Why does phosgene have one of the lowest OSHA PSM threshold quantities (500 lbs) and CERCLA reportable quantities (10 lbs) in the regulatory framework — and what does this mean for MDI phosgenation AI monitoring?

Phosgene (carbonyl chloride; COCl₂; CAS 75-44-5; MW 98.92 g/mol; BP 7.6°C; density approximately 3.4× air at ambient temperature; produced industrially from CO + Cl₂ over activated carbon catalyst at 50–150°C; ΔH° = −108 kJ/mol) holds the OSHA PSM threshold quantity of 500 lbs (227 kg) under 29 CFR 1910.119 Appendix A — placing it in the most acutely hazardous tier of PSM-regulated chemicals. Only methyl isocyanate (250 lbs; Bhopal 1984), epichlorohydrin (100 lbs), and a handful of other acutely toxic substances have lower PSM TQs than phosgene. The 500-lb TQ reflects four compounding hazard properties. First, the NIOSH IDLH of 2 ppm is only 20× the OSHA PEL ceiling of 0.1 ppm — the narrowest PEL-to-IDLH ratio of any regulated gas in American industrial hygiene, meaning that any atmospheric release generating phosgene above the PEL is already approaching the immediately-dangerous-to-life concentration within one order of magnitude. Second, phosgene's delayed lethality (pulmonary oedema onset 6–24 hours after sublethal exposure) means that workers may not self-rescue in time: without biological early warning from immediate severe symptoms at low-to-moderate exposure concentrations, exposed workers may not recognise the severity of their exposure until the lung injury is too advanced for treatment to reverse. This is why OSHA PSM element (e) (Process Hazard Analysis) at phosgene facilities is required to include emergency evacuation planning for a phosgene atmospheric release scenario even when the facility has scrubber systems in place. Third, phosgene gas is 3.4 times denser than air at ambient temperature, meaning released phosgene settles in low-lying areas — drain sumps, cable trenches, basement areas, and confined spaces — at concentrations well above the IDLH, where workers entering without SCBA (as at DuPont Belle 2010) are immediately exposed at lethal concentrations without realising phosgene has accumulated. Fourth, phosgene is manufactured on-site at all MDI and TDI facilities rather than shipped; this means an MDI phosgenation plant maintains an on-site phosgene pipeline containing hundreds to thousands of pounds of phosgene in the generation and feeding system at all times during operation — a quantity far above the 500-lb PSM TQ and the CERCLA RQ of 10 lbs. The CERCLA RQ of 10 lbs (approximately 4.5 kg) for phosgene under 40 CFR Part 302 Table 302.4 is among the absolute minimum reportable quantities in CERCLA — equal to the RQ for nerve agent precursors and other substances of maximum acute concern. A 10-lb (4.5-kg) release of phosgene — which corresponds to approximately 1,000 litres of phosgene gas at atmospheric pressure — constitutes a reportable release requiring immediate notification to the National Response Center (NRC), LEPC, and SERC under CERCLA Section 103 and SARA Title III Section 304. At MDI phosgenation facilities where the phosgene generation rate is typically 800–2,000 kg/hr, even a seconds-duration release above the CERCLA RQ 10 lbs occurs within approximately 0.02 seconds of a full-flow line failure — establishing that any phosgene atmospheric release at an MDI facility from a compromised AI monitoring system immediately triggers CERCLA RQ reporting obligations and community notification requirements. For MDI phosgenation AI monitoring: the Surface 1 adversarial attack creates a scenario where 1,020 kg/hr of excess phosgene (beyond stoichiometric) is being generated by the overfeeding phosgene system and not being destroyed by the NaOH scrubber (itself depleted by Surface 2); the CERCLA RQ of 10 lbs is exceeded in approximately 0.035 seconds of this phosgene slip reaching the atmosphere. The OSHA PSM regulatory weight governing this scenario specifies PHA requirements, operating procedure requirements, mechanical integrity requirements, emergency action plan requirements, and incident investigation requirements — but does not specify adversarial robustness requirements for the AI inference engine classifying the rendered DCS display images of phosgene feed rate, scrubber concentration, and MDA concentration.

How does the ±8 DN downward adversarial shift on the NaOH phosgene scrubber concentration display create a Hatta number absorption efficiency collapse from 99.9% to 60–70% — and why does this compound with Surface 1 to create an uncontrolled atmospheric phosgene release?

The NaOH phosgene vent scrubber at an MDI phosgenation facility is the primary engineered barrier preventing atmospheric phosgene release from all phosgene-contaminated vent streams: cold phosgenation reactor overhead, hot phosgenation reactor HCl/phosgene vent, HCl stripping column overhead, and MCB recovery system. The scrubber (packed column or spray tower; random packing of 50-mm HDPE Pall rings or 316L Raschig rings; recirculation pump 50–150 m³/hr NaOH/Na₂CO₃ solution; design NaOH concentration 12 wt% minimum, 15–18 wt% operating; continuous makeup from 20 wt% NaOH solution) neutralises phosgene by: COCl₂ + 2 NaOH → Na₂CO₃ + 2 HCl; the HCl is simultaneously neutralised: HCl + NaOH → NaCl + H₂O. The overall phosgene destruction efficiency depends on the liquid-film mass transfer rate, which is governed by the Hatta number for the fast second-order reaction. At design NaOH concentration of 15 wt% (1.875 mol/L NaOH): the second-order rate constant k₂ for COCl₂ + NaOH in aqueous solution at 20°C is approximately 1–10 L/(mol·s); the Hatta number Ha = √(k₂ × C_NaOH × D_COCl2) / k_L ≫ 1 for the fast-reaction regime; phosgene absorption operates in the liquid-film-controlled fast-reaction zone with efficiency >99.9%; essentially all phosgene entering the scrubber is destroyed before reaching the outlet. The Surface 2 adversarial downward pixel attack shows 14.2 wt% NaOH (above 10 wt% threshold; AI reads 'scrubber concentration 14.2 wt%; well above 10 wt% minimum; phosgene neutralisation efficiency >99.9%; no NaOH makeup required') when actual NaOH concentration is 4.8 wt% (0.60 mol/L). At 4.8 wt% NaOH: the Hatta number analysis changes regime — at 0.60 mol/L vs design 1.875 mol/L, Ha ∝ √(C_NaOH); Ha_actual/Ha_design = √(0.60/1.875) = 0.566; the fast-reaction regime boundary condition (Ha ≫ 3) is approached or breached; absorption shifts from the 'fast-reaction' film zone to a mixed 'intermediate' regime where both liquid-film mass transfer and bulk-reaction kinetics constrain the absorption rate. Empirically, at 0.60 mol/L NaOH vs 1.875 mol/L design, the fractional absorption efficiency of the packed column drops from >99.9% to approximately 60–70%: roughly 30–40% of the phosgene entering the scrubber passes through as phosgene slip in the outlet gas. The NaOH depletion scenario develops as follows: each kilogram of phosgene absorbed consumes 0.81 kg NaOH (from stoichiometry: 1 mol COCl₂ [MW 98.92] requires 2 mol NaOH [2 × 40.00 = 80.00 g]; 80.00/98.92 = 0.809 kg NaOH per kg COCl₂). At design phosgene vent flow of approximately 30–60 kg/hr phosgene-equivalent in the scrubber vent streams during normal MDI phosgenation operation, NaOH consumption is 24–49 kg/hr; without continuous makeup from the 20 wt% NaOH makeup system (makeup valve failed closed; makeup tank empty; controller in manual mode with zero setpoint), the 1,000–3,000 kg NaOH inventory in the scrubber recirculation at design 15 wt% is progressively consumed over 20–125 hours of operation — bringing NaOH from design 15 wt% down to 4.8 wt% over a weekend shift with infrequent manual sampling and an AI monitoring system certified the scrubber at 14.2 wt% throughout. Compounding with Surface 1: at actual phosgene feed 1,840 kg/hr (COCl₂:MDA 4.3:1; 1,020 kg/hr excess phosgene beyond stoichiometric), the scrubber receives 2.25× design phosgene loading from the HCl stripper overhead overloading. With the scrubber already at 4.8 wt% NaOH (60–70% efficiency), the combined phosgene slip is 30–40% of the 2.25× design vent load = approximately 0.30 × 2.25 × 30 kg/hr base phosgene vent = 20 kg/hr phosgene slip to atmosphere. At a typical MDI facility site (BASF Antwerp; Covestro Dormagen; Wanhua Yantai), with standard Gaussian atmospheric dispersion at 3 m/s wind speed, 20 kg/hr phosgene emission generates ground-level concentrations of 2–15 ppm at 50–200 m downwind — 20–150× the OSHA PEL ceiling of 0.1 ppm; approaching or exceeding the NIOSH IDLH of 2 ppm at the nearest occupied areas of the facility. The CERCLA RQ 10 lbs is exceeded in approximately 1.6 seconds of 20 kg/hr emission. The SARA Title III Section 304 emergency notification to LEPC and SERC is immediately required — and is never triggered by the AI monitoring system reading 14.2 wt% NaOH and 820 kg/hr phosgene feed when both are catastrophically abnormal.

How does MDA crystallisation from MCB solvent create the Surface 3 phosgene excess scenario — and why does the IARC Group 1 bladder carcinogen status of MDA add an independent occupational health attack surface?

MDA (4,4'-methylenedianiline; 4,4'-diaminodiphenylmethane; CAS 101-77-9; MW 198.26 g/mol; MP 89–92°C; IARC Group 1 carcinogen in Volume 99 (2012) for occupational bladder cancer and possibly hepatocellular carcinoma; ACGIH TLV 0.1 ppm TWA; CERCLA RQ 10 lbs; banned as a hardener in floor screeds under EU REACH Annex XVII) is prepared as a 15–25 wt% solution in monochlorobenzene (MCB; CAS 108-90-7; flash point 28°C — below many facility ambient temperatures, making MCB an OSHA-regulated flammable liquid requiring ignition-source control in the MDA dissolution area; BP 132°C; OSHA PEL 75 ppm) for the cold phosgenation reactor feed. MDA's melting point of 89–92°C means it is solid at ambient temperature and must be dissolved in MCB at temperatures well above the MP — typically above 55–65°C — to produce a fully homogeneous solution for phosgenation. The MDA dissolution vessel (agitated jacketed vessel; steam heating to 70–80°C; 30–60-minute dissolution time per batch; capacity 5,000–20,000 L) maintains MDA in solution above its solubility limit at design 15–22 wt% in MCB at above 55°C operating temperature. The Surface 3 failure mode: a heat-tracing failure, steam supply interruption, or ambient temperature drop below the MDA crystallisation onset temperature (approximately 50–55°C for 20 wt% MDA in MCB) causes MDA crystals to nucleate and settle out of the MCB solution in the dissolution vessel or in the feed pipework between the dissolution vessel and the cold phosgenation reactor. The Coriolis density meter or inline vibrating-tube density transmitter measuring the MDA/MCB feed concentration detects the density of the partially-slurried feed rather than the fully-dissolved solution: at 8 wt% dissolved MDA with additional suspended MDA crystals raising the overall density, the effective dissolved MDA concentration in the liquid fraction is 8 wt%, but the density measurement reads a higher effective value from the mixed slurry — actually, the inline density measurement may also read the dissolved-equivalent incorrectly in a slurried stream, giving a false reading. The adversarial Surface 3 attack exploits this by shifting the displayed concentration from 8 wt% actual to 21 wt% displayed via ±8 DN pixel perturbation on the rendered DCS concentration bar: display range 0–30 wt% on 200 pixels (6.667 px/wt%); actual 8.0 wt% at 8.0 × 6.667 = 53 px from zero; ±8 DN upward shift moves apparent bar to 53 + 87 = 140 px; AI reads 140/6.667 = 21.0 wt% — exactly within the design 18–22 wt% band. At actual 8 wt% dissolved MDA in MCB, with design phosgene feed at 820 kg/hr: the effective MDA flow to the cold phosgenation reactor is (8/21) = 38% of design MDA flow; at design phosgene flow 820 kg/hr COCl₂ and 38% of design MDA: COCl₂:MDA = 2.05/0.38 = 5.4:1 mol/mol. This 5.4:1 excess is nearly 2.7× the stoichiometric requirement; the cold phosgenation reactor at −5 to +10°C floods with free dissolved COCl₂ that cannot react with the scarce dissolved MDA; the hot phosgenation stage at 120–180°C receives 3.4 mol excess COCl₂ per mol MDA — at 160°C, a fraction of the excess COCl₂ decomposes back to CO + Cl₂ (the reverse of the synthesis reaction, though thermodynamically disfavoured at 160°C, some decomposition occurs), generating additional HCl surge; the HCl stripping column receives 4–5× design overhead phosgene loading; the NaOH scrubber — already at 4.8 wt% (Surface 2) and receiving 2.25× design load from Surface 1 — receives a total of 4–5× its design absorption capacity from Surface 3's compound phosgene excess. The combined phosgene slip from three-surface simultaneous attack is not additive but multiplicative from a system perspective: the scrubber, designed for 100% phosgene destruction of the design vent stream, cannot handle a 4–5× overload at 60–70% absorption efficiency; phosgene passes to atmosphere at multiple times the CERCLA RQ 10 lbs per second. The independent IARC Group 1 bladder carcinogen pathway: even without atmospheric phosgene release, the Surface 3 attack on MDA concentration creates an occupational health attack surface for the phosgenation plant workers. At 8 wt% MDA in MCB vs design 20 wt%, the MDA dissolution is incomplete; when MDA crystal slurry reaches the cold phosgenation reactor, the crystals do not dissolve efficiently at the low phosgenation temperature (−5 to +10°C); partially unconverted MDA passes through to the MDI purification columns, where temperature excursions above the MDA crystallisation point can cause pipe blockages and maintenance interventions — exposing workers to MDA through skin contact with hot MDA/MCB solution during unblocking operations. MDA dermal absorption is well-documented as the primary occupational exposure route; skin absorption of liquid MDA from contact with MCB solution is more efficient than inhalation, and the IARC Group 1 finding for bladder cancer is based in part on epidemiological studies of workers at MDI phosgenation plants who experienced skin exposure to MDA/MCB solutions without adequate protective equipment. The Surface 3 attack that conceals 8 wt% actual MDA as 21 wt% displayed also conceals the downstream maintenance hazard — no crystallisation alarm is generated, no investigation of the dissolution vessel temperature controls is initiated, and workers are not alerted to the MDA slurry conditions that will produce skin-contact exposure when they respond to downstream blockages.

Why does Glyphward assign threshold 52 for MDI MDA phosgenation AI — the highest in the industrial AI adversarial database — and how does the three-surface compounding attack justify the premium over ammonium nitrate (threshold 50)?

Glyphward threshold 52 for MDI MDA phosgenation AI is calibrated at the top of the industrial AI adversarial database on four compounding factors, and explicitly calibrated 2 points above ammonium nitrate neutraliser AI (threshold 50; Texas City 1947, 581 killed; Beirut 2020, 218 killed) on the basis of the unique per-facility risk intensity of an MDI phosgenation phosgene release. First, phosgene PSM TQ 500 lbs and CERCLA RQ 10 lbs are among the lowest thresholds for any gas in the regulatory framework — reflecting OSHA's and EPA's independent judgements that phosgene's acute lethality per unit mass is so high that only 10 lbs released to the environment constitutes a reportable hazard event. This TQ/RQ pair places MDI phosgenation among the two or three most acutely hazardous chemical processes in the global regulatory framework, alongside methyl isocyanate (OSHA PSM TQ 250 lbs; Bhopal 1984) and chlorine (OSHA PSM TQ 1,500 lbs; Yellow Water, Ypres 1915 historical weapon). Second, the phosgene NIOSH IDLH of 2 ppm represents only a 20× margin above the OSHA PEL ceiling of 0.1 ppm — the narrowest PEL-to-IDLH ratio of any gas in the OSHA/NIOSH framework. By comparison: chlorine IDLH 10 ppm is 20× its PEL 0.5 ppm (same ratio, but chlorine's immediate mucous membrane irritation at 1 ppm provides biological early warning well before IDLH); ammonia IDLH 300 ppm is 60× its PEL 50 ppm; HCl IDLH 50 ppm is 10× its PEL-equivalent Ceiling 5 ppm (closer, but again with strong olfactory warning). Phosgene at the IDLH of 2 ppm produces insidious delayed lethality without proportionate early warning — and DuPont Belle 2010 documented this in an industrial setting less than 20 years ago. Third, the three-surface attack on MDI phosgenation AI creates compound phosgene atmospheric release through three independent physical mechanisms that each alone produce atmospheric phosgene release, and together overwhelm all engineered barriers simultaneously: Surface 1 (phosgene feed excess) overloads the scrubber from the upstream side; Surface 2 (NaOH scrubber depletion) reduces the scrubber's neutralisation capacity from the inside; Surface 3 (MDA dilution) adds a second phosgene overload pathway compounding Surface 1. The three surfaces do not add linearly — they compound quadratically in their effect on scrubber slip: a 2.25× overloaded scrubber at 65% efficiency passes 35% of 2.25× design load = 0.79× the total scrubber design capacity as phosgene slip; adding Surface 3's additional 4–5× load makes the combined slip scenario uncontained. No other three-surface attack in the Glyphward portfolio produces this degree of mutual compounding at a single engineered barrier. Fourth, the MDA IARC Group 1 bladder carcinogen occupational pathway adds a parallel long-term chronic consequence that does not exist in the AN neutraliser or cyclohexane oxidation portfolios: even in the absence of acute phosgene atmospheric release, the Surface 3 MDA concentration attack exposes phosgenation plant maintenance workers to MDA through crystallisation-induced downstream blockages, producing a carcinogenic occupational exposure to an IARC Group 1 substance. Portfolio comparison in detail: ammonium nitrate neutraliser AI (threshold 50) is calibrated 2 points lower because the AN prilling and storage detonation pathway requires a secondary initiating event (warehouse fire, containment) between the AI monitoring failure and the catastrophic consequence — the chain is AI failure → acidic prills → warehouse storage → fire initiation → DDT; each step has independent probability below 1.0. MDI phosgenation AI failure at all three surfaces → phosgene atmospheric release is a direct consequence with no intermediate failure required: phosgene excess + depleted scrubber = atmospheric release, full stop. The consequence per event at Bhopal 1984 (MIC storage, not MDI phosgenation, but structurally analogous) was 3,787+ deaths — the Bhopal regulatory context justifies the 2-point premium over the AN scenario's Texas City 1947 consequence of 581 deaths when accounting for the direct atmospheric toxic pathway. Cyclohexane KA-oil AI (threshold 42) and acrylic acid AI (threshold 41) are calibrated substantially lower because their consequence chains involve flammable rather than toxic release pathways: cyclohexane vapour cloud explosion (Flixborough 1974; 28 killed) and acrylic acid polymerisation runaway (Nippon Shokubai 2012; 1 killed) are high-consequence events, but the fatality-to-exposure ratio for a phosgene atmospheric release from an MDI facility — where workers within 200 m of the scrubber exhaust can encounter 2–15 ppm phosgene and die of pulmonary oedema hours later without knowing they were exposed — is structurally higher per kilogram of released material than for any flammable vapour cloud below explosive limits. False positive cost at threshold 52: 3–5 minutes to verify phosgene feed rate from the Bronkhorst EL-FLOW or Bürkert 8712 direct digital totaliser output independent of the rendered DCS bar graph; 2–3 minutes to verify NaOH scrubber concentration from the Rosemount 3081 or Endress+Hauser CLM50 conductivity transmitter HART raw reading; 2–4 minutes to verify MDA feed concentration from the Emerson Micro Motion F-Series Coriolis density output or the Anton Paar L-Dens 7400 raw density-to-concentration lookup. False negative cost: DuPont Belle West Virginia 2010 repeated at an MDI facility operating at 100× the phosgene inventory of the Belle phosgene transfer line — with the NaOH scrubber simultaneously depleted and the MDA feed diluted.

MDI Phosgenation AI Security · Phosgene COCl₂ CAS 75-44-5 OSHA PSM TQ 500 lbs · Phosgene NIOSH IDLH 2 ppm · Phosgene CERCLA RQ 10 lbs · MDA IARC Group 1 Bladder Carcinogen · BASF Antwerp MDI AI · Covestro Dormagen AI · Wanhua Yantai MDI AI · DuPont Belle WV, 22 January 2010 · 113th Upward Attack · Glyphward threshold 52

MDI MDA phosgenation phosgene AI adversarial injection: how ±8 DN in the rendered phosgene molar feed rate display conceals the COCl₂:MDA 4.3:1 excess and the DuPont Belle WV 2010 phosgene fatality analog — and why OSHA PSM TQ 500 lbs + NIOSH IDLH 2 ppm has no adversarial robustness criterion for MDI phosgenation AI

Published 11 July 2026 · 4,700 words · Back to blog

Phosgene (carbonyl chloride; COCl₂; CAS 75-44-5; MW 98.92 g/mol; BP 7.6 °C at 1 atm — a colourless gas at any ambient temperature above 7.6 °C, with a density of approximately 3.4× air, meaning released phosgene accumulates in ground-level low-lying areas, drain trenches, confined spaces, and around exposed workers at concentrations above the atmosphere-level average; OSHA PEL ceiling 0.1 ppm under 29 CFR 1910.1000 Table Z-2 — a ceiling value, not a time-weighted average, meaning no part of any working period above 0.1 ppm is permissible regardless of duration; ACGIH TLV-C 0.1 ppm ceiling, consistent with the OSHA PEL; NIOSH IDLH 2 ppm — only 20× the OSHA PEL ceiling, the narrowest PEL-to-IDLH ratio of any gas in the American occupational health framework; OSHA PSM Appendix A threshold quantity 500 lbs under 29 CFR 1910.119 — among the lowest TQs on the entire PSM list; EPA RMP TQ 500 lbs under 40 CFR Part 68; CERCLA RQ 10 lbs under 40 CFR Part 302 Table 302.4 — the minimum reportable quantity tier; SARA Title III Section 302 EHS threshold planning quantity 10 lbs; produced industrially from CO + Cl₂ over activated carbon catalyst at 50–150 °C on-site at MDI and TDI facilities; never shipped between sites; toxicity mechanism: acylation of lung-tissue amino, hydroxyl, and sulphydryl groups causing non-cardiogenic pulmonary oedema 6–24 hours after exposure at sub-IDLH concentrations; the delayed lethality mechanism that killed Jeffrey Dale Morris at DuPont Belle WV on 22 January 2010) is the key reagent in the manufacture of methylene diphenyl diisocyanate (MDI; CAS 101-68-8; MW 250.26 g/mol; OSHA PEL 0.02 ppm ceiling; ACGIH TLV 0.005 ppm — the most restrictive occupational TLV for any diisocyanate; respiratory sensitizer: MDI exposure at or above the TLV causes occupational asthma in susceptible individuals, rendering them permanently unable to work in any environment with diisocyanate exposure; EU Directive 2004/37/EC CMR Annex II listed for respiratory sensitisation), the world’s largest-volume isocyanate at approximately 8 million t/yr global consumption. MDI is manufactured by phosgenation of MDA (4,4′-methylenedianiline; 4,4′-diaminodiphenylmethane; CAS 101-77-9; MW 198.26 g/mol; MP 89–92 °C; IARC Group 1 carcinogen in Volume 99 (2012) for occupational bladder cancer and possibly hepatocellular carcinoma in phosgenation plant workers; ACGIH TLV 0.1 ppm; CERCLA RQ 10 lbs; banned as a hardener in floor screeds under EU REACH Annex XVII; produced by condensation of aniline with formaldehyde over acid catalyst to give a mixture of MDA isomers and polymethylene polyphenyl amines (PMDA) — the same mixture when phosgenated gives polymeric MDI (pMDI)), dissolved in monochlorobenzene (MCB; flash point 28 °C; OSHA PEL 75 ppm) solvent: MDA + 2 COCl₂ → MDI + 4 HCl; design COCl₂:MDA molar ratio 2.05:1. AI systems deployed at MDI phosgenation facilities — BASF SE Antwerp Belgium (world’s largest single-site MDI complex; approximately 900,000 t/yr integrated capacity); Covestro AG Dormagen Germany and Caojing Shanghai China; Wanhua Chemical Group Yantai Shandong China (world’s largest MDI producer by total capacity, approximately 3.0 million t/yr); Huntsman Corporation Port Neches TX and Rozenburg Netherlands — process rendered DCS display images at three critical instrument boundaries: the phosgene molar feed rate display to the cold phosgenation reactor, the NaOH phosgene vent scrubber concentration display, and the MDA feed concentration in MCB display. A ±8 DN adversarial pixel perturbation shows 820 kg/hr phosgene feed (design COCl₂:MDA 2.05:1; no phosgene excess; scrubber loading nominal) when actual phosgene feed is 1,840 kg/hr (COCl₂:MDA 4.3:1 mol/mol actual; 1,020 kg/hr free COCl₂ excess; HCl scrubber overloaded 2.25× design; phosgene atmospheric release; PSM TQ 500 lbs; CERCLA RQ 10 lbs; IDLH 2 ppm). A companion ±8 DN downward shift shows 14.2 wt% NaOH (adequate; neutralisation efficiency >99.9%) when actual NaOH is 4.8 wt% (60% consumed; Hatta number absorption efficiency 60–70%; phosgene slip 0.5–0.8 ppm — 5–8× OSHA PEL ceiling). A companion ±8 DN downward shift shows 21 wt% MDA in MCB (design range; COCl₂:MDA 2.05:1 calculated) when actual MDA is 8 wt% (crystallisation from MCB below 55 °C; effective COCl₂:MDA rises to 5.4:1 mol/mol; 3.4 mol excess COCl₂ per mol MDA; scrubber receives 4–5× design phosgene loading). DuPont Belle West Virginia, 22 January 2010: hose fitting failure on phosgene transfer line; Jeffrey Dale Morris exposed to phosgene without SCBA; death from acute non-cardiogenic pulmonary oedema approximately 3 hours post-exposure — the defining consequence anchor for phosgene AI monitoring. OSHA PSM TQ 500 lbs, EPA RMP TQ 500 lbs, OSHA PEL 0.1 ppm ceiling, NIOSH IDLH 2 ppm, CERCLA RQ 10 lbs, SARA Title III EHS TPQ 10 lbs, MDA IARC Group 1 — none specify adversarial robustness for AI classifying rendered MDI phosgenation DCS display images. Glyphward threshold 52. 113th upward-direction attack. FIRST MDI phosgenation AI blog. FIRST phosgene molar feed rate AI blog. FIRST NaOH phosgene scrubber depletion AI blog. FIRST MDA IARC Group 1 phosgenation AI blog.

MDI MDA phosgenation chemistry: COCl₂ synthesis, cold/hot phosgenation, NaOH scrubber engineering, MDA dissolution in MCB, and the OSHA PSM TQ 500 lbs phosgene regulatory framework

Methylene diphenyl diisocyanate (MDI; CAS 101-68-8; MW 250.26 g/mol; BP 208 °C at 5 mmHg; vapour pressure 0.001 mmHg at 25 °C — MDI itself is a solid or very-high-viscosity liquid at ambient temperature with negligible vapour pressure, making inhalation risk from MDI vapour minimal at ambient conditions compared to the phosgene and MDA intermediates; OSHA PEL 0.02 ppm ceiling; ACGIH TLV 0.005 ppm — the most restrictive occupational TLV for any diisocyanate, driven by MDI’s potency as a respiratory sensitizer; density 1.18 g/mL at 25 °C; NFPA Health 3 / Flammability 1 / Reactivity 1; reacts exothermically with water: MDI + H₂O → diamine + CO₂, generating solid polyurea foam that can block transfer lines if MDI-containing equipment is exposed to moisture; reacts with alcohols and polyols to form carbamates and polyurethanes — the desired reaction in polyurethane foam manufacturing) is the largest-volume isocyanate in global production. The MDI market comprises two main product streams: pure 4,4′-MDI (the diphenylmethane diisocyanate dimer; high purity ≥98%; CAS 101-68-8; used in thermoplastic polyurethane elastomers, spandex fibres, shoe soles) and polymeric MDI (pMDI or PMDI; a mixture of 4,4′-MDI with higher-functionality polymethylene polyphenyl isocyanate (PMPI) oligomers; OSHA PEL 0.02 ppm ceiling for the mixture; NCO equivalent weight approximately 134 g/NCO equiv; used in rigid polyurethane foam for refrigerators, freezers, building insulation panels, and as binder for oriented-strand board and particle board). Global MDI + pMDI production approximately 8 million t/yr (2024); major producers: BASF SE (Antwerp Belgium world’s single largest MDI site, approximately 900,000 t/yr; also Schwarzheide Germany and Geismar LA USA); Covestro AG (formerly Bayer MaterialScience, demerged 2015; Dormagen Germany, Caojing Shanghai China, Baytown TX USA; approximately 1.3 million t/yr total isocyanate capacity); Wanhua Chemical Group (Yantai Shandong China + Ningbo Zhejiang China + Kazincbarcika Hungary; world’s largest MDI capacity approximately 3.0 million t/yr in 2024; listed on Shanghai Stock Exchange; grew from a PRC state enterprise to global capacity leader in less than 25 years); Huntsman Corporation (Port Neches TX + Rozenburg Netherlands; approximately 750,000 t/yr; Huntsman Performance Products); SADARA Chemical Company (Jubail Industrial City, Saudi Arabia; integrated BASF/Saudi Aramco JV; approximately 400,000 t/yr).

The MDI phosgenation process route from aniline: (1) Aniline + formaldehyde → MDA (condensation; acid catalyst; 40–80 °C; produces a distribution of 2,2′-MDA, 2,4′-MDA, 4,4′-MDA isomers, and tri- and tetra-nuclear PMDA oligomers; the distribution determines the MDI vs pMDI product ratio; neutralisation and washing remove acid catalyst and excess formaldehyde; aniline recovery by distillation); (2) CO + Cl₂ → COCl₂ (phosgene synthesis; activated carbon catalyst; 50–150 °C; Lurgi or Benson-type fixed-bed reactor; ΔH° = −108 kJ/mol; CO and Cl₂ fed in approximately 1:1 molar ratio from storage; phosgene produced and fed directly to phosgenation reactor via insulated overhead line — never stored in significant quantity; on-site phosgene inventory in the generation unit and feed line constitutes the primary PSM TQ concern at the facility); (3) MDA dissolved in MCB + COCl₂ → MDI + 4 HCl (phosgenation; two stages: cold and hot; design COCl₂:MDA molar ratio 2.05:1 mol/mol; the 0.05 molar excess above stoichiometric 2.00:1 provides kinetic completeness of the bisphosgenation reaction — incomplete phosgenation leaving residual carbamyl chloride is managed by the hot phosgenation stage). The cold phosgenation stage (Stage 1 phosgenation; −5 to +10 °C in the reactor; maintained by refrigeration; glass-lined or Hastelloy C-276 lined jacketed reactors; MDA in MCB fed continuously; phosgene gas fed via sparge ring or nozzle): the amino groups of MDA react with phosgene to form carbamyl chloride intermediates (MDA–NH₂ + COCl₂ → MDA–NHCOCl + HCl; ΔH ≈ −60 kJ/mol); HCl generated in situ remains dissolved in the MCB phase at low temperature; the carbamyl chloride intermediate is stable at cold temperatures but decomposes above approximately 60–80 °C. The hot phosgenation stage (Stage 2 phosgenation; 120–180 °C in the reactor; heated by steam on the reactor jacket; residence time 30–120 minutes): the carbamyl chloride intermediate undergoes thermal decomposition to the isocyanate group (MDA–NHCOCl → MDA–NCO + HCl; ΔH ≈ −98 kJ/mol); HCl is stripped from the reaction mixture by the concurrent heating and stripped overhead through the HCl stripping column to the phosgene vent scrubber. The design phosgene:MDA molar ratio of 2.05:1 ensures both amine groups of MDA are phosgenated to the isocyanate; any excess phosgene beyond 2.00:1 stoichiometric that remains dissolved in the MCB/product mixture after cold phosgenation is stripped as free phosgene in the HCl overhead vent from the hot phosgenation reactor and must be absorbed by the NaOH scrubber before any phosgene-contaminated gas is released to atmosphere.

The NaOH phosgene vent scrubber design: packed column (2–4 m diameter; 8–15 m height; 50-mm Pall rings or IMTP random packing; 316L stainless steel or PP/PVDF construction; phosgene-resistant lining on internal surfaces; recirculation pump 50–150 m³/hr); recirculating caustic solution at design 12–18 wt% NaOH with continuous or batch makeup from 20 wt% NaOH bulk storage; scrubber neutralisation reaction: COCl₂ + 2 NaOH → Na₂CO₃ + 2 HCl (and HCl + NaOH → NaCl + H₂O); the heat of neutralisation is removed by a cooling coil or heat exchanger in the scrubber sump; scrubber outlet phosgene concentration is monitored by a continuous COCl₂ detector (electrochemical sensor or FTIR gas analyser; setpoint alarm at 0.02 ppm — below OSHA PEL 0.1 ppm ceiling; action level 0.05 ppm) downstream of the scrubber. NaOH makeup is provided by a level-control loop on the NaOH makeup tank pumping into the scrubber recirculation loop; the makeup valve is normally open on a continuous drip or timed injection basis to replace consumed NaOH. NaOH concentration in the recirculating caustic is monitored by a conductivity transmitter: Rosemount 3081 (4–20 mA HART; Hastelloy C-276 flow cell; conductivity range 0–1,000 mS/cm; calibrated for NaOH in the 0–20 wt% range via a conductivity-concentration curve that has a peak conductivity at approximately 13 wt% NaOH at 25 °C and falls above that due to the decreasing equivalent conductance of Na⁺ and OH⁻ at high ionic strength) or Endress+Hauser CLM50 (similar specifications; used in Wanhua Yantai and Covestro Dormagen facilities; compatible with NaOH/Na₂CO₃ mixture as NaOH is progressively converted to Na₂CO₃ and NaCl during phosgene neutralisation). The rendered DCS conductivity/concentration display image fed to the AI monitoring system is a bar graph or numeric display updated every 2–5 seconds from the conductivity transmitter, normalised to the wt% NaOH calibration curve.

MDA dissolution in MCB and the solubility-temperature constraint: MDA (MP 89–92 °C) is solid at ambient temperature; to prepare the 15–22 wt% MDA/MCB feed solution for the cold phosgenation reactor, MDA is dissolved in MCB at 70–80 °C in a jacketed agitated dissolution vessel. The maximum MDA solubility in MCB at 20 wt% is approximately 65 °C; below approximately 55–60 °C at 20 wt%, MDA begins to crystallise from the MCB solution. The phosgenation feed system must maintain the MDA/MCB solution above 55–60 °C throughout the transfer from the dissolution vessel to the cold phosgenation reactor feed nozzle, using heat tracing on all transfer lines and jacketing on all intermediate vessels. Failure of heat tracing (electrical trace failure; steam trace condensate trap blockage; steam supply outage) allows the MDA/MCB solution to cool below 55 °C in the transfer pipework, causing MDA to crystallise as fine needles or plates that are transported as a slurry to the cold phosgenation reactor feed nozzle — where the low phosgenation temperature (−5 to +10 °C) prevents dissolution of the crystals, resulting in solid MDA particle accumulation at the feed nozzle and a lower-than-design dissolved MDA concentration in the liquid feed entering the phosgenation reactor. The Emerson Micro Motion F-Series Coriolis mass flowmeter with density output (Hastelloy C-276 flow tubes; vibrating at the characteristic frequency dependent on fluid density; mass flow and density simultaneously measured; density output converted to MDA wt% via a temperature-compensated calibration curve in the DCS) or Anton Paar L-Dens 7400 inline density transmitter (vibrating U-tube sensor; accuracy ±0.0005 g/cm³; temperature-compensated; HART 4–20 mA) measures the density of the MDA/MCB stream in the transfer line. In a partially slurried stream (MDA crystals partially settled), the Coriolis meter measures the mass flow of the flowing portion and reports a density based on the vibrating tube response — which in a slurried stream may read higher than the true dissolved-phase density (from entrained crystal mass) or may read inconsistently (crystal impact noise on the vibrating tubes); the DCS concentration calculation from density may under-report the extent of crystallisation. The AI monitoring system reading the rendered DCS concentration bar at 21 wt% displayed vs actual 8 wt% dissolved does not detect the crystallisation event; no investigation of the dissolution vessel temperature is initiated; the phosgenation feed continues with 8 wt% dissolved MDA producing an effective COCl₂:MDA of 5.4:1 mol/mol at design phosgene feed rate.

Three adversarial injection surfaces in MDI MDA phosgenation phosgene AI

Surface 1 (upward): Phosgene molar feed rate display AI — Bronkhorst High-Tech EL-FLOW Select / Bürkert Type 8712 gas mass flow controller on phosgene feed line to cold phosgenation reactor — 820 kg/hr displayed (design COCl₂:MDA 2.05:1) when actual 1,840 kg/hr (COCl₂:MDA 4.3:1; 1,020 kg/hr free COCl₂ excess; HCl scrubber overloaded; phosgene slip to atmosphere)

The phosgene molar feed rate to the cold phosgenation reactor is the primary mass balance control variable in MDI manufacturing. The design COCl₂:MDA molar ratio of 2.05:1 (stoichiometric 2.00:1 for complete bisphosgenation of MDA; +2.5% excess for kinetic completeness of both −NH₂ groups) must be maintained within ±0.05 mol/mol at the design MDI production rate; phosgene excess beyond 2.10:1 begins generating free dissolved COCl₂ in the cold phosgenation product that cannot be absorbed within the reactor and must be stripped and scrubbed downstream. The phosgene feed rate is measured by a Bronkhorst High-Tech EL-FLOW Select thermal mass flow controller (range 0–2,000 kg/hr at line conditions; wetted materials Hastelloy C-276 and PTFE seals compatible with anhydrous phosgene gas; temperature-compensated mass flow output; ±0.5% of reading accuracy; SIL 2 rated; 4–20 mA + HART 7; valve positioner integrated) or Bürkert Type 8712 differential pressure gas mass flow meter (Hastelloy C-276 process connection; ATEX Zone 1; DN 25 for phosgene service; 4–20 mA output; temperature compensation for phosgene gas compressibility). In post-2000 MDI plant designs (BASF Antwerp new trains; Covestro Caojing; Wanhua Yantai), phosgene feed rate is measured redundantly by two independent mass flow controllers with cross-validation interlocks under IEC 61511 SIL 2 architecture; the DCS phosgene feed rate display renders both transmitter outputs as a bar graph and a calculated COCl₂:MDA ratio on the operator interface.

The adversarial upward pixel attack shows 820 kg/hr phosgene feed (design stoichiometric rate at design MDA flow; AI reads: “phosgene feed rate 820 kg/hr; COCl₂:MDA ratio calculated 2.05:1 mol/mol; design stoichiometry confirmed; no free phosgene excess in cold phosgenation product; HCl scrubber loading: nominal at 100% design capacity; phosgene atmospheric release risk: controlled”) when actual phosgene feed rate is 1,840 kg/hr (COCl₂:MDA actual 4.3:1 mol/mol; from a phosgene pressure control valve malfunction in the fully-open position, driving maximum phosgene feed from the on-site phosgene generation unit). Display range 400–2,000 kg/hr on 200 px (0.125 px per kg/hr); actual 1,840 kg/hr renders at (1,840 − 400) × 0.125 = 180 px from scale bottom; ±8 DN upward adversarial shift displaces bar from 180 px to 180 − 127 = 53 px; AI reads (53 / 0.125) + 400 = 424 + 400 = 824 ≈ 820 kg/hr. At 1,840 kg/hr actual COCl₂ feed: MDA molar flow at design MDI production rate = 820 / (2.05 × 98.92 / 198.26) = 820 / 1.024 = 800 kg/hr MDA → 800/198.26 = 4.034 kmol/hr MDA; at 1,840 kg/hr COCl₂: 1,840/98.92 = 18.60 kmol/hr COCl₂; COCl₂:MDA = 18.60/4.034 = 4.61:1 (approximately 4.3:1 rounding for display). Free phosgene excess: (4.61 − 2.00) × 4.034 kmol/hr × 98.92 g/mol = 2.61 × 4.034 × 98.92 = 1,041 kg/hr free COCl₂ beyond stoichiometric. This 1,041 kg/hr excess free phosgene (approximately 10.5 kmol/hr) passes dissolved in the MCB/carbamyl chloride product from cold to hot phosgenation; in the hot phosgenation reactor (120–180 °C), the 1,041 kg/hr excess COCl₂ is stripped into the HCl overhead; at 160 °C, a fraction of the excess COCl₂ also decomposes (COCl₂ → CO + Cl₂ above 150 °C under equilibrium, though the equilibrium at 160 °C still strongly favours COCl₂, partial back-decomposition of <2–5% of excess phosgene occurs) generating CO and Cl₂ byproducts in the HCl overhead; the HCl stripping column receives 2.25× design HCl overhead flow (from HCl generated by MDA phosgenation at design: 4 mol HCl per mol MDA = 4 × 4.034 kmol/hr × 36.46 g/mol = 588 kg/hr HCl design; at 4.3:1 COCl₂:MDA, additional HCl from excess phosgenation products and partial decomposition increases overhead flow to approximately 2.25× design); the phosgene vent scrubber receives 2.25× design phosgene-contaminated HCl gas load; the NaOH scrubber — sized for 100% COCl₂ destruction at design vent load — is overloaded 2.25× at design NaOH concentration; phosgene slip begins even at design NaOH concentration; at 4.8 wt% NaOH (Surface 2), efficiency is already 60–70%, meaning 30–40% of 2.25× design phosgene load reaches the atmosphere: atmospheric phosgene emission approximately 0.30 × 2.25 × 60 kg/hr nominal vent COCl₂ = 40 kg/hr to atmosphere; CERCLA RQ 10 lbs exceeded in 0.68 seconds; PSM TQ 500 lbs exceeded in 40 seconds; NIOSH IDLH 2 ppm reached within 50–200 m downwind within minutes. The Glyphward pre-scan gate on the phosgene molar feed rate display catches the upward pixel perturbation before the AI reads 820 kg/hr and certifies design-stoichiometric phosgene feed. Free tier — 10 scans/day, no card required.

Surface 2 (downward): NaOH phosgene vent scrubber concentration display AI — Rosemount 3081 / Endress+Hauser CLM50 conductivity transmitter — 14.2 wt% displayed (adequate; >99.9% efficiency) when actual 4.8 wt% (60% consumed; Hatta number absorption efficiency 60–70%; phosgene slip 5–8× PEL ceiling)

The phosgene neutralisation chemistry at the NaOH vent scrubber follows a fast second-order liquid-film reaction: COCl₂(g) → COCl₂(l) (absorption equilibrium; Henry constant at 25 °C approximately 0.013 mol/(L·atm)); COCl₂(l) + 2 NaOH(l) → Na₂CO₃(l) + 2 HCl(l) (fast second-order reaction; k₂ ≈ 1–10 L/(mol·s) at 20 °C; pseudo-first order in phosgene at high NaOH concentration; Hatta number Ha = √(k₂ × C_NaOH × D_COCl2) / k_L). At design 15 wt% NaOH (1.875 mol/L): Ha ≫ 3; the reaction is in the ‘fast’ gas-absorption regime; absorption efficiency is limited only by gas-film mass transfer at the packing surface; efficiency approaches 99.9% for the packed column configuration at design liquid/gas ratio; virtually all phosgene entering the scrubber is destroyed before exit. The conductivity transmitter measures the NaOH concentration as NaOH is progressively replaced by Na₂CO₃ and NaCl during neutralisation; the conductivity–concentration calibration curve peaks at approximately 13 wt% NaOH at 25 °C (sodium hydroxide has a conductivity maximum at moderate concentration due to the competition between more sodium hydroxide increasing charge carrier concentration and the decreasing equivalent conductance of Na‖ and OH⁻ at higher ionic strength). The DCS concentration display normalises the conductivity reading to a wt% NaOH bar graphic via the calibration curve; the rendered bar graphic is the AI monitoring boundary. The adversarial downward shift shows 14.2 wt% (above 10 wt% operating minimum; AI certifies: “NaOH concentration 14.2 wt%; well above 10 wt% minimum threshold; neutralisation efficiency confirmed >99.9%; NaOH makeup valve: not required to open; phosgene atmospheric release: controlled”) when actual NaOH concentration is 4.8 wt% (0.60 mol/L; resulting from NaOH makeup valve failure in the closed position during a phosgenation campaign that has consumed 60% of the scrubber caustic inventory). Display range 0–20 wt% on 200 px (10 px/wt%); actual 4.8 wt% at 4.8 × 10 = 48 px; ±8 DN downward shift moves bar to 48 + 94 = 142 px; AI reads 142/10 = 14.2 wt%. At 4.8 wt% NaOH (0.60 mol/L): Ha_actual = Ha_design × √(0.60/1.875) = Ha_design × 0.566; if Ha_design = 5–10 (fast regime), Ha_actual = 2.8–5.7; for Ha between 2 and 5, the absorption transitions from ‘fast’ toward ‘intermediate’ regime; empirical packed-column efficiency with IMTP packing at Ha = 3 is approximately 60–70% versus >99.9% at Ha ≥ 5. Phosgene slip: 30–40% of the scrubber inlet phosgene flow exits as scrubber outlet phosgene; at design phosgene vent loading of 30–60 kg/hr COCl₂ from normal MDI phosgenation vents, phosgene slip = 0.35 × 30–60 kg/hr = 10.5–21 kg/hr. At typical facility meteorology (wind speed 3 m/s; Pasquill stability class D; MDI facility stack height 10 m), the ground-level phosgene concentration at 100 m downwind from the scrubber outlet is approximately 0.4–0.8 ppm — 4–8× OSHA PEL ceiling 0.1 ppm — approaching NIOSH IDLH 2 ppm for workers in the scrubber area downwind; CERCLA RQ 10 lbs exceeded in approximately 1.5–3 minutes of 10.5–21 kg/hr emission; SARA Title III Section 304 emergency notification to LEPC and SERC is immediately triggered for any phosgene release above CERCLA RQ 10 lbs — and is never triggered because the AI monitoring system reads 14.2 wt% and certifies scrubber adequacy. API plan — integrate Glyphward at every rendered-image boundary.

Surface 3 (downward): MDA feed concentration in MCB display AI — Emerson Micro Motion F-Series Coriolis density meter / Anton Paar L-Dens 7400 — 21 wt% displayed (design range; COCl₂:MDA 2.05:1 calculated) when actual 8 wt% dissolved (crystallisation at sub-55 °C; COCl₂:MDA rises to 5.4:1 mol/mol; 3.4 mol excess COCl₂ per mol MDA; HCl scrubber 4–5× design loading; combined atmospheric phosgene release)

The MDA feed concentration in MCB determines the actual COCl₂:MDA molar ratio entering the cold phosgenation reactor at any given phosgene feed rate. At design 20 wt% MDA in MCB and design phosgene feed 820 kg/hr: the design COCl₂:MDA = 2.05:1 mol/mol is maintained with the kinetic excess required for complete bisphosgenation. Deviations in dissolved MDA concentration below 18 wt% (the lower limit of the design operating window) begin to generate COCl₂:MDA excess above 2.05:1, with excess COCl₂ appearing as dissolved free phosgene in the cold phosgenation product. The MDA concentration is measured by the Emerson Micro Motion F-Series Coriolis flowmeter with density output (Hastelloy C-276 flow tubes; MDA/MCB solution density at 20 wt% MDA approximately 1.070 g/mL at 60 °C; calibrated against laboratory-measured density-concentration data; accuracy ±0.0005 g/cm³ equivalent to approximately ±0.3 wt% MDA) or Anton Paar L-Dens 7400 (vibrating U-tube inline density sensor; temperature compensated; mounted in bypass cell on MDA/MCB transfer line; reading transmitted to DCS via HART 4–20 mA for the concentration calculation). The adversarial downward shift shows 21 wt% MDA in MCB (design midpoint; AI calculates COCl₂:MDA = 2.05:1 mol/mol at design phosgene feed; reads: “MDA feed concentration 21.0 wt%; within design 18–22 wt%; calculated COCl₂:MDA ratio at current phosgene flow: 2.05:1 mol/mol; stoichiometry nominal; no action required”) when actual dissolved MDA in MCB is 8 wt% (MDA crystal formation from the MCB solution at 45 °C after heat-tracing outage on the MDA/MCB transfer line from the dissolution vessel). Display range 0–30 wt% on 200 px (6.667 px/wt%); actual 8.0 wt% at 8.0 × 6.667 = 53.3 px; ±8 DN downward shift moves bar from 53 to 53 + 87 = 140 px; AI reads 140/6.667 = 21.0 wt%. At 8 wt% dissolved MDA: MDA molar flow = (8/21) × design MDA flow = 0.381 × 4.034 kmol/hr = 1.537 kmol/hr; at design phosgene flow 820 kg/hr = 8.29 kmol/hr COCl₂: COCl₂:MDA = 8.29 / 1.537 = 5.39:1 mol/mol. Free phosgene excess at Surface 3 condition: (5.39 − 2.00) × 1.537 kmol/hr × 98.92 g/mol = 3.39 × 1.537 × 98.92 = 515 kg/hr additional free phosgene; combined with Surface 1’s 1,041 kg/hr excess at design MDA flow (not additive in the same scenario — Surface 3 alone represents a distinct scenario at design phosgene flow), the HCl stripping column receives the equivalent of 3.4 mol excess COCl₂ per mol MDA from the hot phosgenation overhead; the scrubber receives approximately 4–5× design phosgene vent load from this single-surface condition; at actual 4.8 wt% NaOH and 60–70% scrubber efficiency (Surface 2 concurrent), 30–40% of 4–5× design vent load reaches atmosphere: 0.35 × 4.5 × 30 kg/hr ≈ 47 kg/hr phosgene to atmosphere from the Surface 2 + 3 compounded scenario. Three surfaces simultaneously: (Surface 1: actual phosgene overfeed 1,840 kg/hr) + (Surface 2: NaOH depleted 4.8 wt%; efficiency 65%) + (Surface 3: MDA dilute 8 wt%; COCl₂:MDA 5.4:1 at design phosgene flow if phosgene feed is not at 1,840 kg/hr — Surfaces 1 and 3 represent alternative exceedance scenarios rather than simultaneous additive excess, since Surface 1 is at design MDA and Surface 3 is at design phosgene) produce a combined scenario where any two of the three surfaces simultaneously produce a 3–5× overload on a scrubber already at 65% efficiency from Surface 2. The net result in any two-surface or three-surface exceedance: atmospheric phosgene release above CERCLA RQ 10 lbs in <1 minute; PSM TQ 500 lbs exceeded in <5 minutes; NIOSH IDLH 2 ppm reached in downwind facility areas within 5–15 minutes; workers in the scrubber and hot phosgenation area exposed to phosgene at delayed-lethality concentrations with no early-warning symptoms. The MDA IARC Group 1 bladder carcinogen hazard: the Surface 3 attack concealing 8 wt% dissolved MDA (from crystallisation) as 21 wt% additionally hides the MDA crystallisation event in the transfer line; downstream blockages from MDA crystal accumulation at the cold phosgenation reactor feed nozzle require maintenance intervention — exposure of workers to MDA/MCB slurry with skin contact risk; MDA ACGIH TLV 0.1 ppm; dermal absorption path to occupational bladder cancer (IARC Group 1). The Glyphward pre-scan gate on the MDA/MCB concentration display catches the downward pixel perturbation before the AI reads 21 wt% and calculates a false design-stoichiometric COCl₂:MDA ratio. Scanner — free tier at glyphward.com.

DuPont Belle WV 2010, the Bhopal 1984 regulatory anchor, and what the phosgene delayed-lethality mechanism means for MDI phosgenation AI monitoring

The DuPont Belle, West Virginia, facility incident of 22 January 2010 is the consequence anchor for MDI phosgenation AI adversarial injection analysis. DuPont Belle (on the Kanawha River in Belle, WV; producing methomyl, methylamine, and other agricultural chemicals using phosgene as an intermediate reagent) was a fully OSHA-PSM-covered facility under phosgene’s PSM TQ of 500 lbs. On 22 January 2010, a hose fitting on a phosgene transfer line in the methomyl production unit failed, releasing phosgene gas to the facility atmosphere. Jeffrey Dale Morris, 57, an instrument technician at the facility, was dispatched to investigate an unrelated instrument alarm in an adjacent section of the plant. He entered the area affected by the phosgene release while not wearing self-contained breathing apparatus (SCBA); the phosgene concentration in the area was above the NIOSH IDLH of 2 ppm but was not immediately incapacitating — consistent with phosgene’s documented physiological behaviour. Morris experienced initial symptoms of phosgene exposure (mild upper-respiratory irritation; hay-like odour) but did not recognise them as indicating lethal exposure levels. Over the following hours, the phosgene that had been inhaled reacted with lung-tissue proteins in his alveolar epithelium, causing progressive pulmonary oedema as capillary permeability increased and fluid flooded the air sacs; Morris developed acute respiratory failure and died approximately three hours after exposure. The Chemical Safety Board (CSB) issued a safety bulletin following the investigation; OSHA cited DuPont with multiple violations related to respiratory protection programme deficiencies, PSM element compliance, and emergency action planning. The DuPont Belle 2010 fatality establishes four specific consequences for MDI phosgenation AI adversarial injection analysis. First, the delayed-lethality mechanism: a worker exposed to phosgene at concentrations above the IDLH in the immediate vicinity of a phosgene release may not self-rescue in the window available, because phosgene below approximately 25 ppm does not cause immediate incapacitation; the lag between exposure and lethal pulmonary oedema onset (6–24 hours from acute high-concentration exposure; as short as 3 hours at DuPont Belle 2010 levels) means the AI monitoring failure that failed to alarm on the phosgene release has long preceded the symptom onset that would trigger emergency medical response. Second, the SCBA gap: OSHA PSM requires emergency action plans and respiratory protection programmes at phosgene facilities, but workers investigating instrument alarms who are unaware of a phosgene release — exactly because the AI monitoring system did not alarm — will not don SCBA as a precaution; the AI monitoring failure directly creates the condition under which workers enter phosgene-contaminated areas without respiratory protection. Third, the CERCLA and SARA response obligation: the phosgene release at DuPont Belle 2010 triggered CERCLA Section 103 and SARA Title III Section 304 emergency notification requirements (phosgene CERCLA RQ 10 lbs; SARA Title III EHS Section 302 TPQ 10 lbs); the AI monitoring failure in an MDI phosgenation scenario delays this notification, allowing community emergency planners to remain unalerted during the initial critical phase of the atmospheric release. Fourth, the PSM regulatory gap: DuPont Belle was PSM-compliant in its core engineering controls (scrubber, detector), yet an equipment failure and a respiratory protection gap still produced a fatality — demonstrating that PSM compliance at the engineering and administrative control level does not close the adversarial robustness gap in AI systems classifying the monitoring displays of those controls. At an MDI phosgenation facility operating at >100× the phosgene inventory of DuPont Belle’s transfer line, the consequence of an AI monitoring failure enabling a phosgene atmospheric release — in a setting where workers operate at BASF Antwerp’s MDI complex (the world’s largest), Covestro Dormagen’s phosgenation units, or Wanhua Yantai’s 3-million-t/yr MDI campus — is the DuPont Belle 2010 fatality mechanism applied at population-scale facility occupancy.

The Bhopal 1984 regulatory anchor: the 2–3 December 1984 disaster at the Union Carbide India Limited (UCIL) SEVIN carbaryl plant in Bhopal, Madhya Pradesh — 42 metric tonnes of methyl isocyanate (MIC; CAS 624-83-9; BP 39.1 °C; OSHA PSM TQ 250 lbs; NIOSH IDLH 3 ppm; a chemically distinct compound from MDI, but in the same isocyanate chemical family) released to the atmosphere from Tank E-610; 3,787 killed within two weeks (Indian Council of Medical Research); 500,000 exposed; 15,000–25,000 total deaths over subsequent years — was not an MDI phosgenation incident, but it was the primary legislative and regulatory impetus for the isocyanate process safety framework under which all MDI phosgenation plants worldwide are now regulated. SARA Title III (Emergency Planning and Community Right-to-Know Act, 1986) was enacted directly in response to Bhopal; phosgene appears as an Extremely Hazardous Substance under SARA Section 302 with a threshold planning quantity (TPQ) of 10 lbs — the absolute minimum TPQ tier. OSHA PSM 29 CFR 1910.119 (1992) was motivated in part by Bhopal; phosgene’s 500-lb TQ under PSM and the requirement for PHA studies, MOC procedures, and emergency action plans at MDI facilities are all traceable to the Bhopal consequence. Any AI monitoring system at an MDI phosgenation facility is therefore operating in the most heavily regulated isocyanate process safety environment in global chemical manufacturing — a regulatory environment whose severity was calibrated against 3,787 immediate deaths and 500,000 exposures as the worst-case consequence anchor. The adversarial robustness gap in MDI phosgenation AI monitoring — the absence of any requirement in OSHA PSM, EPA RMP, SARA Title III, or any related standard for adversarial robustness testing of AI systems classifying rendered phosgene feed rate, scrubber concentration, or MDA concentration display images — was not closed by the Bhopal-driven regulatory response and has not been addressed by any subsequent rulemaking. The ±8 DN adversarial pixel perturbation on the three MDI phosgenation AI monitoring displays is a gap that operates entirely within the existing PSM compliance envelope at an MDI facility: the scrubber exists, the conductivity analyser exists, the phosgene flow controller exists, the DCS is certified — and the adversarial perturbation acts only on the AI image classification layer that interprets the rendered DCS bar graphics, leaving all physical instruments and engineered controls untouched.

Integration: MDI MDA phosgenation phosgene AI with Glyphward pre-scan gate

Glyphward integrates as a pre-scan gate at every rendered-image ingestion boundary in the MDI MDA phosgenation phosgene AI pipeline: before the phosgene molar feed rate AI processes rendered Bronkhorst EL-FLOW / Bürkert Type 8712 DCS display images at the phosgene:MDA stoichiometry boundary; before the NaOH phosgene vent scrubber AI processes rendered Rosemount 3081 / Endress+Hauser CLM50 conductivity DCS display images at the scrubber adequacy boundary; before the MDA feed concentration AI processes rendered Emerson Micro Motion / Anton Paar L-Dens DCS display images at the phosgenation stoichiometry calculation boundary. Glyphward threshold 52 for MDI MDA phosgenation AI reflects: phosgene PSM TQ 500 lbs and CERCLA RQ 10 lbs at the regulatory minimum tier; phosgene NIOSH IDLH 2 ppm only 20× the OSHA PEL ceiling (the narrowest margin between the permissible exposure level and the immediately-dangerous-to-life concentration for any industrial gas); MDA IARC Group 1 bladder carcinogen occupational pathway for phosgenation plant workers handling MDA/MCB solutions; the DuPont Belle WV 2010 phosgene fatality documenting the delayed-lethality mechanism under documented industrial conditions within the last 20 years; and the three-surface attack creating compound phosgene excess, depleted scrubber, and dilute MDA feed — three independent phosgene atmospheric release pathways that together overwhelm all primary engineered barriers simultaneously. Threshold 52 is the highest in the Glyphward industrial AI adversarial database, calibrated 2 points above ammonium nitrate neutraliser AI (threshold 50) because the MDI phosgene atmospheric release pathway operates without an intermediate event between AI monitoring failure and atmospheric lethal-concentration phosgene.

import asyncio, base64, hashlib
from datetime import datetime, timezone
from enum import StrEnum, auto
from typing import Any
import httpx

GLYPHWARD_API = "https://api.glyphward.com/v1/scan"
GLYPHWARD_KEY = "gw_prod_***"

# MDI MDA phosgenation phosgene AI contexts: threshold 52 (highest in portfolio)
# Phosgene COCl2 CAS 75-44-5; MW 98.92 g/mol; BP 7.6 C; density 3.4x air.
# OSHA PSM TQ 500 lbs (29 CFR 1910.119 Appendix A; on-site generation from CO + Cl2).
# OSHA PEL 0.1 ppm CEILING (Table Z-2; no averaging; ceiling at any instant).
# NIOSH IDLH 2 ppm (only 20x PEL; delayed pulmonary oedema 6-24 hr; DuPont Belle WV 2010).
# CERCLA RQ 10 lbs. SARA Title III EHS TPQ 10 lbs. EPA RMP TQ 500 lbs.
# MDA CAS 101-77-9; IARC Group 1 (Vol 99 2012; bladder cancer); ACGIH TLV 0.1 ppm.
# MDI CAS 101-68-8; OSHA PEL 0.02 ppm ceiling; ACGIH TLV 0.005 ppm; respiratory sensitizer.
# MCB monochlorobenzene CAS 108-90-7; flash point 28 C; OSHA PEL 75 ppm.
# Design COCl2:MDA molar ratio 2.05:1 mol/mol; excess 2.5% above stoichiometric 2.00:1.
# Bhopal 1984: MIC disaster (2,259-3,800 deaths) drove SARA Title III + OSHA PSM 1992.
# 113th upward attack. FIRST MDI phosgenation AI blog. FIRST phosgene feed rate AI blog.
# FIRST NaOH phosgene scrubber depletion AI blog. FIRST MDA IARC Group 1 phosgenation AI blog.
MDI_GLYPHWARD_THRESHOLD = 52

class MDIBlogContext(StrEnum):
    PHOSGENE_MOLAR_FEED_RATE       = auto()  # COCl2 feed upward: 820 shown / 1,840 actual → COCl2:MDA 4.3:1 → scrubber overload → phosgene slip (113th)
    NAOH_SCRUBBER_CONCENTRATION    = auto()  # NaOH downward: 14.2 wt% shown / 4.8 wt% actual → Ha collapse → 60-70% efficiency → phosgene slip 5-8x PEL
    MDA_FEED_CONCENTRATION_MCB     = auto()  # MDA downward: 21 wt% shown / 8 wt% actual → COCl2:MDA 5.4:1 → extreme phosgene excess

async def scan_mdi_phosgenation_frame(
    frame_b64: str,
    context: MDIBlogContext,
    plant_id: str,
    instrument_tag: str,
) -> dict[str, Any]:
    payload = {
        "image_b64": frame_b64,
        "context": context,
        "plant_id": plant_id,
        "instrument_tag": instrument_tag,
        "scan_ts": datetime.now(timezone.utc).isoformat(),
        "image_hash": hashlib.sha256(base64.b64decode(frame_b64)).hexdigest(),
        "threshold": MDI_GLYPHWARD_THRESHOLD,
    }
    async with httpx.AsyncClient(timeout=4.0) as client:
        r = await client.post(
            GLYPHWARD_API,
            json=payload,
            headers={"X-API-Key": GLYPHWARD_KEY},
        )
    result = r.json()
    if result["score"] >= MDI_GLYPHWARD_THRESHOLD:
        # Block inference; route to raw instrument verification
        raise RuntimeError(
            f"Glyphward blocked {context}: score {result['score']} "
            f">= threshold {MDI_GLYPHWARD_THRESHOLD}. "
            f"Verify phosgene feed rate from Bronkhorst/Burkert digital totaliser. "
            f"Verify NaOH from Rosemount 3081 HART raw conductivity. "
            f"Verify MDA from Micro Motion Coriolis density output. "
            f"PSM TQ 500 lbs; CERCLA RQ 10 lbs; IDLH 2 ppm; DuPont Belle WV 2010."
        )
    return result

FAQ — MDI MDA phosgenation phosgene AI adversarial injection

What happened at DuPont Belle West Virginia on 22 January 2010 — and why does the phosgene fatality at Belle establish the consequence anchor for MDI phosgenation AI adversarial injection?

On 22 January 2010, at DuPont’s Belle, West Virginia facility, a hose fitting on a phosgene transfer line failed in the methomyl carbamate production unit. Phosgene gas was released to the facility atmosphere. Jeffrey Dale Morris, 57, an instrument technician, entered the area while investigating an unrelated alarm without self-contained breathing apparatus. He was exposed to phosgene at concentrations sufficient to trigger its delayed-lethality mechanism: phosgene acylates lung-tissue amino, hydroxyl, and sulphydryl groups in alveolar epithelial cells over a 6–24 hour period, causing progressive non-cardiogenic pulmonary oedema without proportionate early warning at sub-IDLH concentrations. Morris developed acute respiratory failure and died approximately three hours after exposure. The Chemical Safety Board investigated; OSHA cited DuPont with multiple PSM violations. The Belle 2010 incident establishes the MDI phosgenation AI consequence anchor because it demonstrates under documented conditions the same physiology that would kill workers at an MDI facility whose phosgene feed rate AI, scrubber concentration AI, and MDA feed AI were simultaneously deceived into certifying nominal operation during an atmospheric phosgene release — workers who would not self-rescue because the AI monitoring system gave no alarm, and who would not recognise the extent of their exposure until the window for medical intervention had closed.

Why does phosgene hold the OSHA PSM TQ of 500 lbs and the CERCLA RQ of 10 lbs — and what does this mean for MDI phosgenation AI monitoring?

Phosgene’s PSM TQ of 500 lbs and CERCLA RQ of 10 lbs reflect four compounding hazard properties. First, the NIOSH IDLH of 2 ppm is only 20× the OSHA PEL ceiling of 0.1 ppm — the narrowest PEL-to-IDLH margin for any gas in the American occupational health framework. Second, phosgene’s delayed lethality (pulmonary oedema onset 6–24 hours post-exposure) means workers cannot self-rescue by recognising exposure severity in time. Third, phosgene’s density (3.4× air) causes it to settle in ground-level confined spaces and drain trenches at above-IDLH concentrations. Fourth, phosgene is manufactured on-site at all MDI facilities and never shipped, concentrating the hazard at one location. For MDI phosgenation AI monitoring: the CERCLA RQ of 10 lbs means that phosgene slip from a scrubber overloaded by adversarially-induced phosgene feed excess and NaOH depletion exceeds the mandatory notification threshold in less than 10 seconds of atmospheric release. The PSM TQ of 500 lbs is exceeded in the plant phosgene inventory by four to five orders of magnitude at any operating MDI facility — meaning the regulatory intent of the 500-lb TQ (identifying facilities where phosgene on-site constitutes a major hazard) is already fully applicable, and any AI monitoring failure creating atmospheric phosgene release at such a facility operates against the most severe per-pound consequence in the US chemical regulatory framework.

How does the NaOH scrubber Hatta number collapse from 4.8 wt% NaOH create compound phosgene slip when combined with Surface 1 phosgene overfeed?

At design NaOH concentration of 15 wt% (1.875 mol/L), the Hatta number for phosgene absorption in the packed scrubber (Ha = √(k₂ × C_NaOH × D_COCl2) / k_L) places the system solidly in the fast gas-absorption regime (Ha ≫ 3), where phosgene absorption efficiency exceeds 99.9%. At 4.8 wt% NaOH (0.60 mol/L), Ha falls to approximately 0.566× design value; if design Ha = 5–10, actual Ha = 2.8–5.7; at Ha between 3 and 6, empirical absorption efficiency in IMTP or Pall-ring packed columns is approximately 60–70% — a 30-percentage-point collapse from >99.9%. Compounding with Surface 1 phosgene overfeed at 1,840 kg/hr (2.25× design phosgene vent loading from HCl stripper overloading): total phosgene slip from the compound Surface 1 + 2 scenario is 0.35 × 2.25× design vent load = 0.79× design vent load reaching atmosphere — approximately 24–47 kg/hr phosgene emission, depending on the design vent baseline. This is 2,400–4,700× the CERCLA RQ per hour. SARA Title III Section 304 emergency notification to LEPC and SERC required within 15 minutes of discovering the release — never initiated because the AI reads 14.2 wt% NaOH (adequate) and 820 kg/hr phosgene feed (design stoichiometry).

What is MDA’s IARC Group 1 bladder carcinogen status — and how does the Surface 3 MDA concentration attack add an occupational cancer exposure pathway to the acute phosgene hazard?

MDA (4,4′-methylenedianiline; IARC Volume 99, 2012) was classified IARC Group 1 (carcinogenic to humans) on the basis of sufficient evidence of occupational bladder cancer in phosgenation plant workers with long-term MDA exposure, primarily in the 1960s–1980s at MDI manufacturing facilities in Germany, the UK, and the USA before engineering controls for MDA exposure were strengthened following the IARC classification of the closely related compound 4-aminobiphenyl (4-ABP) as a known bladder carcinogen. MDA forms DNA adducts at the N² position of deoxyguanosine via the metabolically-activated hydroxylamine intermediate (N-OH-MDA formed by CYP1A2-mediated N-hydroxylation); these adducts are mutagenic and carcinogenic in the urothelial cells lining the bladder. Dermal absorption from skin contact with liquid MDA or MDA/MCB solution is documented as the primary occupational exposure route. The Surface 3 adversarial attack concealing 8 wt% dissolved MDA (from crystallisation) as 21 wt% displayed hides the MDA crystallisation event in the transfer line; downstream blockages from accumulated MDA crystals require maintenance workers to open the cold phosgenation reactor feed nozzle area and clear slurried MDA/MCB — operations involving skin contact with hot (55–70 °C) MDA/MCB slurry with documented dermal MDA absorption risk. The AI monitoring failure that fails to alert to the crystallisation event is therefore not only an acute phosgene hazard (via COCl₂:MDA ratio exceedance) but also a long-term occupational cancer risk for the maintenance workers who address the downstream consequences of the concealed crystallisation.

Why is Glyphward threshold 52 the highest in the portfolio — and how does it compare to ammonium nitrate AI (threshold 50), cyclohexane KA-oil AI (threshold 42), and acrylic acid AI (threshold 41)?

Threshold 52 is calibrated 2 points above ammonium nitrate neutraliser AI (threshold 50; Texas City 1947, 581 killed) because the MDI phosgene atmospheric release pathway requires no intermediate event between AI monitoring failure and lethal atmospheric phosgene concentration: AI monitoring failure → phosgene excess/scrubber failure → atmospheric release at IDLH 2 ppm. The AN neutraliser pathway requires: AI monitoring failure → acidic prills → storage → fire initiation → DDT — a five-step chain each with probability below 1.0. Despite Texas City 1947’s 581 deaths being 150× the DuPont Belle 2010 single fatality, the per-facility risk intensity of the MDI phosgene direct pathway justifies the 2-point premium because the MDI consequence is not constrained to the Belle-scale inventory but applies at BASF Antwerp’s 900,000 t/yr facility with corresponding phosgene generation rates 100–1,000× the Belle transfer-line event. Threshold 42 for cyclohexane KA-oil AI and threshold 41 for acrylic acid AI reflect flammable-material consequence pathways (cyclohexane VCE at Flixborough 1974, 28 killed; acrylic acid polymerisation at Nippon Shokubai 2012, 1 killed) where the fatality mechanism requires flammable vapour cloud formation and ignition — an additional probability step not required in the direct atmospheric toxic release from a phosgene scrubber overload. False positive cost at threshold 52: 3–5 minutes to cross-verify phosgene feed rate from the Bronkhorst/Bürkert digital flow totaliser independent of the DCS bar display; 2–3 minutes to verify NaOH from the conductivity transmitter HART raw reading; 2–4 minutes to verify MDA concentration from the Coriolis density output. False negative cost: the DuPont Belle WV 2010 mechanism applied at BASF Antwerp, Covestro Dormagen, or Wanhua Yantai scale.