Casthouse Melt Refining: Hydrogen Degassing and Ceramic Foam Filtration (CFF) | 铸造车间熔体精炼：铝液旋转除气与陶瓷泡沫过滤（CFF）技术

Melt Refining & Filtration · Published 2026-10-30

1. Three contaminants, three mechanisms

The literature divides aluminium melt contamination into three categories, not two:

Class	Examples	Phase	Removed by
Dissolved gas	Hydrogen	Dissolved atomic	Rotary degassing
Non-metallic inclusions	Oxides, carbides, nitrides, borides, chlorides, spinel	Solid particles / films	Filtration (+ flotation)
Alkali / alkaline earth metals	Na, Ca, Li	Dissolved elemental	Chlorine / flux (+ filter adsorption)

The third category is where most short summaries stop short. Sodium at 5–10 ppm degrades mechanical properties in aluminium alloys, and chlorine degassing is described as the most effective alkali removal method available. Alumina filters provide a secondary effect here — reducing dissolved Na and K through surface chemistry, not mechanical capture.

2. Why hydrogen is the only gas that matters

Hydrogen is described as the only gas with meaningful solubility in liquid aluminium under standard atmospheric conditions. It enters through moisture on scrap, damp fluxes, furnace atmosphere humidity, and the direct reaction with water vapour:

3 H₂O + 2 Al → 6 [H] + Al₂O₃

Note what that equation does: the reaction that creates the gas problem simultaneously creates the inclusion problem. Every kilogram of moisture that reaches the melt generates both dissolved hydrogen and alumina.

State	H₂ solubility	Consequence
Liquid aluminium, 750 °C	≈ 0.690 mL/100 g	Hydrogen stays in solution
Solid aluminium, 660 °C	≈ 0.036 mL/100 g	A 19-fold drop across the solidification front

That solubility leap is the porosity mechanism: the melt cannot retain what it held, so atomic hydrogen recombines into molecular H₂ and forms pores. One source states the liquid-to-solid ratio as “roughly 20 times”; the measured figures give 19.2×.

The inclusion interaction is not incidental. A contaminated melt contains more nucleation sites on which bubbles can form. One study frames it precisely: in industrial-purity aluminium, pinholes appear above about 0.1 mL/100 g, whereas high-purity aluminium shows no pores until about 0.4 mL/100 g — a four-fold difference attributable to oxide inclusions acting as pore nuclei. This is why degassing and inclusion removal are described as operations that should be performed together rather than as independent steps.

3. How rotary degassing actually works

The mechanism is a partial pressure gradient, not a chemical reaction:

• In a bubble of pure argon just introduced into the melt, the hydrogen partial pressure is essentially zero.
• In the surrounding melt, dissolved hydrogen corresponds to a finite equilibrium partial pressure — calculable from Sieverts’ law: [H] = K × √(P_H₂)
• That gradient drives hydrogen atoms to diffuse from melt into bubble. Flux is described by Fick’s first law, with a hydrogen diffusion coefficient in liquid aluminium of approximately 3.2 × 10⁻³ cm²/s at 700 °C (Eichenauer and Markopoulos, 1974).

Why the rotor exists: interfacial area

Total hydrogen removal rate depends on the gas–liquid interfacial area available for mass transfer. For a fixed volume of gas:

A = 6 × V_gas / d_bubble

Halving bubble diameter quadruples the interfacial area for the same gas volume. This is the entire reason a spinning rotor outperforms a lance:

Injection method	Bubble diameter	Note
Lance injection	5 – 20 mm	Baseline
Rotary degassing rotor	0.5 – 3 mm	A 5–40 fold diameter reduction → comparable increase in mass-transfer area per unit gas

Jahn and Schwerdtfeger (Metallurgical Transactions B, 1978) established bubble size distribution in liquid aluminium as a function of rotor design and speed, finding bubble diameter scales approximately with rotor tip speed to the power of −0.6 — with the caveat that secondary coalescence limits further reduction at high speed.

4. What a rotor achieves in practice

Parameter	Documented value	Source
Rotor speed	200 – 600 rpm (one source 300–600)	/
Treatment temperature	720 – 760 °C	; 720–750 °C
Gas flow	15 – 30 L/min per rotor
Treatment time	10 – 20 min
Diffusion coefficient DH	≈ 3.2 × 10⁻³ cm²/s at 700 °C

Hydrogen removal: the numbers

Configuration	H₂ reduction	Post-treatment level
Argon alone, single-pass inline	50 – 70%	0.08–0.12 mL/100 g from 0.20–0.30 initial
Argon alone, optimized dual-rotor	up to 80%
Argon / nitrogen + 2–5% Cl₂	70 – 82%
Bath treatment, 720–760 °C	—	0.25–0.35 → 0.08–0.12 mL/100 g
General rotary degassing	—	0.10–0.15 mL/100 g
Optimized documented run	—	0.40 → 0.10 mL/100 g (400 rpm, 4 L/min, 15 min, 745 °C, 2 t)

Nitrogen is not equivalent to argon in every alloy. Nitrogen achieves similar degassing efficiency at 30–60% of argon cost, but can form nitrides with lithium, with magnesium at high concentration, and with some rare earth additions. For Li-containing alloys (2xxx and 8xxx series with Li) and very high-Mg alloys (>4% Mg), argon is preferred.

5. Chlorine: what it does, and why it is being dropped

The draft this page replaces described rotary degassing as injecting “argon/chlorine gas mixtures”. That is half right, and the half that is wrong matters.

Chlorine is an additive at 2–5%, not the degassing agent. The inert bubble does the hydrogen removal by partial pressure gradient. Chlorine adds three separate benefits:

1. Inclusion agglomeration. Chlorine reacts with aluminium to form AlCl₃ vapour, which rises through the melt and collects non-metallic inclusions by flotation. Granger at Pechiney (Light Metals, 1998) measured average inclusion size rising from approximately 8 μm to 25 μm — a three-fold increase.
2. Alkali removal. Chlorine reacts with dissolved sodium and calcium to form volatile chlorides that escape with the gas bubbles.
3. Enhanced hydrogen removal — the mechanical and chemical action together reach 70–82%, versus 50–70% for argon alone.

Why many plants stopped using it

Chlorine is toxic — OSHA PEL 1 ppm, IDLH 10 ppm. The aluminium chloride vapour generated during treatment is toxic and corrosive. Installations using chlorine-bearing gas require fume extraction, gas detection and emergency procedures. One source states that many European facilities have moved to argon-only degassing, accepting the performance trade-off. Another describes chlorine-based degassing as hazardous and restricted in many jurisdictions.

Note the consequence: a plant that drops chlorine loses the 8→25 μm agglomeration step. Its filter then sees a finer inclusion population, for which capture efficiency is lower. The trade-off is not simply “less hydrogen removal”.

6. CFF: pore size is not capture size — and “sub-micron” is wrong by three orders of magnitude

Pore size vs effective capture size

PPI	Pore opening	Effective capture (80% efficiency)	Typical application
10	≈ 2.59 mm	> 100 μm	Pre-filtration; very large inclusions
20	≈ 1.27 mm	> 50 μm	Standard casting; general cleanliness
30	≈ 0.84 mm	> 20 μm	Good cleanliness; billet casting
40	≈ 0.58 mm	> 10 μm	High cleanliness; aerospace billet
50	≈ 0.42 mm	> 5 μm	Very high cleanliness requirements
60	≈ 0.42 mm	> 3 μm	Ultra-clean applications

Pore-opening and capture figures are manufacturer-published and derived from PoDFA (Porous Disc Filtration Apparatus) testing. “Effective capture size” is defined there as the inclusion size at which removal efficiency exceeds approximately 80%; smaller inclusions are captured at lower efficiency due to reduced inertial impaction.

Cross-check: an independent plant measurement

Vendor efficiency tables should not be taken alone. A peer-reviewed study reporting on-line monitoring at a working plant, using LiMCA inclusion counting, gives CFF performance by size band:

Inclusion size band	CFF efficiency	Note
20 – 29 μm	≈ 68%	Efficiency rises with size; the study notes the plant’s performance on small inclusions was “less optimal” than on large ones. N20 count fell from 33.62 to 6.82 k/kg across CFF — a 79.7% reduction.
up to 60 μm	≈ 60%
70 – 79 μm	≈ 88%
> 80 μm	≈ 97%

This is the pattern that matters: CFF is confident above roughly 80 μm and merely partial in the 20–60 μm band. Any specification implying near-total capture of fine inclusions is not describing a ceramic foam filter.

7. How a filter captures: three mechanisms, not one

A ceramic foam filter is an open-cell network of ceramic struts with roughly 75–90% porosity. Metal travels a path typically 3 to 5 times the filter thickness. Capture occurs by:

1. Mechanical interception (sieving) — inclusions larger than the minimum pore throat are physically stopped. Dominant for coarse inclusions and low PPI.
2. Inertial impaction — inclusions with sufficient mass cannot follow the curved streamlines around a strut, and instead strike and adhere to it. More effective at higher velocity and for denser particles.
3. Surface adhesion — the alumina surface has chemical affinity for aluminium oxide; contact can retain particles with no mechanical trapping. This is the mechanism that captures fines below 10 μm, which is why capture size is far below pore size.

Captured inclusions accumulate on struts and at pore throats, progressively narrowing flow passages. This improves efficiency during a campaign, as the accumulated layer acts as an additional filter medium — but excessive build-up raises resistance and can block the filter entirely.

8. Selecting PPI: the trade-off is flow, not cleanliness

PPI	Filtration efficiency	Velocity reduction	Blockage risk
10	40 – 55%	minimal	very low
20	55 – 70%	5 – 10%	low
30	70 – 82%	10 – 20%	low–medium
40	82 – 91%	20 – 35%	medium
50	88 – 95%	35 – 55%	medium–high
60	92 – 97%	55 – 75%	high

Manufacturer data. Note that efficiency figures in §6 and §8 come from different vendors with different bases; the shape of the curve is consistent, the absolute values are not directly comparable across tables. This is itself a reason not to treat any single table as specification.

The failure mode of over-specification is worse than under-specification. Too fine a PPI blocks before the mould fills — the result is a misrun, which is 100% scrap, and no cleanliness benefit is achieved if the casting never fills. The correct selection requires knowing three things: the cleanliness level entering the filter, the section thickness and volume that set flow rate, and the quality specification of the finished casting.

9. Operating discipline: what actually decides real-world performance

Parameter	Documented value	Consequence if violated
Preheat	≥ 460 °C, for 15 – 30 min	Thermal shock cracking; retained moisture
Operating head	must never exceed 250 mm	Impending blockage; bypass
Linear velocity	Gravity launder 5–15 cm/s (max 25); DC billet 8–20 (max 30); strip 10–25 (max 35); LPDC 3–10 (max 15)	Above maximum: inclusion re-entrainment — captured particles are dislodged downstream, causing sudden quality loss
Seal	Gasket must be tight	Metal bypasses the filter entirely, rendering the process useless
Flux upstream	Never add flux upstream of the filter	Listed as a prohibited action
Filter life	Single use	CFF is disposable; change every cast or when pressure drop indicates blockage
Rotor life	Graphite 3–6 weeks; silicon nitride 6–12 months	Worn rotor: gas-shear efficiency collapses, hydrogen drifts up

10. The TiB₂ problem: an additive that behaves like an inclusion

The draft this page replaces listed TiB₂ alongside Al₂O₃ as a non-metallic inclusion to be removed. That is the wrong category, and the error has practical consequences.

TiB₂ is deliberately added, as Al–Ti–B grain refining master alloy, to produce an equiaxed grain structure. It is a process input, not contamination. But it then does three things that make it behave like an inclusion:

Behaviour	Mechanism	Consequence
Settling and agglomeration	Settling is much faster than Stokes’ law predicts, because particles self-agglomerate — consistent with CFD showing agglomerates of several tens of micrometres	Grain refinement fades with holding time; fewer effective nuclei
Reduces filtration efficiency	The “inclusion bridge” mechanism: inclusions bridge across ceramic cells and raise efficiency; TiB₂ and TiC particles are thought to prevent or destroy those bridges	Adding grain refiner lowers the efficiency of the filter on high-inclusion-load melts
Can be over-removed	Oxide bifilms act as collection sites; TiB₂ adheres to them. Filtration and gas flotation then carry the bifilm — with its attached TiB₂ — out of the melt	Excessive TiB₂ removal, i.e. loss of grain refinement. Boron enrichment has been measured in dross after rotary degassing of grain-refined 5083

11. Sequence: why the order of operations is not arbitrary

Putting the interactions above together gives a defensible order, and — more usefully — shows what each step does to the next:

1. Skim / remove dross before treatment. Removing slag and inclusions is described as the foundation of effective degassing.
2. Degas (with chlorine, if used). Chlorine’s flotation effect triples average inclusion size from ~8 to ~25 μm — which improves what the filter downstream can catch.
3. Filter. Coarse stage first if the melt is dirty; fine stage for the specification.
4. Add grain refiner after degassing and filtration, to avoid stripping it into the dross and to avoid depressing filter efficiency.
5. Protect the surface and cast promptly. Hydrogen re-gassing from humid ladle transfers is a documented cause of post-treatment hydrogen rise.

12. What neither process does

• CFF does not remove dissolved hydrogen. Filtration removes solids. The literature is explicit that, in contrast to purging gas treatment, filtration removes even the finest oxides — and the existence of active research into reactive filter coatings designed to add hydrogen removal capability confirms standard CFF lacks it.
• Degassing does not finish inclusion removal. Rotary degassing floats inclusions toward the surface as a secondary benefit, but the fine population passes through. One industry source quantifies rotary degassing as removing 10 μm and above; finer material remains for the filter.
• Finishing one does not compensate for failing the other. Because inclusions nucleate pores, a melt with low hydrogen and high inclusion content still forms porosity — and vice versa.
• The filter itself can contaminate. Standard phosphate-bonded filters use an AlPO₄ binder and are documented at 1.0–3.0 ppm phosphorus pickup — described as marginal for EC grade and not acceptable for 4N purity. Phosphate-free colloidal-alumina-bonded filters exist for this reason.

13. What this site could not verify

• A single authoritative PPI-to-capture-size standard. Two vendors publish different pore sizes for the same nominal PPI (30 PPI appears as 0.84–0.85 mm and as 0.8–1.0 mm) and different capture bands (50 PPI is given as both “>5 μm” and “10 μm”). We present both and do not select a canonical figure.
• The “0.1 micron” claim. Reported as vendor-published and in contradiction with the same vendor’s own table. Not resolved here.
• Chlorine’s precise regulatory status by jurisdiction. Sources describe it as hazardous, restricted, and abandoned by many European facilities, but we found no primary legal instrument to cite. Readers requiring compliance certainty should consult the applicable national regulation.
• Efficiency figures are largely vendor-published. The LiMCA plant study in §6 is the exception and is the only independent measurement reported here.
• Source-character limitation, stated plainly: for PPI selection, filtration mechanisms and operating windows, the most detailed published engineering data is manufacturer literature. We have cited it as such and have linked to none of it.

14. What this page is not

• Not a supplier recommendation. No manufacturer is recommended and no vendor site is linked. See Mission & Independence.
• Not specification or design data. Pore sizes, efficiencies and operating windows are reported from named sources, not issued as procurement or engineering specification. See Terms of Use §6.
• Not a substitute for measurement. Hydrogen is measured by Telegas/Alspek/Hydris (±0.02–0.03 mL/100 g) or reduced-pressure test; cleanliness by PoDFA, Prefil or LiMCA. This page describes mechanisms, not how to qualify a specific melt.

Related: Melt Refining & Filtration · Casthouse Equipment · Five Dross Routes Compared · Editorial Standards

---

中文摘要 · 熔体精炼：除气与过滤各自除去什么

除气除去的是溶解氢——溶液中的气体。陶瓷泡沫过滤除去的是固态非金属夹杂物。二者是不同相态、受不同机理支配、用不同仪器测量，且不能互相替代。陶瓷泡沫过滤器不脱除溶解氢；旋转除气也除不掉残留的夹杂物。铸造车间熔体精炼中几乎一切混淆，都源自把这两件事当成了一个叫”精炼”的操作。

1. 三类污染物，三种机理。文献把铝液污染分为三类而非两类：溶解气体（氢）→ 由旋转除气去除；非金属夹杂物（氧化物、碳化物、氮化物、硼化物、氯化物、尖晶石）→ 由过滤去除（辅以浮选）；碱金属与碱土金属（Na、Ca、Li）→ 由氯气/熔剂去除（辅以过滤吸附）。第三类是多数简略摘要的盲区：钠在 5–10 ppm 即会劣化铝合金力学性能，而氯气除气被描述为现有最有效的除碱方法。氧化铝过滤板在此提供次级效果——通过表面化学降低溶解态 Na、K，而非机械捕集。

2. 为什么氢是唯一要紧的气体。氢被描述为标准大气条件下唯一在液态铝中有显著溶解度的气体。它经炉料表面水分、受潮熔剂、炉气湿度以及与水蒸气的直接反应进入：3H₂O + 2Al → 6[H] + Al₂O₃。请注意这个方程式的双重后果：制造气体问题的反应，同时制造了夹杂物问题——每一份进入熔体的水分，既产生溶解氢，也产生氧化铝。溶解度跃变就是针孔机理本身：750 ℃ 液态铝 ≈0.690 mL/100 g，660 ℃ 固态铝 ≈0.036 mL/100 g，凝固前沿骤降 19 倍（文献常述”约 20 倍”）。熔体保不住它原本溶解的氢，原子氢复合成分子氢并形成气孔。夹杂物的交互作用不是附带的：受污染的熔体含有更多气泡形核位点。工业纯铝在约 0.1 mL/100 g 以上出现针孔，而高纯铝要到约 0.4 mL/100 g 才出现气孔——这四倍差异归因于氧化物夹杂作为气孔核心。这正是除气与除渣被描述为应当合并进行而非各自独立的原因。

3. 旋转除气究竟如何工作。机理是分压差，不是化学反应：刚注入的纯氩气泡内氢分压基本为零；周围熔体中溶解氢对应一个有限的平衡分压，可由西华特定律 [H] = K × √(P_H₂) 计算；该梯度驱动氢原子从熔体扩散进入气泡，通量由菲克第一定律描述，700 ℃ 时氢在液态铝中的扩散系数约 3.2 × 10⁻³ cm²/s（Eichenauer 与 Markopoulos，1974）。转子为什么存在——界面面积：总脱氢速率取决于可用于传质的气液界面面积，对固定气量有 A = 6 × V_气体 / d_气泡；气泡直径减半，同样气量的界面面积翻两番。这就是旋转转子优于喷枪的全部原因：喷枪注入产生 5–20 mm 气泡；设计良好的除气转子产生 0.5–3 mm 气泡——直径缩小 5–40 倍，对应单位气量传质面积的同等提升。Jahn 与 Schwerdtfeger（《Metallurgical Transactions B》，1978）确立了液态铝中气泡尺寸随转子设计与转速的分布，发现气泡直径约与转子叶尖速度的 −0.6 次方成比例，并指出高速时二次聚并会限制进一步细化。

4. 转子实际达到什么水平。转速 200–600 rpm（另有来源记 300–600）；处理温度 720–760 ℃（另记 720–750 ℃）；气量 15–30 L/min/转子；处理时间 10–20 min。脱氢率：纯氩单程在线 50–70%，处理后 0.08–0.12 mL/100 g（初始 0.20–0.30）；纯氩优化双转子最高 80%；氩/氮 + 2–5% Cl₂ 70–82%；另有记载 720–760 ℃ 下 0.25–0.35 → 0.08–0.12 mL/100 g；一般性记载 0.10–0.15 mL/100 g；一份有完整参数的优化记录：0.40 → 0.10 mL/100 g（400 rpm、4 L/min、15 min、745 ℃、2 t）。氮并不在所有合金中等效于氩：氮的除气效率相近、成本仅为氩的 30–60%，但会与锂、高浓度镁及部分稀土添加元素生成氮化物。含锂合金（2xxx、8xxx 系含 Li）与高镁合金（>4% Mg）宜用氩。

5. 氯气：它做什么，以及为什么正在被弃用。本页所替换的原稿把旋转除气描述为注入”氩氯混合气”——这只对一半，而错的那半恰恰要紧。氯气是 2–5% 的添加剂，不是除气剂本身。惰性气泡靠分压梯度完成脱氢；氯气另外提供三项独立收益：①夹杂团聚——氯与铝反应生成 AlCl₃ 蒸气，上升过程中通过浮选收集非金属夹杂物；Granger（Pechiney，《Light Metals》，1998）实测平均夹杂尺寸从约 8 μm 增至 25 μm，放大三倍。②除碱——氯与溶解态钠、钙反应生成随气泡逸出的挥发性氯化物。③强化脱氢——机械与化学作用合计达 70–82%，对比纯氩的 50–70%。团聚收益正是连接两个工序的那一条：把平均夹杂尺寸从 8 μm 放大到 25 μm，等于把相当一部分夹杂群体推进了陶瓷泡沫过滤器高效捕集的尺寸带。氯气除气与 CFF 过滤不是并列选项——前者改善后者的进料。为什么许多厂停用：氯气有毒（OSHA PEL 1 ppm，IDLH 10 ppm），处理中产生的氯化铝蒸气有毒且腐蚀性强，使用含氯气体的装置必须配备烟气抽吸、气体探测与应急程序。有来源称许多欧洲工厂已改用纯氩除气，接受性能上的折让；另有来源称氯基除气在许多辖区受限。请注意这个后果：停用氯气的工厂同时失去了 8→25 μm 的团聚步骤，其过滤器随后面对的是更细的夹杂群体，而细夹杂的捕集效率更低。这个折让并不只是”脱氢少一点”。

6. CFF：孔径不等于捕集尺寸，”亚微米”错三个数量级。更正：陶瓷泡沫过滤器能捕集”亚微米级”夹杂的常见说法没有依据。孔口本身就是数百微米量级——30 PPI 过滤板孔口约 840 μm，比 1 微米大约三个数量级。过滤板确实能捕集远小于孔口的颗粒，靠的是惯性撞击与表面粘附而非筛分；但高效捕集带是几十微米，不是亚微米。孔径 vs 有效捕集尺寸（80% 效率）：10 PPI ≈2.59 mm / >100 μm；20 PPI ≈1.27 mm / >50 μm；30 PPI ≈0.84 mm / >20 μm（良好洁净度、棒材铸造）；40 PPI ≈0.58 mm / >10 μm；50 PPI ≈0.42 mm / >5 μm（极高洁净度要求）；60 PPI ≈0.42 mm / >3 μm。孔径与捕集数据为制造商发布、源自 PoDFA（多孔盘过滤装置）测试；”有效捕集尺寸”定义为去除效率超过约 80% 的夹杂尺寸，更小的夹杂因惯性撞击减弱而捕集效率较低。独立交叉验证——一项工厂实测：厂商效率表不应单独采信。一项报告了在役工厂在线监测的同行评议研究，用 LiMCA 夹杂计数给出 CFF 的分尺寸带表现：20–29 μm ≈68%；至 60 μm ≈60%；70–79 μm ≈88%；>80 μm ≈97%。N20 计数经 CFF 从 33.62 降至 6.82 k/kg，降幅 79.7%。该研究明确指出工厂对小夹杂的表现”不如对大夹杂理想”。这个规律才是要紧的：CFF 在约 80 μm 以上很有把握，在 20–60 μm 带只是部分有效。任何暗示细夹杂近乎全捕集的说法，描述的不是陶瓷泡沫过滤器。一处我们不裁决的厂商内部矛盾：某制造商页面称陶瓷泡沫过滤器在”特殊高纯应用中精度可达 0.1 微米”，而同一页的表格却把 50 PPI 写为 0.42 mm。本站两组都列，不代为挑选。真正能去除 5 μm 以下物质的通常是滤芯式过滤器（cartridge filter）而非 CFF——一份政府技术报告记载滤芯过滤器对 50 μm 以下夹杂的去除效率达 95% 以上。

7. 过滤板如何捕集：三种机理，不是一种。陶瓷泡沫过滤板是开孔网络结构，孔隙率约 75–90%，金属流经的路径通常为过滤板厚度的 3 至 5 倍。捕集通过三种同时发生的机理：①机械拦截（筛分）——大于最小孔喉的夹杂被物理挡住，对粗大夹杂与低 PPI 占主导；②惯性撞击——质量足够的夹杂无法跟随绕过支杆的弯曲流线，而是直行撞击并粘附于支杆，流速越高、颗粒密度越大越有效；③表面粘附——氧化铝表面对氧化铝夹杂有化学亲和力，接触即可滞留而无需机械截留，这正是捕集 10 μm 以下细颗粒的机理，也是捕集尺寸远小于孔径的原因。被捕集的夹杂在支杆表面与孔喉处累积，逐渐收窄流道，这在一次浇注周期内提高了效率——累积层充当了附加过滤介质；但过度累积会抬高阻力并最终完全堵塞。实践中的悖论：正在工作的过滤板会逐渐变得更具阻滞性。加拿大政府技术文献一方面指出滤饼过滤在铝工业中并非有意采用（因为累积物最终会堵塞），另一方面又记载滤饼是被认为的主要机理。两句话都成立：滤饼会形成、在一定程度内有益、是被管理而非被设计的。

8. 选择 PPI：权衡的是流量，不是洁净度。效率与流速折让：10 PPI 40–55%（流速损失极小）；20 PPI 55–70%（5–10%）；30 PPI 70–82%（10–20%）；40 PPI 82–91%（20–35%）；50 PPI 88–95%（35–55%）；60 PPI 92–97%（55–75%）。过度选型（选太细）的失效模式比选型不足更糟：过细的 PPI 会在型腔充满之前就堵塞——结果是浇不足（misrun），属于 100% 废品，而如果铸件根本没充满，任何洁净度收益都不存在。正确选型需要知道三件事：进入过滤板的洁净度水平、决定流速的壁厚与总体积、以及成品的质量规范。脏料要用更粗的 PPI，不是更细的。高夹杂负荷配细过滤板会过早堵塞。这正是关键工况推荐两级过滤的原因：上游粗级 20–30 PPI 捕集大颗粒并保护下游，随后 40–50 PPI 处理细颗粒。

9. 操作纪律：真正决定现场表现的是什么。预热 ≥460 ℃、15–30 min（否则热震开裂、残留水分）；操作压头 任何时刻不得超过 250 mm（否则预示堵塞与旁路）；线速度——重力流槽 5–15 cm/s（上限 25）、DC 棒材 8–20（上限 30）、带材连铸 10–25（上限 35）、低压压铸 3–10（上限 15），超速发生夹杂再夹带：已捕集颗粒被冲脱带入下游，造成突发性质量劣化；密封——垫片必须严密，密封失效则金属完全旁通过滤板，使整个工序失去意义；过滤板上游严禁加熔剂；过滤板为一次性，每浇次更换或压降指示堵塞时更换；转子寿命——石墨 3–6 周，氮化硅 6–12 个月，转子磨损后气体剪切效率崩塌、氢含量回升。值得直接引用一句制造商判断：“安装与预热质量对过滤表现的影响，远大于过滤板品牌或等级之间的差异。”规范执行据称可比通常的随意做法把现场效率提高 300–500%。你买什么等级，不如你怎么装它。

10. TiB₂ 问题：一种表现得像夹杂的添加剂。本页所替换的原稿把 TiB₂ 与 Al₂O₃ 并列为要去除的非金属夹杂物。这是错误的归类，而这个错误有实际后果。TiB₂ 是人为加入的——以 Al–Ti–B 晶粒细化中间合金的形式加入，用于获得等轴晶组织。它是工艺投入，不是污染物。但它随后做了三件使它表现得像夹杂的事：①沉降与团聚——沉降速度远快于斯托克斯定律预测，因为颗粒会自团聚（CFD 显示熔体中存在数十微米的团聚体）；后果是晶粒细化随保温时间衰退、有效晶核减少。②降低过滤效率——所谓”夹杂架桥”机理：夹杂在陶瓷孔格间架桥从而提高效率，而 TiB₂ 与 TiC 颗粒被认为会阻止或破坏这些架桥；后果是加入晶粒细化剂会降低过滤器的效率（在高夹杂负荷熔体上）。③可能被过度去除——氧化双膜（bifilm）充当收集位点，TiB₂ 粘附其上，过滤与气体浮选随后把双膜连同其附着的 TiB₂ 一并带出熔体；后果是 TiB₂ 被过度去除，即晶粒细化失效。已有研究在晶粒细化 5083 合金旋转除气后的浮渣样品中测到硼富集。操作上的推论——加入时机：由于除气会把双膜浮选至浮渣，而双膜携带 TiB₂，在旋转除气之前或期间加入晶粒细化剂，可能使其相当一部分被带入浮渣。这正是晶粒细化剂通常在除气单元之后加入的原因。

11. 顺序：为什么工序次序不是任意的。把上述交互作用串起来，可得到一个站得住的次序，更有用的是它说明了每一步对下一步做了什么：①处理前先扒渣/除渣——除渣被描述为有效除气的基础；②除气（若用氯，则在此）——氯的浮选效应把平均夹杂尺寸从约 8 μm 放大到 25 μm，从而改善下游过滤板能捕集到的东西；③过滤——脏料先粗级，再按规范用细级；④除气与过滤之后再加晶粒细化剂——避免其被带入浮渣，也避免压低过滤效率；⑤保护液面并尽快浇注——潮湿转包导致的再吸氢是被记载的处理后氢回升原因。

12. 两个工序都不做的事。CFF 不脱除溶解氢——过滤去除的是固体；文献明确指出与吹气处理相反，过滤能去除最细的氧化物，而”反应性过滤涂层”这一旨在增加脱氢能力的研究方向的存在，恰好反证标准 CFF 不具备该功能。除气不完成夹杂去除——旋转除气把夹杂浮选至液面是其次级收益，但细颗粒群体仍会穿过；一份行业来源把旋转除气量化为去除 10 μm 及以上，更细的材料留给过滤板。做好一个不能补偿另一个的失败——因为夹杂是气孔的形核核心，低氢高夹杂的熔体仍会形成气孔，反之亦然。过滤板本身也会污染——标准磷酸盐结合过滤板使用 AlPO₄ 结合剂，记载的磷摄入量为 1.0–3.0 ppm，对电工级（EC grade）被描述为”勉强可接受”，对 4N 纯度则不可接受；无磷酸盐的胶态氧化铝结合过滤板正是为此而存在。

13. 本站未能核实的部分。PPI 与捕集尺寸的单一权威标准——两家厂商对同一标称 PPI 给出不同孔径（30 PPI 分别作 0.84–0.85 mm 与 0.8–1.0 mm），捕集带也不同（50 PPI 分别作”>5 μm”与”10 μm”）；本站两者并列，不代为指定标准值。“0.1 微米”这一宣称——已标注为厂商发布，且与同一厂商自身表格矛盾，此处不作裁决。氯气在各辖区的确切法规状态——来源描述其为有害、受限、且被许多欧洲工厂弃用，但我们未找到可引用的原始法律文件；需要合规律确定性的读者应查阅适用的本国法规。效率数字大部分为厂商发布——第 6 节的 LiMCA 工厂研究是例外，也是此处报告的唯一独立测量。来源性质局限，坦白说明：关于 PPI 选型、过滤机理与操作窗口，现有最详细的已发布工程数据均为制造商文献；我们已如此标注，且未链接其中任何一家。

14. 本页不是什么。不是供应商推荐——不推荐任何制造商，未链接任何厂商网站，见《使命与独立性》。不是规格或设计数据——此处引用的孔径、效率与操作窗口报道自具名来源，不作为采购或工程规格发布，见《使用条款》第 6 节。不能替代测量——氢含量用 Telegas/Alspek/Hydris（±0.02–0.03 mL/100 g）或减压试验测定；洁净度用 PoDFA、Prefil 或 LiMCA 测定。本页描述机理，不描述如何判定某一具体熔体是否合格。