Nitrobenzene adiabatic mononitration AI adversarial injection: how ±8 DN in the rendered crude NB cooler cooling-water flow display recreates the Jilin 2005 Songhua River pathway — and why OSHA PEL 1 ppm skin + CERCLA RQ 1,000 lbs has no adversarial robustness criterion for NB production AI

Nitrobenzene adiabatic mononitration chemistry: electrophilic aromatic substitution, the NORAM process, DNB/TNB over-nitration cascade, and OSHA PEL 1 ppm methemoglobin regulatory framework

Nitrobenzene (C₆H₅NO₂; MW 123.11 g/mol; BP 211°C at 1 atm; MP 5.7°C; density 1.204 g/mL at 20°C; vapour pressure 0.3 mmHg at 20°C, corresponding to a saturation concentration of approximately 395 ppm at 20°C; flash point 88°C; autoignition temperature 482°C) is produced industrially by electrophilic aromatic substitution (EAS) nitration of benzene with nitronium ion (NO₂⁺): C₆H₆ + HNO₃ → C₆H₅NO₂ + H₂O (ΔH = −117 kJ/mol; exothermic). The electrophilic agent is nitronium ion generated in mixed acid via: HNO₃ + 2H₂SO₄ → NO₂⁺ + H₃O⁺ + 2HSO₄⁻. Mixed acid composition in the NORAM adiabatic mononitration process: 65–68 wt% H₂SO₄ / 28–32 wt% HNO₃ / 4–8 wt% H₂O. The NORAM Engineering adiabatic mononitration process (developed by NORAM Engineering and Constructors Ltd, Vancouver BC; first commercial installation 1990s; now the most widely deployed modern NB process, operating at 50,000–300,000 t/yr per train): benzene and mixed acid are fed at 50–55°C to an adiabatic tubular reactor (plug flow or multi-stage reactive zones); the exothermic nitration reaction generates an adiabatic temperature rise of approximately 55–65°C, terminating at 105–120°C at the reactor outlet; the crude product mixture (organic phase: crude NB containing 90–95 wt% NB, 3–5 wt% unreacted benzene, <0.05 wt% DNB; aqueous phase: spent acid H₂SO₄ 65–75 wt%) is then separated by gravity in a phase separator at 40–50°C; spent acid is recycled to the acid recovery unit (ARU) where it is concentrated from ~70 wt% back to 96–98 wt% H₂SO₄ in a vacuum falling-film concentrator and re-fortified with fresh HNO₃ and H₂SO₄ for return to mixed acid preparation. Global NB production is approximately 2.5 million t/yr (2024); >95% is feedstock for aniline production via catalytic vapour-phase hydrogenation (NB + 3H₂ → aniline + 2H₂O) over supported Cu/Pd/Ni catalysts; aniline feeds methylene diphenyl diisocyanate (MDI) production via phosgenation, providing the isocyanate half of polyurethane chemistry.

Nitration selectivity control is the central process safety and quality challenge in adiabatic mononitration. At the design HNO₃:benzene molar ratio (HNO₃ limiting; benzene in slight excess, typically 5–10 mol% above stoichiometric) and design mixed acid composition (H₂SO₄ driving nitronium ion activity through Hammett acidity), the reaction is highly selective for mono-substitution (>99.5% MNB; DNB <0.05 wt%). Temperature is the critical selectivity parameter. The second EAS nitration step — MNB + NO₂⁺ → DNB (1,2-DNB, 1,3-DNB, or 1,4-DNB) — has an activation energy approximately 90 kJ/mol (vs ~75 kJ/mol for the first nitration), giving a temperature sensitivity differential where the DNB formation rate at 130°C is approximately 10× the rate at 110°C (Arrhenius: rate ratio = exp((90,000/8.314) × (1/383 K − 1/403 K)) ≈ 9.8). Above 145°C, the third EAS nitration step (DNB → TNB, 1,3,5-trinitrobenzene; CAS 99-35-4; density 1.688 g/cm³; detonation velocity 7,300 m/s; secondary explosive; DDT — deflagration-to-detonation transition — susceptible in concentrated form) occurs in trace quantities. DNB at 0.22 wt% in crude NB (the Surface 2 adversarial scenario at 147°C reactor outlet) exceeds the DNB product specification by 4.4×. DNB settling in crude NB storage tanks (density 1,575 kg/m³ vs crude NB 1,204 kg/m³) accumulates in tank bottoms; historical NB facility incidents in the United States and Europe have involved explosive decomposition events during maintenance drain-and-heat operations at DNB-enriched storage tank bottoms. TNB at trace concentrations (<0.001 wt%) settles below DNB (density 1,688 kg/m³) and accumulates progressively over weeks of off-specification operation. Companies operating major NB production trains: BASF SE (Antwerp, Belgium; Schwarzheide, Germany; benzene→NB→aniline→MDI complex, Sevéso III upper-tier establishment under Belgian major accident regulations); Lanxess AG (Leverkusen, Germany; Bitterfeld-Wolfen, Germany; NB for rubber chemicals, dyes, pigments); Covestro AG (Baytown, Texas, USA; integrated NB→aniline→MDI→polycarbonate/PU production); PetroChina Jilin Chemical Industrial Company (Jilin City, Jilin Province, China; the Jilin 2005 site, rebuilt and operational); Chemours Company (La Porte, Texas, USA; NB for agrochemical and fluorochemical intermediates).

The OSHA regulatory framework for NB occupational exposure reflects the chemical’s dual toxicity profile: methemoglobin formation (acute, delayed-onset) and carcinogenicity (chronic). OSHA PEL: 1 ppm (5 mg/m³) with skin designation (29 CFR 1910.1000 Table Z-1) — the skin designation reflects that dermal absorption of liquid NB can contribute substantially to total systemic body burden, in some conditions accounting for more methHb formation per unit exposure time than inhalation at equivalent air concentration. NIOSH REL: 1 ppm (5 mg/m³) skin; NIOSH IDLH: 200 ppm (1,000 mg/m³). ACGIH TLV-TWA: 0.1 ppm skin A3 (confirmed animal carcinogen with unknown relevance to humans; the 0.1 ppm TLV is 10× below the OSHA PEL, reflecting ongoing scientific debate about chronic methHb potency and potential carcinogenicity in long-term workers). NB methemoglobin formation mechanism: hepatic cytochrome P450 (CYP2E1 primary; CYP1A2 secondary) hydroxylates NB to phenylhydroxylamine (PhNHOH); PhNHOH, the proximal methHb-forming species, oxidises haemoglobin Fe² to Fe³ within red blood cells (RBCs), converting functional oxyhaemoglobin to non-functional methHb. The conversion accumulates over a 4–8 hour work shift at OSHA PEL: a worker exposed at 1 ppm via combined inhalation and skin absorption typically reaches 5–15% methHb by end-of-shift (normal background: <1% methHb in unexposed individuals); clinical symptoms (grey-blue cyanosis, particularly lips and fingertips; “chocolate-brown blood” visible on blood draw) appear at approximately 20–30% methHb; severe hypoxia, tachycardia, and altered consciousness at 30–50% methHb; life-threatening above 50% methHb; potentially fatal above 70% methHb.

EPA CERCLA 40 CFR Part 302 Table 302.4 lists NB with a reportable quantity (RQ) of 1,000 lbs (454 kg): any release of NB to the environment at or above 1,000 lbs requires immediate notification to the National Response Center (NRC; 1-800-424-8802) and may trigger CERCLA response authority. A single phase separator inventory release (crude NB: ~1,200 kg/m³; a 2,000-litre separator hold-up = ~2,400 kg crude NB containing ~85% NB = ~2,040 kg NB = ~4,495 lbs NB) exceeds the CERCLA NB RQ by approximately 4.5× in NB content alone, and exceeds the benzene CERCLA RQ (10 lbs) by approximately 5,000× via the benzene content of the crude NB stream (3–5 wt%). The Jilin 2005 release of ~100 tonnes NB to the Songhua River exceeded the CERCLA NB RQ by approximately 220,000× — a release magnitude placing NB production in the highest-consequence tier for CERCLA notification obligations. EPA RMP 40 CFR Part 68 lists benzene as a flammable substance with TQ 10,000 lbs (4,536 kg): NB facilities consuming benzene feedstock at commercial rates (100,000 t/yr NB requires ~65,000 t/yr benzene) trigger RMP worst-case release scenario analysis obligations. EU Seveso III Directive 2012/18/EU Annex I Part 1 lists NB in Category P1 (acute toxic, Category 1/2) with lower-tier threshold 5 tonnes, upper-tier threshold 20 tonnes — thresholds reached within the liquid inventory of a single crude NB phase separator vessel at any commercial NB facility.

Jilin Chemical Plant, 13 November 2005: the PetroChina Songhua River contamination and the NB production consequence anchor

The Jilin Chemical Industrial Company (Jilin Petrochemical) facility at Jilin City, Jilin Province, China — a PetroChina subsidiary and one of China’s largest integrated chemical complexes, producing nitrobenzene, aniline, dyes, rubber chemicals, and petrochemical derivatives from its location on the southern bank of the Second Songhua River — experienced a catastrophic explosion in its nitrobenzene/aniline production unit at approximately 13:47 local time on Sunday 13 November 2005. The explosion occurred in the No. 101 chemical plant within the Jilin Chemical complex, specifically in the section handling benzene nitration and the subsequent aniline hydrogenation step. Eight workers were killed; more than 70 were injured. The explosion and resulting fire destroyed multiple production units and their containment infrastructure; firewater and process chemical runoff combined to create a drainage stream that bypassed available containment and discharged to the Second Songhua River through the facility’s existing drainage network. The root cause of the explosion — as investigated by the State Administration of Work Safety (SAWS) and the State Environmental Protection Administration (SEPA, the predecessor to China’s Ministry of Environmental Protection/MEP and current Ministry of Ecology and Environment/MEE) — included a chemical reaction upset in the benzene nitration section in which inadequate cooling and temperature control led to an exothermic runaway in the nitration unit, initiating the explosion sequence. The process safety failure at Jilin in 2005 — inadequate thermal monitoring and cooling control in the NB nitration section — is precisely the failure mode that the Surface 1 and Surface 2 adversarial injection attacks are designed to recreate in AI-monitored NB plants: Surface 1 suppresses the cooling water deficiency that prevents the crude NB from reaching its safe handling temperature; Surface 2 suppresses the reactor outlet temperature excursion that indicates over-nitration is occurring.

The environmental consequence of the Jilin explosion was disproportionate to the on-site fatality count because of the facility’s location directly adjacent to the Second Songhua River — a major tributary of the Songhua River (Sungari River; Heilongjiang’s principal tributary, one of China’s seven major river systems). Emergency responders applied an estimated 10,000–15,000 tonnes of firewater over approximately 48 hours to suppress the explosion-triggered fires; the firewater, contaminated with NB, benzene, aniline, and other organic chemicals from the production facility, flowed through the factory drainage network into the Second Songhua River. Environmental monitoring conducted by SEPA and Jilin Province environmental authorities in the days following the explosion detected a contamination plume in the Second Songhua and, downstream, in the main Songhua River extending approximately 80 kilometres in length on 16 November (three days after the explosion) and expanding to approximately 380 kilometres by 20–21 November as the plume moved downstream at the river current velocity (approximately 2–3 km/h under November low-flow conditions). The contamination plume contained benzene at concentrations peaking at approximately 50–100× the Chinese drinking-water standard (GB 5749-2006 benzene limit: 10 μg/L; peak plume concentrations estimated at 500–1,000 μg/L in the Songhua main channel near Jilin City in the first days post-explosion) and nitrobenzene at concentrations significantly above the environmental standard (Chinese surface water Category III standard NB: 0.017 mg/L; peak plume concentrations estimated at 0.1–0.5 mg/L). Total estimated release to the river: approximately 100 tonnes of benzene and nitrobenzene combined, based on mass balance of production unit inventory, recovered contaminated firewater, and river concentration measurements integrated over the plume.

Harbin (Haerbin; capital of Heilongjiang Province; population approximately 3.8 million in the central urban area in 2005; located approximately 380 km downstream from Jilin City on the main Songhua River) began receiving advance warning of the contamination plume on approximately 18–19 November 2005, as SEPA and Jilin Province authorities disclosed (after initially stating there was no environmental impact) that a chemical contamination plume was moving toward the city. The Harbin Municipal Water Authority shut down the city’s entire surface water intake from the Songhua River at approximately noon on 21 November 2005 (some sources indicate the official outage began 22 November), with municipal tap water supply cut to the city for approximately 10 days (12–22 November is the commonly cited outage window in Chinese regulatory documentation). During the outage: bottled water was flown into Harbin by Chinese government and military airlift; emergency water distribution points were established throughout the city; hospitals and industrial users relying on water supply were rationed; social disruption from the sudden 10-day loss of water to a city of 3.8 million in sub-zero November temperatures was significant. The contamination plume continued downstream after passing Harbin, eventually crossing the Chinese-Russian border via the Amur River (Heilongjiang) at Khabarovsk, Russia, in late November 2005. Russian federal environmental authorities (then Rosprirodnadzor) activated emergency monitoring at Khabarovsk water intakes; Russian and Chinese governments exchanged diplomatic communications regarding the cross-border contamination — the Jilin 2005 event becoming the first major cross-border river contamination incident involving China and Russia in the post-Soviet period. The event precipitated significant changes to Chinese environmental emergency response law: the Emergency Response Law of the People’s Republic of China (2007) and the Regulation on Reporting, Investigation and Handling of Environmental Pollution Accidents (2006) were both significantly informed by the Jilin 2005 disclosure failure and the inadequacy of emergency containment at chemical facilities adjacent to major rivers.

The regulatory lesson of Jilin 2005 for NB production process safety: the two primary process control failures that initiated the explosion sequence — inadequate cooling in the crude NB handling section and temperature excursion in the nitration section — were failures of the monitoring and control systems that are now the surfaces targeted by Surface 1 (CW flow suppression) and Surface 2 (reactor outlet temperature suppression) in the adversarial injection scenario. In 2005, these failures involved human operator error and mechanical deficiencies in conventional instrumentation. In 2026, at AI-monitored NB plants operating SCADA/DCS-integrated AI display classification, the Surface 1 and Surface 2 adversarial pixel attacks replicate the same monitoring failure by different means — the AI receives falsified display images and classifies the process as nominal while the identical thermal and chemical hazard development proceeds undetected. The Songhua River consequence anchor remains fully applicable: modern NB/aniline plants operated by BASF (Scheldt River, Antwerp), Covestro (Houston Ship Channel, Baytown), and Lanxess (Rhine, Leverkusen) are each sited on or adjacent to major commercial waterways whose drinking water implications for downstream populations are directly comparable to the Songhua›Harbin relationship.

Three adversarial injection surfaces in nitrobenzene adiabatic mononitration AI

1. Crude NB cooler cooling-water volumetric flow display AI (Emerson Fisher FIELDVUE DVC6200 / Honeywell ST800 / Yokogawa EJX110 DP transmitter rendered DCS display AI — 102nd upward-direction attack; FIRST NB adiabatic mononitration AI attack; FIRST Jilin 2005 Songhua River AI anchor)

The crude nitrobenzene cooler cooling-water volumetric flow rate is the primary temperature-control variable governing safe phase separation in NORAM adiabatic mononitration. After leaving the adiabatic reactor at 105–120°C, the crude product mixture (NB organic phase + spent acid aqueous phase; combined density ~1.35 g/mL; roughly 55 vol% NB organic / 45 vol% spent acid) enters a shell-and-tube heat exchanger (crude NB/spent acid cooler; typically two parallel exchangers; cooling water on shell side at 25–30°C supply) that reduces the mixture temperature from ~120°C to approximately 40–50°C before gravity phase separation. At 40–50°C, the organic phase (crude NB: benzene 3–5 wt% + NB 90–95 wt% + trace DNB) cleanly separates from the spent acid aqueous phase (H₂SO₄ 65–75 wt% + H₂O) by density difference in the settling vessel. Cooling water flow to the crude NB cooler is measured by a differential pressure transmitter (Honeywell ST800 or Yokogawa EJX110 DP transmitter across an orifice plate; 4–20 mA HART output; calibrated 0–80 m³/h; displayed on DCS as m³/h). Design CW flow: 50–60 m³/h at 25°C supply / 35°C maximum return; the design flow provides approximately 2,090 kW of cooling duty (50 m³/h × 1,000 kg/m³ × 4.18 kJ/(kg·K) × 10 K = 2,090 kW) sufficient to cool the crude product mixture from 120°C to 45°C. At 8.4 m³/h actual cooling water flow (severely deficient; consistent with CW pump cavitation, supply strainer blockage, or control valve sticking in partial-close position), the heat removal drops to 2,090 × (8.4/50) = 351 kW — only 17% of design; the crude NB/spent acid exits the cooler at 88°C and enters the phase separator at 88°C.

At 88°C in the phase separator: benzene vapour pressure (Antoine equation; benzene BP 80.1°C; Antoine B = 1196.76, C = 219.161 at T = 88°C) is approximately 0.27 bar (203 mmHg; 270 kPa). In the nitrogen-blanketed separator headspace (operating at approximately 1.05 bar absolute), the equilibrium benzene vapour concentration reaches 0.27/1.05 = 25.7 mol% in the vapour-liquid equilibrium at the interface — significantly above benzene LEL 1.4 vol% and approaching benzene UEL 8.0 vol% in the bulk headspace gas (diluted by nitrogen blanket and sensor dead-zones; realistic bulk headspace benzene concentrations: 5–8 vol%). This is the same precursor condition that characterised the Jilin Chemical Plant explosion of 13 November 2005: a hot crude NB/spent acid mixture in the phase separator vessel with benzene vapour above LEL, plus potential ignition sources from instrument maintenance sparking, electrostatic discharge from liquid surface agitation, or pressure relief chattering. The adversarial upward pixel attack on the crude NB cooler CW flow display shows 52.6 m³/h (within the 50–60 m³/h design range; the AI classifies: “cooling water flow adequate; crude NB cooler operating at design duty; separator inlet temperature nominal ~45°C; benzene vapour in separator headspace estimated 0.4 vol% — well below LEL 1.4 vol%; no action required”) when actual CW flow is 8.4 m³/h. Display: 0–80 m³/h on 200-pixel vertical bar (2.5 px per m³/h); actual 8.4 m³/h at pixel position 21 px from zero; ±8 DN upward perturbation shifts display to 132 px → AI reads 52.6 m³/h. The adversarial attack extends the window during which the benzene-above-LEL separator condition persists undetected: without the attack, the AI detects 8.4 m³/h within one scan cycle (30–60 seconds) and alarms; with the attack, it scans the falsified 52.6 m³/h indefinitely. Any ignition source in the separator area during this extended window — a spark from separator level transmitter maintenance, static discharge from the liquid surface turbulence at the NB/spent acid interface, or a PRV cycling event generating heat — initiates the Jilin 2005 consequence trajectory: separator vessel deflagration, crude NB containment breach, firewater runoff pathway to the adjacent waterway. The Glyphward NB production SEO reference documents this surface as the 102nd upward-direction attack in the Glyphward industrial AI adversarial database.

2. Adiabatic reactor outlet temperature display AI (Emerson 3144P / Honeywell STT850 / Yokogawa EJA210E thermocouple transmitter rendered DCS temperature display AI — downward attack; FIRST NB adiabatic over-temperature AI attack; FIRST DNB/TNB formation AI attack)

The adiabatic reactor outlet temperature in NORAM mononitration is the primary selectivity indicator. The adiabatic temperature rise from reactor inlet (50–55°C mixed acid + benzene) to outlet is driven entirely by the exothermic nitration reaction (ΔH = −117 kJ/mol NB); at design conditions (HNO₃ limiting, 65–68 wt% H₂SO₄ acid composition, design benzene excess) the adiabatic end-point is 105–120°C with MNB >99.5% and DNB <0.05 wt%. The outlet thermocouple transmitter (Emerson 3144P RTD or Honeywell STT850 multi-point array; 4–20 mA HART; PT-100 or Type-K; calibrated 0–200°C; displayed on DCS as °C) provides the real-time signal that the AI monitoring system classifies against the 125°C design maximum: above 125°C, AI is expected to generate a high-temperature alarm and initiate HNO₃ feed reduction; above 130°C, a high-high alarm triggers emergency acid dilution or reactor isolation. At actual 147°C reactor outlet temperature: DNB formation rate at 147°C vs 115°C = exp((90,000/8.314) × (1/388 K − 1/420 K)) ≈ 50× the design DNB formation rate — crude NB DNB content reaching approximately 0.22 wt% (4.4× the 0.05 wt% product specification). Above 145°C: TNB formation in trace quantities (<0.001 wt% in the crude NB stream; settling in storage as density-stratified explosive layer over extended operation). The adversarial downward pixel attack: display 0–200°C on 200 px (1.0 px/°C); actual 147°C at 147 px; ±8 DN downward shift moves apparent thermocouple trace to 112 px; AI reads 112°C (13°C below the 125°C design maximum; “nitration proceeding within design selectivity window; MNB >99.5%; DNB <0.02 wt% estimated; no corrective action required”). At actual 147°C, the crude NB stream entering the phase separator contains 0.22 wt% DNB and trace TNB — neither detectable from the falsified temperature display. The DNB consequence chain: DNB in crude NB carries through the phase separator wash section and to intermediate storage; in storage (temperature 20–40°C), DNB (density 1.575 g/cm³; higher than crude NB 1.204 g/cm³) settles to tank bottoms; over weeks of off-spec operation, DNB-rich bottoms accumulate; maintenance events requiring heating of tank bottoms (steam-jacketed drain valves, heat-traced sump pumps) can bring DNB-rich liquor to temperatures approaching DNB’s thermal decomposition onset; historical NB storage incidents in the US and Europe (not publicly attributed in OSHA/EPA enforcement databases, but documented in CCPS Guidelines for Chemical Reactivity Evaluation and AIChE DIERS literature) have involved DNB-initiated explosive decomposition events during such maintenance operations. TNB settles below DNB (density 1.688 g/cm³); even at <15 wt% TNB (the approximate detonation-sensitivity threshold for dry TNB), accumulated TNB in the concentrated hot H₂SO₄ ARU concentrator sump — created by the Surface 3 spent acid HNO₃ overcarryover scenario — presents a deferred thermal decomposition risk. The phosgene reactor temperature adversarial injection scenario — 162°C activated carbon deactivation displayed as 68°C — provides a directly analogous downward temperature suppression attack at a different PSM TQ tier (COCl₂ TQ 10 lbs vs NB production’s benzene flammable TQ 10,000 lbs).

3. Spent acid recycle HNO₃ concentration display AI (Endress+Hauser Liquiline CM442 / ABB AO2000 / Emerson Rosemount inline NIR/Raman spectrometer rendered DCS composition display AI — downward attack; FIRST spent acid HNO₃ carryover AI attack; FIRST ARU dinitration AI attack)

The spent acid HNO₃ concentration at the NORAM phase separator outlet (before recycle to the ARU) is the critical quality indicator confirming whether the adiabatic nitration reaction was complete (all HNO₃ consumed in benzene nitration to MNB) or incomplete (residual HNO₃ remaining in the spent acid aqueous phase due to insufficient acid residence time, lower-than-design nitronium ion activity, or temperature excursion altering the kinetics). The spent acid composition (H₂SO₄ 65–75 wt% + H₂O 20–30 wt% + trace HNO₃) is measured by an inline NIR spectrometer (Endress+Hauser Liquiline CM442 with optical probe; or ABB AO2000 Uras26 with extractive sampling; or Raman spectrometer calibrated to the 1,048 cm⁻¹ HNO₃ symmetric stretch; 4–20 mA HART output; calibrated 0–10 wt% HNO₃; multivariate PLS calibration; updated every 15–30 seconds). The 0.5 wt% HNO₃ maximum carryover limit is set by two criteria: (a) in the ARU falling-film concentrator at 120–140°C and reduced pressure (150–200 mbar absolute), the Hammett acidity function H₀ of 70–75 wt% H₂SO₄ at 130°C is approximately −9 to −10 — sufficient nitronium ion activity for EAS nitration of any aromatic trace in the feed; HNO₃ above 0.5 wt% in the ARU feed creates an active dinitrating environment for residual organics (dissolved NB at ~200 ppm equilibrium in spent acid); (b) trace dissolved NB can be dinitrated to DNB in the ARU concentrator sump, accumulating as dense DNB-rich sediment at 140–150°C concentrator bottoms temperature — the same DNB thermal decomposition risk described under Surface 2. The adversarial downward pixel attack: display 0–10 wt% HNO₃ on 200 px (20 px/wt%); actual 4.8 wt% at 96 px from zero; ±8 DN downward adversarial shift moves apparent spectrometer indication to 6 px; AI reads 0.3 wt% (below 0.5 wt% limit; “spent acid HNO₃ within ARU feed specification; nitration complete; no corrective action required”) when actual HNO₃ is 4.8 wt% (9.6× the carryover limit). At 4.8 wt% HNO₃ entering the ARU concentrator at 120–140°C: each tonne of spent acid carries 48 kg HNO₃ to the ARU; at 100,000 t/yr NB production (typical train size) requiring ~120,000 t/yr spent acid, the HNO₃ carryover at 4.8 wt% amounts to 5,760 t/yr HNO₃ arriving at the ARU unconsumed — nitrating acid that was not consumed in the reactor, indicating a fundamental process deviation. Additionally: the combination of Surface 2 (reactor temperature 147°C causing over-nitration) and Surface 3 (spent acid HNO₃ 4.8 wt% due to incomplete conversion at the ARU concentrator recycle) is causally self-reinforcing: the same reactor temperature excursion that accelerates DNB formation (Surface 2) also tends to reduce the nitration selectivity and increase HNO₃ slippage to the spent acid (because at 147°C the DNB side-reaction, consuming NB rather than benzene, changes the benzene:HNO₃ consumption ratio, leaving more HNO₃ in excess). The three-surface compound creates a causally consistent over-nitration scenario that the AI monitoring system classifies as normal operation at all three surfaces simultaneously. Sulphuric acid DCDA process AI — SO₃ converter temperature downward attack and acid strength upward attack — provides a structurally parallel acid-plant composition/temperature compound attack documented in the Glyphward adversarial database.

OSHA PEL 1 ppm skin, EPA CERCLA RQ 1,000 lbs, PSM coverage via benzene TQ 10,000 lbs, and the adversarial robustness gap in nitrobenzene production AI

OSHA PSM 29 CFR 1910.119 coverage of nitrobenzene production facilities operates through two Appendix A and section (a)(1)(ii) pathways. First, benzene (C₆H₆; flash point −11°C; GHS Category 1 flammable liquid; OSHA 29 CFR 1910.1200(c) classification) is the primary NB feedstock: at commercial NB plant scale (50,000–300,000 t/yr per train), benzene feedstock is stored and processed in quantities of thousands of tonnes — far exceeding the PSM section (a)(1)(ii) 10,000 lb (4,536 kg) flammable liquid inventory threshold. Benzene also appears in EPA RMP 40 CFR Part 68 as a flammable substance TQ 10,000 lbs, triggering RMP worst-case scenario analysis at all commercial NB feedstock storage installations. Second, nitric acid (HNO₃; 70 wt%; density 1.41 g/mL) is listed in OSHA PSM Appendix A with a TQ of 10,000 lbs (4,536 kg; concentration ≥80 wt%); the higher-concentration (98 wt%) fuming nitric acid used in some mixed acid formulations may reach PSM Appendix A TQ in on-site storage. Mixed acid inventories at commercial NB plants (containing 28–32 wt% HNO₃; total mixed acid inventory typically 50–200 tonnes per plant) are addressed by PSM section (a)(1)(ii) through the benzene feedstock pathway regardless of acid concentration. PSM element (d) (Process Safety Information) requires NB facilities to document: benzene properties (flash point −11°C; LEL 1.2%; UEL 8.0%; autoignition 560°C; vapour pressure 100 mmHg at 26°C; IDLH 500 ppm); NB properties (PEL 1 ppm skin; IDLH 200 ppm; CERCLA RQ 1,000 lbs; methHb formation mechanism; delayed-onset cyanosis timeline; 4–8 hr shift before obvious symptoms); reactor design operating temperature (50–55°C inlet; 105–120°C outlet design maximum), pressure (atmospheric to 1.2 bar), HNO₃:benzene molar ratio, and the DNB onset temperature (130°C) and TNB onset temperature (145°C); crude NB cooler design CW flow and minimum alarm setpoint; spent acid maximum HNO₃ carryover limit (0.5 wt%) and ARU concentrator thermal stability model for DNB accumulation in sump.

PSM element (e) (Process Hazard Analysis) requires PHA-HAZOP studies of NB plants covering: (Guide word MORE on cooling water flow; Node: crude NB cooler CW supply) — the Surface 1 downward deviation (LESS cooling water flow) is the primary HAZOP node for separator temperature excursion; primary safeguard: AI-monitored DCS cooling water flow display and CW low-flow alarm. (Guide word MORE on reactor outlet temperature; Node: adiabatic reactor outlet) — the Surface 2 upward deviation is the primary HAZOP node for DNB/TNB over-nitration; primary safeguard: AI-monitored DCS reactor outlet temperature display and high-temperature alarm. (Guide word MORE on HNO₃ in spent acid; Node: spent acid recycle) — the Surface 3 upward deviation is the primary HAZOP node for ARU dinitrating conditions; primary safeguard: AI-monitored inline NIR/Raman spectrometer DCS display and HNO₃ high-concentration alarm. PSM element (e) identifies the AI-monitored DCS displays for all three Surface scenarios as the primary safeguards in the HAZOP safeguard documentation. PSM element (e) does not specify adversarial robustness requirements for the AI systems classifying those displays. PSM element (j) (Mechanical Integrity) requires inspection and testing of the crude NB cooler (heat exchanger tube integrity, CW flow transmitter calibration verification, CW control valve function test), the adiabatic reactor thermocouple calibration verification, and the NIR/Raman spectrometer HNO₃ calibration verification — all at scheduled intervals. Adversarial pixel robustness of the AI that classifies the calibration-verified instrument outputs is not addressed by Mechanical Integrity. PSM element (o) (Emergency Planning and Response) requires emergency response plans for NB releases and for the Jilin-type scenario of a large-volume NB/benzene release to adjacent waterways, including NRC notification upon CERCLA RQ exceedance; the emergency response plan depends on the monitoring systems identified as primary safeguards in element (e), which the three-surface adversarial attack simultaneously compromises.

EU Seveso III Directive 2012/18/EU (transposed as COMAH 2015 in Great Britain; as STÖRFALL-Verordnung 12. BImSchV in Germany; as Arrêté du 26 mai 2014 in France) lists NB in Annex I Part 1 under Category P1 (acute toxic Category 1 by swallowing — NB LD₆₀ oral 349 mg/kg rat; not Category 1 by acute toxicity criteria but covered under P1 via combined toxicity/environmental group criteria) with quantities 5 tonnes (lower-tier) and 20 tonnes (upper-tier). Every commercial NB plant holds tens to hundreds of tonnes of NB in process inventory — above the 20-tonne upper-tier Seveso threshold. Seveso Safety Reports at NB facilities must identify the three-surface adversarial scenarios — cooling water failure (Surface 1), reactor temperature excursion (Surface 2), spent acid HNO₃ carryover (Surface 3) — as major accident prevention measures and must document the AI-monitored DCS displays as primary risk reduction measures for these scenarios. Seveso Safety Reports under COMAH 2015 / 12. BImSchV do not specify adversarial robustness requirements for the AI systems now classifying rendered DCS display images at the NB production monitoring boundary. The REACH Regulation (EC) No 1907/2006 requires NB registrants (BASF, Lanxess, Covestro, Chemours) to maintain Chemical Safety Reports (CSR) documenting NB’s methemoglobin-forming potency, skin absorption DNEL, and environmental PNEC — all relevant to the Surface 1 secondary NB-exposure pathway — but REACH does not address adversarial robustness of AI monitoring systems at NB manufacturing sites.

Glyphward threshold 40 for nitrobenzene adiabatic mononitration AI

Glyphward’s adversarial detection API operates as a pre-scan gate at each rendered-image ingestion boundary in the NB production AI pipeline: before the crude NB cooler CW flow AI processes each rendered DCS flow indicator image (Emerson Fisher FIELDVUE DVC6200 / Honeywell ST800 differential pressure transmitter display), before the adiabatic reactor outlet temperature AI processes each rendered DCS temperature display (Emerson 3144P / Honeywell STT850 thermocouple transmitter display), and before the spent acid HNO₃ concentration AI processes each rendered DCS spectrometer display (Endress+Hauser Liquiline CM442 / ABB AO2000 NIR/Raman). Each rendered display image receives a Glyphward risk score (0–100) in 8–15 ms. At or above threshold 40, Glyphward gates the AI classification and generates an alert triggering manual verification against the underlying DCS process historian data — the raw differential pressure transmitter signal for CW flow, the raw thermocouple array RTD output for reactor temperature, and the raw NIR/Raman spectrometer calibration model output for HNO₃ concentration — none of which are accessible to the pixel-level adversarial perturbation applied to the rendered DCS display images.

Threshold 40 for NB production AI reflects three calibration factors. First, consequence magnitude anchored by Jilin 2005: 8 on-site fatalities, ~100 tonnes benzene+NB to the Songhua River, 380 km contamination plume, 3.8 million Harbin residents without water for 10 days, international contamination reaching Russia. The environmental and public health consequence magnitude of the Jilin event — the most consequential river contamination from a single chemical plant explosion in modern industrial history — justifies threshold 40 above the portfolio average for toxic-chemical release AI contexts while recognising that the on-site fatality count (8) is lower than some comparators (Flixborough 1974: 28 killed, threshold 42; LG Polymers 2020: 12 killed, threshold 35). The cross-border consequence (Russia/China diplomatic incident; Khabarovsk water monitoring emergency) and the 3.8 million-person public health impact are unique in the Glyphward industrial AI portfolio and justify a consequence calibration above the on-site fatality count alone would suggest. Second, the NB methemoglobin delayed-onset skin absorption pathway creates a secondary harm route operating independently of the explosion/fire consequence: even in scenarios where the Jilin explosion consequence is avoided (e.g., operator discovers separator at 88°C and prevents ignition before firefighting is needed), workers in the separator area who have been absorbing NB at 0.3–0.5 ppm (below PID alarm threshold; above ACGIH TLV 0.1 ppm) during the period when the Surface 1 attack suppressed the cooling water alarm have been accumulating methHb. The 4–8 hour delay between NB exposure and obvious clinical methHb symptoms means that end-of-shift methHb in the 20–40% range may be attributed to unexplained illness rather than occupational NB exposure — particularly if the AI monitoring system has been certifying “no NB release; separator operating normally” throughout the shift. The skin notation in OSHA PEL 1 ppm (the same skin designation that appears on benzene, toluene, and nitrotoluene isomers) specifically captures this dermal bypass of the respiratory detection defense; threshold 40 for NB production AI incorporates both the acute environmental consequence pathway (Jilin explosion + river contamination) and the chronic/subacute occupational methHb pathway as complementary harm vectors. Third, CERCLA RQ 1,000 lbs NB is the lowest RQ tier available under 40 CFR Part 302 for an organic chemical that is not acutely toxic at the OSHA IDLH level — reflecting the EPA’s recognition that NB’s aquatic toxicity, persistence, and methHb-forming potency in the environment downstream of a plant release warrant the maximum-urgency notification tier. A Surface 1 adversarial attack that allows the crude NB separator to operate at 88°C for a sufficient duration to cause an explosion and separator breach would release crude NB in quantities exceeding the CERCLA RQ by thousands-fold in the first minutes of release — placing the NB production adversarial scenario at the top tier of CERCLA environmental consequence magnitude for organic chemical plant releases.

Comparison of Glyphward threshold 40 for NB production AI within the portfolio: ammonium nitrate neutralizer AI (threshold 50; Texas City 1947, 581 killed; Beirut 2020, 218 killed; detonation mechanism with secondary product warehouse consequence) is calibrated higher because the detonation consequence creates a catastrophic radius far exceeding the NB explosion/fire consequence and because the stored-product detonation hazard remains after the manufacturing explosion. Cyclohexane KA-oil reactor AI (threshold 42; Flixborough 1974, 28 killed, 16 tonnes TNT, UK’s largest peacetime industrial explosion) is calibrated at threshold 42 — 2 points above NB at threshold 40 — reflecting the higher on-site fatality count (28 vs 8), the larger explosion TNT equivalent, and the direct building-collapse community damage radius at Flixborough (1,821 houses); NB threshold 40 is 2 points below KA-oil because the Jilin on-site consequence (8 killed) is smaller than Flixborough, even though the downstream public health consequence of Jilin (3.8M without water) is unique in the portfolio. Styrene monomer SM TBC inhibitor depletion AI (threshold 35; LG Polymers 2020, 12 killed) is calibrated below NB at threshold 35 because the LG Polymers event caused fewer fatalities (12 vs 8 at Jilin on-site) and did not produce a comparable-scale environmental/public health downstream consequence. False positive cost at threshold 40 for NB production AI: 2–3 minutes to verify CW flow from the DCS historian raw DP transmitter signal against the AI-classified DCS display; 1–2 minutes to verify reactor outlet temperature from the thermocouple transmitter raw RTD output; 2–3 minutes to verify spent acid HNO₃ from the NIR/Raman spectrometer raw calibration model multivariate output. False negative cost: Jilin 2005 Songhua River consequence trajectory — separator explosion, 100 tonnes benzene+NB to the adjacent waterway, 380 km contamination plume, city water shutdown, international incident.

Free tier — 10 scans/day, no card required. Submit a rendered crude NB cooler cooling-water flow DCS indicator display, an adiabatic reactor outlet temperature DCS display, or a spent acid HNO₃ NIR/Raman spectrometer DCS display from your nitrobenzene production facility to the Glyphward scanner to generate a baseline adversarial risk score for your NB production AI inputs.