normal stress. However, microprobe analyses of the fault surface from this 112 MPa experiment showed a significant amount of fluorine in the fault zone material. We surmise that this fluorine is a thermal breakdown product from our Teflon sliding jackets adjacent to the rock in our sample assembly. Furthermore, the presence of fluorine appears to have induced melting on the fault surface and caused the more dramatic reduction in strength. This suggestion is supported by microstructural observations which reveal the presence of amber colored, transparent glassy material on the fault surface, with the glassy material having apparently been injected into thin cracks in the underlying gouge layer. Microprobe analyses of samples slid at lower normal stresses show no indication of fluorine. Samples from two other experiments at 112 MPa (shown in Fig. 1) - one slid at a lower velocity of 1 mm/s for the first 0.5 m of rapid sliding, then at 3.2 mm/s, and one which was terminated by a gas pressure leak at a displacement of ~0.7 m while sliding at 3.2 mm/s - generated lower temperatures (see temperature calculations below) insufficient for degrading Teflon for most of their displacements, so that the frictional strength remained high. This is supported by microprobe analyses that reveal no fluorine in the samples. Neglecting the results from the 112 MPa sample with fluorine contamination, these results suggest at most a modest effect of normal stress on the magnitude of the dramatic frictional weakening we observe.
We have also conducted rapid sliding experiments under confined conditions on other sample materials of geophysical significance; in addition to quartzite, we studied granite and monominerallic samples of feldspar. The results of these tests, for a normal stress of 28 MPa, are compared with those of quartzite in Fig. 2.
Fig. 2 – Rapid sliding behavior of three different rock
types – Westerly
loaded up at a low sliding speed of 1 m m/s, then slid at a velocity of 3.2
mm/s to large displacements. Granite slid at 28 MPa does not weaken
with displacement whereas granite slid at 112 MPa yields similar results
to quartzite at 28 MPa.
As shown in Fig. 2, a similar weakening effect occurs in feldspar as in quartzite, albeit the weakening is less dramatic than for quartzite. Samples of granite slid at a normal stress of 28 MPa show no weakening. The granite sample slid at 112 MPa (the experiment was terminated by a gas pressure leak), however, shows a nearly identical weakening as quartzite slid at 28 MPa. We have also conducted rapid sliding experiments on simulated gouge layers of quartz and of granite. Although the results are somewhat complex, slip in quartz gouge becomes localized and substantial weakening occurs at 28 MPa normal stress, as in experiments on initially bare quartzite surfaces. Experiments on granite gouge show no weakening at 28 MPa, as is the case for initially bare granite surfaces.
Unconfined tests in our high pressure apparatus. We have completed a series of unconfined experiments in our high pressure apparatus. These experiments serve the important purpose of determining whether the weakening effect we observe at rapid slip rates is an experimental artifact of our confined sample assembly. In these unconfined experiments, samples are placed inside the gas pressure medium, but the sample assembly is without the O-rings and Teflon sliding jackets that isolate the sample from the confining medium as in a confined test. The samples are therefore effectively unconfined. Results of three tests are shown in green in the friction coefficient vs. displacement plot of Fig. 3. The friction coefficient decreases from its initial, low speed run-in value of 0.7-0.9 to a final value of <0.3 in ~9 m of displacement. The frictional resistance is quite variable in the first part of the tests, but the data appear to merge to a common trend at larger displacements.
These results on effectively unconfined samples are compared with results from confined tests in Fig. 3. This plot suggests that the strengths of all of the samples with displacement, for normal stresses in the range 16-112 MPa, follow a similar trend. This favorable comparison, coupled with our microprobe analyses of the fault surfaces of our samples, demonstrates that the weakening we observe in confined tests is not caused by contamination from thermal decomposition of the Teflon sliding jackets. Furthermore, the data suggest at most a very modest effect of normal stress on the magnitude of the observed weakening over the range 16 to 112 MPa.
Calculation of average surface temperature. To determine whether the extraordinary weakening we observe might result from lubrication of the fault surface by melting, the gradually evolving frictionally generated surface temperature of our experimental faults was estimated. During our experiments, initially bare surfaces of quartzite generate a 50-100 m m thick gouge layer. However, to obtain an upper bound on the temperature, we assume a fault zone thickness of zero. The average temperature of the fault surface is calculated by numerically integrating the following equation:
where r is density, C is heat capacity, k is thermal diffusivity, m is the friction coefficient, s n is normal stress, V is velocity, t is time since perturbations started and ti is a dummy variable for time (Carslaw and Jaeger, 1959; Sleep, 1995). Although we are continuing these calculations beyond the expiration of this grant period, the preliminary results of the calculations suggest that the average surface temperatures are insufficient to induce melting in quartz. Thus, it seems difficult to attribute the dramatic weakening we observe to wholesale melting of the fault surface.
Estimation of temperature using thermal dyes. To help constrain the frictionally generated temperature in our experiments, we painted thermal dyes on the outside of our sample rings as well as in small, 1 mm deep wells drilled into the fault surface in the effectively unconfined experiments. The dyes undergo an irreversible phase change at temperatures of 100 and 300 oC. After sliding to a displacement of 9 m at 16 MPa normal stress, at 3.2 mm/s, the 100 oC dye in the well on the fault surface was completely transformed, whereas the 300 oC dye remained unchanged. On the outside of the sample rings, the 100 degree dye was melted in an extremely narrow zone near the fault, and the 300 degree dye was unchanged. These observations are consistent with our temperature calculations, which indicate a fault surface temperature of less than 300 oC.
‘Flash’ temperatures. Though the average surface temperature in our experiments is insufficient for melting, transiently high temperatures, which can be much higher than the average surface temperature, can occur at the contact junctions where energy is dissipated as heat [Archard, 1958; Rice, 1999]. These so-called 'flash' temperatures can be very short-lived; at high sliding speeds, on the order of m/s, flash temperatures can be of only millisecond duration We estimated flash temperatures in our experiments using the flash temperature theory presented by Archard . The calculation assumes that circular contact junctions of average radius a slide on a flat surface and that the average contact junction area is determined by the applied load, the number of contacts, and the yield stress of the material (i.e., the material flows plastically at the contact junctions). Assuming plastic flow of asperities yields an upper bound for the temperature.
At low sliding velocities, there is sufficient time for the temperature distribution in the upper body (asperity side) to be established in the lower body (the flat surface). The flash temperature is then given by
At higher sliding velocities, there is insufficient time for significant lateral flow of heat from the contacts, such that lateral flow can be neglected and a linear heat flow solution applies. The flash temperature is then
The appropriateness of the low vs. high speed flash temperature solutions is determined by the value of a dimensionless parameter, the Peclet number, given by L=aV / (2C ), where C is thermal diffusivity. The Peclet number is the ratio of the time over which heat is applied to the time required to diffuse it away. For L < 0.1, the low speed solution applies, whereas for L > 10, the high speed solution applies. At intermediate Peclet numbers between 0.1 and 10, the flash temperature is determined by interpolation between low and high speed solutions [Archard, 1958].
The Peclet number for the highest velocity in our confined experiments, 3.2 mm/s, is 0.001. For this case we predict a flash temperature of <100 oC above the background average surface temperature in our experiments. Thus, most of the weakening in our experiments occurs at flash temperatures <400 oC, i.e., well below the melting temperature of quartz. Thus, it is difficult to reconcile the weakening we observe with melting at frictional contacts. Furthermore, if the observed weakening was the result of melting at asperities at rapid slip rates, one might expect that a return to a slow velocity of 1 m m/s might result in an almost immediate strengthening, due to rapid cooling of the small asperities. In fact, the complete return with time of the strengths of our confined samples at slow sliding rates after the cessation of rapid sliding, discussed below, occurs over characteristic times of several hours.
Healing with time. At the cessation of rapid sliding, the frictional strength of confined and unconfined samples returns completely to pre-rapid sliding levels, as shown for unconfined samples in Figs. 3 and 4.
In Fig. 4, the increase in frictional strength of two unconfined samples is shown as a function of time. At the cessation of rapid sliding, one sample was slid continuously at 1 m m/s to a total displacement of 700 mm. The other sample was slid periodically at 1 m m/s after varying lengths of time, to a total displacement of 0.2 mm. The strengths of both samples recover nearly identically with time, demonstrating that the recovery depends on time, not displacement. Furthermore, the increase in strength of both samples appears to mimic the decrease in temperature with time, as shown in Fig. 4.
The Nature of the Frictional Weakening Mechanism
Our experiments activate a dramatic frictional weakening mechanism that operates outside the range of displacements available in most rock friction apparatus. Temperature calculations indicate that it is unlikely that the observed weakening results from melt lubrication caused by either wholesale melting of the entire fault surface or local melting at asperities (at least for sliding velocities < 3 mm/s). Furthermore, the slight dependence of the observed frictional weakening on normal stress makes it difficult to attribute the weakening solely to temperature, since the temperature of the sliding surface increases with normal stress. Also, experiments conducted on quartzite at high pressures and temperatures and slow sliding speeds in the triaxial apparatus (and therefore necessarily to small displacements) indicate ‘normal’ values of the friction coefficient of 0.6 or higher [Stesky et al., 1974; Stesky, 1978]. These triaxial experiments are at imposed temperatures that span the range of frictionally generated temperatures in our study.
We have considered that the weakening may be linked somehow to solid state amorphization. Both quartz and feldspar are susceptible to amorphization . Transmission electron microscope (TEM) images (taken by Dick Yund at Brown University) of the fluorine-bearing material from the fault zone in the 112 MPa experiment (Fig. 1) demonstrate that the material is completely amorphous to electron diffraction. We anticipate that the fault zone material from lower normal stress experiments is also amorphous, based in part on previous TEM analyses of samples slid at low velocity to large displacements [Yund et al., 1990]. Additional TEM and x-ray diffraction work will be required to determine whether the material from our low velocity, lower normal stress experiments is also amorphous.
It seems difficult, however, to explain the low strength of the material in the fault zone as being solely the result of amorphization; experiments at low sliding speeds and large displacements also generate amorphous material [Yund et al., 1990] but yield typical values of the friction coefficient. Low speed sliding experiments on silica glass also yield typical friction values [Weeks et al., 1991]. Our temperature calculations show that the frictional healing with time (Fig. 4) occurs over the same characteristic time as the decay in frictionally generated temperature at the cessation of rapid sliding. These observations suggest that the strength of the material on the fault is intimately linked to its temperature. Thus, we believe the weakening we observe is related to profound microstructural changes in the material along the fault that occur with displacement, and to shear heating of this highly altered material.
Though we have not yet positively identified this weakening mechanism, we believe it may have important implications for faulting in the earth. The dramatic weakening we observe occurs at modest normal stresses and sliding velocities. Thus, in the Earth, dramatic weakening might occur at conditions much less severe than those required for melting. At higher sliding velocities and normal stresses, we expect a transition to a melt-lubricated regime, which may result in even lower values of the friction coefficient. We hope in the future to extend our sliding capabilities to 1 m/s slip speeds so that the future data we collect will span the range from quasi-static velocities to truly dynamic slip rates.
It is apparent from our experiments that substantial reductions in shear stress can result from frictional heating at slip rates 2.5 orders of magnitude lower than typical seismic slip rates, even without frictional melting. This suggests that dynamic shear stresses could be even lower during seismic slip due to shear melting. If these large reductions in shear stress are characteristic of earthquakes it would seem to imply that dynamic stress drops may be nearly complete and that accelerations could be quite large unless the initial stress is also small.
We have measured the frictional resistance of rock and find that if slip is fast enough, but still considerably slower than natural earthquake slip, heat produced by friction can be enough to weaken the fault by about a factor of two. The reason for this is not clear, but the temperatures resulting from the frictional heating are too low to have produced melting. In one case where melting apparently occurred for another reason, the friction became very much lower. This suggests that faster slip during earthquakes should generate enough heat to cause melting, and that this melting could cause the resistance to be low during seismic slip. This low resistance could cause unexpectedly strong shaking of the ground and potentially more severe damage than currently expected.