The Science of Glass Fining: How Bubbles Are Removed From Molten Glass, and How Better Fining Can Improve Yield, Energy Use and Profit
Inside the invisible battle every furnace fights before glass reaches a mould, and why fining is now a yield, energy and cost problem, not just a bubble problem
Every sheet of float glass, every bottle, every fibre pulled from a furnace has already survived a problem nobody sees: gas trapped inside the melt. If that gas is not removed before the melt reaches the forming stage, it can remain as a seed, blister or other defect that reduces product quality and saleable tonnage. The industry calls the process of removing such bubbles fining, or refining. The principle sounds simple. Running it profitably at furnace scale is not.
Where the Bubbles Actually Come From
Bubbles do not simply enter glass from the surrounding air. Gas is generated and introduced throughout the batch-to-glass conversion process.
Carbonates such as limestone and soda ash release gases as they decompose. Sulfur-containing batch components participate in reactions that generate sulfur-bearing gases [1]. Moisture and gas physically entrained with the batch add further gas to the system. A furnace is therefore dealing with several gas sources at once: gas generated by chemical reactions, gas dissolved in the melt, and gas physically entrained with the batch.
That distinction matters on the floor, not just in a lab. A defect that looks like one simple “bubble problem” at inspection can trace back to different stages or mechanisms, and the appropriate corrective action depends on which mechanism is actually responsible.
From Batch Melting to Fining: Two Different Engineering Problems
Glass melting is not one uniform event. It proceeds through overlapping stages involving batch reactions, melting, homogenisation and refining, and the dominant engineering problem changes through the furnace.
Early on, the furnace is converting heterogeneous raw batch into a continuous glass melt. Later in the furnace, the dominant challenge shifts toward removing dissolved gas and residual bubbles and producing a homogeneous melt suitable for forming.
This is why a furnace can have enough thermal input to melt the batch and still produce poor-quality glass. Melting, refining and flow are related but distinct process problems. Furnace performance therefore needs to be judged on more than tonnes melted alone; product quality, yield and energy use have to be considered together.
Nucleation: Why Bubbles Form Where They Do
A bubble needs a nucleation event. In glass melts, nucleation can occur at interfaces, inclusions, residual particles or other favourable locations when local chemical and thermal conditions permit a gas phase to form. Experimental work on soda-lime-silica glass has shown that gas release and bubble nucleation depend strongly on temperature, composition and the redox state of the melt [2].
The critical point for a furnace operator is that nucleation temperature is not a universal constant. It is characteristic of a particular glass composition and set of process conditions, rather than a number that transfers directly from one furnace to another. The useful question on the floor is therefore not simply “at what temperature do bubbles nucleate?” It is: under our composition and operating conditions, where and when is gas being generated, and does the furnace give those bubbles enough opportunity to grow and leave the melt?
Bubble population itself is measurable through quantities such as bubble number density, bubble diameter and diameter distribution. A mathematical model of bubble number density in a glass tank furnace has been compared against measurements from an actual flat-glass furnace and used to quantify how fining behaviour and furnace position affect bubble density [3]. That turns bubble population from a research curiosity into a usable diagnostic layer.
Fining Gases: Making Bubbles Large Enough to Escape
Once a bubble exists, waiting for it to disappear on its own is rarely enough. Because molten glass has very high viscosity, small bubbles can have extremely low rise velocities. Fining agents help by generating gases that transfer into bubbles, increasing their size and improving their ability to rise and escape on a practical timescale.
The core sequence is: gas generation, gas transfer into the bubble, bubble growth, buoyant rise, and escape from the melt. The same chemistry can also produce the opposite result. When gas generation is excessive or poorly matched to the furnace’s thermal and flow conditions, the outcome can be unwanted foam rather than efficient refining.
Sulfate Fining: Powerful, but Not Unlimited
Sodium sulfate is a major fining agent in commercial soda-lime-silica glass. Sulfate decomposition releases oxygen and sulfur-oxide species, and the resulting gas chemistry interacts with the melt during refining. Thermodynamic work on soda-lime-silica melts has quantified sulfate-decomposition equilibria, SO2 solubility and sulfur retention as functions of temperature and glass chemistry [1].
This produces a rule worth stating plainly: more sulfate does not automatically mean better fining. A peer-reviewed model of foam formation, stability and breakdown in glass-melting furnaces relates foam height to gas flux and to the drainage of liquid lamellae, with drainage affected by bubble radius, density, viscosity, surface tension and interfacial mobility. That work specifically identifies sodium sulfate and silanol groups among the surface-active species influencing the glass-melt interface [4].
The operating implication is direct: if sulfate addition is increased beyond what the thermal, chemical and flow conditions can accommodate, the additional gas generation can contribute to foaming rather than improving bubble removal.
Redox Chemistry: The Hidden Variable
Sulfate fining never acts alone. Its behaviour is coupled to the melt’s redox state, the balance between oxidised and reduced species, including multivalent elements whose oxidation states respond to furnace conditions.
Industrial research has directly connected redox control with melting and fining performance, relating glass redox conditions to observed refining behaviour and examining SO2 evolution in production furnaces [5].
The practical consequence is that a change in redox conditions can alter the effectiveness of a given sulfate addition even when the nominal batch recipe has not changed. Fining-agent dosage and redox control therefore cannot be treated as independent variables.
Water Vapour’s Double Role
Water vapour can influence both the chemistry and the physical behaviour of a glass melt during refining. Experimental work exposed soda-lime-silica melt to high water-vapour partial pressure at 1400 to 1500 degrees C and found bubble nucleation in both oxidised and reduced melts, high bubble-growth rates and pronounced foaming under the conditions studied, with proposed mechanisms involving interactions between water vapour, sulfur-containing species and the melt surface [6].
The defensible conclusion is narrower than “humidity controls every furnace.” Furnace-atmosphere conditions and raw-material moisture become relevant variables when abnormal bubble growth or foaming appears, and should be investigated alongside temperature, redox and batch chemistry rather than dismissed as background conditions.
From Bubble to Bubble-Free: Growth, Rise and Escape
A bubble that has nucleated and grown still has to physically leave the melt. Rise can be approximated using Stokes-type relationships, where velocity depends strongly on bubble size, density difference and melt viscosity. Real furnace behaviour is more complicated: bubbles grow, shrink, merge, deform and exchange gas with the surrounding melt while also interacting with the furnace flow field.
This is also why laboratory-crucible results cannot simply be scaled onto a production furnace. Furnace geometry, convection patterns, residence-time distribution and thermal gradients all affect the outcome, which is why bubble and foam models treat density, viscosity, surface tension, gas flux and interfacial behaviour as coupled variables [4].
The First Rule of Troubleshooting: Diagnose the Defect Before Changing the Recipe
This is where fining science turns into furnace engineering, and it deserves its own rule because process changes made under production pressure can easily hide the real cause of a defect.
A plant seeing more bubbles should first determine what kind of defect has changed and where it is appearing, before changing the fining recipe. Bubble-population measurements provide one useful diagnostic layer, because number density, bubble diameter and size distribution can be compared against furnace position and expected refining behaviour [3].
- Identify the defect. Seeds, blisters, foam, stones, cords and unmelted batch arise from different mechanisms.
- Locate the origin. Determine whether the problem appears to originate during batch melting, around the hot-spot and refining region, in downstream flow regions, or during a later forming operation.
- Separate chemistry from transport. Ask whether the evidence points primarily toward gas generation and redox, or toward temperature, viscosity, residence time and melt flow.
- Check what changed. Compare current operating conditions against the last period when defect rates were stable.
Skipping this sequence creates an expensive learning problem in continuous production: several variables get changed at once, after which the plant cannot establish which change produced the improvement.
When Fining Goes Wrong: What to Investigate
Persistent small bubbles or seeds. Check bubble number density and size distribution, refining temperature, melt viscosity, residence time, fining chemistry, redox conditions and melt flow. More fining agent is one hypothesis to test here, not the automatic answer. The bubble-density work supports treating both bubble population and furnace position as diagnostic information [3].
Excessive foam. Check gas-generation rate, sulfate and other gas-forming reactions, furnace atmosphere, temperature and viscosity, glass composition, surface chemistry, and gas flux and drainage conditions. The foam model provides a quantitative framework linking foam behaviour to gas flux and melt properties rather than treating foaming as a single-variable problem [4].
Defect rate rising after a production increase. A higher pull rate changes the furnace’s thermal and residence-time balance. A defect spike following a pull increase can indicate that a melting, refining or transport limitation has been exposed, and should trigger diagnosis rather than an automatic change to the fining recipe.
Bubbling and Electric Boosting: Different Tools for Different Problems
Chemical fining is not the only process lever. Bubbling introduces gas into the melt and can alter melt circulation, gas transport and bubble-removal behaviour. Electric boosting introduces heat through submerged electrodes and changes the thermal and flow fields.
A peer-reviewed CFD study of an industrial glass-melting furnace directly compared electrodes and bubblers and found interconnected differences in temperature distribution, melt flow, glass quality and process efficiency, concluding that targeted use of electrodes and gas nozzles can be advantageous [7].
The right question when evaluating either intervention is therefore which identified process limitation it addresses. The two should be selected according to furnace design, operating conditions and the diagnosed problem, not treated as interchangeable.
Batch Composition Is a Furnace-Efficiency Variable Too
The furnace does not operate independently of what goes into it. Batch-to-melt conversion kinetics can themselves become a process limitation, which makes batch optimisation a route to improved melting performance rather than purely a raw-material cost decision.
The commercial implication is straightforward: a cheaper raw material is not automatically the lower-cost raw material if it slows conversion, changes gas generation, increases the thermal burden or demands longer residence time.
The Second Rule of Troubleshooting: Measure Before and After
Every furnace intervention needs a measurable target, tracked across more than one category at once, because fining, transport, thermal conditions and furnace control interact:
- Bubble and quality KPIs: bubble number density, size distribution, defect rate, rejection rate, first-pass yield.
- Production KPIs: tonnes per day, pull rate, saleable tonnes per day, line yield.
- Energy KPIs: fuel and electricity per tonne, specific energy consumption, and ideally energy per saleable tonne rather than per tonne melted.
- Process KPIs: fining-agent consumption, redox indicators, batch moisture, cullet percentage, furnace temperature profile.
For profitability analysis, energy per saleable tonne is more informative than energy per tonne melted, because it incorporates the effect of yield. A process change that lowers energy per tonne melted while increasing rejected glass has shifted the economics rather than improved them.
The Economics of Fining: Why Small Yield Gains Are Worth Chasing
The commercial objective of fining is bigger than making bubbles disappear. It is producing more acceptable glass at the lowest sustainable cost per tonne.
The reason small improvements matter is scale. Continuous furnaces operate at high production volumes, so a modest improvement in yield is multiplied across every tonne produced over a full operating year. The same applies to specific energy consumption: a small reduction in energy per saleable tonne accumulates across continuous production into a meaningful change in the annual energy bill.
What that improvement is actually worth depends on the plant’s production volume, product mix, contribution margin and existing defect baseline. The correct financial calculation has to use the plant’s own figures rather than a generic industry-wide saving.
The objective is not minimum sulfate, minimum fuel or maximum pull in isolation. It is the lowest sustainable cost per tonne of glass that still meets the required quality specification.
A Real Industrial Example: Yield Can Be a Furnace-Control Problem, Not Just a Recipe Problem
An industrial case study on an oxy-fuel fired glass furnace, presented at the Slovak Glass Conference and carried out through TNO, examined a plant losing approximately 10.5 percent of production to glass defects. The investigation identified pull variations that were not adequately compensated by energy corrections, measured the furnace’s response to disturbances and energy input, built a black-box furnace model from that data and used it to improve control of the melting tank. Following changes to process settings, the reported production loss fell from about 10.5 percent to about 5 percent [9].
Two things are worth separating here. The exact percentages belong to that one furnace and should not be read as a benchmark. The sequence, however, generalises: defects lead to lost yield, lost yield triggers process diagnosis, and diagnosis can reveal a furnace-control problem rather than a raw-material or fining-agent problem. This is an industrial conference case study rather than a peer-reviewed journal experiment, and is best read as a documented real-plant example alongside the peer-reviewed mechanism work cited elsewhere in this article.
Furnace Optimisation Is Becoming a Data Problem
The traditional approach to furnace operation leans heavily on recipes, operator experience and incremental adjustment. Modern process engineering is moving toward treating the furnace as an integrated system.
A 2025 study of an electric-boosted natural-gas glass furnace used real operating data, data reconciliation, gross-error detection, regression modelling and constrained optimisation to identify combinations of process variables associated with lower specific energy consumption. It evaluated raw materials, cullet, water, raw-material temperature, electric boosting and fuel together rather than optimising any single variable in isolation [8]. The optimised specific-energy figure reported in that work belongs to the furnace and constraints studied, and is not a general target for other plants.
The transferable method is the sequence: measure, reconcile the data, model the furnace, identify the sensitive variables, then optimise within quality and production constraints.
Where Fining Is Headed Now: Decarbonisation and Emissions
The chemistry above still holds, but the pressure shaping fining research today is no longer only clarity and defect control. It increasingly includes emissions, energy and carbon.
Sulfate fining introduces sulfur into the process and can contribute to sulfur-oxide emissions, with the relevant sulfur balance depending on glass composition, furnace conditions and how much sulfur is retained in the melt versus released [1]. Research is exploring alternative fining systems involving tin, cerium and halide-assisted chemistry, though their effectiveness and environmental trade-offs depend heavily on glass composition and process conditions. Arsenic- and antimony-based fining systems have been widely displaced in many applications on toxicity grounds.
A second, more structural shift sits upstream of fining altogether. Soda ash is an energy-intensive sodium source, and a 2026 peer-reviewed study demonstrated a series of container-glass compositions that can be melted from natural raw materials without any added soda ash, each formulation containing up to 50 percent recycled container cullet. The reported compositions showed temperature-viscosity profiles, liquidus temperatures, electrical resistivities and basic processability comparable to conventional soda-lime-silica glass, with decreased melting-energy consumption and markedly reduced batch-to-melt conversion time [10]. If such compositions scale while holding the required melting, refining and product-performance characteristics, the fining problem itself changes with the batch chemistry.
Furnace practice is also getting more specific about when fining chemistry stops mattering. A process document covering sulfate use in submerged-combustion melted glass describes a minimum glass fining temperature, above which sulfate contributed little additional bubble removal under the conditions investigated and glass viscosity became the dominant factor [11]. That source is a patent document rather than a peer-reviewed study, and the temperature it reports is specific to its own process and measurement point, so it is best read as a process-specific illustration of the principle rather than a number to apply elsewhere.
None of this changes the underlying physics of nucleation, gas diffusion or buoyant rise. It changes how the process gets optimised, now that the cost of a furnace decision is measured in yield, energy, emissions and operating cost at the same time.
India and China Are Not Playing by the Same Rulebook
The scientific literature on fining is international. The regulatory pressure acting on it is not. The practical effect of emissions and efficiency requirements depends on the standards, implementation mechanisms and enforcement arrangements applicable to each plant.
In India, glass manufacturing operates under the CPCB and state pollution-control framework, with national emission standards covering SO₂ and NOx and project-level environmental-clearance requirements applying according to the relevant project criteria. The glass-sector emission standard specifies SO₂ limits of 500 mg/Nm³ for natural-gas firing and 1500 mg/Nm³ for other fuels, with a NOx limit of 1000 mg/Nm³.
China’s 2026–2028 action plan explicitly includes flat glass among nine high-energy-consuming, high-emission industries and establishes sector-specific energy-saving and carbon-reduction measures through 2028. The policy is therefore relevant to furnace efficiency and investment decisions in China’s flat-glass sector, while the consequences for individual plants depend on the applicable implementation and enforcement measures.
For manufacturers in either market, the important business question is how those requirements translate into actual plant costs, investment requirements, operating constraints and product economics.
What a Furnace Operator Should Actually Take Away
The science of fining reduces to one chain: gas is generated or introduced, bubbles nucleate, gas moves between melt and bubble, bubbles grow, bubbles rise, bubbles escape.
Running a profitable furnace runs on a second chain layered on top of the first: measure the defect, locate its origin, separate chemistry from transport, make one targeted change, measure the response, and calculate what it actually did to saleable tonnes and cost per tonne.
The strongest furnace in this business is not the one running the most fining agent, the highest temperature or the newest technology. It is the one that consistently produces the required glass quality at the lowest sustainable cost per saleable tonne, and can demonstrate it with its own process and financial data rather than someone else’s averages. For an industry that looks from the outside like it is simply melting sand, that is the entire difference between a furnace that survives the next efficiency mandate and one that does not.
References
[1] Beerkens, R.G.C. (2003). Sulfate Decomposition and Sodium Oxide Activity in Soda-Lime-Silica Glass Melts. Journal of the American Ceramic Society. DOI: 10.1111/j.1151-2916.2003.tb03578.x
[2] Vernerová, M., Němec, L., Kloužek, J., Hujová, M. (2018). Gas Release Phenomena in Soda-Lime-Silica Glass. Journal of Non-Crystalline Solids. DOI: 10.1016/j.jnoncrysol.2018.07.058
[3] Oda, K., Kaminoyama, M. (2009). Mathematical Model of Bubble Number Density in Glass Tank Furnace. Journal of the Ceramic Society of Japan, 117(1366), 736-741. DOI: 10.2109/jcersj2.117.736
[4] van der Schaaf, J., Beerkens, R.G.C. (2006). A Model for Foam Formation, Stability, and Breakdown in Glass-Melting Furnaces. Journal of Colloid and Interface Science, 295(1), 218-229. DOI: 10.1016/j.jcis.2005.07.068
[5] Manring, W.H., Diken, G.M. (1980). A Practical Approach to Evaluating Redox Phenomena Involved in the Melting-Fining of Soda-Lime Glasses. Journal of Non-Crystalline Solids. DOI: 10.1016/0022-3093(80)90537-2
[6] Vernerová, M., Kloužek, J., Němec, L. (2015). Reaction of Soda-Lime-Silica Glass Melt With Water Vapour at Melting Temperatures. Journal of Non-Crystalline Solids, 416, 21-30. DOI: 10.1016/j.jnoncrysol.2015.02.020
[7] Daurer, G., Raič, J., Demuth, M., Gaber, C., Hochenauer, C. (2023). Detailed Comparison of Physical Fining Methods in an Industrial Glass Melting Furnace Using Coupled CFD Simulations. Applied Thermal Engineering. DOI: 10.1016/j.applthermaleng.2023.121022
[8] Kodak, O. et al. (2025). Improvement of Energy Performance of Glass Furnaces Using Modelling and Optimization Techniques. Processes, 13(11), 3739. DOI: 10.3390/pr13113739
[9] Koch, R., de Leede, G.L.A. (2000). Yield Improvement for an Industrial Oxy-Fuel Fired Glass Furnace. Slovak Glass Conference, Trenčín, pp. 35-41. Industrial conference case study.
[10] Engelmann, C., Roos, C. (2026). Soda Ash Free Batches for Novel Container Glass Compositions. International Journal of Applied Glass Science, 17(3), e70036. DOI: 10.1111/ijag.70036
[11] Utilization of Sulfate in the Fining of Submerged Combustion Melted Glass. US Patent Application 2021/0094860. Patent document, not peer-reviewed literature.
Zest Glass Markets
