Casting 6.5 meter mirrors for the MMT Conversion and Magellan

B. Olbert, J. R. P. Angel, J. M. Hill and S. F. Hinman

Steward Observatory Mirror Lab, University of Arizona
Steward Observatory, Tucson, AZ 85721
email: bolbert, rangel, jhill, shinman@as.arizona.edu

http://medusa.as.arizona.edu/lbtwww/tech/spiecast.htm

Proceedings of SPIE conference on Advanced Technology Optical Telescopes V, 2199, p. 144 (1994)

Abstract

1. Introduction

2. Mirror Blank Handling and Cleanout

3. Magellan Mold Construction

4. Conclusions and Future Plans

5. Acknowledgements

6. References

List of Figures:

Figure 1 Figure 2 Figure 3
Figure 4 Figure 5 Figure 6

Abstract

We report on the casting of two 6.5 meter diameter borosilicate honeycomb mirrors at the Steward Observatory Mirror Laboratory. These F/1.25 primary mirror blanks are for the MMT Conversion Project in Arizona and the Magellan Project in Chile. The first 6.5 meter mirror blank was cast in April 1992. After a three month annealing cycle it was removed from the furnace using a fixture glued to the upper surface of the blank. It has since been stripped of its mold material in preparation for polishing. The mold material is broken up with a high pressure water spray. The second identical 6.5 meter was cast in February 1994. Each honeycomb 6.5 meter is cast from over 10 tons of E6 borosilicate glass manufactured by Ohara. This glass is melted into a mold constructed of aluminosilicate fiber to produce a honeycomb structure with roughly 20% of solid density. The hexagonal voids in the honeycomb are produced by ceramic fiber boxes bolted to the bottom of the mold with SiC bolts. The furnace rotates at 7.4 rpm during the casting process to produce a F/1.25 parabola on the front surface. This rough parabola minimizes the amount of glass which must be removed during the grinding process. The front faceplate of the mirror will be 28 mm thick after generating. After the second 6.5 meter blank has been cast, minor modifications will be made to the furnace to allow casting of the first 8.4 meter mirror for the Large Binocular Telescope.

1. Introduction

In the early 1980's Angel and Hill 1 began developing the technology required to produce honeycomb mirror blanks up to 8 meters in diameter. The honeycomb structure provides a high stiffness to weight ratio while allowing ventilation to achieve a short thermal time constant. After producing a number of small blanks including three 1.8 m and three 3.5 m mirrors, the Steward Observatory Mirror Lab cast the first 6.5 m diameter borosilicate honeycomb mirror blank in April 1992. The progress of that blank through the opening of the oven after annealing has been described by Hill and Angel 2 . A description of the process of lifting the mirror and removing the mold material is given below in Section 2. This first 6.5 m mirror blank has now moved on to the polishing laboratory for finishing 3 . The first 6.5 m F/1.25 parabola is destined for the MMT Conversion on Mt. Hopkins 4 . In February 1994, a second 6.5 m mirror blank was cast. That mirror is currently annealing in the furnace. A view of the second 6.5 m mirror in the mold at 1180 C ° is shown in Figure 1. Four views of the chunks of glass flowing into the mold are shown in Figure 2. This second 6.5 m mirror is destined for the Magellan telescope in Chile 5 . Section 3 describes in more detail the process of casting these 6.5 m honeycomb mirror blanks. We also describe below some of the technical details which we have learned about the casting process.

View Figure 1 here

View Figure 2 here

2. Mirror Blank Handling and Cleanout

The MMT casting cycle was completed on June 29, 1992. After the 25-ton upper furnace enclosure was removed and the inner and outer walls and tension bands were disassembled, the mirror blank was thoroughly inspected. The mirror remained on the surface of the oven from July 20 until October 15 awaiting tests and modifications to the 6.5 meter lifting/turning fixture.

2.1 Handling Fixture

A new handing fixture composed of a 10 m diameter lifting frame and turning ring was designed and built to lift, support and flip blanks up to 8.4 m diameter. The design goal was to insure that tensile stresses in the blank stayed less than 0.7 MPa at every point in handling. The first lift of the mirror is the largest stress it is ever likely to see because the honeycomb structure is carrying the extra 10 ton load of the refractory mold material. Some analysis of the handling support is described by Parodi, Hill and Salinari$^6$. Because proper handling is so important, deflections of the lifting frame during a vertical lift and partial rotation of a specially-made 6.5 m concrete dummy were measured experimentally to 10 micron accuracy with LVDTs. In the first step of lifting, 36 mild steel pads 60 cm in diameter were positioned and attached to the blank's faceplate with a 1 cm thick, special patchwork layer of RTV silicone sealant (clear GE Silicone II) and silicone rubber tubing. The tubing acted as a portal for easy removal of reaction products relased during cure, letting the silicone cure as rapidly and uniformly as possible throughout the layer. Shear strength, tensile strength, and creep response of steel-RTV-glass sandwiches were measured to assess the short and long term performance of the pad attachment method. After a week's cure, the blank was lifted and placed in the handling ring on October 16, 1992 and rotated to a horizon-pointing direction a few days later. While a full three months were used up in the schedule, the safety of the mirror was our first concern, especially for this first lift. The mirror in the turning fixture is shown in Figure 3.

View Figure 3 here

View Figure 4 here

2.2. Mold Removal

To clean the blank, the ring was rotated until vertical, with the faceplate facing the wall. A special washout stand with two independent elevating work platforms was assembled and moved within easy reaching distance of the backplate. Removal of the hexagonal SiC tiles is shown in Figure 4. Working from top to bottom, the cleaning crew removed all nuts and tiles. Then working from bottom to top, each cell was breached and the bolt and washer removed. Using a high pressure (11 MPa) water jet ``cleaning wand'', the core material was sliced into chunks and the pieces of soft refractory mold removed by hand. The water blasting of the refractory material trapped inside the cores began in late December, 1992 and was completed on February 8, 1993. The backplate of the mirror after the mold material has been removed is shown in Figures 5 and 6. The blank remained in the handling ring until the end of July when it was moved face down (still glued to the handling fixture) into the polishing area.

View Figure 5 here

View Figure 6 here

Inspection of the MMT casting showed that small gaps opened between the post collars on the cores and floorboard in many locations. Glass melt slowly flowed through the gaps and onto the bolt, nut, tile and furnace hearth, foaming whenever it contacted SiC. Only about 50 kg of glass leaked out so the effect on the blank was negligible except that the glass coating bound the SiC parts together, making part removal and cleanup difficult and time consuming. Approximately 50% of the bolts were broken during disassembly because threads along bolt/nut contact region were fouled by glass.

2.3 Blank Quality

The general quality of the MMT 6.5 m blank was excellent, but not perfect. The paragraphs below describe some of the areas where we have tried to improve the next casting.

Rib Cracks

Seventeen rib cracks were discovered in the MMT mirror during cleaning and inspection. Fourteen were single, slit-like cracks that started at or very near the cross pin hole and ran into the rib, the lengths ranging from ~1 cm to ~12 cm. The rest had more than one crack branch with the longest branch being ~12 cm. Two rib sections containing cracks were cut out for analysis. Fractography studies done on the two sections by Dr. Edwin Beauchamp of Sandia National Lab showed that at least one crack was associated with an impact as evidenced by a Hertzian-type cone fracture on the surface opposite the crack origin. It was then suspected that all cracks were due to an inadvertent, intense impact of the cleaning wand on the rib. Working under this assumption, the wand was encased in soft rubber and the jet pressure was reduced by 25%. With these changes, and the cleaning crew with heightened awareness of the problem, no more rib cracks were observed while cleaning the remaining half of the blank. The crack tips of all rib cracks were eliminated by stop-drilling, and the drill-hole boundary was etched with HF paste. Once treated to round the crack tip, the cracks do not threaten the integrity of the honeycomb. A detailed inspection of the blank and stop-drilling of the cracks was completed on May 30, 1993.

Geometry

The rib and core geometry of the MMT 6.5 meter blank turned out very well. All the ribs in the blank are within a few millimeters of their expected absolute positions. The faceplate thickness was measured to be 32.4 ±0.7 mm, somewhat thinner than the 42 mm expected, but still more than the 28 mm required. It appears that the cores compress slightly and creep under the hydrostatic pressure causing the ribs to be slightly thicker than the original mold geometry. This results in the finished blank being 2% heavier than nominal. Faceplate curvature indicates that the rotation speed of 7.40 rpm was held to within 0.5%.

Stress Spot measurements of residual stress in 47 ribs showed an average stress-induced retardation of 4.8 nm cm -1 (20 psi tension) with a standard deviation of 2.2 nm cm -1 . The maximum observed retardation was 7.2 nm (30 psi). This is consistent with the expected amount of residual annealing stress combined with some small amount of inhomogeneity of the expansion coefficient. The stress magnitude was calculated from the measured stress-optic coefficient of E6 = 4.22 psi nm -1 cm. Additional measurements will be made after both sides of the mirror have been polished.

Bubbles Approximately 150 bubbles of ~1 cm diameter were found in the upper 5 cm of the MMT blank. These bubbles are located mainly in the outer third of the faceplate. These seem to have been caused by not holding at maximum temperature long enough to allow all the trapped air to escape. An additional hour of cooking time would probably have cured the problem. The number of 1 mm size bubbles is noticeably lower than was seen in the 3.5 m mirrors. While many of the bubbles will be ground away during generating, about two dozen will intersect the final surface. These large bubbles will be plugged prior to polishing to avoid trapping polishing compound as has been done on occasional bubbles in previous mirrors.

3. Magellan Mold Construction

Once the oven was cleaned and inspected, floor tile layout began for the Magellan 6.5 meter mold on February 8, 1993. Tub walls, inconel tension bands, and pneumatic cylinders were installed and the oven was fired to seat the inconel bands in March 1993. After the 2-week firing, the tub was lined with refractory and machined to size. Core machining and installation began on May 26, 1993 and was completed by the end of July after which the mold was inspected and fired once again to 1180 C °.

3.1. Ceramic Materials and Mold Assembly

The 6.5 m Magellan mold design was an exact replica of the successful 6.5 m MMT mold. A drawing of the mold is shown in Figure 5 of Hill and Angel2. The ``hard'' refractory mold parts used in the casting tub were made from injection molded, clay-bonded SiC by Ferro Corp., East Rochester, NY (regular hex tiles, bolts, nuts, washers), or from Carborundum Carbofrax SiC-based castable refractory (tub walls, irregular partial tile shapes). ``Soft'' refractory parts used in the honeycomb mold were vacuum-cast into pre-formed shapes from an aluminosilicate ceramic fiber mixture by Rex Roto Corp., Fowlerville, MI.

As with the MMT mold, the Magellan mold was built within the furnace area on the 10 m turntable. The base of the casting tub was constructed from a close-packed array of individual hexagonal-shaped floor tiles, all tiles installed with high-accuracy templates. The nuts used to engage the bolts were individual hex-shaped parts that were positioned with the tiles. To prevent rotation as the bolt was engaged, the nut meshed into the tile feet. All tiles were made undersized by ~0.8 mm to guarantee no overlap problems during layout. To insure stability, the tile corners were pinned together by pyramidal wedges cemented into the array's tri-corner junctions. The nearest neighbor tile separation distance was 192.2 ±0.13 mm. Over the mold radius of 3.25 m, the p-v spread in tile separation distance was 1.8 mm.

Inner and outer ``tub wall'' sections were assembled around the tile base in the manner of barrel staves. Inner walls were locked into position by keystone pins. Pre-rolled inconel 601 bands were wrapped around the outer wall and the band ends hooked to air cylinders shackled to steel posts located outside the furnace shell. The posts were effectively tied to the the main I-beam spokes of the turntable. Band tension was applied by feeding a well-regulated supply of 2.4 MPa nitrogen to the cylinders. Tension of 500 pounds on each of 80 bands serves to constrain the tub against the hydrostatic pressure of liquid glass. As usual, the casting tub was prefired to 1180 C ° to seat the bands and fix the tub geometry.

The tub was lined with soft refractory fiberboard, and the floor machined flat. The tub sidewall liner was machined to the skirt contour of the blank. The soft refractory hex boxes, cross pins, post collars and box caps needed to make the honeycomb structure were cut and machined from various preforms. All regular hex core shapes were NC machined from one hex preform shape. The irregular partial cores along the inner hole boundary and outer boundary were made by cementing pieces of flat board to various hex box sections.

Each machined box was guided to its particular hex tile with a special box placement tool. The tool was located by locking it into the center holes of two nearest neighbor tiles. In this way, the core positions were referenced directly from the tile positions. The box was pushed snugly into a V-shaped guide plate on the tool, and slid down until the post collar made contact with the floorboard. The post collar separates the core from the floorboard to form the perforated backplate of the mirror. The box sidewall was clamped to the tool, the bolt and washer inserted into the box cavity, and the bolt tightened to a fixed torque. Ceramic fiber cross pins are used to stabilized the cores laterally. The cross pin holes were reamed by hand to fit, and the tapered fiber pin cemented in place. The cores were capped after all cross pin connections were made. The assembly process is shown on Figure 3 of Hill and Angel2.

3.2. Process Changes and Observations

With some minor changes, the same tooling and techniques used to make the MMT mold were used for Magellan. One change was modification of the post collar and the floorboard joining method. Just before placement in the mold, the free surface of the post collar is always smeared with adhesive. The machined box, with attached post collar, is immediately pressed onto the floorboard using the placement tool. The adhesive used in the MMT mold was Carborundum {\em Fibermax} patching compound. To help stop glass leaks in the Magellan mold, a 100 mm hole was bored into the floorboard down to the tile surface. A stub of strong post collar stock was cemented directly to the tile and machined flush to the floorboard. The free surface of the post collar was now smeared with cement and the box glued to the strong stub. Because of the high strengths of the material and the cement joint, this composite joint alone was theoretically able to carry the maximum (upward) core buoyancy force of 450 N. This new joining method should help insure that a strong, gap-free joint is maintained during the glass melt flow stage, and so fewer leaks are expected in the Magellan mold.

Since both molds were geometrically identical, all fit SiC parts recovered from the MMT mold were used in the Magellan mold.

The most significant departure from the MMT involved use of soft refractory material prefired to 1180 C ° by the manufacturer and use of a new, specially post-treated hex box material. The MMT parts were prefired in the 10 m furnace. The new box material gave cores with much higher compressive and flexural (tensile) strength and reduced core-to-core strength variability. A strength improvement was needed for Magellan because the compressive washer/core contact (bearing) stress in some MMT partial cores exceeded the design limit stress by ~40 %. A selected property data summary for MMT and Magellan hex preforms used to make cores is shown in Table 1.

Preform Avg. Compressive Strength
at Damage Limit
(MPa)
Avg. Flexural Strength
(3 point bend)
(MPa)
Avg. Axial Burst Force
(corresponding to buoyancy
force at failure)
(N)
MMT 0.101 0.384 975
Magellan 0.384 0.7941,827

Table 1: Selected Material Properties of 6.5 m Hex Box Preforms

After the 2-week prefire of the assembled mold, an inspection revealed delaminated partial cores, which necessitated a 4-month mold repair cycle, much of which was failure analysis and testing of repair processes. This was completed on January 11, 1994. There was no observable delamination of the one-piece hexagonal cores.

Approximately two thirds of the Magellan partial cores showed a delamination-type crack in the board along the board-hex box or board-board glue joints. The delamination occurred sometime during or after the mold prefire step. A few delaminations were found in the core caps along the cap-core glue joint.

Crack growth was empirically proven time-dependent. Forty-five cracks were found in partial cores during inspection completed a few days after the prefire. Thirty-seven previously undelaminated partials showed cracks along the glue joint after a second inspection was completed several weeks later. Because of time-dependent crack growth, every partial core and every core cap were considered potentially defective at some level.

The cracks were believed caused by a destructive superposition of cement shrinkage and material property differences between the mated box and board sections. Specifically, the expansion coefficient, high temperature shrinkage rate, stiffness and strength of board were different from that of the box at some variable level, with the board always weaker than the box. Although the property differences between box and board were small and some apparently changed in sign (e.g., expansion coefficient and shrinkage rate), an unfavorable superposition could open and drive a delamination-type crack. Time-dependent crack propagation is well known in this material. Comparison of core tops with failed and good glue joints showed that failure occurred most commonly with board lacking a strong ``rind'' which is naturally created by the manufacturer's post-treatment process.

Manufacture of all-new hex boxes, boards and glue to make new cores that wouldn't delaminate was considered to be a time-consuming and risky option because the manufacturer had no experience in solving this type of problem. Rebuild of delaminated cores with some ``all-new'' cores made from in-house stock was considered most attractive, and rebuild methods were developed at top priority. From this priority effort, new patching and part construction techniques were perfected and the finished parts were aggressively proof tested and qualified for use.

With exception of the triangle-shaped cores made from board stock only, all partial cores with joints that could potentially be breached and filled by glass melt were pulled from the mold. The residual joint strength was tested by inserting a flexible air bladder into the core interior and pressurizing it to induce an average tensile joint stress of 14 kPa. If the board split from the box section, the gluing surfaces on box and board were impregnated with rigidizer to give a strong, thick rind layer to resist delamination stresses. The board was then re-glued to the box section and the two edges pinned through and into the box with four sets of mullite ceramic pins (four pins along each edge for a total of eight pins). Boards that remained attached to the core after pressurization were pinned-only. The sidewalls of the triangular cores were pinned together at the corners while still in the mold because they were easily accessed and removal would likely damage them beyond repair. All new and patched cores were pre-fired to 1180 C ° and re-tested to a 14 kPa joint stress with the same pressurized bladder method used in the patching pre-test. Cores judged too damaged for patching or cores that failed the proof test were replaced by new cores. All caps were pinned through to the core with three mullite pins placed symmetrically around the edge at 120 degree intervals.

After all repairs were completed, the mold was inspected, cleaned and prepared to accept 10,233 kg of E6 borosilicate glass chunk. Once the glass was loaded on top of the mold, the furnace and lid and were lowered into position and the casting started. Glass loading is shown in Figure 6 and the casting and annealing cycle is shown in Figure 8 of Hill and Angel2. Based on real-time video images of the casting, the Magellan mold held up well with no evidence of significant glass leaking or of significant bubble content.

4. Conclusions and Future Plans

Two successful 6.5 meter mirror blanks have now been cast at the Steward Observatory Mirror Laboratory. These castings have demonstrated that the technology to cast honeycomb blanks of this size is now in hand. Once the Magellan blank is moved into the washout fixture, work will commence to upgrade the rotating furnace to 8.4 meter capability. This effort should be complete by November 1994 at which point construction of the 8.4 meter F/1.14 LBT primary mirror mold will begin. Casting of this blank, the world's largest honeycomb monolith, is expected to begin in late summer 1995. Current plans call for casting a second 8.4 meter LBT blank following the first one, although the possibility exists for casting one or more additional 6.5 meter blanks, depending on funding at the Mirror Lab.

5. Acknowledgements

The production of the 6.5 meter mirrors is a team effort in which dozens of people are involved. In addition to the authors, the following Steward Observatory staff members were involved in the cleanout of the MMT mirror and the casting of the Magellan mirror:

E. Anderson, J. Astier, D. Baxter, D. Campbell, M. Cline, J. Collins, R. Cordova, S. DeRigne, D. Dan, W. Davison, K. Duffek, P. Esterline, M. Hunten, K. Johnson, K. Kenagy, D. Ketelsen, A. Klocko, R. Kraff, R. Lutz, R. Meeks, P. Muir, B. Powell, S. Schaller, E. Smith, W. Stoss, P. Strittmatter, T. Trebisky, D. Watson and S. West.

Thanks to Lori Stiles for providing the photographs in this paper.

6. References

  1. Angel, J. R. P. and Hill, J. M. 1982,
    ``Manufacture of large glass honeycomb mirrors'',
    S.P.I.E.,332, pp. 298-306.

  2. Hill J. M. and Angel, J. R. P. 1992,
    ``The casting of the 6.5m Borosilicate Mirror for the MMT Conversion'',
    {\it Proceedings of the ESO Conference on Progress in Telescope and
    Instrumentation Technologies}, ed. M.-H. Ulrich, (Garching:ESO), pp. 57-66.

  3. Anderson, D. S., Martin, H. M., Burge, J. H. and West, S. C. 1994,
    ``Rapid fabrication strategies for primary and secondary mirrors at Steward
    Observatory Mirror Laboratory'',
    S.P.I.E., 2199 (These proceedings).

  4. Chaffee, F. H., Foltz, C. B. and Williams, J. T. 1994,
    ``Progress report on the MMT/6.5-m conversion project'',
    S.P.I.E., 2199 (These proceedings).

  5. de Jonge, P. 1994,
    ``Status of the Magellan Project'',
    S.P.I.E., 2199 (These proceedings).

  6. Parodi G., Hill J. M. and Salinari P. 1992,
    ``Supporting the 8.4m honeycomb mirrors of Columbus'',
    Proceedings of the ESO Conference on Progress in Telescope and
    Instrumentation Technologies
    , ed. M.-H. Ulrich, (Garching:ESO), pp. 301-306.