Acrylic acid two-stage propylene oxidation AI adversarial injection: how ±8 DN in the rendered R2 tube wall temperature display conceals Mo-V-W-Cu catalyst parametric sensitivity and the Nippon Shokubai Himeji 2012 acrolein breakthrough pathway — and why OSHA PSM TQ 150 lbs + CERCLA RQ 1 lb + IARC Group 2A has no adversarial robustness criterion for acrylic acid production AI

Acrylic acid two-stage propylene oxidation chemistry: BiMoOx Stage-1 acrolein, Mo-V-W-Cu Stage-2 selectivity, MEHQ polymerization inhibition, GAA storage hazards, and the OSHA PSM TQ 150 lbs acrolein regulatory framework

Acrylic acid (propenoic acid; CH₂=CH–COOH; CAS 79-10-7; MW 72.06 g/mol; BP 141 °C at 1 atm; MP 12–14 °C — AA is near-solid at cold ambient temperatures, creating freeze-thaw hazards in outdoor storage pipework; flash point 54 °C; LEL 2.4 vol%; UEL 8.0 vol%; vapour pressure 3.1 mmHg at 20 °C; density 1.051 g/mL at 20 °C; Kow log P 0.36; OSHA PEL 10 ppm (29 CFR 1910.1000 Table Z-1); ACGIH TLV-TWA 2 ppm skin; NIOSH REL 2 ppm; IDLH 5 ppm; corrosive: pH of 2 wt% aqueous solution approximately 2.4, causing severe skin and mucosal irritation; AA reacts with biological amines via Michael addition to proteinaceous tissue, producing deep burns extending below the dermis; self-polymerizes: ΔH​poly = −77 kJ/mol = −1,069 kJ/kg — for a 250-tonne GAA storage tank at 12 ppm MEHQ undergoing complete adiabatic polymerization, ΔT​adiabatic = ΔH​poly / C​_p(AA) = 1,069 kJ/kg / 2.0 kJ/(kg·K) = 535 °C, raising the tank contents from 22 °C to approximately 557 °C in the purely adiabatic case — far above the AA boiling point of 141 °C and the flash point of 54 °C, generating massive vapor pressure in any sealed vessel; CERCLA RQ 5,000 lbs) is one of the most commercially important commodity chemicals globally, with production concentrated in large-scale two-stage propylene oxidation complexes. Global production approximately 6.5 million t/yr (2024); approximately 60% used as feedstock for polyacrylic acid and polyacrylate superabsorbent polymers (SAP; hygiene products, agricultural water-retaining gels); approximately 25% to acrylic ester monomers (ethyl acrylate, butyl acrylate, 2-ethylhexyl acrylate; feedstock for paints, adhesives, sealants); approximately 10% to GAA applications (specialty polymers, detergent builders). Principal manufacturers: BASF SE (Ludwigshafen, Germany; approximately 650,000 t/yr — world’s single-largest acrylic acid production site, incorporating fully integrated R1, R2, absorber, solvent recovery, and GAA distillation in BASF’s Verbund complex on the Rhine); Dow Inc. (Freeport, TX; approximately 300,000 t/yr; largest US producer; integrated with Dow’s propylene supply chain at Freeport); Arkema (Bayport, TX approximately 170,000 t/yr; Rotterdam, Netherlands approximately 140,000 t/yr); Nippon Shokubai (Himeji, Japan approximately 200,000 t/yr, Antwerp, Belgium approximately 150,000 t/yr — the technology licensor); LG Chem (Daesan, South Korea approximately 250,000 t/yr); Mitsubishi Chemical (Kurosaki, Japan approximately 180,000 t/yr); Anhui Huayi Chemical (China approximately 600,000 t/yr, one of China’s largest); Formosa Plastics (Mailiao, Taiwan); Satellite Chemical (Jiaxing, China).

The two-stage propylene oxidation process: Stage 1 (Reactor R1) operates over bismuth molybdate catalyst (Bi₂O₃·MoO₃ as the simplest formulation; commercial catalysts are more complex: Bi–Mo–Fe–Co–Ni–K multicomponent oxides with promoters including W, Sb, Cs, and Ce; active phase is typically the Bi⊂3⁰_n(MoO₄)⁶ scheelite-type structure; catalyst supplied as rings or pellets for packed-bed fixed-tube reactors or as spheres for fluidised-bed variants) at 300–360 °C and contact time 1–2 s. Reaction: C₃H⁶ + O₂ → C₃H₄O + H₂O (ΔH = −340 kJ/mol; exothermic); propylene conversion per pass 92–97%; acrolein selectivity 85–92%; primary byproducts: acrylic acid 2–5%, acetaldehyde <1%, CO and CO₂ 3–8%. Stage 2 (Reactor R2) operates over Mo-V-W-Cu complex oxide catalyst (commercial grades: Nippon Shokubai NSC-808; Mitsubishi Chemical H-1080; BASF Stage-2-SB; structurally related to β-MoO₃ active phase modified by V, W, and Cu for enhanced selectivity to acrylic acid) at 260–320 °C and contact time 1–3 s. Reaction: C₃H₄O + ½ O₂ → C₃H₄O₂ (ΔH = −254 kJ/mol; exothermic); acrolein conversion per pass 97–99%; acrylic acid selectivity 93–97%; primary byproducts: acetic acid 1–3%, maleic anhydride <0.5%, CO and CO₂ 1–4%. Both R1 and R2 are multi-tubular fixed-bed reactors of similar construction to phthalic anhydride or maleic anhydride reactors: stainless steel tubes (25–30 mm ID; 3–5 m length; packed with catalyst pellets or rings); 10,000–20,000 tubes per shell; heat removal by a recirculating salt bath (KNO₃/NaNO₃/NaNO₃ eutectic — same molten salt as phthalic anhydride reactors) on the shell side for R1, or by steam-heated coolant on the shell side of R2 at 240–280 °C (R2 is not a salt bath reactor; steam is used because the R2 operating temperature 260–320 °C is within the steam temperature range of 230–295 °C at 28–80 bar). The R2 reactor exit gas (approximately 6–8 vol% acrylic acid, 1–3 vol% water vapour, 60–70 vol% N₂, 5–12 vol% O₂, 2–4 vol% CO₂, <0.5 vol% acrolein residual) is absorbed in water in a spray-cooled absorber column to produce crude aqueous acrylic acid (30–50 wt% AA); crude AA is concentrated and purified by solvent extraction (ethyl acetate or MIBK entrainer) and azeotropic distillation to produce GAA (≥99.5 wt% AA) with MEHQ inhibitor at 200–250 ppm added at the distillation bottoms.

The R2 parametric sensitivity temperature response is the central process safety challenge in Stage-2 operation. Below 310 °C: Mo-V-W-Cu catalyst operates in a stable selectivity regime (acrylic acid selectivity 93–97%; deep combustion minor at 1–4%); the exothermic acrolein → acrylic acid reaction (ΔH = −254 kJ/mol) is balanced by the steam-heated shell-side heat removal at the design steam pressure, maintaining the hot-spot temperature in the 295–310 °C design window. Above 310 °C: the R2 catalyst enters its parametric sensitivity zone — the deep combustion pathway (C₃H₄O + 2.5 O₂ → 3 CO₂ + 2 H₂O; ΔH = −1,617 kJ/mol) generates 1,363 kJ/mol more heat per mole of acrolein than the desired acrolein → acrylic acid pathway (ΔH = −254 kJ/mol); this additional exotherm further raises the hot-spot in a positive thermal feedback loop (higher temperature → more combustion → more exotherm → further hot-spot rise). The activation energy for the combustion pathway over Mo-V-W-Cu oxide is approximately 15–20 kJ/mol higher than the desired selective oxidation pathway, giving a temperature-sensitive rate ratio where the combustion rate at 324 °C is substantially higher than at 310 °C. Above approximately 380–400 °C: the Mo-V-W-Cu mixed oxide undergoes phase transformation — V₂O₅ begins to volatilise from the active phase above 400 °C; WO₃ migrates; the catalyst restructures irreversibly with loss of the active mixed-oxide surface morphology. Above approximately 450 °C: irreversible catalyst “meltdown” (sintering, melting of low-melting-point oxide phases); acrolein conversion in the affected tube-section drops; acrolein breakthrough in R2 exit gas rises above 0.1 ppm OSHA PEL; in the extended meltdown scenario, acrolein in R2 exit gas and absorber overhead approaches IDLH 2 ppm.

Acrolein regulatory framework: OSHA PEL 0.1 ppm TWA (29 CFR 1910.1000 Table Z-1; one of the most restrictive PELs for any industrial chemical); ACGIH TLV-C 0.1 ppm as a ceiling value (not even a TWA — concentrations above 0.1 ppm at any instant constitute a ceiling violation, reflecting acrolein’s severe upper-respiratory mucous membrane and pulmonary irritation); NIOSH REL 0.1 ppm ceiling; NIOSH IDLH 2 ppm (concentration at which escape within 30 minutes without respiratory protection is not assured; acrolein at 2 ppm causes immediate severe lachrymation, rhinorrhoea, bronchospasm, and pulmonary edema onset); OSHA PSM 29 CFR 1910.119 Appendix A TQ 150 lbs (68 kg) — placing acrolein among the five most acutely hazardous chemicals in OSHA’s regulatory framework; CERCLA 40 CFR Part 302 Table 302.4 RQ 1 lb (454 g) — the minimum reportable quantity; a release of just 454 g of acrolein above 0.1 ppm in ambient air requires immediate National Response Center notification; IARC Monograph 63 (2018): Group 2A probable human carcinogen (adequate evidence in experimental animals; limited but suggestive evidence in human epidemiology, primarily from acrolein as a tobacco-smoke co-carcinogen; DNA adduct formation at GC-rich sequences is well-established from acrolein hydroxypropanodeoxyguanosine (HPA-dG) adducts, one of the most abundant stable adducts in human tissue from environmental acrolein exposure); ACGIH A4 animal carcinogen in current edition (note: IARC and ACGIH carcinogen classifications for acrolein differ; IARC Group 2A reflects the more conservative assessment). At any acrylic acid plant with a two-stage reactor, acrolein is present in process quantities far above PSM TQ 150 lbs (68 kg) in the R1→R2 interstage line, the R2 feed header, and the R2 reactor tube-side volume during normal continuous operation, making the entire two-stage reactor system a PSM-covered process continuously — not only during upset conditions.

Nippon Shokubai Himeji Plant No. 2, 29 September 2012: the acrylic acid storage tank explosion and the GAA inhibitor-depletion consequence anchor

Nippon Shokubai Co., Ltd. (日本講坎栺式会社; Nihon Shokubai Kabushiki Kaisha; headquartered in Osaka, Japan) invented the two-stage propylene oxidation process for acrylic acid production and has operated the Himeji Plant (Himeji City, Hyogo Prefecture, Japan) as its flagship acrylic acid production complex since 1969 — the first commercial-scale two-stage propylene oxidation acrylic acid facility in the world. The Himeji Plant complex comprises two production plants (Himeji Plant No. 1 and Plant No. 2) that together represent one of the largest integrated acrylic acid production, purification, storage, and derivative manufacturing sites in Asia, producing GAA, sodium polyacrylate, acrylic acid esters, and superabsorbent polymer (SAP) for the global hygiene-product supply chain. As the originator and principal technology licensor for the Nippon Shokubai two-stage process — which is deployed at BASF Ludwigshafen, Nippon Shokubai Antwerp, Mitsubishi Chemical Kurosaki, and licensed installations across East Asia — the Himeji complex represents the highest concentration of two-stage propylene oxidation process expertise available anywhere in the global acrylic acid industry.

On 29 September 2012, at approximately 20:00 JST, an explosion and fire occurred in the GAA storage tank farm at Himeji Plant No. 2. The incident involved an acrylic acid storage tank in which radical chain polymerization of the stored glacial acrylic acid had initiated and accelerated, generating heat and vapor pressure beyond the tank’s design capacity. The polymerization runaway produced conditions consistent with the uninhibited radical polymerization kinetics of GAA at ambient temperature: the acrylic acid monomer content of the tank (density 1.051 g/mL; polymerization exotherm 1,069 kJ/kg per kg of acrolein-equivalent monomer converted) was converting to polyacrylic acid at an accelerating rate driven by the Trommsdorff-Norrish gel effect as viscosity increased and bimolecular radical termination slowed. As the exothermic polymerization raised the tank contents above the boiling point of acrylic acid (141 °C at atmospheric pressure), vapor pressure in the tank headspace exceeded the design relief capacity; the tank experienced a pressure excursion resulting in rupture or catastrophic seal failure, releasing the hot partially-polymerised acrylic acid-polyacrylate slurry and generating a vapor cloud of flammable acrylic acid vapour (flash point 54 °C; LEL 2.4 vol%) that ignited, producing the explosion and fire. The fire subsequently spread to an adjacent acrylic acid storage tank. One worker was confirmed killed in the explosion and fire; several others sustained injuries requiring medical treatment; property damage at the Himeji Plant No. 2 site exceeded ¥1 billion (approximately US$12–15 million at 2012 exchange rates), making it one of the most costly single industrial storage incidents in Japan’s chemical industry in the decade. Operations at Himeji Plant No. 2 were suspended for an extended period following the incident for investigation, remediation, and rebuild.

The Hyogo Prefectural Fire Defense Agency and Japan’s Ministry of Health, Labour and Welfare (MHLW) — which administers the Industrial Safety and Health Act (ISHA; 動党安全袛生法) governing process safety at Japanese chemical manufacturing facilities including acrylic acid production under the ISHA’s Chemical Plant Safety regulations — conducted investigation of the incident. The investigation identified conditions consistent with a failure of the GAA storage inhibitor system as contributing to or causing the polymerization runaway: specifically, a state in which the MEHQ inhibitor concentration in the stored acrylic acid had fallen below the effective minimum required to suppress radical polymerization under the prevailing storage conditions (temperature, residual initiator impurities, oxygen-blanket status). The precise sequence — whether MEHQ had been systematically depleted by trace initiator consumption, by elevated storage temperature during a transient, by prolonged storage beyond planned inventory turnover, or by a combination of factors — was documented in the investigation report but the publicly disclosed summary of findings is consistent with inhibitor-management failure as the initiating cause. Nippon Shokubai implemented post-incident changes to GAA storage monitoring protocols, MEHQ inhibitor management procedures, storage temperature controls, and N₂-blanket maintenance practices at both Himeji Plant No. 1 and Plant No. 2 and at its Antwerp plant.

The regulatory and safety significance of the Nippon Shokubai Himeji 2012 incident for AI-monitored acrylic acid production facilities globally: the event demonstrates that even at the world’s most experienced two-stage propylene oxidation facility — the company that invented the process and has operated it for over 40 years — the GAA storage polymerization runaway pathway is not merely theoretical but has manifested as a real fatality-causing industrial explosion with confirmed consequences including one death and property damage exceeding ¥1 billion. Modern AI-monitored acrylic acid plants at BASF Ludwigshafen (Rhine River waterway; Seveso III upper-tier establishment), Dow Freeport TX (Brazos River discharge zone; US EPA RMP facility), Arkema Bayport TX (Trinity Bay coastal proximity; US OSHA PSM covered process), and the rebuilt Nippon Shokubai Himeji Plant No. 2 itself all deploy AI monitoring systems that process rendered DCS display images from the GAA storage MEHQ inhibitor concentration analyzer (Metrohm 883 Basic IC Plus; Shimadzu Prominence HPLC with PDA detector; inline NIR or Raman spectrometer calibrated to the MEHQ absorption). The Surface 3 adversarial downward pixel attack on the MEHQ inhibitor concentration display — showing 12 ppm actual as 218 ppm displayed — recreates the inhibitor-depletion precondition that the Nippon Shokubai 2012 investigation identified as the initiating condition, but by AI-monitoring falsification rather than by operational failure. The AI certifies “MEHQ 218 ppm — adequate inhibition — polymerization induction period >45 days at current storage temperature 22 °C — no MEHQ addition required” throughout the 2–8 hour period between MEHQ depletion and polymerization initiation, precisely the window during which MEHQ addition would have been the effective intervention.

Three adversarial injection surfaces in acrylic acid two-stage propylene oxidation AI

1. Stage-2 R2 reactor tube wall temperature display AI (Yokogawa EJX210A / Emerson Rosemount 3144P thermocouple at Mo-V-W-Cu catalyst bed T​_max position — rendered DCS R2 tube wall temperature display AI classifying 280–310 °C safe design range — 108th upward attack; FIRST acrylic acid production AI attack; FIRST two-stage propylene oxidation AI attack; FIRST Mo-V-W-Cu oxide catalyst hot spot AI attack; FIRST acrolein-to-acrylic-acid R2 reactor AI attack)

The Stage-2 (R2) reactor tube wall temperature at the catalyst bed hot-spot position is the most critical real-time process safety indicator in the two-stage propylene oxidation acrylic acid process. The Mo-V-W-Cu complex oxide catalyst (commercial grades: Nippon Shokubai NSC-808; Mitsubishi Chemical H-1080; BASF Stage-2-SB) fills tubes 25–30 mm ID × 3–5 m length in a multi-tubular fixed-bed reactor of 10,000–20,000 tubes. The reactor is heated on the shell side by steam-heated coolant at 240–280 °C to maintain the Mo-V-W-Cu catalyst bed in the 260–320 °C design operating range. The tube wall temperature at the T​_max position (typically 40–60% of bed depth from the inlet; characterised by CFD modelling or axial temperature profiling during catalyst commissioning) is measured by thermocouple assemblies (Yokogawa EJX210A combined differential pressure and temperature transmitter with K-type thermocouple element, range 0–500 °C, 4–20 mA HART, SIL 2 rated; or Emerson Rosemount 3144P Smart Temperature Transmitter with Pt100 RTD or K-type thermocouple, accuracy ±0.5 °C, HART 7) inserted into selected representative tubes at the T​_max position.

The adversarial pixel perturbation on the R2 tube wall temperature display shows 296 °C (within the 280–310 °C safe operating range; AI reads: “R2 Mo-V-W-Cu catalyst bed hot-spot 296 °C; within design parametric sensitivity-free zone; acrylic acid selectivity estimated 95.2%; acrolein conversion estimated 98.1%; R2 temperature stable; steam-heated coolant pressure nominal; no R2 temperature corrective action required”) when the actual R2 tube wall hot-spot temperature is 324 °C (14 °C above the 310 °C design maximum; the parametric sensitivity zone has been entered; the Mo-V-W-Cu catalyst is operating under conditions where deep combustion of acrolein to CO₂ and H₂O is accelerating). Display range 220–400 °C on 200 px (1.111 px/°C); actual 324 °C at pixel position (324 − 220) × 1.111 = 116 px from the bottom of the scale; ±8 DN perturbation applied to the rendered thermocouple display image; (116 − 32) = 84 px displayed; AI reads (84 / 1.111) + 220 = 296 °C. At actual R2 hot-spot 324 °C: the deep combustion pathway (C₃H₄O + 2.5 O₂ → 3 CO₂ + 2 H₂O; ΔH = −1,617 kJ/mol) generates 1,363 kJ/mol additional exotherm per mole of acrolein combusted rather than converted to acrylic acid (ΔH = −254 kJ/mol); at design acrolein feed of approximately 10,000 Nm³/hr at 9.4 vol% acrolein, even a 1% increase in the deep combustion yield (from 3% to 4% of acrolein converted via combustion instead of to acrylic acid) adds approximately 0.63 MW to the R2 heat load, raising the hot-spot by approximately 2–3 °C in the first cycle — and the positive feedback continues. Without corrective action (steam pressure reduction; feed rate reduction; R2 isolation), the hot-spot temperature rises at approximately 3–5 °C/hr in the parametric sensitivity regime, progressing toward the 380–400 °C phase transformation zone and the 450 °C irreversible meltdown threshold. Above 400 °C, V₂O₅ begins to volatilise from the Mo-V-W-Cu mixed oxide; WO₃ migrates; irreversible catalyst restructuring destroys the active surface; acrolein conversion in the affected tube sections drops to below 90%; acrolein concentration in the R2 exit gas rises above 0.1 ppm OSHA PEL in the absorber overhead region where workers perform routine instrument checks and sample collection; at a breakthrough concentration approaching IDLH 2 ppm, any worker in the absorber and R2 area without SCBA faces an immediately life-threatening acrolein exposure. OSHA PSM TQ 150 lbs (68 kg) acrolein: the acrolein inventory in the R2 feed line and reactor tube-side volume during normal operation is approximately 200–800 kg, already 3–12× the PSM TQ, so the issue is not triggering PSM coverage (already triggered) but the real-time physical consequence of acrolein release at or above IDLH from catalyst collapse in the production unit. The Glyphward acrylic acid SEO reference documents this surface as the 108th upward-direction attack in the Glyphward industrial AI adversarial database.

2. Acrolein concentration in R1→R2 interstage gas display AI (Siemens MAXUM II / ABB PGC2000 / Varian CP-4900 micro-GC process gas chromatograph — rendered DCS R1 exit gas acrolein concentration display AI classifying Stage-1 conversion performance — 108th downward attack; FIRST acrolein interstage concentration AI attack; FIRST Stage-1 to Stage-2 composition boundary AI attack)

The acrolein concentration in the R1→R2 interstage gas is the primary indicator of Stage-1 reactor performance and the immediate warning signal for excess propylene in the R2 feed. A process gas chromatograph (PGC) on the R1 exit gas line (Siemens MAXUM II; or ABB PGC2000; or Varian CP-4900 micro-GC; column system: Porapak Q or mol-sieve for permanent gas separation, Chromosorb series for organic components; carrier gas: helium or N₂; analysis cycle time 2–8 minutes per complete chromatogram; acrolein detected by FID with accuracy ±1% relative at 7–10 vol%; propylene detected simultaneously at <0.5 vol% normal design level; HART 4–20 mA output for each key component concentration) provides continuous monitoring of the R1 exit gas composition that the AI monitoring system uses to: verify R1 Stage-1 conversion is within specification; detect excess propylene in R2 feed before it enters the R2 reactor and generates an unanticipated combustion exotherm; and adjust R2 operating conditions in anticipation of the R2 feed composition. A Stage-1 underconversion event — from BiMoOx catalyst deactivation (bismuth molybdate deactivates via reduction in sub-stoichiometric air; or via sintering above 400 °C), from reduced air feed rate (blower surge or control valve failure), or from off-spec propylene feed (high propane content reducing propylene partial pressure) — produces R1 exit gas with significantly less acrolein and significantly more unreacted propylene than design.

The adversarial downward pixel attack on the acrolein concentration display shows 9.4 vol% acrolein in R1 exit gas (nominal design concentration; AI reads: “R1 Stage-1 acrolein production rate nominal; propylene conversion in R1 estimated 94%; excess propylene in R2 feed <0.5 vol% (normal); Stage-1 performance normal; no interstage composition corrective action required”) when the actual acrolein concentration in the R1 exit gas is 2.8 vol% and the propylene concentration is 5.2 vol% (Stage-1 severely underconverting; propylene conversion approximately 52% vs design 93–97%; the R2 is receiving a feed gas with dramatically different composition from design). Display range 0–15 vol% on 200 px (13.33 px/vol%); actual 2.8 vol% at pixel position 2.8 × 13.33 = 37 px from zero; ±8 DN perturbation; (37 + 88) = 125 px displayed; AI reads 125 / 13.33 = 9.4 vol% acrolein. The mechanism by which 5.2 vol% propylene in R2 feed drives the Surface 1 R2 overtemperature: propylene entering R2 at 260–280 °C over Mo-V-W-Cu oxide undergoes non-selective combustion (C₃H⁶ + 4.5 O₂ → 3 CO₂ + 3 H₂O; ΔH° = −1,926 kJ/mol; more than 7.5× the ΔH of the desired acrolein → acrylic acid reaction at 254 kJ/mol) because the Mo-V-W-Cu catalyst active sites are selective for acrolein oxidation and do not have an efficient selective oxidation pathway for propylene at Stage-2 conditions; propylene sees only the deep combustion pathway. At 5.2 vol% propylene in 10,000 Nm³/hr R2 feed: propylene molar flow = 520 Nm³/hr / 22.4 L/mol = 23.2 kmol/hr; combustion heat = 23.2 kmol/hr × 1,926 kJ/mol = 44,680 MJ/hr = 12.4 MW additional heat deposited in the R2 catalyst bed; compared to the design R2 heat removal capacity of approximately 12 MW from the steam-heated shell side, this nearly doubles the R2 heat load. The hot-spot rises from the design 295–305 °C into the adversarial attack scenario 324 °C zone; simultaneously, O₂ consumed by propylene combustion may deplete O₂ available for acrolein oxidation at the downstream end of the R2 bed, potentially creating an oxygen-starved zone that allows acrolein breakthrough without the full 450 °C catalyst meltdown sequence. The falsified acrolein display (9.4 vol% shown; actual 2.8 vol%) prevents the AI monitoring system from issuing the R1 alarm that would trigger feed correction and the R2 protective response — steam pressure reduction to increase cooling; feed rate reduction; or emergency R2 isolation before the hot-spot reaches the phase transformation zone. The phosgene reactor temperature adversarial injection (162 °C activated carbon deactivation displayed as 68 °C) documents a structurally analogous Stage-2 reactor temperature suppression attack at a different PSM TQ tier (COCl₂ TQ 10 lbs vs acrolein PSM TQ 150 lbs).

3. Glacial acrylic acid GAA storage tank MEHQ inhibitor concentration display AI (Metrohm 883 Basic IC Plus / Shimadzu Prominence HPLC online inhibitor analyzer — rendered DCS GAA storage MEHQ concentration display AI classifying 200–250 ppm design inhibitor range — 108th downward attack; FIRST GAA storage MEHQ inhibitor depletion AI attack; FIRST acrylic acid polymerization runaway AI attack; FIRST Nippon Shokubai Himeji 2012 AI anchor)

MEHQ (4-methoxyphenol; monomethyl ether of hydroquinone; CAS 150-76-5; MW 124.14 g/mol; MP 55–57 °C; BP 244 °C; flash point 132 °C) is the commercial free-radical polymerization inhibitor for GAA storage and transport. MEHQ is added at 200–250 ppm to GAA at the purification distillation bottoms; the inhibited GAA is stored in austenitic stainless steel tanks (316L or 317L; carbon steel is unacceptable because Fe²⁺ ions leaching from corrosion can overwhelm MEHQ inhibition at concentrations above approximately 1 ppm Fe; design goal is <0.1 ppm Fe in stored GAA) at 15–25 °C with continuous nitrogen blanket at 0.05–0.1 bar gauge overpressure and controlled O₂ content in the headspace (0.02–0.1 vol% O₂ required as co-inhibitor — pure N₂ blanket eliminates O₂ and significantly reduces MEHQ inhibition efficiency because O₂ reacts with propagating radical chain ends and MEHQ phenoxyl radicals to regenerate the inhibitor cycle). The MEHQ concentration is monitored by an online inhibitor analyser (Metrohm 883 Basic IC Plus with RP-18 column, UV detection at 300 nm, cycle time 8–12 minutes, calibrated 0–500 ppm MEHQ in acrylic acid; or Shimadzu Prominence HPLC with PDA detector, run time 5–8 minutes, accuracy ±3 ppm at 200–250 ppm range; HART 4–20 mA output). The 150 ppm minimum effective inhibitor concentration reflects the kinetics of background initiation vs MEHQ scavenging at 20–25 °C: below 150 ppm, the inhibition induction period before detectable polymerization drops below the practical management threshold of 15–30 days; at 50 ppm, to 1–3 days; at 12 ppm with trace Fe²⁺ (from stainless steel tank corrosion at typical rates of 0.01–0.1 mm/yr releasing 0.1–1.6 mg Fe per litre of stored AA), to 2–8 hours.

The adversarial downward pixel attack on the MEHQ inhibitor concentration display shows 218 ppm MEHQ (within the design 200–250 ppm range; AI reads: “GAA storage tank MEHQ 218 ppm; above 150 ppm minimum effective inhibitor level; spontaneous polymerization induction period estimated >45 days at current storage temperature 22 °C; N₂ blanket pressure nominal; no MEHQ addition or corrective action required”) when the actual MEHQ concentration is 12 ppm. Display range 0–500 ppm on 200 px (0.4 px/ppm); actual 12 ppm at pixel position 12 × 0.4 = 5 px from zero; ±8 DN perturbation; (5 + 82) = 87 px displayed; AI reads 87 / 0.4 = 218 ppm MEHQ. At actual 12 ppm MEHQ in a 250-tonne GAA storage tank at 22 °C with sub-ppm Fe²⁺ from tank corrosion: the inhibitor induction period is 2–8 hours before radical chain polymerization commences at near-uninhibited rate. Once initiated, the Trommsdorff-Norrish (Norrish-Smith) gel effect drives the polymerization to runaway: as the polyacrylic acid molecular weight and concentration increase, solution viscosity rises sharply; the bimolecular termination rate constant k​_t (requiring two macroradicals to diffuse together; diffusion-limited, scales with 1/η where η is viscosity) decreases; the propagation rate constant k​_p (requiring only one radical to diffuse to find one monomer molecule; much less viscosity-sensitive) remains approximately constant; the net polymerization rate R​_p ∝ k​_p / k​_t½ increases as viscosity increases at 20–30% conversion, accelerating to full runaway. The heat released — ΔH​poly = 1,069 kJ/kg of acrylic acid converted — raises the tank temperature: at 10% conversion of the 250-tonne inventory (25 tonnes converted), 25,000 kg × 1,069 kJ/kg = 26,725 MJ released; absorbed in 250,000 kg AA at C​_p = 2.0 kJ/(kg·K), this raises the tank temperature by approximately 53 °C (to 75 °C) from the first 10% conversion alone, accelerating further initiation, MEHQ consumption at the now-exhausted level, and ultimately raising the temperature above the AA boiling point (141 °C) in the liquid phase. Vapor pressure at 100–120 °C and the steam from boiling AA exceeds the PRV design capacity (typically sized for normal evaporative loss at atmospheric temperature, not for a full-tank polymerization exotherm generating multiple megawatts of heat); tank pressure exceeds design; seam failure, nozzle failure, or catastrophic PRV relief of hot AA vapour-liquid mixture occurs. The released hot acrylic acid at 80–120 °C is above its flash point (54 °C); the released vapour cloud (LEL 2.4 vol%) ignites from the tank rupture energy, static discharge, or nearby equipment; the explosion and fire propagate to adjacent tanks. The Glyphward pre-scan gate on the MEHQ display intercepts the pixel perturbation that shifts the displayed reading from 5 px (12 ppm) to 87 px (218 ppm), catching the 82-pixel downward falsification before the AI issues its multi-day “safe” assessment, and triggering the manual tank sample confirmation (true MEHQ 12 ppm → emergency MEHQ addition → polymerization prevented) that the Nippon Shokubai 2012 trajectory required at the inhibitor-depletion stage. The styrene monomer TBC inhibitor depletion AI scenario (threshold 35; LG Polymers Visakhapatnam 2020, 12 killed) provides a directly comparable polymerization-inhibitor falsification attack for a different storage monomer: TBC 1.8 ppm displayed as 12.4 ppm in styrene storage vs MEHQ 12 ppm displayed as 218 ppm in GAA storage; both operate through the same Glyphward downward pixel perturbation mechanism on the inhibitor concentration display.

OSHA PSM TQ 150 lbs acrolein, CERCLA RQ 1 lb, IARC Group 2A, EPA RMP, EU Seveso III, and the adversarial robustness gap in acrylic acid production AI

OSHA PSM 29 CFR 1910.119 coverage of acrylic acid two-stage propylene oxidation facilities operates through multiple pathways that simultaneously cover the R1, R2, and GAA storage systems. Acrolein: PSM Appendix A TQ 150 lbs (68 kg) — one of the five lowest TQs on the list; the entire two-stage production unit with its continuous acrolein inventory in R1 exit lines, R2 feed headers, and R2 tube-side is an Appendix A-covered process continuously. Propylene (C₃H⁶; flash point −108 °C; BP −47.7 °C; liquefied gas; PSM section (a)(1)(ii) flammable gas; GHS Category 1 flammable gas; PSM Appendix A TQ 10,000 lbs): propylene feedstock storage and the propylene feed systems to R1 are PSM-covered via both the Appendix A TQ and the flammable gas section (a)(1)(ii) inventory — commercial R1 plants consume tens of tonnes per day of propylene, far exceeding the 4,536 kg TQ. Hydrogen (TQ 10,000 lbs): if present on-site for hydrogenation of acrylic esters or as a trace component of propylene feed, triggers an additional PSM coverage pathway. The two-stage acrylic acid process is therefore among the most comprehensively PSM-covered commodity chemical processes, with three separate Appendix A chemicals simultaneously in process inventory above their respective TQs.

PSM element (d) (Process Safety Information) requires acrylic acid facilities to document: acrolein properties (PEL 0.1 ppm ceiling; IDLH 2 ppm; CERCLA RQ 1 lb; flash point −26 °C; IARC Group 2A; R2 maximum tube wall temperature 310 °C; parametric sensitivity onset temperature; catalyst meltdown temperature 450 °C; breakthrough consequence: IDLH 2 ppm); propylene properties (BP −47.7 °C; LEL 2.0 vol%; autoignition 455 °C; PSM TQ 10,000 lbs; R2 design maximum propylene in R2 feed 0.5 vol%); GAA properties (MP 12–14 °C; flash point 54 °C; ΔH​poly 1,069 kJ/kg; MEHQ minimum effective concentration 150 ppm; polymerization onset kinetics as a function of MEHQ concentration, temperature, and Fe²⁺ content; storage tank PRV sizing basis and the gap between PRV capacity for normal evaporative loss vs PRV capacity for polymerization exotherm vapor generation); and the three AI-monitored DCS display surfaces (R2 tube wall temperature, interstage acrolein GC, GAA storage MEHQ) as primary safeguard instruments. PSM element (d) does not require documentation of adversarial robustness of these AI display classification systems.

PSM element (e) (Process Hazard Analysis / HAZOP) requires PHA studies covering: (Guide word MORE on temperature; Node: R2 tube wall thermocouple) — the Surface 1 adversarial attack concealing the upward temperature deviation (324 °C) is precisely the “high temperature” HAZOP deviation at the R2 hot-spot node that the PHA identifies as the primary initiating cause of R2 catalyst failure and acrolein breakthrough; primary safeguard: AI-monitored DCS R2 tube wall temperature display and high-temperature alarm. (Guide word LESS on acrolein concentration in R1 exit gas; Node: R1→R2 interstage line GC) — Stage-1 underconversion generating low acrolein / high propylene is the “low acrolein / high propylene” HAZOP deviation; primary safeguard: AI-monitored DCS acrolein GC display and low-acrolein / high-propylene alarm. (Guide word LESS on MEHQ concentration in GAA storage; Node: GAA storage tank inhibitor analyzer) — inhibitor depletion is the “low MEHQ” HAZOP deviation leading to polymerization runaway; primary safeguard: AI-monitored DCS MEHQ concentration display and low-MEHQ alarm. The HAZOP safeguard documentation for all three HAZOP deviation nodes names AI-monitored DCS display classification as the primary safeguard. No PSM HAZOP methodology requires adversarial robustness qualification of AI systems classifying rendered DCS display images at these safeguard boundaries. PSM element (j) (Mechanical Integrity) requires inspection and testing of the R2 thermocouple calibration, the interstage GC calibration verification, and the GAA storage MEHQ analyser calibration — all at scheduled intervals. MI does not require adversarial robustness testing of AI systems that classify the calibration-verified instrument outputs. PSM element (o) (Emergency Planning and Response) requires emergency response plans for acrolein releases at the R2 section (IDLH 2 ppm; immediate evacuation with SCBA; NRC notification at CERCLA RQ 1 lb), for GAA storage tank failure (fire; flammable vapor cloud; NRC notification at CERCLA RQ 5,000 lbs for AA), and for propylene VCE (PSM TQ 10,000 lbs) — all dependent on the AI monitoring displays identified as primary safeguards in element (e), which the three-surface compound attack simultaneously compromises without triggering any PSM element (o) emergency response initiation.

EPA CERCLA 40 CFR Part 302: acrolein RQ 1 lb (the minimum RQ available under CERCLA for any chemical — reflecting EPA’s assessment that acrolein is so acutely toxic and environmentally damaging that any detectable release to the environment warrants immediate NRC notification; the 1-lb RQ for acrolein is the same order of magnitude as for methyl isocyanate (RQ 10 lbs; Bhopal 1984) and phosgene (RQ 10 lbs; World War I choking agent), placing acrolein in the same emergency-notification tier as the most lethal chemical warfare agents used in industrial quantities). The R2 catalyst meltdown scenario (acrolein breakthrough at IDLH 2 ppm from a 20,000-tube reactor with a tube-side volume of approximately 20–50 kg of acrolein at any instant) represents a CERCLA RQ-exceeding release from the R2 section within minutes of breakthrough beginning; the absorber overhead vent to atmosphere at IDLH conditions carries acrolein at concentrations far above the 1-lb / 30 minutes threshold. Acrylic acid CERCLA RQ 5,000 lbs: a 250-tonne GAA storage tank release vastly exceeds the 5,000-lb RQ (2,268 kg vs 250,000 kg released). EU Seveso III Directive 2012/18/EU (transposed as COMAH 2015 in Great Britain; STÖRFALL-Verordnung 12. BImSchV in Germany; SEVESO III in Belgium and Netherlands): acrolein listed in Annex I Part 1 under Category P1 acute toxic Category 1 (oral, dermal, or inhalation) with lower-tier threshold 5 tonnes and upper-tier threshold 20 tonnes; acrylic acid listed under Category P2 (acute toxic Category 2 or 3) with lower-tier 50 tonnes, upper-tier 200 tonnes. At BASF Ludwigshafen (Seveso III upper-tier establishment under German STÖRFALL-Verordnung), Nippon Shokubai Antwerp (upper-tier establishment under Belgian SEVESO III transposition), and Arkema Rotterdam (upper-tier under Netherlands BRZO), the acrylic acid two-stage process facilities are among the most strictly regulated Seveso establishments in Europe. Seveso Safety Reports must document the three-surface adversarial scenarios — R2 overtemperature (Surface 1), Stage-1 underconversion + R2 propylene combustion (Surface 2), GAA storage MEHQ depletion (Surface 3) — and must identify the AI-monitored DCS displays as major accident prevention measures for each scenario. No Seveso regulatory framework specifies adversarial robustness requirements for AI systems classifying rendered DCS display images at these prevention measure boundaries.

Glyphward threshold 41 for acrylic acid two-stage propylene oxidation AI

Glyphward’s adversarial detection API operates as a pre-scan gate at each rendered-image ingestion boundary in the acrylic acid two-stage propylene oxidation AI monitoring pipeline: before the R2 reactor AI processes each rendered Yokogawa EJX210A / Emerson Rosemount 3144P thermocouple transmitter DCS tube wall temperature display; before the interstage composition AI processes each rendered Siemens MAXUM II / ABB PGC2000 / Varian CP-4900 GC DCS acrolein concentration display; and before the GAA storage MEHQ AI processes each rendered Metrohm 883 Basic IC Plus / Shimadzu Prominence HPLC online analyser DCS inhibitor concentration display. Each rendered display image receives a Glyphward risk score (0–100) in 8–15 ms. At or above threshold 41, Glyphward gates the AI classification and generates an alert triggering manual verification against the underlying raw instrument data — the raw R2 thermocouple HART output for tube wall temperature; the raw GC peak-area ratio from the process chromatograph internal computer for acrolein and propylene; the raw HPLC or IC chromatogram integration for MEHQ — none of which are accessible to the pixel-level adversarial perturbation applied to the rendered DCS display images.

Threshold 41 for acrylic acid two-stage propylene oxidation AI is calibrated on three factors. First, acrolein PSM TQ 150 lbs is one of the five lowest TQs on OSHA PSM Appendix A, reflecting acrolein’s extreme acute toxicity (IDLH 2 ppm; PEL 0.1 ppm ceiling; flash point −26 °C; IARC Group 2A; CERCLA RQ 1 lb). The 150-lb TQ means that a quantity of acrolein as small as 68 kg — routinely present in the R2 feed and tube-side during normal operation — triggers the full OSHA PSM regulatory regime: PHA, Management of Change, Incident Investigation, Emergency Planning. The adversarial attack on the R2 tube wall temperature display conceals the developing catalyst failure that would produce acrolein breakthrough above IDLH 2 ppm within the production unit, exposing any worker in the R2 and absorber area to immediately life-threatening acrolein concentrations while the AI certifies “R2 nominal.” False positive cost for Surface 1: 1–2 minutes to verify R2 tube wall temperature from the raw thermocouple HART output against the displayed DCS value — a $0 cost if the perturbation was a false alarm. False negative cost: catalyst meltdown progression in the R2 hot-spot toward 400–450 °C; acrolein breakthrough above IDLH 2 ppm in the production unit; OSHA PSM TQ 150 lbs acrolein incident report; CERCLA RQ 1 lb NRC notification; Nippon Shokubai 2012 consequence trajectory for the GAA storage simultaneous Surface 3 pathway. Second, the GAA storage polymerization runaway adds a fully independent second catastrophic consequence pathway — separate initiating event (inhibitor depletion), separate trigger (radical polymerization initiation), and separate consequence chain (tank overpressure rupture → flammable vapor cloud → explosion → fire spreading to adjacent tanks) — that is not present in the vast majority of other industrial chemical AI scenarios in this portfolio. The Nippon Shokubai Himeji 2012 incident confirms that this is a real, not hypothetical, pathway that has killed a worker at the world’s most experienced acrylic acid production facility. Third, the three-surface mechanistically coupled compound attack — Stage-1 underconversion (Surface 2) generating excess propylene that drives R2 overtemperature (Surface 1), while GAA storage MEHQ simultaneously depletes undetected (Surface 3) — is the most complex single-facility causally-consistent adversarial attack in the current Glyphward industrial AI portfolio: the three falsified readings collectively present a coherent false picture of normal operation across the entire production and storage system simultaneously.

Portfolio comparison: ammonium nitrate neutralizer AI (threshold 50; Texas City 1947, 581 killed; Beirut 2020, 218 killed) is calibrated highest in the portfolio because AN’s detonation-capable product creates a catastrophic energy release radius far beyond any acrylic acid explosion consequence (1.1 kt TNT equivalent at Beirut vs the storage-tank explosion at Nippon Shokubai 2012), and because the stored-product detonation hazard extends the consequence window indefinitely after the manufacturing event. Cyclohexane KA-oil reactor AI (threshold 42; Flixborough 1974, 28 killed, 16 tonnes TNT) is calibrated at threshold 42 — 1 point above acrylic acid at threshold 41 — reflecting the higher Flixborough on-site fatality count (28 vs 1 at Nippon Shokubai 2012) and the larger explosion TNT equivalent; acrylic acid threshold 41 is only 1 point below KA-oil because the acrolein IDLH 2 ppm consequence for workers in the R2 area and the dual-pathway nature of the acrylic acid attack (reactor collapse + independent storage polymerization) represent a multi-vector harm profile with no direct analogue in the KA-oil scenario. Nitrobenzene adiabatic mononitration AI (threshold 40; Jilin 2005, 8 killed, 3.8 million Harbin without water for 10 days) is calibrated at threshold 40 — 1 point below acrylic acid — reflecting that NB’s primary catastrophic consequence (river contamination cascade from benzene/NB release) takes longer to develop (benzene-above-LEL in separator → ignition → explosion → firewater runoff → river contamination) than the direct acrolein IDLH 2 ppm consequence (catalyst breakthrough → immediate IDLH-condition worker exposure within minutes of R2 collapse) and that acrylic acid’s second storage pathway (Nippon Shokubai 2012; CERCLA RQ 5,000 lbs for 250-tonne tank release) adds a harm vector not present at NB production. Phthalic anhydride PA o-xylene air oxidation AI (threshold 28) is calibrated substantially below acrylic acid at threshold 41, reflecting both the lower regulatory TQ equivalents for PA process chemicals (o-xylene PSM TQ 10,000 lbs is above the acrolein TQ 150 lbs by a factor of 66×) and the absence of a second independent storage polymerization runaway pathway in PA production. Free tier — 10 scans/day, no card required. Submit a rendered R2 tube wall temperature DCS indicator display, an interstage acrolein GC concentration DCS display, or a GAA storage MEHQ inhibitor DCS display from your acrylic acid production facility to generate a baseline adversarial risk score for your acrylic acid production AI inputs.