 When dams retain reservoirs, they establish a new flow pattern for the water beneath the ground surface around the lake. As the depth of water in the reservoir increases, the pressures that cause flow increase. These flows occur in the soil and rock materials that contain the reservoir and under the dam. If it is an embankment dam, water also flows through the dam. These conditions exist for the entire lifespan of the reservoir, which may exceed 100 years. Dam designers provide filter and drain zones in an embankment dam and between the dam and the foundation to control these flows and protect the dam structure. The use of these zones has long been recognized as essential for safe embankment dams. This is because they control the possibility of piping, the movement of small particles resulting in erosion within the embankment or foundation of the dam. If the potential for piping is not controlled, it may result in failure of the dam after years of apparently satisfactory operation. Not all older dams contain these zones. Even in dams where the designers included filters and drains, not all designs have been effective and some sinkholes and failures have resulted. In some cases, properly designed filters have failed to function because they were constructed improperly. Also, it is economically prohibitive to locate and construct filters to protect all the natural materials that contain the reservoir. In this program, the Federal Interagency Committee on Dam Safety is presenting Mr. John Lowe III, who will review filter design criteria and present two-case histories where sinkholes developed in modern projects. Mr. Lowe, a member of the National Academy of Engineers, has worked in geotechnical engineering his entire career. He did his graduate studies at MIT and in 1945 he joined NAP Engineering Company, which later became Tippets, Abbott, McCarthy, Stratton, commonly called TAMS. As head of the saw mechanic and rock engineering department, he established their first soils laboratory. He eventually became a partner, retiring in 1983. During his career, he has worked on numerous large civil works projects, including ports, highways, airports, and dams. His experience with dams includes active participation from early feasibility stage through design and construction into operation and surveillance. In fact, his role in Tarbelladam in Pakistan, the world's largest embankment dam, is an example of his long-term involvement and broad experience. Also, the use of an excess of one million cubic yards of roller-compacted concrete at Tarbella exemplifies his role in technology development and application. He has served as secretary of the U.S. National Committee of the International Society of Saw Mechanics and Foundation Engineers and as president of the United States Committee on Large Dams. For many years, he was a member of the Saw Mechanics Advisory Board of the Corps of Engineers, Office Chief of Engineers. Contemporary members of the board were Arthur Casagrandi, Ralph Peck, and Stanley Wilson, among others. His experience includes working on dam projects in the Mediterranean and Middle East, the Far East, South America, and the United States. For his contributions in Morocco, the king decorated him as commander of the Order of the Alluit. Since his retirement from Tams in 1983, Mr. Lowe has been a consultant on a number of projects around the world. Some of his assignments included working in Bulgaria, Chile, Malawi, Jordan, Greece, and the United States. May I present Mr. Lowe. Sinkholes have developed in the impervious cores and blankets of what presumably were well-designed earth dams. Projects which have experienced such problems include Baldur Head Dam, England in the 1960s, Tabela Dam, Pakistan in the 1970s, Luddington Pump Storage Project, Michigan in the 1980s, and W. A. C. Bennett Dam, British Columbia in the 1990s. In his ICOTS video and type of seepage and piping, Professor Ralph Peck showed examples of sinkholes and piping tunnels. To create such features, material must be eroded from the dam or from its foundation. Filters are the key measures incorporated in the dam project to prevent such erosion. Here is a typical cross-section of an earth dam. Water seeps through the core of the dam into the chimney drain. The chimney drain must be designed to be much more pervious than the core and, at the same time, have interstices fine enough so that material from the core will not enter them. Similar design is required for the blanket drain with respect to seepage from the foundation. Also, filters are frequently required between the upstream shell of a dam and the overlapping rip-rap slope protection. Upon drawdown of a reservoir, water will seep out of the upstream shell. Filters are required to prevent washing of upstream shell material through the rip-rap and into the reservoir. Filters may consist of one layer or two or more layers as shown in this sketch. The material being protected is termed the base material. If the material is the core of a dam, first a fine filter would be required, such as sand and then a coarse filter of gravel. The design of filters has been well understood since about the 1940s. First, I will review briefly the development of filter design. Most of my talk, however, will be devoted to a discussion of examples of widely graded base materials for which standard filter requirements have not worked. We use the term non-self-filtering for these materials. I believe that this non-self-filtering behavior was first recognized in 1974 at the Tabela Dam Project. I will describe the Tabela Dam Project situation in some detail and also a similar situation at the Luddington Pump Storage Project. After discussing these two projects, I will summarize my conclusions regarding filter design for both self-filtering and non-self-filtering base materials. Let us start with a brief review of the development of filter design. From the early days of earth dam design, engineers have generally appreciated the importance of proper filter design. Many early dams included filters, although they may not have been designed as we would design them today. In his textbook Fundamentals of Soul Mechanics, published in 1948, Taylor discusses filter design. One of the points which he makes is the following. Given a stack of tightly nested spheres as illustrated by these tennis balls, the largest-sized sphere which can pass through the interstices of such a stack would have a diameter somewhat less than one-sixth that of the nested spheres. In 1940, George Bertram carried out laboratory investigations of filters under Professor Arthur Casagrandi. His investigations were on materials of uniform grain size, both for the filter material and for the base material. He found that there was not any appreciable washing of the base material into the filter until the grain size of the base material was finer than one-tenth the size of the filter material. For comparison, the gradation curve for a widely graded or well-graded material is also shown. Tasagi advanced the criterion that the D15 size of the filter should be less than four to five times the D85 size of the base material. The concept is that the D15 size of the filter is an approximate measure of the size of the pore openings of the filter. Also, that if the D85 size of the base material cannot enter the pores of the filter, all of the base material will be held back from entering the filter. Tasagi also proposed that the D15 size of the filter be four to five times greater than the D15 of the base material. Purpose of this was to ensure that the filter had appreciably greater permeability than the base material. We need to recognize also that over the years, considerable research on filters has been carried out by the U.S. Corps of Engineers, the US Bureau of Reclamation, and the Soil Conservation Service. Now let me describe the sinkhole problem which developed at the Tabela Dam project. Tabela Dam is the largest hydro project dam in the world. It is located on the Indus River in northern Pakistan. The dam was completed in 1974 and the first power went online in 1976. A geologic section along the baseline of the main embankment looking downstream is shown here. As shown, the central 6,000 feet of the dam is founded on river alluvium. The left 1,000 feet and the right 2,000 feet are founded on pond rock. On the right side, the alluvium is as much as 700 feet thick. Because of the great depth of alluvium, it was not feasible to make a positive cutoff. To control under seepage, an impervious upstream blanket was connected to the inclined impervious core of the dam. The impervious blanket extended about 6,800 feet upstream. At the toe of the dam, it is 46 feet thick. It then tapers to 3 feet at the upstream end. A horizontal drainage blanket was provided under the downstream half of the dam and relief wells were provided at the downstream toe. These are typical seepage measures as discussed by Professor Peck. The only impervious material available for the blanket and core in the vicinity of the site was flood plain silt. Such material would be difficult to compact, and unless well compacted, might be subject to liquefaction under earthquake loading. The site is located about 60 miles from the Himalayan thrust vault, and ciliary faults occur much closer to the site. Thus, the dam had to be designed to withstand severe earthquake loading. In order to obtain a suitable impervious material, the design required that the flood plain silt be mixed with well graded slope wash found a short distance from the site. The grain size distributions of the silt and of the slope wash are shown here. The slope wash was separated into the fraction between the six inch and the three-quarter inch screens and the minus three-quarter inch material. These fractions were then combined with the silt in proper proportions. The resulting core and blanket material is shown on this grain size plot. The average material and the 80% limits are shown. The resulting material is well graded, has about 35% passing the number 200 sieve. The blended material is similar to a well graded glacial till. Such glacial materials have low permeability, high shear strength, and low compressibility. Engineers have generally considered them to be ideal for use in impervious zone of a dam. The blended material was used in the core of the dam and in part of the upstream impervious blanket. The portion of the blanket between the upstream toe of the dam and the upstream coffer dam was required to have 50% of its thickness of such material. The remaining blanket material had to have low permeability, was not required to be so well graded. The foundation on which the blanket was placed consisted of alluvium deposited by a fast flowing river. The slope of the Indus River at the site is six feet per mile. The alluvium is composed of two materials, cobble gravel and fine sand. Here are their gradations. The cobble gravel consists almost entirely of 12 inch to 3 quarter inch material. The sand consists of about two-third fine sand and one-third medium sand. The cobble gravel was carried downstream by the river as bed load and the fine sand was carried down as suspended load. Strata in the foundation consists of three possible combinations of these two materials, that is open work cobble gravel, fine sand strata and cobble gravel choked with fine sand. The cobble gravel does not at all meet filter requirements with respect to the fine sand, that is the d-15 size of the cobble gravel is much coarser than four to five times the d-85 of the fine sand. The fine sand can readily pass through the interstices of the cobble gravel. On the other hand, the blended impervious material readily meets filter requirements with respect to the cobble gravel. The d-15 size of the cobble gravel is definitely smaller than four times the d-85 of the blanket material. The first filling of the tabella reservoir took place the summer of 1974. As the so-melt runoff season progressed, the four tunnels being used for diversion were progressively closed. A problem developed with the last of the three gates on the last tunnel to be closed, tunnel two. The gate would close only about halfway and became stuck in this position for a period of about 25 days. Severe cavitation damage occurred to the tunnel immediately downstream of the gate and a 200 foot long section of the tunnel collapsed. During the closing of the tunnel gates, the reservoir had risen about 340 feet above riverbed. The gates of the other three tunnels were opened and the reservoir lowered as quickly as possible. When the top of the upstream blanket was exposed, everyone was surprised to see that it contained 362 sinkholes. About 140 tension cracks were observed also as well as several linear compression zones. The sinkholes were generally located along the cracks. Here are three photographs which give a panoramic view of the blanket after a drawdown of the reservoir. The first view is along the upstream toe of the dam and was taken from the left abutment. The upstream face of the dam is evident on the left hand side of the picture. In the far distance are the intake structures for the four tunnels through the right abutment. The rectangular black area is the cut face of an excavation made on the right abutment for the approach channel to the tunnel intakes. The cut is black because it was made in carbonaceous schist. The second picture is of the central portion of the blanket. In the foreground, black clay silt deposited during the filling of the reservoir is evident. In the distance, the placement of additional blank material over the original blanket is evident. This is the area where the alluvium is the deepest and where most of the sinkholes occurred. The third picture is of the left hand side of the blanket. An individual sinkhole is evident in the foreground. The blanket had settled appreciably under the difference in water pressure between that on the top of the blanket and that on the bottom. This difference in pressure varied from zero at the upstream end of the blanket to close to 300 feet of water at the downstream end because of the non-uniformity of the foundation strata and undulating settlement pattern developed. Tension cracks developed along the centerline of the ridges in the pattern and linear compression zones along the valleys in the pattern. The compression zones were evident only in the soft clay silt covering the blanket. This photograph shows an elongated sinkhole following a presumed tension crack. A backhoe trench has been dug across the sinkhole. The layer of silt which was deposited while the reservoir was up is evident. In the dried out state, the silt has a great color. Another example of an elongated sinkhole is shown in the next photograph. In this instance, there seems to have been a branching of the tension crack. Here is a photograph of a deep exploratory trench across the previous sinkhole. One can barely make out vertical cracks extending downward from the vertical sides of the sinkhole. The next photograph is a close-up of the upper part of the sinkhole. The cracks extending downward from the sides of the sinkhole are more evident in this photograph. This is a photograph of a crack extending down to the bottom of the blanket. A pencil was placed at the interface of the blanket and the foundation. The material along the crack is coarser than the adjacent material, presumably because some of its fine fraction has been washed out. Openwork type alluvium is evident at the bottom of the picture. Here is another photograph taken of the bottom of the blanket. Only coarse blanket material remains, just above where the blanket material contacted the foundation cobble gravel. Void space exists between the washed blanket material and the intact blanket. Washing out of fines from the blanket presumably will continue with time and a sinkhole will become evident at the surface. This is a photograph of a sinkhole which developed in the material surrounding the casing of an instrumentation drill hole. Apparently the installation of the casing disturbed and loosened the surrounding blanket material making it susceptible to leaching out of fines from the blanket. There are two basic questions to be covered. Where did the fines from the blanket go and to why did they get washed out of the blanket? The river cobble gravel on which the blanket was placed was choked with river sand at least for the top two or three feet. However, as mentioned earlier the sand fraction can readily pass through the interstices of the cobble gravel fraction. Water seeping under the blanket could wash the sand fraction out of the cobble gravel immediately under the blanket to open work cobble gravel nearby. Also water seeping down cracks would assist in the washing down of this fine fraction. We need then to look at the possibility of the blanket material washing into the open work cobble gravel. As I mentioned earlier the open work cobble gravel readily satisfies filter criteria for the blanket. The D-15 size of the cobble gravel is much less than four times the D-85 of the blanket material. The comparison of the gradation of washed blanket material which has been found in the top of the foundation cobble gravel and the gradation of nearby intact blanket material is shown here. Practically all of the silt fraction was washed out as well as a very high percentage of the sand fraction. To analyze the situation it is interesting to divide the blanket gradation into two fractions and to check whether the coarse fraction satisfies filter requirements with respect to the fine fraction. The separation into coarse fraction is made at a grain size slightly coarser than the one-tenth millimeter. The D-15 size of the coarse fraction is about 12 times the D-85 size of the fine fraction. Normal filter requirements are required to be less than four or five times. On the percent finer scale the separation between the D-15 of the coarse fraction the D-85 of the fine fraction is 15 percent. To satisfy normal filter requirements the slope of the gradation curve should be no flatter than four times change in grain size for this 15 percent change in percent finer. Professor Cameron Kenny who carried out laboratory tests at the University of Toronto however has found that some materials were not self-filtering unless their gradation curve was flatter than a four time change in grain size for a 22 percent change in percent finer. That is the same as a two and a half times change in grain size for a 15 percent change in percent finer. It would appear that anyone using widely graded non-plastic materials for impervious sections in an earth dam should check the self-filtering characteristics of the material in the laboratory. The sinkholes found in the tabella blanket after the first filling and emptying were repaired as indicated in this drawing. Filter material was placed in the sinkhole and a blanket two feet thick and five feet out from the sinkhole. Then this was overlained with 15 feet of regular blanket material. Since the upstream blanket had not been replaced the top of the blanket has been monitored each succeeding year of operation using side scan sonar. Here is what the results of the side scan sonar look like. No picture is obtained directly underneath the side scan unit. This is the white band in the picture. The side scan unit is towed 10 to 15 feet above the bottom of the reservoir. It takes a picture at a flat angle out from both sides of the instrument. The three white dots evident in the lower part of the picture are sinkholes. From time to time the depth of sinkholes was determined by towing a special sounding device called an ORE profiler across the top of the sinkhole. When a sinkhole was found about 50 barge loads of wet self filtering well graded gravel sand and silt were dumped on it. The barges used for this operation were the bottom dump type as shown in this picture. The barges had a capacity of 105 cubic yards. Side scan sonar pictures taken after dumping on a sinkhole confirmed that the dump material stayed together as a mass as it sank through the reservoir. Here is a picture of one of the barges being loaded and here a picture of two barges as they traversed the reservoir. It is significant to note that the open part of the sinkholes penetrated no more than about one third the blanket. This is the result of the loss of the fine fraction from the blanket. The remaining course fraction occupies the lower two thirds of the sinkhole. Leakage through a sinkhole certainly was much more than leakage through a corresponding zone of undisturbed blanket material. Nevertheless as Arthur Casagrande who was on the board of consultants pointed out the sinkholes amounted only to the equivalent of pinpricks in the blanket. Upstream of the blanket the reservoir is in direct contact with the alluvium. This area is vastly greater than the total area of all the sinkholes. Thus the contribution of the sinkholes to the quantity of under seepage is negligible compared to the quantity of seepage coming from upstream of the blanket. Nevertheless it appeared prudent to monitor the development of sinkholes and to treat any that developed with barge dumping. During the filling and emptying of the reservoir in 1974 362 sinkholes developed. During the filling in 1975 440 sinkholes developed and were treated by barge dumping. The reservoir was filled to a higher level in 1975 than 1974 and then held at that level. In 1976 29 sinkholes developed in 1977 9 and in 1978 10 11 developed in 1980 and since then only an occasional sinkhole has developed. Now let us consider the sinkhole trenches which developed at the Luddington Pump Storage Project in Michigan. The upper reservoir of the Luddington Project experienced many trench type sinkholes similar to those found in the upstream blanket of the Tobelodam Project. As shown on the plan view the upper reservoir was located by a dike. The reservoir is located on a bluff about 250 feet above Lake Michigan. The sinkhole trenches are indicated on the plan. The sinkhole trench at the southern end of the reservoir is about 3,600 feet long. Most of the bottom of the reservoir is covered with a five foot thick blanket of impervious material. In some areas impervious glacial till was present and no blanket was required. The blanketed area is underlain by outwashed sand. The groundwater table in the outwashed sand occurs at elevation 750 feet which is about 90 feet below the bottom of the reservoir. The sinkhole trenches developed primarily through the blanket. In one or two instances however where the outwashed sand contained an appreciable percentage of minus 200 material the sinkhole trench penetrated several feet into the outwashed sand. During operation the reservoir fluctuated from 100 feet to 30 feet deep. This is the range in gradation of the outwashed sand. Material coarser than the number 20 sieve practically always is less than 10 percent. Generally material finer than the number 200 sieve amounts to 5 to 35 percent. Occasionally larger amounts occur because of its very small percentage the plus 20 sieve material occurs as float in the rest of the material and has no significant effect on the characteristics of the material. Generally the gradation curve of the minus 200 material is very flat. At least a 10 time change in grain size occurs for a 15 percent change in percent finer. Because of this flatness the coarse fraction does not at all act as a filter for the fine fraction. Under the 100 foot water load imposed upon the blanket the blanket settled due to the glaciation which occurred at the site and the foundation had linear features of higher and lower compressibility material. Thus the settlement pattern which developed had ridges and valleys. As at the tabella blanket tension cracks developed along the ridges. At both projects the tension cracks developed first at the crests of the ridges. They then readily penetrated through the blanket by hydraulic fracturing. The reservoir pressure which entered the initial shallow crack was definitely greater than the horizontal earth pressure in the blanket. Because of this unbalance in pressures the crack readily penetrated to the bottom of the blanket. Considering the porosity of the outwash sand about 25 percent by weight of minus 200 material is required to fully choke the outwash sand. A large portion of the outwash sand contains only 5 to 15 percent of minus 200 material and thus is not fully choked. A very high gradient is imposed upon the outwash sand foundation where crack completely penetrates the blanket. Under this gradient whatever minus 200 material is in the voids of the outwash sand in the upper part of the foundation is washed downward into not fully choked material in the lower part of the foundation. Thus progressively ample space is made available for minus 200 material from the blanket to be washed into the foundation. The blanket material generally consists of over 80 percent minus 200 material. It is interesting to note that the sinkhole trenches which developed in the tabella and Luddington blankets both had more or less parallel vertical sides. Probably two parallel tension cracks developed a few fruit apart along a settlement ridge. Material is first lost from the bottom of the blanket adjacent to the bottom of the cracks. In the upper part of the crack water is seeping into the blanket. This seepage stabilizes the sides of the upper part of the crack. At the bottom where material is lost the minus number 200 fraction of the blanket is washed into the voids of the outwash sand. The plus 200 fraction which is only a small percentage of the total blanket is left behind. A void space developed above the plus 200 material and with time works its way toward the surface. Also there would be a tensi for the blanket between the two cracks to settle into the void space. A sinkhole trench then becomes evident at the surface. The upper parts of the trench walls tend to remain vertical because of seepage into them from the trench. The treatment proposed for the Luddington sinkhole trenches was to fill the trenches and to cover the area adjacent to the trenches with a special filter material. The gradation of the 30 and 200 fraction of the outwash sand is shown on the left hand side. The D85 size of the special filter material should be no smaller than 1 fourth the D15 of the 20 to 200 fraction of the outwash sand. A reasonable band for the special filter is shown shaded. Gradations of several blanket samples are shown on the right hand side. The D15 size of the special filter readily satisfies filter requirements for the D85 size of the blanket material. Tailings for several nearby mining operations were found to meet the special filter requirements. This concludes my remarks on the sinkholes which have developed in the tabella and Luddington blankets. Now let me summarize the key items in the design of filters. One, materials whose minus 200 sieve fraction is composed of non-plastic silt and rock flour require much more stringent filter criteria than materials containing plastic fines. The fines for both the tabella and Luddington blankets were non-plastic fines. Two, a widely graded material containing gravel, sand and non-plastic fines must meet self-filtering requirements to be stable. Gap graded materials are especially suspect. Applying standard filter requirements, one would expect that a material whose grain size curve had a slope which was no flatter than a four-time size change for a 15 percent finer change would be stable. However, laboratory tests performed by Kenny indicate that sand and gravel materials may not be self-filtering when the grain size curve has a slope as steep as a four-time size change for a 22 percent finer change. Three, in view of Kenny's test results, it appears desirable to lean toward somewhat more conservative zine of filters. Thus it is desirable that the D-15 size of filter material be no more than four times rather than five times the D-85 percent size of the material being protected. Four, when well-compacted, non-self-filtering material is much more resistant to loss of fines than when poorly compacted. The additional compaction reduces the size of voids through which the fines can pass. Thus, such material should be well-compacted whenever possible. The development of sinkholes through non-self-filtering impervious blankets is greatly facilitated by the development of cracks in the blanket and by loose zones that may exist around instrumentation installations through the blanket. Six, transverse cracks frequently occur across impervious cores of a dam where the dam overlies a steep abutment. The filtered downstream of the core must be designed to prevent loss of fines from the core if the core is not self-filtering. As a second line of defense, a layer of filter material can be placed on the upstream side of the core. Then, if a transverse crack developed through the core, this filter material would be washed into the crack and greatly reduce flow through the crack. As a result, there would be less tendency for the loss of fines. For materials containing fines which are plastic, fine concrete sand makes a suitable filter even if the grain size curve indicates that the impervious material is not self-filtering. To avoid segregation during transport and placement, filter materials coarser than concrete sand should have narrow gradations. The hundred percent size of the filter should be no more than two to three times the zero percent size. Concrete sand can have a gradation much wider than this because when placed damp, it has enough apparent cohesion to keep it from segregating. Generally, our current knowledge is adequate for us to design proper filters. The principles need to be thoughtfully applied. Care must be exercised to ensure that segregation of materials does not occur during construction.