Glass-contact refractory of the nuclear waste vitrification melters in the United States: a review of corrosion data and melter life

ABSTRACT

The performance of the refractory lining in glass melters used for nuclear waste vitrification is critical to the melter reliability for long-term continuous operation. Monofrax® K-3, a high Cr₂O₃ fused cast refractory material, has been widely used to build the liners of nuclear waste glass melters in the United States. Corrosion behaviour of Monofrax® K-3 refractory has been evaluated based on crucible-scale testing, inspection of the refractory components following scaled melter testing, and inspections of the Defense Waste Processing Facility (DWPF) melter refractory after service. The literature generally consists of empirical models based on short-term testing to describe refractory corrosion dependence on glass composition. Corrosion data from tests with longer testing times, at various temperatures, in the presence of molten salts, and with different redox reactions in the plenum atmosphere exist, may be insufficient to provide accurate refractory service life estimates. Additionally, the corrosion data collected under actual and scaled melter operating conditions are limited. Recommendations to achieve more direct correlation between the laboratory refractory corrosion data predictions and the observed melter service life are discussed to allow for more accurate predictions of the useful life of melter refractory linings.

KEYWORDS:

• Refractory
• nuclear waste
• glass melter
• corrosion
• waste treatment
• vitrification
• glass property model
• glass formulation

Acronyms

CLSM	=	continuous laboratory-scale melter
DFLAW	=	direct feed low activity waste
DM	=	DuramelterTM
DOE	=	U.S. Department of Energy
DWPF	=	Defense Waste Processing Facility
HLW	=	high-level waste
IDMS	=	Integrated DWPF Melter System
INL	=	Idaho National Laboratory
JHCM	=	joule-heated ceramic melters
LAW	=	low-activity waste
LFCM	=	Liquid Fed Ceramic Melter
LLW	=	low-level waste
LSFM	=	Large Slurry Fed Melter
ORP	=	Office of River Protection
PNNL	=	Pacific Northwest National Laboratory
RSM	=	research-scale melter
SCM	=	Small Cylindrical Melter
SGM	=	Scale Glass Melter
SRNL	=	Savannah River National Laboratory
SRS	=	Savannah River Site
VITPP	=	Vitrification Pilot Plant
VSL	=	Vitreous State Laboratory at The Catholic University of America
WTP	=	Waste Treatment and Immobilization Plant
WVDP	=	West Valley Demonstration Project

Introduction

The United States (U.S.) Department of Energy (DOE) is responsible for managing high-level waste (HLW) generated since the 1940s from nuclear materials processing. The HLW originated as liquid process effluent and is stored in underground steel tanks to await final treatment and disposition. The majority of the HLW is primarily at four sites: the West Valley Demonstration Project (WVDP) near West Valley, New York; the Idaho National Laboratory (INL) near Idaho Falls, Idaho; the Savannah River Site (SRS) near Aiken, South Carolina; and the Hanford Site near Hanford, Washington [1]. Approximately 4 × 10⁵ m³ (96 million gallons) of HLW has been generated and stored as Tank Waste; however, only a small fraction of the volume of Tank Waste under DOE control has been treated to date [1–3]. DOE has ongoing missions to retrieve, treat, and dispose of these wastes in a safe and efficient manner [4,5].

The DOE processed HLW at the WVDP Site between 1996 and 2001, converting the Tank Waste into borosilicate glass that is canistered in stainless steel and currently stored at the site [6–9]. At the INL Site, the majority of tank waste has been converted (calcined) to a solid material that is stored in stainless steel bins to await final treatment and disposition, while the remaining portion of the liquid HLW at INL is to be converted (steam-reformed) to a solid material that will be packaged in stainless steel canisters and stored in concrete vaults at the site [10]. The tank waste at both the Hanford Site and SRS represents the majority of the HLW remaining across the DOE sites that requires treatment and disposition [11–17]. After decades of storage and conditioning, the tank waste is generally in the form of mixtures of solid (sludge) and liquid (salt solution); in many tanks, the oversaturated solution formed a salt cake layer [18–20].

Treatment of tank waste involves two primary processes: pretreatment and immobilization [21]. Pretreatment involves certain unit operations to separate the low-activity waste (LAW) fraction from the HLW components and to prepare the liquid waste for feeding to immobilization processes. Immobilization is the process that converts the liquid waste into a stable solid form. The purpose of separating LAW and HLW is to reduce the volume of HLW. Immobilized HLW will be shipped to a geologic repository for disposal, while immobilized LAW is suitable for on-site disposal [15,21]. It is important to note that LAW should not be confused with low-level waste (LLW) as defined by the International Atomic Energy Agency [22]; LAW is the waste that remains following the process of separating as much radioactivity as is practicable from HLW [15].

Vitrification is the established technology to immobilize HLW into a glass waste form that possesses the requisite chemical durability and radiation stability for long-term storage and disposal [23–28]. Vitrification was selected for HLW treatment at WVDP, SRS, and the Hanford Site, [1,2,4,5]. At INL, vitrification is being considered to immobilize the Tank Waste that was calcined and currently stored in bins below ground at the INL Site [29,30]. Aside from these HLW vitrification projects, two other treatment projects in the U.S. have applied vitrification process to immobilize mixed low-level waste and geological mill tailings [4,5,31–36].

Table 1 lists the major nuclear waste vitrification projects in the US. A general joule-heated ceramic melter (JHCM) design with continuous feeding and glass discharging layout (Figure 1) has been adopted for the large-scale vitrification facilities given the high production capacity and stable processing capability [37–45]. These JHCMs are lined with ceramic refractory brick to contain the glass melt. The glass melt is heated by an electric current between electrodes submerged in the glass pool. Melter feed, which is a mixture of waste and additives in slurry form, is charged into the melter and partially covers the melting surface forming a ‘cold cap’ as ambient temperature feed is introduced on top of the molten glass. A vertical temperature gradient exists that extends from the glass pool up through the cold cap region. Convection can be also forced through the glass pool through differential control of the upper and lower electrodes. The cold cap region controls the feed-to-glass conversion reactions and is critical to managing volatile species during glass production [46–53]. Glass is drawn from near the bottom of the melter and discharged be an overflow weir. Bubbling systems can be another important feature used to increase the glass production rate [54]. Bubbler systems are planned for the WTP and have been in place in the DWPF for over a decade [16, 55–57]. The design and operations of these glass melters influence attainment and life of the melter.

Figure 1. Cross-sectional view of Defense Waste Processing Facility (DWPF) melter representing a typical JHCM design. Reprinted with permission from Ref. [40].

Table 1. Major U.S. nuclear waste treatment projects using JHCM for vitrification. Monofrax® K-3 is used as the refractory lining of all the JHCMs in the table.

Project	Waste inventory	Operation timeline	Melter size and production rate in metric tons of glass per day (t/day)	Waste glass produced	Inspection of melter refractory	Notes
West Valley Demonstration Project (WVDP), near West Valley, New York [1, 4-9]	4 tanks, 2,200 m^3 HLW with high Fe, Al	1995-2002 Melter began non-radioactive operations in October 1995 and radioactive operations began in June 1996.	Glass pool surface area: 2.2 m^2 (23.7 ft^2, rectangular) Operation Volume: 0.86 m^3 Production rate: 0.7-1.2 t/day	∼ 570 t of glass was produced in 275 canisters (∼ 2.1 t glass per canister).	Cross-sectioning samples of the Monofrax^® K-3 bricks were taken for analysis. See section “Refractory corrosion in production melters”	Zeolite was used to remove the Cs-137 from the liquid waste (salt solution) separated from HLW; the effluent solution was concentrated and processed into cement waste form which was deposed as LLW.
Defense Waste Processing Facility (DWPF) at Savannah River Site (SRS), near Aiken, South Carolina [1, 4, 5, 11-14, 55]	51 tanks, 148,000 m^3 HLW with high Fe, Al	1994-present Melter began non-radioactive operations in April 1994 and radioactive operations began in March1996. Melter 1 serviced from 1994 to 2002; Melter 2 serviced from 2003 to 2017; and Melter 3 was installed in 2017 and is currently in service	Glass pool surface area: 2.6 m^2 (6 ft diameter cylindrical) Operation Volume: 2.2 m^3 Production rate: 2.2 t/day (1.3-1.4 t/day before bubblers installed in 2010)	As of 2022, over 7,700 t glass had been produced in 4313 vitrified waste canisters (∼ 1.8 t glass per canister). The total number of canisters needed to vitrify all the DWPF waste estimated by model is 8,170.	Both DWPF Melter 1 and Melter 2 were visually inspected after service. See section “Refractory corrosion in production melters”	The liquid waste (salt solution) separated from HLW is immobilized into grout form (salt stone)
Waste Treatment and Immobilization Plant (WTP) at Hanford Site, near Hanford, Washington [1, 11, 15-17]	177 tanks, 210,000 m^3 HLW with various chemical compositions in different tanks	The first phase of LAW vitrification, i.e., direct feed LAW (DFLAW), is projected to start in 2023 Vitrification of HLW is anticipated to start in 2033	LAW Melter: Glass pool surface area: 10 m^2 (rectangular) Production rate (estimated average): 15 t/day	N/A	N/A	The HLW and LAW (salt solution separated from HLW) will be vitrified in two separate facilities. There will be two parallel-operating LAW melter systems for the LAW vitrification facility and two parallel-operating HLW melters for the HLW vitrification facility.
			HLW Melter: Glass pool surface area: 3.75 m^2 (rectangular) Production rate (estimated average): 4.2 t/day
Vitrification Pilot Plant (VITPP) at Fernald Site, near Fernald, Ohio [4, 5, 31-35]	9,000 m^3 Geological mill tailings include high sulfur and high lead wastes	1996 Melter was “baked up” in May 1996; and failed after processing non-radioactive surrogate waste for seven months.	Glass pool surface area: 0.84 m^2 (3 ft by 3 ft rectangular) Operation Volume: 0.76 m^3 Production rate: 1 t/day	N/A	The leakage was caused by dissolution of molybdenum disilicide bubbler tubes installed in the melter floor and the molten glass flowed through the bubbler hole in the Monofrax^® K-3 layer. The inspection did not indicate any problems with the Monofrax^® K-3 layer in contact with the glass.	The operating temperature was 1350-1450 °C, while other vitrification melters worked at 1150 °C. After the melter failed, the waste that was planned to be vitrified was shipped off-site.
SRS M-area melter at Savannah River Site (SRS), near Aiken, South Carolina [4, 5, 36]	2,500 m^3 Mixed low-level waste	1996-1999 The first melter DM5000 was replaced after 5 months of operation to avoid potential glass penetration; the new melter DM5000A served from 1997 to 1999, which successfully completed the mission.	Glass pool surface area: 5 m^2 (rectangular) Production rate: 5-15 t/day	Over 4,500 m^3 (1.2 million gallons) of feed was processed and ∼ 1,000 t (2.2 million pounds) of glass was produced.	Visual inspections indicated that hot spots formed in the DM5000 melter were a result of refractory corrosion by molten glass near the refractory joints. Molten liquid had leaked from the refractory assembly (i.e., joints) during non-radioactive startup testing. The feeding of anhydrous borax directly to the melter when the melter was at high temperatures (1000 °C) resulted in the borax migrating, causing electrical shorts, localized joule heating, and damage to the refractory. DM5000A melter was inspected after service. See section “Refractory corrosion in production melters”

Melter life is of significant prominence among the many factors that influence waste vitrification facilities operational efficiency. The long duration (decades) of typical operations requires these JHCMs to possess a commensurate operational lifetime to reach attainment for the duration of the mission. Corrosion of the JHCM refractory lining in contact with the glass melt is a critical factor in estimating melter operating life [58]. Monofrax® K-3¹ refractory (Table 2), a fusion-cast refractory consisting primarily of chromium spinel and chrome-alumina corundum, has been selected to line all the JHCMs that have been, are, or will be used to treat nuclear waste in the U.S [58–65]. The operational experience, although limited, demonstrated that the refractory lining lifetime in the JHCMs can exceed ten years [66,67]. However, laboratory corrosion data collected for Monofrax® K-3 refractory, including crucible and scaled melter tests, vary widely; high corrosion damage (measured as corrosion depth at the melt line) observed in a few scaled melter tests has been used to suggest a much shorter service life; (defined as the service time when half of the brick gets corroded) on the order of 2–6 years (two years of lifetime was considered as a conservative limit for melter design) [67]. Taken together, operational experience indicates a deficiency exists in the methods used to estimate refractory lifetime from laboratory corrosion test methods. The ability to accurately predict refractory life in a JHCM using a scaled test is needed to accelerate the development and design of more efficient JHCMs. In this article, the literature on corrosion of Monofrax® K-3 refractory is reviewed as it relates to JHCMs used for nuclear waste vitrification in the United States. Technical challenges relating corrosion behaviour, especially at the melt line, to refractory service life and recommendations for future work are presented.

Table 2. Compositions of Monofrax® K-3 [64–66].

	Wt-%	Note
Chemical Composition
Al2O3	58.6
Cr2O3	27.1
Fe2O3	5.9	Analysis showed that the measured redox ratio Fe(2+)/ΣFe ≈ 92–94%, that is 5.4–5.5 wt-% of FeO of the reported 5.9 wt-% Fe2O3 [64].
MgO	6.1
SiO2	1.6
Phase Composition
Major Phase I: Cr2O3-Al2O3 solid solution (corundum structure)	60 - 65	18 wt-% none-spinel Cr2O3
Major Phase II: Complex spinel	35 - 40

Background: refractory loss mechanisms

Refractory lifetime in a glass melter is determined by the aggregate effects of all the mechanisms acting on the refractory during operation that contribute to material loss. Refractory loss typically occurs by mechanisms including dissolution, physical cracking, spallation/grain pullout, and upward drilling, all of which can be enhanced by thermal cycling and temperature gradients [58, 68–72]. Figure 2 schematically presents these important mechanisms. The design and construction of the refractory package must address the combination of these factors expected to be present when operating a glass melter. Many of the loss mechanisms exhibit significant random variability and are impractical to directly quantify. Instead, gross corrosion tests are the single most widely accepted method for measuring refractory loss. Models based on crucible corrosion tests are used to evaluate the refractory loss, but a direct correlation between crucible-scale corrosion tests and corrosion occurring in the melter has not been established [72]. Quantifiable correlations are likely ambiguous owing to different conditions at the melt line in the melter compared to that in the typical crucible corrosion test. For this reason, standard crucible corrosion test results cannot be used to directly extrapolate corrosion rates in melters or to quantitatively estimate melter refractory lifetime.

Figure 2. Representation of typical wear in refractories described in the literature [58, 71].

Corrosion at the melt line is known to be driven by surface tension gradients, which is the primary mechanism simulated by the crucible-scale tests and is observed as ‘neck corrosion’ [73]. Dissolution at the melt line is sensitive to conditions in the molten glass, including convection (forced or otherwise), causing dynamic thermal and chemical driving forces as the refractory is continually exposed to fresh molten glass, which may accelerate the refractory dissolution rate [74]. Vapour phase interactions with the refractory above the melt line are also known to contribute to refractory loss [75].

Fused cast refractory is particularly susceptible to material loss by spallation of bulk pieces, which generally fall into the melter [76]. Spalling implies the loss of fragments from a refractory shape caused by stress produced inside a refractory body [68,72]. Upward drilling occurs where a downward-facing refractory surface is in contact with gas bubbles trapped against it and takes place where open pores or cracks form in the refractory lining surface [68,77–79]. These structure anomalies include material defects introduced during the manufacturing process (casting the refractory bricks and lining the melter) such as pores, inclusions, and cracks, and the stress from thermal gradients (thermal shock) during melter operation [80,81]. Most catastrophic thermal shock effects occur during initial melter heat up and can be managed by proper control of temperature during the initial heat up. Cracks and fracturing arising from thermal shock are usually not a significant concern unless large pieces are displaced or the cracks between fractured pieces increase surface area or expose refractory to glass penetration [72]. When physical changes to the refractory become excessive, rising gas bubbles can be trapped on the underside of structural anomalies where they slowly continue to drill through the refractory over time [71]. This effect leads to enhanced refractory damage as corrosion and thermal shock continue to act on compromised areas.

Molten glass will also penetrate seams between refractory brick and enter existing cracks or those developed during operation. Depending on the geometry of the crack and the glass composition, such cracks can be sites for increased corrosion, but the glass can also serve to infiltrate and seal the refractory lining by the combination of the glass liquidus temperature, T_L, in the crack and the thermal gradients extending away from the melt. The penetration depth is limited by the tortuous path of the crack, the thermal gradient across the brick (resulting in increased viscosity as the temperature is reduced), and the increased viscosity from dissolution of chromium and aluminium oxides into the glass as it flows into the cracks and seams [69,82]. Melter refractory lifetime can be reduced by corrosion if fresh glass continually penetrates and washes out between the cracks and seams, dissolving refractory material into the molten glass in the process.

Some waste streams have high concentrations of salts, which promote the segregation of a molten salt phase during melting [83]. Refractory interaction with the molten salts must also be considered when considering refractory loss mechanisms [58,84]. In waste glass melters, the refractory package is often mechanically constrained to keep refractory blocks compressed and seams tight. Excess molten salts (low viscosity phase) can penetrate seams and cracks (more easily compared to the molten glass) in the refractory and cause electrical disturbances in a joule-heated melter and, in extreme cases, an inability to heat the glass [85]. For example, sulphate salts, if present in large quantities, can eventually lead to excess migration throughout the refractory package [86]. For these reasons, melter feed processing and glass formulations are designed to limit the formation of a segregated molten salt phase (also known as a sulphate gall containing mostly sodium sulphate with small amounts of chloride and chromate) [83,87–90]. However, salts are ubiquitous in nuclear waste glass processing, including the nitrate/nitrite salts (the major components of the LAW) which decompose quickly and the sulphate which is low in concentration but tends to form a gall layer; therefore, their effect cannot wholly be mitigated [91,92].

Corrosion data from Monofrax® K-3 literature survey

During the selection process of refractory materials for the vitrification JHCMs [60,62,63], several candidate refractory materials were tested and evaluated for corrosion resistance, electrical resistivity, thermal shock resistance, thermal expansion and conductivity, ease of fabrication, and the cost and availability of the materials. Based on these results, Monofrax® K-3 was selected as the refractory lining material for waste vitrification melters [60].

Bingham et al. [58] reviewed relevant literature to familiarize the reader with the background into corrosion behaviours and degradation mechanisms of glass-contact refractory materials used in nuclear waste vitrification applications. Much of the relevant literature is from tests performed at, or under the auspices, of the US national laboratories. In this section, a review of applicable crucible tests, scaled melter tests, and production melter inspections are presented. Most corrosion data of Monofrax® K-3 were collected using the crucible test method by various investigators. Each crucible-scale test showed experimental variation in test conditions, including the crucible size, glass melt volume, refractory coupon size, glass chemistry, testing temperature, and test duration. Not all test conditions were fully recorded in all the literature reviewed, making direct comparison of corrosion data difficult. Nevertheless, these experiments have demonstrated the effects of melt temperature, test duration, and glass chemistry on refractory corrosion. These test results are summarized in the following discussion, along with the limited data provided for the small-scale and production melter studies of the corrosion of refractory lining. The nature of the scaled and production melter evaluations precludes a similar level of detailed assessment of critical variables related to refractory degradation and are categorically summarized.

Crucible tests

Crucible tests refer to those performed in accordance with ASTM C621, Standard Test Method for Isothermal Corrosion Resistance of Refractories to Molten Glass [93]. Figure 3 shows a typical schematic of the crucible test. In a typical test, refractory samples (coupons) are suspended in molten glass, in an inert crucible (e.g. Pt) for a given period of time (generally days), after which the coupons are removed and examined for chemical and physical changes. The most intensive corrosion/material loss is observed at the melt line resulting in a ‘neck’ in the coupons. The half-down corrosion depth and the altered zone thickness generally exhibit much less material loss than at the neck region. Corrosion rates are commonly estimated by dividing the material loss at the melt line by the duration of the crucible test and are reported in units such as μm/day or cm/year.

Figure 3. Schematic representation of general crucible test method. (a) crucible setup (b) cross-section of the altered refractory test coupon after material loss. The measured ‘neck loss’ is calculated by $Gc=(G-g)/2$, where Gc is the neck corrosion, G is the width of the coupon before test, and g is the width of the coupon at the melt line after test.

In practice, many variables will affect the ASTM C621 method results. Experimental conditions, such as time, temperature, redox of atmosphere, and forced convection, as well as glass compositions impact the measured corrosion rate. Moreover, corrosion in real systems (e.g. operational melters) is dynamic, resulting from the influence of those combined parameter effects. As such, a variety of crucible studies have been performed that deviate from the ASTM C621 to simulate prototypic melter conditions and control individual parameters applicable to nuclear waste glass processing. The influence of these parameters on the corrosion of refractory using crucible tests with nuclear waste glass simulants is summarized below.

Effect of temperature, test time, and redox

Early work on the DWPF melters explored the effect of a limited number of parameters, such as temperature, but even these were generally qualitative in nature. It wasn’t until later years that parameters including time and redox were experimentally investigated to better understand refractory loss. Crucible tests with glasses typical for DWPF HLW glass are summarized in Table 3. General conclusions from those tests indicate the Monofrax® K-3 corrosion rate measured at the melt line increases with increased temperature in the range of 1100–1300°C [60]. However, the temperature effect was not quantitatively assessed. Carborundum (a division of Saint-Gobain Company) performed tests for Savannah River National Laboratory (SRNL) in the 1980s that looked at the effect of temperature on the corrosion rate over a range from 1150 to 1400°C. The conclusion was that the corrosion rate doubled for every 50°C increase in temperature [94].

Table 3. Monofrax^® K-3 Corrosion data from crucible tests with glass formulations typical for DWPF HLW following the ASTM-C621 methods.

Reference	Test condition		Mass% of components of interest. wt %		Monofrax^® K-3 neck loss at melt line, average loss rate (μm/day) **
	Temperature (°C)	Time (days)	Na2O	Li2O
[60]	1100	6	13.4	2.9	0.00
	1200				14.82
					4.23
					8.47
	1300				25.40
					8.47
					8.47
[62]	1150	NA	13.7	4.1	16.26
[96, 99, 100]	1150	7	11-13	Not specified	16.26
					18.54
					5.84
					17.27
		5			20.07
					39.88
[96]*	1150	0.167			151.89
		1			37.34
		5			40.89
		0.167			182.88
		1			70.36
		5			24.64
		0.167			253.24
		1			114.81
		5			52.32
		0.167			223.27
		1			58.67
		5			23.88
		0.167			254.00
		1			46.23
		5			28.19

*Reducing additives were included in the batch to control the redox.

** The average loss rate data were calculated based on the originally reported data in non-SI units.

Lu et al. [95] tested Monofrax® K-3 coupons in two alumino-borosilicate glasses with ∼ 17 wt-% of Na₂O, ∼ 2.5 wt-% of Li₂O, and ∼ 2 wt-% K₂O at 1208°C for 1, 2, 3, and 6 days. The results showed that the chromium oxide and aluminium oxide dissolved in the glass melt leading to an increase in viscosity, which resulted in a decreasing dissolving rate of the coupon over time. Williams et al. [96] reported an average corrosion rate that decreased inversely with test time from 4 to 120 h. This was attributed to the formation of a passivating spinel layer between the glass melt and the refractory and the saturation of cations, such as Al and Cr at the reaction layer between the refractory and the glass melt [72,76].

The refractory corrosion rate depends upon the oxygen fugacity of the melt, or redox conditions [97]. This dependence is expected given the redox active species present in the refractory, such as Cr and Fe. In practice, formic acid (reducing) and nitric acid (oxidizing) were used as additives to control the redox of DWPF glasses [98]. However, the effects of oxidizing or reducing additives were not performed during materials of construction testing for DWPF. Pilot-scale melter tests were principally performed under reducing conditions, which were assumed to be representative of normal DWPF operations. During the first two years of non-radioactive DWPF operations, the acid additions to the feed would have favoured oxidizing conditions due to a single acid tank and a focus on startup (not redox). The attack of the refractory was noticeably more rapid than observed in the previous pilot-scale melter tests. This result suggested that oxidizing conditions were potentially more aggressive than reducing conditions.

Due to this observation, untreated refractory corrosion coupons made from Monofrax® K-3 were tested with a DWPF feed under oxidizing conditions in a crucible and a pilot-scale melter [96,99,100]. The corrosion rates were approximately 1.8–2.8 times greater in oxidizing conditions than those measured previously under reducing conditions. They were also greater than those measured in the pilot-scale melters. The corrosion mechanism was shown to be dominated by depletion of Al₂O₃ from the Monofrax® K-3 allowing reaction with Na₂O, NiO, and Fe₂O₃ from the glass, which is accelerated by the oxidative dissolution of reduced metals present in the Monofrax® K-3.

Effect of segregated molten salt phase

The presence of high concentrations of salts during waste processing is known to lead to a molten salt, or ‘gall’ layer, that segregates from the molten glass phase [101]. For Hanford LAW glasses, the salt layer segregated in the melter primarily consists of sodium sulphate [83,87–90,102,103], formed through excess sulphate salts present in the feed. Small proportions of chloride and chromate salts are also present in the salt layer [104]. Table 4 summarizes corrosion test results for glasses formulated with excess salt or those that exhibit distinct salt phases in/on the glass melt.

Table 4. Monofrax^® K-3 Corrosion data from crucible tests with a molten salt phase.

Reference	Test condition		Salt phase	Monofrax^® K-3 neck loss at melt line, average loss rate (μm/day) ***
	Temperature (°C)	Time (days)
[85]*	1100	3	0.7 wt% sulfate in feed	73.07
	1300			430.76
	1100		0.5 wt% chloride in feed	160.05
	1300			707.72
	1100		1.5 wt% chloride in feed	218.51
	1300			522.61
	1100		0.7 wt% sulfate, 1.5 wt% chloride, 4.5 wt% phosphate, and 0.6 wt% chromate in feed	50.80
	1300			527.48
[86]**	1208	3	∼ 50 wt % sulfate mixed with a pre-melted glass	151.81
				87.50
		6	Sulfate layer formed by SO2 bubbling in the glass melt	88.69
				26.33
				16.51

*The change of salt layer was not described during the tests.

**Two necks distinguishable in one coupon; one at the air/salt interface and one at the salt/glass melt interface.

***The average loss rate data were calculated based on the originally reported data in non-SI units.

Marra et al. [85] tested Monofrax® K-3 corrosion in glass with added sulphate, chloride, chromate, and phosphate, but the formation of a salt layer was not clearly described. The data show sulphate and chloride were correlated to greater corrosion depth. When chromate and phosphate were added to sulphate and chloride, the corrosion rate decreased [85]. Gan et al. [86] performed Monofrax® K-3 tests in simulated LAW glass with excess sulphate added. The loss of material was not significant in those cases; however, a detailed examination showed the formation of a very loose, sponge-like layer on the surface of the refractory sample. Gan et al. [86] also tested Monofrax® K-3 coupons in a series of mixed sodium and lithium sulphate (without glass in the crucible). The Monofrax® K-3 coupons showed very little dissolution after the test (no neck loss data was available) but cracking in the material was observed. The different reported corrosion behaviours of Monofrax® K-3 coupons in molten salt suggest that neck loss measurement may not be a representative way to characterize corrosion damage in glass melts with high salt concentrations, other methods, such as microscopic analysis of the penetration layer, should perhaps be employed.

Effect of glass melt composition

Leading up to the startup of WVDP and DWPF, limited experiments existed to quantify the effects of large changes in glass composition on the refractory corrosion rates. Historically, the DWPF and WVDP were designed to produce glass with a relatively narrow, consistent composition range. However, to process future wastes at SRS and Hanford with maximum waste loadings (the fraction of glass originating from waste), the corrosion behaviour is needed for glass composition envelopes outside those nominally processed through the DWPF. Recently, Monofrax® K-3 corrosion tests were conducted on more than 300 glass compositions projected to be processed through the WTP [86,87,105–126]. Those tests were specifically intended to provide information to develop a model to predict the effects of glass components on the Monofrax® K-3 corrosion rate within a glass composition envelope representative of Hanford LAW formulations. Those tests were performed using a modified ASTM C621 method where each test was carried out at 1208°C for six days and air was bubbled through the glass melt at a rate of 8 cc/min for the duration of the test. The full dataset of 344 glasses is not given in this paper but can be found in Muller et al. [126]. The results are compiled and summarized in Table 5. There are two major groups of LAW glasses (1) WTP baseline which were tested to support plant commissioning with reasonably modest waste loadings, and (2) ORP (DOE’s Office of River Protection) which were tested to help optimize operation of the plant over the mission lifetime. The computed Monofrax® K-3 corrosion rates from this data set span more than two orders of magnitude across the composition matrix and a cursory review of the data in Figure 4 suggests a correlation between the alkali content and higher observed corrosion rates. Several models have been developed based on this dataset to predict the effects of glass components on Monofrax® K-3 corrosion [126–128]. The models generally include a strong effect of Na₂O on promoting Monofrax® K-3 corrosion (i.e. positive coefficient for neck loss) and a strong effect of SiO₂ on repressing Monofrax® K-3 corrosion (i.e. negative coefficient for neck loss) [128]. The compositional models will be further discussed in the Models for Crucible Test Data section.

Figure 4. Monofrax_® K-3 corrosion rate and normalized alkali oxide in mass % of the LAW glasses in the datasets of WTP baseline and ORP [126].

Table 5. Summary of Monofrax® K-3 corrosion data by a modified ASTM C621 method tested at 1208°C for 6 days with air bubbling in the glass melt. The average corrosion rate was converted from the reported 6-day corrosion depth at the neck/melt line in inch [126].

Dataset (number of tests)	Monofrax® K-3 corrosion rate (neck loss), μm/day			Mass% of components of interest
	minimum	maximum	median	Na2O	NAlk*	SO3	Cr2O3	Cl
All (344)	4	592	120	2.5–26	14.3–27.2	≤ 2.5	≤ 0.6	≤ 1.2
WTP Baseline (157)	4	494	98	2.5–25	14.3–26.2	0.01–2.5	≤ 0.6	≤ 1.2
ORP (178)	4	592	130	5.0–26.0	16.7–26.5	0.16–1.5	≤ 0.6	0.01–0.73
DFHLW (9)	63	539	143	17.1–24.1	21.5–27.2	≤ 0.57	≤ 0.2	≤ 0.19

* NAlk (normalized alkali oxide in mass %) = Na₂O + 0.66 K₂O + 2.07Li₂O.

Scaled melter tests

From the 1970s to the 1990s, scaled melters built at Pacific Northwest Laboratory (now Pacific Northwest National Laboratory, PNNL) and Savannah River Laboratory (now Savannah River National Laboratory, SRNL) processed simulated DWPF HLW glasses to support DWPF design [37,61,66,67,94,129–131]. Starting from the 1990s, Vitreous State Laboratory (VSL) at The Catholic University of America, cooperating with Duratek Inc. (now Atkins), and PNNL built scaled metres to support the optimized operation of the WTP [44, 132–139]. Most of the VSL/Duratek and PNNL melter tests were on simulant WTP LAW glasses which might cause higher corrosion of the refractory due to the higher alkali-containing glass compositions compared with the DWPF HLW glass compositions [44,126]; and some tests of simulant WTP HLW glasses were also conducted to investigate the crystal formation and accumulation in melters [137,138]. These scaled melters processed simulated nuclear waste feeds for extended periods (i.e. typically years) under prototypical processing conditions. After operation, visual inspections of the glass contact refractory were made for the melters to assess vulnerable areas for attack. Corrosion data obtained from these pilot-scale operations are summarized in Table 6. Figure 5 shows photos of Monofrax® K-3 refractory from disassembled melters.

Figure 5. Monofrax® K-3 refractory from disassembled scaled melters. (a) PNL-LFCM. Cross-section of the glass discharge showing glass penetration into the refractory brick joints and cracks [61]. (b) DM3300. View of the inside of the melter after the melter lid was removed. Picture shows the plenum refractory, tuckstone (yellow brick on ledge) and the K-3 refractory (black brick). The bubblers and temporary riser can also be seen [135]. (c) DM3300. K-3 glass contact refractory that was broken during the removal process. Picture shows approximately 10 cm (4 inches, the measuring tape is in inch) of loss at the melt line [135]. Reprinted with permission from refs [61] and [135].

Table 6. Refractory corrosion data from scaled melter Tests.

Melter*	Glass Pool Surface Area	Operational time (not continuous) of the inspected Monofrax® K-3 refractory		Average corrosion rate		Glass Composition
				Neck (melt line)	Riser/pour spout
PNL-LFCM [37, 61]	1.05 m^2 (rectangular, 122 cm × 86 cm)	3 years	1977–1980	23 μm/day (0.85 cm/year)	52 μm/day (1.9 cm/year)	Simulant DWPF HLW glasses (11–14 wt-% Na2O)
Project 1941 [66, 67, 94, 129]	1.17 m^2 (cylindrical, 4 ft diameter)	2 years	1978–1981	Not measured	70 μm/day (2.5 cm/year)
SCM [66, 67, 129, 130]	0.12 m^2 (roughly octagonal, 1.3 ft^2)	2 years	1979–1982	191 μm/day (7 cm/year)	279–330 μm/day (10-12 cm/year)
LSFM [66, 67, 129, 131]	1.17 m^2 (octagonal, 4 ft diameter)	2 years	1982–1985	25 μm/day (0.91 cm/year)	Not measured
SGM [66, 67, 129]	1.17 m^2 (cylindrical, 4 ft diameter)	2 years	1986–1988	found slight deform of the bottom	Not measured
IDMS [66, 67, 129]	0.29 m^2 (cylindrical, 2 ft diameter)	7 years	1988–1995	Not measured	As much as 5 cm (2 inch) Monofrax® K-3 loss was found near the drain valve
DM1200 [132, 133]	1.2 m^2 (rectangular, 34 inch × 54 inch)	5 years	2001–2006	Good condition	Not measured	Simulant WTP HLW and WTP LAW glasses
DM3300 [36, 134, 135]	3.3 m^2 (rectangular, 80 inch × 64 inch)	5 years	1998–2003	70 μm/day (10-13 cm in 5 years)	Not measured	Simulant WTP LAW glasses
RSM [136–139]	0.017 m^2 (circular, 15.2 cm diameter)**	11 weeks	2016	No measurable neck	Not measured	Simulant high Fe-Ni HLW glass

* PNL-LFCM = Pacific Northwest Laboratory – Liquid Fed Ceramic Melter, Project 1941 = a prototype melter system tested at SRS, SCM = Small Cylindrical Melter (SRS), LSFM = Large Slurry Fed Melter (SRS), SGM = Scale Glass Melter (SRS), IDMS = Integrated DWPF Melter System (SRS), DM1200 = HLW Pilot Melter (VSL/Duratek), DM3300 = LAW Pilot Melter (VSL/Duratek), RSM = Research-Scale Melter (PNNL).

** The Monofrax® K-3 lining of the RSM is exchangeable. The melter diameter is 15.2–25.4 cm (6-10 inch) surface area is 0.017–0.05 m² when using different sizes of linings.

The Pacific Northwest Laboratory – Liquid Fed Ceramic Melter (PNL-LFCM) project operated 1977–1980 was developed to demonstrate the feasibility of producing borosilicate glasses within a joule-heated ceramic melter [61]. Corrosion of the refractory bricks was carefully examined for PNL-LFCM after three years of operation [61]. It was concluded that the observed corrosion was < 0.85 cm/year. Glass-filled cracks were found in the refractory (Figure 5(a)); however, the examination did not show that the cracks enhanced the corrosion depth. Analysis indicated that the cracking of the refractory was caused by thermal stress which can probably be prevented by allowing sufficient space for refractory thermal expansion in melter design, reducing the thermal gradients in the refractory by increasing refractory wall thickness, and reducing the severity of thermal transients such as using a slower heating rate during the startup. It is worth noting that extensive cracking of the Monofrax® K-3 refractory caused by stress or thermal shock was not observed for the scaled melters tested later.

Figure 6. Average corrosion rate (neck or melt line corrosion depth per day) as a function of time in SRNL’s crucible tests [60, 62, 96, 99, 100] and scaled melter tests [61, 66, 129, 131]. Results of crucible tests with different glass at 1100–1200°C and scaled melter tests are shown, with the rates averaged over time.

Five scaled melters were operated at SRS in support of the design of DWPF between 1978 and 1995 [66,67]. Typically, the scale of these SRS test melters was based on the melt pool surface area and therefore the melting rate achieved is based on this area in the DWPF melter design basis. The majority of those SRS scaled melters were operated with a reducing flowsheet, similar to DWPF, but some were operated with calcine (Project 1941 and Small Cylindrical Melter (SCM)) or oxidizing Integrated DWPF Meter System (IDMS) flow sheets. Corrosion rates were quantified in three of the five SRS scaled melters. In general, two areas of the refractory were attacked significantly in each case − the melt line and the riser or pour spout area (Figure 1). The pour spout area typically had the highest rate of attack (70–330 µm/day) as measured by the 1941 project and the SCM [94,130]. For the 1941 melter, the corrosion rates were higher in this area because the pour spout wall was in the direct path of any electric current passing from pool electrodes to the riser heater. The pour spout region for the SCM, on the other hand, appeared to erode due to high molten glass velocities. Corrosion was so severe in this region that a hole formed in the 10 cm thick refractory after approximately nine months of operation. After observing the high Monofrax® K-3 corrosion damage at those areas from the scaled melter tests, Inconel 690, a Cr–Ni–Fe alloy with very high corrosion resistance to glass melts was used to make an insert of the pour spout for DWPF melters [63,140].

Corrosion rates at the melt line were measured for the Large Slurry Fed Melter (LSFM) and the SCM and ranged between 25 and 191 µm/day. Refractory degradation in the LSFM occurred in the blocks located closest to the slurry feed tube. The degradation was characterized by spalling that likely occurred due to thermal shock fracturing where the slurry feed water contacted the hot refractory. Since this mechanism is not operable in the DWPF (much larger size), this rate was not considered for the design basis. For the SCM, the majority of the attack occurred 5–8 cm below the melt line at a rate of 191 µm/day (0.0075 inch/day). This corrosion rate was assumed for the DWPF design basis [141]. The average sidewall corrosion rate, including the melt line, for the SCM was approximately 140 µm/day (0.0055 inch/day), which is slightly lower [66].

However, the corrosion rates from the SCM do not distinguish between an ‘average loss of material’ and a ‘selective penetration’ depth, but rather consider the sum of these two (total penetration) measurements as the overall corrosion rate. Therefore, the total penetration rate is significantly influenced by high flow rate conditions and therefore may be conservative. Corrosion rates in more quiescent environments, such as coupon tests, are expected to be lower. Results of crucible tests on similar glasses are also shown in Figure 6, where the average crucible test corrosion rates were found to trend lower than the refractory corrosion data from scaled melters as the crucible test time increased [96,99,100]. This is likely due to (1) the flow of glass melt being more vigorous in scaled melter tests, (2) spalling at the melter melt line that cannot be completely simulated in crucible studies, and (3) buildup of corrosion-inhibiting components (Cr, Al, Fe, Mg) in the crucible tests.

Following the successful operation of DWPF, JHCM designs for the WTP aimed to increase many melter performance metrics, including waste processing rate and waste loading [142,143]. Facing the challenges of chemical complexity and oxidizing conditions of the highly variable Hanford wastes, scaled melters were designed to collect processing data and establish melter performance against simulated wastes. From the 1990s to the 2000s, VSL and Duratek built the Duramelter^TM-10 (DM10), -100 (DM100), -1200 (DM1200), and -3300 (DM3300) scaled melters, using Monofrax® K-3 as the glass contact refractory [44]. For DM1200 and DM3300, the corrosion of refractory lining material was inspected and reported (Table 6) [36].

DM1200 HLW has a glass pool surface area of 1.2 m² [133]. It was inspected in 2006 [36,133] and the visible part of the Monofrax® K-3 lining was in good condition. The DM3300, known as the LAW Pilot Melter, had a glass pool surface area of 3.3 m² [134]. The melter was commissioned in 1998 and shut down in 2003 [135]. Following shutdown, the melter lining was disassembled and the inspection showed that spalling on the Monofrax® K-3 refractory occurred above the melt line with a material loss of approximately 10–13 cm (4–5 inch) as shown in Figure 5(b) and Figure 5(c) [36,135]. Glass was found to have penetrated cracks and seams into the outer layers of the refractory; however, the extent was limited, and no glass was found to have completely penetrated the refractory package [135]. The melter floor, also Monofrax® K-3, displayed 2.5 cm (1 inch) of material loss, significantly less than the walls. Bubbling was not found to significantly increase material loss even in the case where one bubbler was purposely placed near the wall. It should be noted that DM3300 is the largest-scaled melter built for vitrification research to date. The DM3300 design was based on the SRS M-area DM5000 series of melters commissioned between 1996 and 1997 and used in production through 1999 [36]. During the operational life of DM3300, 4000 m³ of simulated feed was processed and over 3,500 tons of glass was produced, which is comparable to the full-scale production melters such as the DWPF/WTP melters. The DM3300 refractory corrosion results are important data in reference to the long-term operation of large melters.

At PNNL, several scaled melter systems have been running vitrification tests, including the laboratory-scale melter, the continuous laboratory-scale melter (CLSM), and the research-scale melter (RSM) [136–138,144,145]. The Monofrax® K-3 lining of RSM was inspected after an 11-week run of a high Fe-Ni simulant HLW glass (17.81 wt-% Na₂O, 7.63 wt-% B₂O₃, 38.47 wt-% SiO₂, 7.84 wt-% Al₂O₃ and 17.5 wt-% Fe₂O₃) [138]. Optical and scanning electron microscopy were used to characterize the cross-section sample cut from the melter lining [139], providing information on morphology and elemental distribution at the refractory surface in contact with the glass melt. No clear neck corrosion was identified, and the loss of material is minor, which should be attributed to the low corrosivity of the high-Fe and low-Na glass and the relatively short test duration. The observed reaction layer was less than 0.5 mm.

Refractory corrosion in production melters

Inspection of full-scale melters after the completed mission provides valuable information to evaluate the material performance against recommendations resulting from corrosion models. In situ inspection and quantification of refractory performance are difficult due to radiation and contamination levels of a melter in service for many years [146]. Visual inspection of the refractory is more often the primary means of evaluation of off-line melters after the end of service [147].

The WVDP melter was inspected in 2014 (12 years after its completion of vitrification and shut down in 2002) to characterize the isotopic distribution and remaining radioactivity [148]. However, no corrosion or glass penetration depth data was collected.

The DWPF replaced two melters − one in 2002 and one in 2017 (Table 1). Each was inspected by a remote camera after shutdown [149–151]. Because some glass remained in the melter, the corrosion of Monofrax® K-3 bricks below the melt line could not be seen, however, severe spalling of the Monofrax® K-3 lining was found at the melt line and near the electrodes (Figure 7(a)). The corrosion was less in other places along the melt line. Overall, the Monofrax® K-3 lining was in acceptable condition at the end of the 8-year and 14-year service periods and would not have presented an immediate challenge to reliable melter operations.

Figure 7. Corrosion of the Monofrax® K-3 refractory lining of production melters. (a) DWPF Melter 2, reprinted with permission from ref [151]. (b) SRS M-Area DM5000A, reprinted with permission from Ref. [36].

The Duramelter^TM (DM) 5000A, constructed by GTS Duratek, was operated by SRS between 1997 and 1999 to vitrify mixed low-level waste at the site’s material fabrication facility [4,5]. As noted in Table 1, a DM5000 melter failed during simulant operations due to incompatibility with the initial feed. After replacement of DM5000, a new DM5000A handled the radioactive waste and operated for approximately two years after which it was drained and inspected. At the melt line, spalling and corrosion resulted in a loss of material of 1–2 inches in depth and 4 inches high. (Figure 7(b)). Examination of Monofrax® K-3 samples taken from the disassembled melter showed dissolution of the chromia-alumina solid solution and penetration of the glass into the refractory to a depth of about 2.5 cm (1 inch). The report estimated that an additional four to seven years of service life could have been expected for the DM5000A [36].

Indeed, studies on refractory corrosion depth as a function of time for production-scale melters are limited. Melter service life typically lasts for only several years and it is often technically difficult or cost prohibitive to collect samples after the melter’s end of life. Nevertheless, a few data that exist are plotted in Figure 8, for which corrosion depth has been estimated from operational melters. For comparison, Figure 8 also shows the DWPF melter refractory design basis. In-service melter corrosion depths are well below the corrosion limit estimated from design basis crucible tests. However, since the commissioning of DWPF, many additional crucible tests have been conducted with different glass compositions [126]. At the bottom of Figure 8, these crucible data conducted at 1208°C for 6 days using the bubbling method are shown. The red dashed line in the plots presents the proposed limit for the 1208°C bubbling test results (set as 0.04 inch neck corrosion depth for the coupon after the 6-day test, equals to 0.0067 inch/day or 169 μm/day); this limit, similar to the DWPF design basis based on the scaled melter corrosion data, was proposed to be used in glass formulation as a guide for acceptable refractory corrosion characteristics [113,126]. Taken at face value, 31% of the tested glasses exceeded the proposed limit (also see Figure 4), which would suggest that a large portion of glass compositions would need to be adjusted (likely at the expense of waste loading) to lower the glass melt corrosivity. For example, reducing of Na₂O can effectively decrease the predicted refractory corrosion; however, since Na₂O is a primary component from the waste stream, the adjustment leads to a lower waste loading [143]. In addition, Cr₂O₃ as a minor component (≤ 0.6 wt%) can effectively reduce the corrosion according to the crucible data, which should be considered as an important factor in adjusting glass compositions (quantitative analysis will be discussed in the following modeling section) [126–128]. However, it is noted that crucible tests are known to be used historically for qualitative comparisons, are not generally prototypic of operational melters, and data from decommissioned melters indicates that corrosion limits determined from raw crucible test data are conservative [66]. Additional work is needed to understand the scaling factors between crucible tests and operational melters as well as the conditions needed to more reliably and accurately correlate refractory corrosion data to melter refractory lifespans.

Figure 8. Summary of Monofrax® K-3 refractory corrosion data from melter and crucible tests [36, 60–62, 66, 96, 99, 100, 126, 129, 131, 149–151]. The limits of corrosion rate are shown in both of the long-term melter data plot (top, > 300 days) and the short-term crucible data plot (bottom, ≤ 7 days) for comparison. The non-SI unit inch is used in the plots as it is used in all the data source reports. The DWPF design basis of 0.0075 inch/day equals to 191 µm/day and the proposed limit of crucible test of 0.0067 inch/day equals to 169 µm/day. It should be noted that the two DWPF data points are very roughly estimated based on visual inspection of DWPF Melter 1 [149, 150] and DWPF Melter 2 [151] after end of their service.

Modelling approaches

Theoretical corrosion models have been derived to understand the dissolution of refractory in contact with molten glass at the melt line [152–155]. However, melt line dissolution models have limited application to production melters due to the complexity of the refractory wear mechanisms during operation. In nuclear waste glass production, the composition of the glass is not fixed, and empirical approaches based on glass composition, one of the most important controlling variables, have been developed in place of theoretical models [127,156,157]. These empirical methods are developed through laboratory data and extrapolated to fit the application. The aim of collecting laboratory-scale melt-line corrosion data of the Monofrax® K3 refractory in contact with nuclear waste glass compositions has been to develop a predictive model for refractory lining service life during operation [126,127]. However, the development of a model for full-scale, operational melters is challenging due to the available data. Moving from an empirical understanding of crucible neck-line corrosion data to a predictive capability of refractory lining service life in nuclear waste glass melters is a fruitful area for future study.

It may be possible to better understand the factors that contribute to corrosion at multiple scales and thus develop models that capture the complex behaviours in nuclear waste glass melters by pursuing various modelling approaches in concert. For example, empirical models based on laboratory-scale tests to correlate glass composition with refractory loss are extremely relevant as a gross approximation. However, these approaches would benefit from theoretical models of melt line dissolution that include relevant physical phenomena. Lastly, a corrosion model that includes the coupled physics mechanisms besides melt line dissolution (e.g. spalling) occurring in a production nuclear waste glass melter over long-term operation should be considered. These approaches are discussed in the following subsections.

Empirical models based on laboratory-scale test data

One basic approach to correlate refractory life to glass composition is to empirically measure the corrosion with various glass compositions and extrapolate a corrosion rate. This approach uses test data from the widely accepted crucible scale tests that are considered qualitative. Basic modelling tools provide empirical fits based on glass composition. For example, Muller et al. [126], reported crucible test data of Monofrax® K-3 corrosion in their database (Table 5 and Figure 4). Based on this dataset, three models were developed using different mathematical tools (Figure 9). Muller et al. [126] fitted the data with a 28-term partial quadratic mixture model. Vienna et al. [127] presented a reduced linear mixture model on a primary set of 11 glass components and reached similar predictions (Figure 9(a)). Smith-Gray et al. [128] used machine learning methods with 16 different oxides in the glass composition and found that Bayesian Ridge regression showed the best comparison for predictions of neck loss data. The linear mixture model and the best-fit regression obtained from machine learning are compared in Figure 9(b). Figure 9(c) displays the response trace plot for the 13-term model by Vienna et al. [127]. Cr₂O₃, Al₂O₃, V₂O₅, SiO₂, and ZrO₂ are predicted decrease the corrosion depth while Li₂O and Na₂O are predicted to increase the corrosion most. The curvature in Li₂O effect is also evident with a high slope at low Li₂O concentrations and lower slope at higher Li₂O concentrations.

Figure 9. Example of empirical models of Monofrax® K-3 corrosion by crucible test. (a) linear mixture model fit. (b) machine learning fit. (c) response trace plot for 13-Term reduced partial quadratic mixture model on the naturallogarithm of melt line corrosion depth. Data adapted from Muller et al. [126], Vienna et al. [127], and Smith-Gray et al. [128]. The non-SI unit inch is used in the plots as it is used in all the data source reports.

Although the composition models fit the test crucible data well, they are insufficient to predict refractory lifetime during melter operation, in which composition and operational parameters change frequently. Rather, these models provide a practical, but conservative operational limit based on the most aggressive corrosion depth (e.g. the neck loss) measured by standardized tests. For example, referring to Figures 4 and 8, a design life prediction can be made based on material loss measured in length per year. This approach is necessarily conservative and further evidenced by melter life in the DWPF. It is important to be able to define a more accurate refractory service life limit to maximize waste throughput while meeting all the processing and quality constraints place upon the glass composition. This will require more theoretically-based approaches than the existing empirical methods.

Melt line dissolution models

There are models that explore the fundamental driving forces for refractory corrosion in contact with molten glass. In particular, melter refractory lifetime is considered largely dependent on corrosion behaviour at the three-phase boundary, i.e. the melt line (also referred to as flux line or slag line). Hrma’s work [152] studied free surface convection Marangoni flows as the cause of neck-line corrosion and developed an equation for the dissolution rate with a simple parameter based on local concentration and velocity in a laminar flow regime. Tsotridis [153] noted the phenomenon of flux line corrosion occurring in the refractory at the gas-melt interface and used 2-D Navier Stokes equations to derive a simple model to test the driving forces for corrosion, such as thermal, concentration gradients, diffusion rates and determined surface tension forces are the most influential. Pötschke and Brüggmann [154] discussed the influence of Marangoni convection currents caused by concentration or temperature gradients along the interface of the gas–liquid interface on refractory corrosion kinetics. While the kinetic corrosion rate equation was developed for steel melting, the authors note that the method is applicable to other high temperature molten liquids, such as glass melting. Lian et al. [155] also noted the Marangoni effect as a direct cause of premature destruction of alumina refractories. A geometrical model of the corrosion groove at the surface was developed and the calculated results were compared with experimental measurements, showing good agreement [155]. Figure 10(a) shows a diagram of the coupon corroding at the neckline, and Figure 10(b) shows the flow occurring from this convection.

Figure 10. Diagram of dissolution at surface (a) neck corrosion on sample (1: solid, 2: liquid, 3: gas) and (b) Marangoni flow schematic showing development of corrosion groove. Adopted from Hrma [152] and Pötschke and Brüggmann [154].

The melt line dissolution models discussed above showed that surface tension-driven flow has a significant influence on material loss at the melt line, whereas other free and forced convection flows are relatively insignificant. Local corrosion of the refractory occurs at a higher rate at the melt line, where components from the refractory dissolve into the glass and alter the interfacial tension and viscosity. This results in a surface tension gradient that induces Marangoni convection, which accelerates corrosion and results in the formation of a corrosion groove or ‘neck’ which is the most severe form of refractory corrosion in glass melters. Several sources have compiled linear fits for the surface tension of glasses based on measurements, though these are generally limited to simple forms with fewer oxide components than used in waste glass [158, 159]. Typically, in silica melts, K₂O, B₂O₃, and PbO decrease the surface tension, while Li₂O, Al₂O₃, MgO, CoO and ZnO increase the surface tension [160]. This is of interest because Al₂O₃ is a major component present in Monofrax® K-3, and localized corrosion at the melt line could increase the local surface tension, amplifying the Marangoni effect.

Surface tension is influenced by temperature, chemical composition, and atmosphere [161]. Bonds between the molecules become weaker at higher temperatures, which decreases the surface tension; industrial glasses typically show a surface tension reduction of 4–10 mN/m for each 100 K temperature increase [160]. The atmospheric water vapour is shown to have an effect between dried and normal air of about 5–13 mN/m, depending on temperature. Measurements of droplet shape using the sessile drop method are commonly used to determine surface tension for glass melts. The Bashforth and Adams method based on the Young-Laplace equation can be used with measurements of droplet geometry to determine contact angle, surface tension, and drop volume [162]. Additionally, images can be analyzed computationally with higher accuracy using numerical procedures that derive a best fit to the drop profile over the entire meridional section instead of at just a few points [163].

Corrosion models based on melter operation

An approach to circumvent the shortcomings of scaled crucible tests and the dynamics in operational melters has been to develop numerical models to predict corrosion rates and extrapolated melter refractory lifetime based on the observed phenomena in operational melters. Brosnan [164] introduced models to correlate the corrosion rate with chemical compatibility (between the refractory and the melt) and temperature gradients. Nishikawa et al. [165] developed a temperature gradient model to predict the long-term steady-state corrosion rate. Nishikawa’s model derived parameters from the corrosion rate of AZS (Al₂O₃–SiO₂–ZrO₂) refractory determined by diffusion and mass transfer, which are used to calculate a change in temperature at the glass/refractory interface with material loss, i.e. the temperature at the interface decreases as the interface moves outward due to material loss. The Nishikawa model has been shown to simulate corrosion rates that decrease with increasing time, in agreement with experimental data of a one-year test. It is generally understood that the corrosion rate of Monofrax® K-3 refractory in production melters decreases with operation time (Figure 8). These models derived from operational data are useful, but they do not explicitly capture the mechanistic understanding needed to predict refractory loss as a function of glass composition. Rather, these types of models incorporate a multitude of variables inherent to operational melters that contribute to refractory corrosion behaviour. Furthermore, models based on melter operation data are often dependent on parameters specific to a system, that are not easily obtained and may not be universal across melter designs. The challenge remains to identify the most significant factors controlling refractory corrosion in operational melters and to utilize the mechanistic models elucidate the contribution and influence of those factors.

Future modelling outlook

Key factors controlling refractory corrosion are the temperature, refractory composition, glass chemistry, and mass transport near the refractory surface. Researchers have studied corrosion effects with other fluids. Xu and Cheng [166] describe a computational fluid dynamics (CFD) simulation for flow-assisted corrosion of aluminium alloy in a glycol water solution. This work couples the electrochemical reactions that cause corrosion of aluminium with the local flow fields to show the impact of convective currents on corrosion rates. Hu and Cheng [167] used a CFD model to simulate corrosion of a pipeline in an oil–water two-phase system with CO₂. An empirical corrosion rate based on local CO₂ concentration is coupled with the CFD model to show the impact of flow velocity on local corrosion. These studies are possible due to the ability of the CFD model to resolve the near-wall flow. Jantzen et al. [99] tested the corrosion of Monofrax® K-3 samples in simulated HLW melter feed and from that testing proposed oxidation and reduction reactions that were an important factor in corrosion. A similar methodology could be employed with CFD simulations of waste vitrification melters to couple the methods of Xu and Chen [166] and Hu and Cheng [167] with associated reactions with model corrosion rates at the melt line.

Simple, yet informative, thermodynamic calculations can also be performed using a material-to-material calculator, such as that available in Thermo-Calc software [168]. This method is applicable where different materials are in contact and will react (or gradually transition into) with each other (i.e. molten glass and refractory). These calculations use thermodynamic equilibrium boundary conditions, so they are much faster computationally than diffusion calculations. Equilibrium calculations are performed along linear composition gradients and along the composition profiles. This method can provide information about the stable phases that are expected when two materials are in contact and transition into each other with linear gradients. The change with temperature of the phases and their compositions, along with the formation of any parasitic phases, can be predicted in this way. This type of analysis is particularly useful in high-temperature applications where materials may be subjected to extreme conditions that can lead to phase transformations, corrosion, or other types of degradation. Coupling this type of analysis with experiments to determine kinetic rates could lead to a more rigorous corrosion model than simple polynomial (or-otherwise) fitted correlations.

Summary and recommendations for future work

Monofrax® K-3 is a high chromia refractory that is used in the waste vitrification melters operating in the United States. These vitrification facilities must reliably operate for decades while the melters are expected to be replaced every several years. Thus, accurate prediction of the useful service life of the refractory lining is necessary to plan for maintenance and replacement of the melters. This review summarizes relevant corrosion data from crucible tests and inspection reports of post-production melters lined with Monofrax® K-3 refractory. A challenge to predicting refractory life is estimating corrosion as a function of time, which is much longer than the duration of a practical crucible test. It is known that the corrosion rate decreases with increasing time for both melter tests/operation and crucible tests. Therefore, investigating the corrosion of the refractory lining of operating melters remains very important. In addition, many factors influence refractory life, including glass composition, temperature, mass flow of glass, plenum atmosphere, salt formation, and glass redox conditions. To understand each of these variables, they should be evaluated systematically in controlled laboratory-scale crucible tests, which is crucial to developing predictive models to estimate refractory lifetime from material loss by corrosion. Relating these results to scaled melter tests will be a necessary further step in evaluating a predictive model for estimating refractory lifetime and corrosion rate.

It is well known that laboratory scale refractory corrosion tests (e.g. ASTM C621) do not reliably predict refractory lifetime in a waste glass melter. Monofrax® K-3 corrosion data collected to date using laboratory scale crucible tests are insufficient to make accurate predictions of melter refractory lifetime. Furthermore, comparisons of corrosion data between crucible tests, scaled melter tests, and production melters, are generally ambiguous. Relating crucible tests to scaled melter tests and the observed behaviour in production melters is a necessary step to develop a predictive model for estimating refractory lifetime and operational limitations. The following recommendations are provided to facilitate scaling of laboratory scale corrosion testing results to prototypic melters:

1. Collect corrosion data under prototypic melter conditions to determine the effects of critical factors such as melt resonance time, melt temperature, redox conditions, and presence of salt phases in contact with the melt and refractory. Collecting corrosion data with various redox conditions (the redox state of the atmosphere in the melter depends on the melter feed composition) and with the presence of molten salt phases are needed to evaluate the effects of environmental factors prototypic of melters. In addition to the glass chemistry, viscosity is a critical property of the melt that affects corrosion. Investigations should also include viscosity under controlled test conditions by selecting and testing glasses with similar viscosity behavior as a function of melt temperature. These types of controlled tests will facilitate assessment of the changes to the refractory from chemical reaction, phase evolution, and mass transfer that occur over the life of the melter.

2. Conduct long-term corrosion studies to understand long-term corrosion damage of the melter lining and improve estimation of long-term material loss. Crucible tests may have practical limitations owing to equilibrium in a closed system. However, one approach to overcome these limitations is to combine computational modeling with experimental data (long-term melter tests would provide the most useful data) to elucidate key drivers affecting refractory corrosion. It may be feasible to develop fitted correlations that can be incorporated into CFD simulations under differing flow regimes to predict corrosion rates in different regions of full-scale melters for refractory life estimates. A conservative range of flow regimes can be used to easily estimate expected melter conditions including the influence of pronounced melt line fluctuations on corrosion.

3. Investigate the impact on the efficiency of the ‘integrated flowsheet’ when applying the corrosion model in glass formulation (see for example [169]). If using the corrosion model in a glass formulation, it would likely result in reduced alkaline and increased Cr₂O₃, which may lead to a lower waste loading. In addition, other properties such as viscosity, electrical conductivity, and sulphur solubility could be impacted as well. The constraints of potential corrosion models and the overall impact on operational efficiency must be evaluated.

Acknowledgements

The authors would like to thank Pavel Hrma of AttainX, Support Services Contractor to the Office of River Protection, U.S. Department of Energy and Chris Chapman of Pacific Northwest National Laboratory for criticism and discussion. The authors would also like to thank Renee Russell and Emily Nienhuis of Pacific Northwest National Laboratory and Richard Wyrwas of Savannah River National Laboratory for their help in technical review of this article.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

Funding for this work was supported by the U.S. Department of Energy Office of River Protection Waste Treatment & Immobilization Plant Project through Department of Energy Work Authorization M0SRV00101. Pacific Northwest National Laboratory is a multiprogram national laboratory operated for U.S. Department of Energy by Battelle Memorial Institute operating under Contract No. DE AC05-76RL0-1830. Contributions from the Savannah River National Laboratory were produced by Battelle Savannah River Alliance, LLC under Contract No. 89303321CEM000080 with the U.S. Department of Energy. Contributions from Idaho National Laboratory were performed by Battelle Energy Alliance, LLC under DOE Idaho Operations Contract DE AC07 05ID14517.

Notes

1 This material was called Carborundum Monofrax^® K-3 in some reports in the 1970s–1990s. The Carborundum Company in Falconer, NY used Monofrax as a registered trademark for the product; after years the company has changed its name to Monofrax LLC. The company has been acquired by Saint-Gobain in 2022.