Thermal cycle stability of glass-to-metal seals with glass preforms produced via powder-metallurgy and casting-machining methods

In order to evaluate the influence of preform preparation processes on thermal cycle stability of glass-to-metal seals, this work embraced two different methods to produce the preform for seals. For the conventional powder metallurgy (PM) method, the molten glass was quenched to form frits, then the frits were ball milled to prepare glass powders. These glass powders were pressed into green bodies and heated to prepare preforms. While for the casting-machining (CM) method, the molten glass was cast into a graphite mold and annealed before accurate machining to preforms. In contrast to the PM method, the CM method provided an ultralow-porosity preform structure and a low porosity glass seal region. Field emission scanning electron microscope (FE-SEM) was conducted to investigate the bubbles and cracks in glass region. Furthermore, thermal cycling tests confirmed that these two tremendously different glass regions strongly affected the thermal cycle stability of the seals. To support the understanding of cracking in seals, the damage features of the samples were observed by FE-SEM and the extended finite element method (XFEM) was used to simulate the crack initiation and propagation. The experimental results demonstrated that cracking in the seals made from CM preforms occurred in the glass region near the sealing interface. However, cracks initiated from the bubbles in the seals made from PM preforms, which was verified by the XFEM simulation results. In addition, the CM seals demonstrated little degradation of the leakage rate until 105 thermal cycles, while cracking was found in the PM seals after 70 thermal cycles, indicating a decreased thermal cycle stability and resulting in hermetic failure.


Introduction
Glasses can be bonded to metals to form hermetic seals for use in a variety of applications, such as electrical penetration assemblies (EPAs) [1], medical implants [2], solid oxide fuel cells (SOFCs) [3], etc. Among them, EPAs have recently garnered broad interest in regard to both fundamental scientific research and practical technological applications [4]. As a typical glass-to-metal seal, EPAs have been successfully employed to prevent gas leakage and provide electrical insulation while allowing electric conductors to pass through the containment structure [5]. Maintaining hermeticity has always been a challenge for EPAs due to their high operating temperature during long-term service [6,7]. It has been generally recognized that hermeticity is related to the stress distribution, cracks, and interfacial defects [8,9]. Once one of the flaws achieves a critical condition, catastrophic failure will occur.
Most of the glass preforms for glass-to-metal seals are produced traditionally by the powder metallurgy (PM) method, normally including raw powder mixing, melt quenching, ball milling, pressing, binder burnout, and the final sintering process [10][11][12]. As with many glass-to-metal seals prepared by the PM method, numerous bubbles and microflaws inevitably appear in the glass region after sealing [1,10]. These bubbles within the bulk could potentially decrease the tolerance of the seal to thermal cycle stress [11]. In addition, the bubbles located in Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. the glass region are possibly hazardous, which may result in the loss of hermetic sealing and an increase in the corrosion susceptibility of the seals due to fracture path generation. Therefore, alternative methods capable of reducing bubble formation and producing dense glass are deemed essential for improving the thermal cycle stability of seals.
Over the past few decades, extensive research has been carried out on eliminating bubbles within glass for glass-to-metal seals [13,14]. Following the work by Peitl et al [15], crystallization-induced bubbles need to be considered for designing and manufacturing low porosity glass. Although bubbles in the local region can be reduced using the doping method [14,16,17], the density of sealed glass is still far from the impressive high density of other kinds of glasses that have ultralow porosity [18,19]. The casting-machining (CM) method is one of the most commonly used methods owing to its simple processability [20] and effectiveness in reducing deleterious bubbles.
Since there were numerous bubbles in the glass region prepared by the traditional PM method, a novel CM method, which provided a simple process, low cost, and easy realization of an industrialization preparation, was performed for the preparation of glass preforms in this work. To contrast the thermal cycle stability of glass-tometal seals, glass preforms were manufactured using both the CM method and the conventional PM method. The porosity of the preforms, glass seals, and interface regions of the seals were morphologically analyzed using a series of techniques. In particular, the effect of two vastly different glass preforms on the thermal cycle stability of the seals was studied over a temperature range from room temperature to 350°C. As an effective numerical simulation method for cracks [21], the extended finite element method (XFEM) was used to simulate the stress concentration near a bubble to help understand the crack initiation mechanism. The objective of this work is to investigate the influence of the glass preform production methods on the porosity and thermal cycle stability of glass seals. The comparison of porosity and thermal cycle stability between these two methods were systematically discussed, which can broaden the manufacturing process for glass preforms and become a provision for reference to practical production.

Experimental procedures 2.1. Sample preparation
The sealing glass was fabricated by a thoroughly mixing process that has been previously described in detail [22]. Figure 1 shows a schematic diagram of the production processes using the PM and CM methods.
The process parameters of these two methods were determined through the previous studies of our group [22][23][24]. Briefly, molten glass with the appropriate composition was prepared using reagent-grade oxides and carbonates as raw materials. Planetary ball mill was used to obtain homogeneous mixtures of raw materials. The mixtures were melted in batches of 300 g using platinum crucibles at a temperature of 1500°C. After a holding time of 30 min, one part of molten glass was rapidly poured into cold deionized water to form frits for the PM method. The obtained glass frits were ball milled in a milling container and mixed with a suitable amount of organic binder. These glass powders were pressed into green bodies and heated to 600°C for 6 h to remove the organic binder.
For the CM method, the remaining portion of molten glass was cast into a graphite mold and transferred to a resistance furnace for annealing heat treatment. The furnace was heated to 530°C and held for 24 h to reduce the residual stress of the glass. Finally, the casting glass with less residual stress was machined into preforms with dimensions of 40 mm in diameter and 20 mm in height. Before the sealing process, 304 stainless steel (304SS) shells were subjected to a preoxidation process at 1050°C under an oxygen partial pressure of 228 mTorr for 45 min, which was reported in our previous publication [25]. Then, all the glass preforms and preoxidized 304SS shells were thoroughly cleaned prior to assembly. The assembled components were heated to 930°C in a tube furnace with a flow rate of N 2 at 1000 ml min −1 .

Thermal cycling test
To determine the thermal cycle stability of the PM and CM seals, thermal cycling tests were conducted in a muffle furnace. After heating to 350°C, the samples were cooled to room temperature.

Characterization
Each sample batch was ground on 500-2000 SiC papers and then mirror-polished, followed by ultrasonic cleaning using ethanol to remove the surface impurities. Cross-sections of all mirror-polished seals were microscopically imaged by a field emission scanning electron microscope (FE-SEM) (SU8220, HITACHI, Japan) operating at 20 kV. The areal porosity (ratio of the area of the pores or bubbles to the total area of the sample) of the glass was measured by the area analysis method conducted on at least ten micrographs (obtained at a magnification of 500X) using Image-Pro software.
At each interval of the thermal cycling test, hermeticity tests were performed using a helium mass spectrometer leak detector (SFI-231, WAYEE, China) to detect the hermeticity degradation of the seals. Then, the final damage features of the samples after the thermal cycling tests were observed using FE-SEM.

Finite element analysis
To explore the influence of bubbles on cracks, a finite element model was established. XFEM was used to simulate crack initiation and propagation, as it allows the position of the crack in the material to be completely free, rather than relying on mesh generation. The glass was simulated as a linear elastic material and would break after a certain tensile strength was reached, with a Young's modulus of 76300 MPa measured by compression method using a universal testing machine (SHIMADZU, WDW-100, Jinan, China) at a loading rate of 0.2 mm min −1 and Poisson's ratio of 0.33, which had been confirmed in our previous study [4,26]. Excessive first principal stress is the main cause of glass fracture, so the maximum principal stress (MAXPS) criterion was selected. The MAXPS was set as 40 MPa in this study.

Porosity analysis
Representative cross-sectional images together with the average porosity of the preforms, glass seals, and glass/ metal interface using the PM and CM methods are shown in figure 2. Before sealing, numerous irregular pores, 1 ∼ 3 μm in size, were visible in the PM preforms, resulting in an average porosity of ∼0.3% ( figure 2(a)). After sealing, 6 ∼ 9 μm bubbles appeared in the glass seals, showing a significantly increasing average porosity from 0.3% to 3.9%, while the number of bubbles was 40% less than that of the PM preforms ( figure 2(b)). Close inspection of the cross-sectional images revealed that these bubbles in the glass seals had a higher degree of sphericity than that of the pores in the PM preforms. Generally, a porosity as low as possible is desired. However, the pressing and binder burnout processes inevitably result in residual pores and supersaturated gas. These small pores in the preforms, as bubble nucleation sites, can grow into larger bubbles due to the escape of dissolved gas in molten glass, expansion of the glass by high temperature, and agglomeration among adjacent bubbles. It is important to note that the bubbles in figure 2(c) were not located only at the glass/metal interface. Instead, they were uniformly located over the glass region.
For the CM preforms ( figure 2(d)), one of the striking differences observed between these two preforms was the porosity. No pores were found in the CM preforms, which showed an ultralow average porosity (near zero). Notably, there was no organic binder, which could bring a number of pores and bubbles when burnout [10], introduced in the preparation process of preforms by the CM method. In addition, before the casting process, the glass was melted at 1500°C for 30 min. The low viscosity of glass at 1500°C promoted the escape of bubbles from molten glass. The subsequent annealing heat treatment and machining processes are hardly capable of introducing pores into the preforms. After the sealing process, the average porosity of the glass seals (figure 2(e)) was also near zero, indicating that no bubbles formed during the sealing process. Figure 3 shows the leakage rates of the seals produced by the PM and CM methods as a function of thermal cycles. As illustrated, before the thermal cycling test, all the as-fabricated seals had great hermeticity, and the initial leakage rates of all seals were 1 × 10 -13 Pa·m 3 s −1 . After 70 thermal cycles, the leakage rate of the PM seal dramatically increased to 1 × 10 −3 Pa·m 3 s −1 , and the seal almost failed. After 75 thermal cycles, the PM seal remained at 1 × 10 −3 Pa·m 3 s −1 . Compared to the PM seal, the CM seal exhibited better thermal cycle stability; it remained at 1 × 10 −13 Pa·m 3 s −1 until 100 thermal cycles and even remained at 1 × 10 −4 Pa·m 3 s −1 after 110 thermal cycles.

Failure analysis of seals
To reveal the initial cause for this difference in thermal cycle stability, FE-SEM was used to observe the seal microstructures after leakage. Figure 4(a) shows the FE-SEM images of cracks in the glass seals prepared by the PM method. As illustrated, there were cracks on the glass surface that propagated along the bubbles in all directions, which was due to the stress concentration around the bubbles. Upon thermomechanical fatigue, micro-and macroflaws may experience crack initiation and propagation first. Eventually, the numerous bubbles, such as macroflaws, and thermomechanical fatigue synergistically promote cracking in the bubble-rich glass region. Figure 4(b) shows the XFEM simulation results of the stress distribution in the glass prepared by the PM method. The stress concentration occurred around the bubble, which played a decisive role in crack initiation. The combined effect of stress concentration and thermal-mechanical fatigue led to cracks in all directions and finally induced failure in hermeticity. Figure 4(c) shows the microstructure of a seal prepared by the CM method after the thermal cycling test. Interestingly, an obvious circumferential crack appeared in the glass area along the interface, and no bubbles were found by further amplification, which means that the macroflaws in the CM seal were almost nonexistent.  Although the CM seals also suffered thermomechanical fatigue, dense structures without macroflaws, such as bubbles, in the CM seal prevented microcrack formation and propagation. Accordingly, the CM seal exhibited better hermiticity than the PM seal. However, the CM seal failed after undergoing more thermal cycles, mainly due to the mismatched thermal expansion coefficient (TEC) between the metal shell (TEC = 19.3 × 10 −6 /K, 25 ∼ 350°C) and glass (TEC = 7.6 × 10 −6 /K, 25 ∼ 350°C), leading to thermomechanical fatigue in the seal. Normally, when the seals achieved the fatigue limit, cracks first occurred in the region with the largest dimensional change, namely, the interfacial region. However, the adhesion at the interface was enhanced because of the preoxidation of steel and the dissolution of the oxide [25,27]. Moreover, 304SS with a higher TEC contracted faster during cooling [28], and inward compression led to compressive stress in the glass, which was concentrated in the glass region near the interface. The glass near the interface could not withstand this excessive compressive stress, and finally, thermomechanical stress cracks appeared inside the glass, which is in good agreement with the results reported by Li et al [29].

Conclusions
In summary, glass-to-metal seals with higher thermal cycle stability were successfully produced via the CM method. After the sealing process, the CM seals showed a low porosity glass structure due to the few residual pores in the preforms. Upon thermomechanical fatigue, the denser glass seals without macroflaws prevented crack initiation and propagation induced by thermomechanical stress, strongly increasing the thermal cycle stability of the seals. For the PM method, there was stress concentration around the bubbles. The cracks initiated and propagated in any direction once the maximum stress exceeded the mechanical strength of the glass, which was also confirmed by the XFEM simulation results. By comparing these two different glass preform processing methods, this study sheds light on the key microstructural variables responsible for the thermal cycle stability of glass seals.