Prepared for presentation at "6th Conference on AE/MS Activity in Geologic Structures and Materials" Pennsylvania State University June 11-13, 1996.
Earthquakes account for more loss of life and property than any other natural phenomena. In spite of this fact, and the fact that we know why and how earthquakes occur, there is a great deal of pessimism from both the scientific community and Government agencies concerning one's ability to accurately predict earthquakes. This pessimism is evidenced by the large amount of funds expended in earthquake preparedness programs compared to the funds available for research concerning earthquake prediction. The primary reason for the lack of an earthquake prediction model is the inability of low frequency surface mounted seismic instrumentation to detect the higher frequencies associated with small fractures that occur prior to a large movement of a fault. The high frequencies associated with these small events are attenuated by the upper mantel and never make it to the surface.
It is shown in this report that the use of Acoustic Emission techniques for predicting failure in materials and structures with the use of high frequency sensors can find parallel applications in predicting movements of a fault in the earth. The primary difference is simply a matter of scale Acoustic emission applications utilize frequencies to 1Mhz, to detect events on the order of 1 micron. Its parallel in geophysical applications would be the use of frequencies of 1 Hz to detect events on the order of 1 meter. The primary difference with this simple analogy is that most acoustic emission applications involve detecting stress waves in plates. In this situation stress waves due to crack growth from a one micron area can produce plate waves of much lower frequency as the wave propagates and therefore can be detected with sensors in the Khz range of frequencies. Whereas a fracture from a fault producing a 1 meter area would have a fundamental frequency of approximately 2Khz and would not undergo the same type of dispersion and mode conversion observed in plates. We know from experience that a 1Hz ground seismometer cannot detect a 1 meter fracture for any practical distance. The only hope of detecting frequencies of 1Khz or more in the earth is to place very sensitive sensors in deep wells and space them in a 5 to 10 kilometer grid in order to get adequate coverage of an area of interest (such as downtown Los Angeles)
Once the higher frequency events due to small fractures occurring from a fault are detected, it is proposed that the data be handled in a fashion similar to that used for acoustic emission data from man made structures. Therefore some discussion of the procedures used for handling acoustic emission data will be presented, and analogies drawn to their use for earthquake prediction.
Acoustic emission technology has been used for many years for the testing of pressure vessels, piping and other man made structures. The foundation for much of this testing can be traced to the research performed by the author and coworkers at Lawrence Livermore Laboratories over 25 years ago. This research was able to establish a relationship between the Stress Intensity Factor (K) at a crack tip and acoustic emission signals that would allow failure prediction models to be established. Figure 1 (Dunegan 1969) shows the acoustic emission counts as a function of stress intensity factor K for several single edge notch fracture toughness specimens with different crack lengths. The specimens with long cracks failed at low loads while those with short cracks failed at higher loads as would be expected. The fourth power curve through the data points shows that the stress intensity factor K is the normalizing factor for the data.
Figure 2 (Dunegan 1971) shows acoustic emission data from a linear compliance fracture toughness specimen of high strength steel that has been hydrogen charged and tested under dead weight load in a creep frame. This type of specimen has been designed such that for a given dead weight load, K remains constant and independent of crack length over several inches of crack growth. A crack opening gage was also used to provide information on the amount of crack growth occurring due to the hydrogen embrittlement cracking occurring in the specimen.
Note that the summation of acoustic emission counts and the crack opening gage data can be fit with a straight line for a given value of the stress intensity factor K. When K is increased (by adding load) both the acoustic emission rate and crack growth rate increase but remain constant for that value of K.
Diffusion of hydrogen to areas of high stress near the crack tip and subsequent crack extension is the mechanism responsible for both the acoustic emission signals and the crack growth. It was observed during these tests that quite periods would occur without much crack growth, and then a lot of activity would suddenly occur. The longer the quite period the more the activity when it did occur. This can be observed in figure 2 from the acoustic emission data between 50 and 60 minutes for a K value of 22.5. Following a long period of small activity a rush of signals occurred over a very short time period, which caused the AE data to match up with the straight line drawn through the data points. This phenomena is what one would expect from activity along a fault in the earth. If one assumes that the fault movement is occurring under constant displacement rate, it would be expected that uniform movement would result in a constant signal rate due to many small fractures. If an area within detection range of the transducers begins to be locked uniformly over a large area, stress will be uniformly built up over this area with a minimum of small fractures until the stress intensity factor reaches a critical value. At this time a domino effect will occur and a rush of signals will be present due to failures occurring over the large area, resulting in stress relief of that area.
An example of how crystalline materials behave under very high pressures when subjected to constant displacement rate loading is shown by figure 3 (Dunegan 1967 unpublished). In this example, acoustic emission signals (counts/sec) were recorded in a frequency range of 100-300Khz from a Quartzite specimen undergoing confining pressure of 3.5 Kbar and axial load. Increasing strain results in increasing activity of the acoustic emission signals then a sudden drop in activity occurs. The drop in activity corresponded to a drop in load from the test machine. The test machine was operating in a constant displacement rate mode and therefore as displacement continued, stress buildup again occurred in the specimen, followed by another drop in load and acoustic emission activity... This type of failure mode called "Stick Slip" is common in hard crystalline materials under confining pressure such as Quartzite. It's interesting that the acoustic emission rate where load drops occurred for each instance in figure 3 had approximately the same value. One could therefore predict from the acoustic emission data for this specimen when the load drops would occur. The reason for the drop in load and subsequent drop in acoustic emission activity is due to the constant displacement rate of the crosshead in the stiff test machine. The sudden slip occurring in the specimen relieves the stress, and therefore stops the permanent deformation process until enough displacement is again imposed to cause further deformation to occur. This is an example of what is happening at the interfaces of a fault in the earth. Fault movement is a constant displacement rate process. If blockage occurs, the same deformation processes and build up of stresses as shown in figure 3 occurs resulting in stress waves having a very broad frequency bandwidth. If we had also used a low frequency sensor during these tests its primary response would have occurred when the load drop occurred. This is analogous to the present use of low frequency sensors at the surface of the earth. They record the stress drops after they occur but give no information of the deformation processes leading up the stress drop.
A model for failure prediction utilizing acoustic emission data has been developed (Dunegan 1988) for man made materials and structures. It has been observed that materials or structures containing cracks that are loaded to failure, such that failure occurs below general yield exhibit a power law increase in activity prior to failure (see figure 1). Dunegan found that curve fitting routines could be used with special software algorithms to recognize the change in slope corresponding to imminent failure of a specimen or structure. The procedure involves first fitting the summation of acoustic emission counts with a 6th order polynomial, and taking the first derivative of the data. The slope of the cumulative amplitude distribution of the signals (b) is then calculated. The first derivative values for each data point is then divided by (b). The resulting factor for each data point defined as (Zfactor) is again fit with a 6th order polynomial as the test proceeds and first and second derivatives are calculated from the fitted curve. The following algorithm is applied to the data in real time.
if Zn>Zn-1>Zn-2 and 2nd derivative >=0 then alarm
Acoustic emission data from several sources in the literature were replotted and curve fit and the above algorithm applied. Table 1 lists the references, type of test and percent of failure where an alarm was indicated by the model.
|REFERENCE||TYPE OF TEST||%OF FAILURE|
|Harris-1974||Low cycle fatigue 7075-T6||79%|
|Harris-1984||High cycle fatigue-Rotor Steel||74%|
|Harris-1984||High cycle fatigue-Rotor Steel||80%|
|Harris-1972||Fatigue of Wire Rope 40dB gain||96%|
|Harris-1972||Fatigue of Wire Rope 60dB gain||87%|
|Komura-1979||Combined SCC and fatigue-304SS||67%|
|Dunegan-1971||Hydrogen Embrittlement Cracking||90%|
|Percentage of failure that an automatic alarm would sound utilizing the failure model presented in this report.|
One might tend to argue that since the first derivative of the cumulative counts as a function of time or cycles is simply the count rate and counts per cycle, why not have a counter give you these values as the test proceeds and forget about the complexity of curve fitting? This will not work for the following reason: The smoothing provided by curve fitting is needed for a machine to unambiguously decide that a trend is developing. Cracks do not grow in a smooth uniform fashion but by jumps and steps (see figure 2). If one were simply to record the count rate or counts per cycle the large jumps in the data throughout the test would cause alarms to constantly occur. The same augments hold for applications to activity from a fault. One would expect from time to time to see swarms of activity created by small blockage at a particular station, but in the case of earthquake monitoring one needs to see the trends in adjacent stations in order to accurately predict whether or not large movements (over several stations) is imminent. This is why it is important in the applications to earthquake monitoring to have adjacent stations equally spaced, and restrict the data base for a particular station to a limited range through frequency, spatial or amplitude ratio filtering.
Another argument one could make concerning the model applied to earthquake monitoring is the number of data points used for the curve fitting and the time interval between data points. We have found that a minimum of 20 data points is needed for obtaining a good fit to the data and a maximum of 40 data points was used to curve fit the data in table 1. It is anticipated that adding the cumulative data on a daily basis from each station would be adequate, after 40 days the data would then be averaged back to 20 data points and fitting resumed on a daily basis until 40 data points were accumulated. If the same type of accuracy can be obtained as that in table 1, and a single station alarmed, a few days warning of a minor earthquake would be indicated. If the alarm occurred in a remote area it is unlikely that any damage would result. On the other hand if say three adjacent stations alarmed within a few hours of each other this could indicate a much larger anticipated movement of the fault. If these three stations are in a populated area warnings should be sent out that a large earthquake is likely in the next few hours or days.
In order to set up an effective monitoring system all of the stations should be capable of digitizing the AE/MS data and transmitting the data over a digital network to a central processing center. In this manner a predictive model can be applied to all stations and combinations of adjacent stations, and in the event of an earthquake in the network, one should be able to see the immediate results of stress redistribution and its effects at each station in the network.
The data in figure 3 shows that high frequency stress waves are generated by irreversible deformation processes in rock like materials. Earth quakes along strike slip faults such as the San Andreas and San Jacinto fault in Southern California typically occur at depths ranging from 5 to 20 km. The high frequencies present due to deformation processes are attenuated severely before reaching the surface and are not detected. One can improve the detectability of the higher frequencies by placing instrumentation in deep wells.
Abercrombie (1995) is the first to the authors knowledge to obtain higher frequency data by using a very deep well. Experiments were conducted at 2.4km in depth at Cajon Pass near the San Andreas and San Jacinto faults using a triaxial geophone with a bandwidth from 1 to 200hz. The signal to noise ratio of the earthquakes recorded at this depth showed a vast improvement over the same earthquakes recorded at the wellhead as evidenced by figure 4 which shows a comparison of the data from a 1.7ML earthquake occurring at 10km distance at 2.4 Km depth with that obtained from a sensor at the wellhead. Over a period of 2 years Abercrombie (1995) recorded several thousand tectonic earthquakes. Approximately 90% of these were too small to trigger the Southern California Seismic Network (SCSN) or to be observed above surface noise by the sensor installed at the wellhead. During this monitoring period hammer blows were made to the wellhead and were not detected by the sensor at 2.4 Km. This further illustrates the value of obtaining data at great depths. This data illustrates the feasibility of placing deep well sensors throughout an area such as the Los Angeles basin without interference from traffic and other surface noise. Making measurements at this depth is not an easy matter. Manov (1996) reported that temperatures of 105 C and pressures of 26 MPa were present at the 2.4km depth. This was the primary reason for selecting a passive sensor as opposed to an active sensor which would require supplying power from the surface. In November of 1993 the 2.4 km sensor was removed and replaced with 2 sensors, one at 1.5km and the other at 3km (Manov 1996). At 1.5km the temperature was reported at 65 C, and pressure at 15MPa. At the 3km depth the temperature was 120 C and the pressure 29MPa.
The traditional thinking concerning failure along a strike slip fault is that friction stresses create enough contact stress to cause shear failures to occur parallel to the strike on one of the plates. From a fracture mechanics viewpoint it appears that if an obstacle is encountered such that the two plates lockup at one point and the fault surfaces on one side of the blockage remains fairly stress free, a mode II failure condition exists. As stress continues to buildup due to displacement of one of the plates the stress intensity factor K becomes critical and fast fracture is initiated. One could also develop a model for the same condition as above that would predict that a mode I (tension perpendicular to crack surface) might also exist in the vicinity of the blockage of a strike slip fault. We will start this development with the following assumptions:
1. The fault is restricted from slipping along a finite section.
2. One plate is moving while the other is stationary.
3. At a depth of 10 to 15 kilometers viscoelastic flow of material continues in the moving plate according to Tse and Rice (1987).
4. The plate is continuing to move upstream and downstream of the blocked section.
5. Surfaces of the fault in contact below the elastic region are only able to transmit short range elastic shear stresses due to the viscoelastic nature of the material in contact in this region.
6. The blocked section is prohibited from moving by frictional forces or geometric obstacles.
Figure 5a shows a plan view of a strike slip fault with the stationary plate being east of the moving plate, with the moving plate moving north at a fixed rate. Three sections of the fault are shown with movement occurring in section I and III, reacted by a blocked section II.
Figure 5b shows a section just inside the west portion of the fault line, with the forces created by the slip occurring on both sides and below the blocked section shown by small arrows, with slip indicated by large arrows. Since the plate is continuing to move on the south side of the blocked section, compressive stresses will be created at the boundary between section II and III. This will cause compressive strain energy to be accumulated in both sections II and III
The tension forces seen acting on the blocked section arise from the continuing movement of section I relative to section II. These forces must be zero at the surface, otherwise cracks would form at the surface at the interface of section I and section II. Since no cracks of this type are usually observed, one way of meeting these boundary conditions is for tension cracks to form at the interface of the elastic and viscoelastic region and near the fault surface's as shown by figure 5b.
Figure 5c shows a section taken through the east side of the fault near the fault line. The forces acting on the blocked section are due to elastic shear stresses transmitted across the boundary due to friction stress and/or obstacles mentioned previously. These stress tend to move the blocked section north, creating compressive stresses between boundaries I and II, and tension stresses between boundaries II and III. The movement of the blocked section creates shear forces at the interface between the elastic portion and viscoelastic material below the blocked region. These forces would tend to create tension cracks at the southern boundary of this plate. Note that there is no viscoelastic flow occurring in figure 4c, only reaction shear forces. This is due to the assumptions 2 and 5 given previously.
Rocks are very weak in tension, especially when saturated with gases under very high pore pressures. Cracks forming at 10 to 15 kilometers in depth with temperatures of 300 C could produce rather dramatic results. The vacuum initially created by the formation of a crack would cause diffusion of gases and water vapor to this region in an explosive manner, which could cause spalling of material from the crack surfaces. Once an opening is available, the viscoelastic material at the interface would tend to flow into and plug the crack. As pressure continues to build along the crack surfaces additional crack propagation could occur until the pressure inside the crack comes into equilibrium with the pore pressure. Any water vapor entering the crack will cause stress corrosion at the crack tip, lowering the fracture toughness and stimulating additional crack growth. In addition convection currents would be setup that would raise the temperature of the crack tip, which could cause creep effects at the crack tip. The new surface area created by the cracks would lower the surface area resisting the movement of the blocked section along the fault line. As new cracks form parallel to the first crack due to a local decrease in pore pressure and continuing tensile forces due to viscoelastic flow along the boundary, the shear stresses along the fault will be relieved enough for a seismic event to occur.
When a portion of the blocked section slips, the sudden release of shear stress along the fault will cause the cracks to close, and the tension forces will be transmitted to the interface between the slipped region, and the region remaining blocked. In addition the gases in the crack will suddenly be pressurized to a very high-pressure due to this closure. These high pressure gases will cause additional mode 1 tensile forces to be present at the crack tip. This could cause additional crack propagation until equilibrium can be established by diffusion. The high pressure gases will also be forced into the gouge, providing a lubricating effect.
When blockage first occurs in a fault the seismic event mentioned previously should occur very deep and on the north end of the blocked section due to the higher tensile forces created by the viscoelastic flow not present on the east side. After a seismic event occurs, the portion of the blocked plate that has slipped will be dormant for a period, and the blocked portion south of this section will to tend to crack until a seismic event occurs in this region. It is postulated that a large earthquake produced by movement along the total length of the strike of the blocked section will be preceded by a progression of small events preceding south on the plate side and north on the continent side due to decoupling created by tension cracks as proposed. The rate of occurrence of these events and their magnitude should increase prior to a large movement along the strike. This increase in rate and amplitude is one of the basis for failure prediction model proposed in an earlier section of this report. The other is the shear failures that will occur along strike due to a sudden movement of one section of the fault.
It has been reported by Rikitake, 1976 (Internet) that the ratio of the P wave velocity to the S wave velocity decreases prior to a large earthquake by approximately 10% and returns to normal immediately before and following the earthquake. The diffusion of gases into the crack and subsequent contact of the crack surfaces to ground water could account for the presence of radon gas in deep wells reported prior to an earthquake. If hydrocarbons are present in the chamber created by the crack, the high pressures generated during a seismic event could possibly force these hot gases to the surface where ignition would occur. This could account for the "earthquake lights" reported by many observers during an earthquake. The presence of hydrocarbons in gaseous form under such high pressures and temperatures could produce a higher chain molecule through polymerization which would account for the presence of petroleum bearing rocks in regions of high fault activity. These cracks could also account for the decrease in electrical resistance of the earth prior to a large earthquake.
If the above model is correct, and tension cracks develop for the reasons given, high frequency acoustic emission signals will be generated prior to the seismic event. To test the model the optimum placement of acoustic emission sensors would be on the northern portion of the west blocked section as shown by figure 5, and on the southern portion of the eastern blocked section. Placement of the sensors in deep wells based on an isotherm of 100 C or less should allow measurement of the high frequency portion of the P and S wave generated due to small fractures occurring prior to a large movement of the fault. From considerations of the P wave velocity in granites and assuming that a crack propagates at 1/2 the shear velocity, a 1 meter jump in crack length would create an acoustic emission signal in the 2 kilohertz range. It will be shown that this high frequency signal can be detected at a distance of 10 kilometers in homogeneous rock under high pore pressure.
The 100Khz and higher data shown in figure 3 would not be able to propagate 10km and still be detected, even though the Q of granite at depths of over 1km is very high. An expression for calculating the displacement at the surface from a fracture along a fault line, including the effects of attenuation was derived by Konamuri (personal communication) as follows: <>
Using values from Abercrombie (1995), Q=1000, c=6 km/sec, stress
= 100bars, f = 200hz, density = 2.8 g/cm3 and r = 10 kilometers,a
displacement of 2.5E-01 microns is calculated which is within
an order of magnitude of that observed by Abercrombie at the 2.4
km depth (figure 3) for a 1.7ML event at 10 km distance.
The unknown value in the above equation is the stress. Additional calculations were done increasing the stress value until displacements comparable with figure 3 were obtained. The frequency was then increased in order to measure the effects on the displacement values at the 10 km distance. Figure 7 shows the displacement values obtained to frequencies up to 1,700Hz, showing that a displacement of 1 picometer would be present at the 2.4 km depth used by Abercrombie (1995). A value of 1 picometer was chosen as the stopping point for the data in figure 7, since this is the value of displacement that can be easily obtained with present transducer technology.
Calculations were performed using equation 1 and the conditions for figure 7, to determine the boundaries on displacement and frequency to produce 1 picometer displacement at the borehole. The results of these calculations are shown in figure 7a. It appears from the data that an order of magnitude decrease in distance will allow an order of magnitude increase in the monitoring frequency. These results show that it may be possible to monitor the high frequency deformation processes due to stick-slip shown by figure three, up to a distance of 1 km.
Historically strong motion transducers have been used to measure seismic events. In order for these devices to operate, the container housing the device must be induced to move by the event to be measured. It is doubtful that 1 picometer of movement at the interface between the granite wall and the sonde in a borehole containing such a device would induce enough movement at 1700 Hz to create enough signal to be measured from a velocity or acceleration transducer, especially if the deep well contains a steel casing between the wall and sonde. It appears that a more sensitive method of measuring these small high frequency events would be to measure the displacement of the wall of the well directly.
This could be accomplished by mounting a displacement transducer on a stainless steel waveguide designed to penetrate the wall of the sonde with "O" ring seals, and make direct contact with the granite. A pointed carbide tip on the end of the waveguide would assure sufficient coupling to the granite. A schematic of such a device is shown by figure 8. Once the Sonde is in the hole either a spring release or small motor would drive the waveguides through the wall until they made contact with the granite. The symmetrical arrangement shown in figure 8 for north, east, south, west and the vertical waveguides would center the Sonde and transmit the first P wave arrival from an event directly to the first transducer struck before the wave interacts with the Sonde. Once this interaction occurs the fidelity of the signal from all transducers will be somewhat compromised. This is one reason that the first hit channel along with the time of the hit is captured and recorded. This information along with the same type of information from adjacent boreholes will be used to accurately locate the source of the signal. One can also measure the time difference between the P and S wave to compute a hypocentral distance, but this calculation will not be as accurate as using the first arrival P wave at several stations.
Deep boreholes are likely to be filled with water. An alternative method of utilizing high frequency detection is to place high frequency sensors at even intervals in the water column and utilize the waveguide properties of the water column. Cross correlation of the direct received signal at each transducer with the delayed signal in the water column (due to slower velocity in water) would allow the location of a seismic event to be calculated. Another possibility would be to bond a high frequency sensor to the inside wall of the sonde and record the high frequency dilational waves transmitted through the water into the sonde.
1. That the configuration shown by figure 8 is installed in boreholes with an isotherm of 100C, or less spaced in a square grid of 10 kilometer between stations.
2. Instrumentation is utilized that will digitize and store all data at a 50Khz sampling rate.
3. A master clock signal is available to link all stations.
4. Each station has the necessary software to locate a source by utilizing data from four adjacent stations. Spatial filtering of the data will limit locations for analysis to seismic events only occurring in its sphere of influence (5 km radius approximately 10km depth). a. Software to be written to accumulate a separate database on more than 5 events (cluster) occurring within a 100 meter distance from each other.
4. The summation of amplitude, counts or energy from all events within the sphere of influence will be tabulated and graphed as a function of time. a. Separate graphs will be made from areas showing clustering.
5. Tabulations and graphics will be transmitted to the central station on a daily basis.
The central station will curve fit each database transmitted
from the individual stations and apply the failure algorithm mentioned
previously. Special attention will be paid to the following.
1. Are any stations showing alarm status?
a. Is any clustered data at a station showing alarm status? b. Is the data as a whole showing alarm status?
2. Are any two or more stations showing exceptionally quiet periods? 3. Are any two or more stations as a whole showing alarm status?
Figure 9 shows an example of two conditions that might exist for data from 5 adjacent stations. 9A shows a "healthy" condition for the faults nearby in that the activity is constant over time indicating that stress relief is occurring uniformly over all stations. On the other hand 9B shows a situation that should be cause for concern. Three of the stations adjacent to each other are showing long quite periods, which indicate that strain energy is being accumulated on all three stations. If one were to observe a pickup in activity on one station, followed by a pickup on another, followed by a pickup on the third within a matter of a few days, this increase in activity on all three would probably create an alarm condition at all three stations. This is an indicator that a large movement covering 30 km is likely and a warning should be sent out if these stations are in a populated area.
The primary value of having predictive capability for earthquakes is to give people some warning so that they can find shelter, or remove themselves from an area where structural failure could be life threatening. A joint program between The California Institute of Technology (Caltech) and the United States Geological Survey (USGS) will begin testing real-time earthquake monitoring with a small network of 20 digital seismometers linked to the Pacific Bell digital telecommunications service called Frame Relay (Internet 1996). The justification for the system is to cut as many as 30 minutes off the time needed to collect, calculate and broadcast vital data for major earthquakes, and ultimately helping public safety agencies respond more quickly to potential injuries and damage (Internet 1996). Although this system is very sophisticated, in the author's opinion it will not provide an adequate warning to populated areas, since the same strong motion instruments are being used as input to the system. A few seconds of warning could be given from such a system if a large earthquake occurred a large distance from populated areas. For instance, if it occurred 60 miles from downtown Los Angeles, and considering that the shock wave is traveling at 2 miles per second only 30 seconds of warning could be given, assuming that the data could be analyzed instantly. If the Earthquake occurred beneath the city of Los Angeles no warning at all could be given using present technology and methods.
The ideas presented in this report will have their greatest utility for monitoring large populated areas such as the City of Los Angeles. History has shown that the city can survive quite large earthquakes with only minor damage when they occur at some distance. Since strong motion seismometers are very sensitive to background noise, they cannot be located in highly populated areas, even if they could be located in these areas, there is presently no way one can predict an earthquake from such limited bandwidth instruments. They only measure the large events after they happen and are incapable of measuring the small events that precede them. This analogous to using a 1 mile fishing line to catch fish. If you catch a whale you can feel it, but you could never detect a "nibble" or a "strike" or a small fish if it gets on the line. The higher frequencies associated with these events are attenuated before reaching the fishing pole.
To implement such a system as proposed in this report will not be inexpensive. The drilling of deep wells will comprise a large percentage of the cost. Some efforts. such as "frame relay" mentioned previously are already underway to solve the communications problem in transferring the data to a central location. The only other component left is to provide the high frequency data as input to this system and the failure model software to go with it.
It is shown in this report that permanent deformation processes in rocks produce high frequency signals prior to fast fracture. An analogy was drawn between the use of high frequency sensors to detect Acoustic Emissions due to crack growth in materials and man made structures and the deformation processes and crack growth occurring in geologic materials along a fault line. It is shown that the Acoustic Emission signals from crack growth in man made materials can be correlated with the stress intensity factor K at the crack tip, and inference is made that the same correlation exists in geologic materials along a fault line. Special emphasis is placed on using high frequency sensors in deep boreholes in order to detect small seismic events that precede larger movements of a fault line. It is shown that the signal to noise ratio obtainable in a deep bore hole is excellent compared to that obtained from surface instruments. Calculations show that frequencies of 1,700 cy/sec can be detected at a distance of 10km from a seismic event, and at 14,000 cy/sec from events of 1km. A failure prediction model is presented based on detecting high frequency events from boreholes evenly spaced on a 10km grid. Data from several types of crack propagation in man made materials are presented which show the failure model can accurately predict failure from crack growth in these materials. Inference is made that the same criterion can be used to predict earthquakes in time to give hours or days of warning.
The author wishes to express special thanks to Baxter Armstrong for his help and advise in preparing this report, and Rachael Abercrombie for her helpful email communications.
Abercrombie R.E. "Earthquake Source Scaling Relationships from -1 to 5 ML, Using Seismograms Recorded at 2.5-km Depth." Journal of Geophysical Research Vol. 100, No. B12, 1995, pp 24,003-24,014.
Dunegan, H.L. and Harris D.O., "Acoustic Emission, A New Non-Destructive Testing Tool, Ultrasonics, 1969, Vol. 7, No. 3.
Dunegan, H.L. and Tetelman, A.S. "Nondestructive Characterization of Hydrogen Embrittlement Cracking by Acoustic Emission Techniques." Engineering Fracture Mechanics, 1971, Vol. 2, pp. 387-402.
Dunegan, H.L. "A Model for Automatic Recognition of Incipient Failure Using Continuous AE Monitoring" Dunegan Engineering Consultants Inc. Report #8808, August 1988.
Harris, D.O. and Dunegan, H.L. "Continuous Monitoring of Fatigue Crack Growth by Acoustic Emission Techniques." Experimental Mechanics, Vol. 14, No 2, Feb. 1974, pp 71-91.
Harris, D.O. and Dunegan, H.L. "Acoustic Emission Testing of Wire Rope," Materials Evaluation, Vol. 34, No. 1, January 1974, pp 1-6.
Harris, D.O., Dedhis, D.D., Mamaros, T.C., "Acoustic Emission from Crack Growth in Steam Turbine Rotor Steels, EPRI report CS-3426 , Project 734-3, March 1984.
Komura, I.I., Uchida K., Kashiwaya M. and Arii M."Acoustic Emission Characteristics of Stress Corrosion Cracking in Type 304 Stainless Steel, Advances in Acoustic Emission - Proceedings of International Conference on Acoustic Emission, Anaheim, CA., 1979.
Li V.C, Rice J.R. "Crustal Deformation in Great California Earthquake Cycles" Journal of Geophysical Research, Vol. 92 No. B11, pp 11,533-11,551, Oct. 1987.
Manov, Derek V., Abercrombie R.E., and P.C. Leary, "Reliable and Economical High-Temperature Deep Borehole Seismic Recording." Bulletin of the Seismological Society of America, Vol. 86, No. 1A, pp. 204-211, February 1996.