Usually the epoxy adhesive’s nominal shear strength is somewhat lower than its tensile strength, so several samples were designed for individual tensile and shear tests, as shown in Figure 1, with a bonding size of ∼500 mm² for the former, and ∼350 mm² for the latter.

The axial samples are designed as those of the TMT M1 and the shear samples follow the Chinese National Standard for testing an adhesive’s shear strength GB/T 7124–2008 (same as ISO 4587:2003). For glass-metal joint applications, high strength, low stress, and viscous epoxy types are preferred (here, EC 2216 B/A Gray from 3M Corp. was used) and controlled to a bond line thickness of ∼0.25 mm. The glass is fused silica from Corning and the metal pads are made of invar 4J32.

For avoiding the undesirable three damage styles in the glass-metal hetero-bonding interface, including the separation of glass/epoxy interface, the separation of metal/epoxy interface and the strength failure of glass sub-surface damage layer, namely, pursuing higher epoxy potentiality, a universal high quality bonding process is strongly recommended to be carried out in a clean room with a venting system to protect operators from hazardous chemicals such as HF and acetone, with as less pollution as possible. Operators should wear safety glasses, a face mask with organic/acidic filters, and plastic apron/gloves for protection. The work flow is outlined in Figure 2 and more detailed operations are presented by Laiterman et al. (2010). Some important notes for the key bonding processes of Figure 2 are given below.

1) Dust cleaning and optics protection: clean the glass and the metal with ethanol; paste protecting masks onto the optics, leaving the bonding regions uncovered.

Based on verifying the success of the coupons obtained from the above processes, mechanical strength tests can be performed, followed by bonding technique evaluation and optimization when needed.

Strength tests were carried out by a Chinese National Standard testing machine DDL10 with a load capacity of 10 kN at a resolution of 0.1 N, as shown in Figure 3. As the force was applied along a relatively long axis, two spherical hinges were involved to avoid creating any moment load.

Sample tensile strength values a _i are assumed to be normally distributed, and the corresponding population average a ¯ would follow the t-distribution, as explained in the Appendix. Then the limits of the average a ¯ L and the lower single strength L T could be determined by Eq. (1): a ¯ L = E a − T p , v S a / n L T = a ¯ L − T p , m − 1 S a , (1) where n is the sample number, m is the number of pads to be bonded, a is a vector consisting of tested strength values, E( a ) and S( a ) are the mean value and the standard deviation of a respectively, T _p,v is the single-side strength distribution region factor at probability p for n samples with v = n − 1 degrees of freedom (DOF), and T p , m − 1 is the factor at probability p for m pads. Here it defines p = 1 − 1 / m for presenting the marginal reliability, whose physical effect means that less than one (namely, none) of the m pads should break under the allowed stress (this stress itself usually has a safety factor of ≥4 in engineering application). It’s noted that for the tensile test m = n, and for the shear test m = 2n

In double-lap shear tests, breakage occurs at one end and the strength x _i of the other end is unknown. The only knowledge is that it should be higher than the broken end’s strength a _i , expressed as x i = a i + t i T p , v S a , with  t i ∈ rand 0 , 1 . (2)

By setting up a parameter optimization model, the lower limit of shear strength L _S can be predicted as Find  t = t 1 , t 2 , · · · , t n M in . L S = E ¯ L t − T p , m − 1 S t St .     t i ∈ rand 0 , 1 E ¯ L t = E t − T p , 2 v + 1 S t / 2 n E t = ∑ i = 1 n a i + x i 2 n S t = ∑ i = 1 n a i − E t 2 + x i − E t 2 2 n − 1 , (3) where E t is the n tested sample average strength, E ¯ L t is the expected average limit for the whole 2n strength values(Note: n tested ends and n opposite ends), and S t is the 2n values’ variance.

The SSD layer was removed by the etching process to strengthen the bonding.

Two pieces of φ150 mm glass were tested for evaluating the etching performance of the acid cream: one was etched horizontally, representing the case of axial pads, and the other was etched vertically, representing the case of lateral pads. Both were masked and a φ25 mm circular region was left open in each quadrant, as shown in Figure 4A, with the numbers 1, 2, 3, and four representing both the quadrant sequence and etching cycles. A Stylus Profilometer NanoMap 500LS from Aep Corporation was used to test the top-etched and the bottom-ethced regions for comparison, as shown in Figure 4B.

Surface profile information is provided in Table 1, where the variables hi and Ri represent the pit depth and surface roughness of the ith quadrant, respectively. Generally speaking, the vertical etching is weaker than the horizontal, not only in depth, but also in roughness. The pit depth increases in approximately linear fashion with the number of etching cycles. It is also noted that a higher number of etching cycles will produce a much rougher surface.

The final figure after the fourth etching is shown in Figure 5. It shows that vertical 4-cycle etching generates a pit depth of 29–36 µm with a roughness of 9.5 µm, while horizontal 4-cycle etching achieves a 35–38 µm pit depth with a roughness of 12.7 µm. According to the widely-accepted assumption that the thickness of the SSD layer is about 1.5–3 times that of the final grit size used for material surfacing, four, etch cycles are sufficient to thoroughly remove the SSD layer of glass surfaces ground/polished using a grit size smaller than 9 µm.

An unanticipated and interesting effect is found: the step width of the, etch zone edges is ∼1 mm for the vertical case, and is ∼3 mm for the horizontal case. This may be attributed to the fact that as we apply a large amount of, etch cream, with much of it exceeding the mask hole, the effective chemical component (such as HF) continuously flows into the etched zone under gravity, supplementing the consumed chemicals during the corrosion process.

Conservatively, this leads to the requirement that the mask hole be ∼5 mm larger than the bonding edge, which has been described in item Eq. 1 of the notes in Section 2.1, as some additional space is needed to compensate for mask locating errors. Bonding the pads to the relatively flat etched zone will produce more uniform glue thickness and strength.

Hollow 250 µm glass beads were scattered onto the pad-applied adhesive for thickness control at a weight ratio of approximately 0.5%. A Coordinate Measuring Machine(CMM) test of six bonded samples was carried out to check the bond line thickness and uniformity, as shown in Figure 6 and Table 2.

The largest relative error of the bond line thickness measured is less than 10%, which may contribute to achieving a relatively uniform bonding strength.

Priming is required to minimize the contamination risk of the glass bonding surface and to further guarantee good bonding. The primer was applied by optical swabs (TX-714A form Texwipe), and the primer thickness t _p was evaluated roughly by computing the fringes of the sodium light (wave length λ = 546 nm), as shown in Figure 7, where (a) illustrates the test principle of interferometry, (b) shows the initial fringe number N ₀ ≈ 3 from a viewing angle of 45°, and (c) shows the fringe number N ₁ ≈ 12 after priming viewed at the same angle. With the application of interferometry, t _p can be calculated by Eq. 4 as t p = λ ∆ N ⁡ sin   θ / 2 (4)

Substituting ΔN = N ₁-N ₀≈9 and θ ≈ 45° into the equation, the result is t _p ≈ 1.74 µm. As the primer works best at thicknesses in the range 1.5–2.5 µm, this control method proves satisfactory.

The universal scheme for locating pads precisely onto optics is shown in Figure 8. For positioning the axial pads in Figures 8A, B, the magnet attracts the metal pad, stiffened by an iron head (optional, but important for diamagnetic material) and the pad position is adjusted by the double precision screws, each of which is fastened by a spring, according to the reading of the tracker ball. In Figures 8C, D, the lateral pads are adjusted by four precision screws, two of which are vertical and the other two are horizontal. Another two plastic screws are installed to provide pre-pressure during adhesive curing. In practice, as shown in Figure 8D, an additional elastic buffer is placed between the pad and one horizontal screw to avoid applying too high a fastening force by restraining the clearance of the horizontal screws, similar to the function of the springs in Figure 8A.

Locating precision tests for such jigs was performed for 21 pads (both axial and lateral), and all results are shown in Figure 9. The location error of axial pads ranges from −0.063 mm to 0.061 mm, averaging 0.003 mm, with a variance of 0.038 mm. The lateral pads location error ranges from −0.058 mm to 0.119 mm, averaging 0.013 mm, with a variance of 0.042 mm.

All errors are within the distribution region of 3σ, so it is believed that no unacceptable value exits. Such small position errors have little effect on bonding strength uniformity.

For predicting the bonding strength limits, six samples (n = 6, ν = 5) were prepared for tensile and shear tests, and all 12 strength values are listed in Table 3, with all undefined variables presented in the mm-N-s unit system hereafter.

For the tensile case, Eq. 2 can be applied directly for the limit calculation, and the results are shown in the second row of Table 4, where E is the sample average strength, E ¯ L is the expected average limit for the whole population, S is the sample variance, and L is the population strength limit.

For the shear case, the optimization model defined by Eq. 3 is adopted for the limit searching procedure. The iterations were driven by the Multi-Island Genetic Algorithm followed by Sequential Quadratic Programming algorithm. Based on successful optimization implementation, all inputs and outputs for predicting shear strength limits are listed in the third row of Table 4 and shown in Figure 10, with p = 0.9444 for an 18-point Whiffletree structure (m = 18), p = 0.9630 for a 27-point Whiffletree (m = 27), and p = 0.9995 for 82 mirror segments, which is 1/6 symmetry of TMT’s primary mirror and are all mounted using 27-point Whiffletrees (m = 2,214).

Table 4; Figure 10 indicate that the strength limits strongly depend on the reliability related to the quantity of pads involved in the opto-mechanical structure.

The proposed bonding technique is expected to have an average tensile strength of 14 MPa, a lower limit of 12.0 MPa for the 18-point Whiffletree application, 11.7 MPa for the 27-point Whiffletree, and 6.67 MPa for the 2,214 pads in the 82 copies of 27-point Whiffletrees. The corresponding lower shear strength limits in these applications are 10.3, 9.99, and 4.28 MPa, respectively, which demonstrates much stronger reliability dependence on the quantity of pads.