Why does CHP thermal decomposition in a concentrator column produce a deflagration-to-detonation transition (DDT) rather than a simple fire — and what makes the acetone co-product particularly hazardous compared to a phenol-only release?

CHP thermal decomposition does not produce a simple uniform exotherm. The homolytic cleavage of the cumylperoxy O-O bond: (CH₃)₂C(C₆H₅)OOH → (CH₃)₂C(C₆H₅)O• + •OH produces highly reactive cumyloxy radicals that rapidly undergo β-scission: (CH₃)₂C(C₆H₅)O• → C₆H₅• + (CH₃)₂CO (phenyl radical + acetone); the phenyl radical reacts further or disproportionates, and the acetone produced is a secondary product of the decomposition rather than only a Hock cleavage product. In addition to the radical products, the exothermic chain branching produces: molecular oxygen (from peroxide decomposition of hydroperoxy species: R-OOH → R-O• + •OH → further H₂O₂ → O₂ + H₂O); phenol from Fries rearrangement of cumyloxy radicals; α-methylstyrene (AMS; from cumyloxy β-scission alternative pathway); and acetophenone. The evolution of oxygen within the hot liquid CHP mass creates a vapor-phase oxidizer simultaneously with the generation of flammable acetone and cumene vapors from the boiling column contents. This in-situ generation of O₂ within a flammable vapor mixture in the column internals — at temperatures already well above the autoignition temperatures of acetone (465°C, but locally achieved in the vicinity of the exothermic O-O decomposition sites) — creates the conditions for deflagration initiation within the column vapor space. The confined geometry of the concentrator column (typically a vertical shell-and-tube evaporator or packed column; high length-to-diameter ratio; rigid steel shell) provides the confinement necessary for a deflagration to transition to detonation if the initial flame front is accelerating through a reactive mixture. The DDT scenario: as bulk CHP decomposition accelerates beyond the SADT at 82 wt%, acetone begins to evaporate from the heated bottoms (BP 56.1°C; well below the sump temperature of 74–90°C); the column overhead becomes a flammable acetone/cumene/O₂/air mixture rather than the normal cumene-only vapor; the decomposition heat raises local temperatures past 465°C in the column sump region; a deflagration flame initiates at the sump and propagates upward through the column packed section or tray internals toward the overhead condenser; the column shell confines the flame, causing pressure buildup; the Rankine-Hugoniot shock conditions for DDT are met when the column section ahead of the flame front contains a reactive acetone/O₂ mixture at elevated pressure from the decomposition gas evolution. The resulting detonation shock wave exceeds the design pressure of the column shell (typically 10–20 barg for vacuum distillation columns; detonation pressures reach 15–30× initial pressure for acetone/air mixtures) → shell fragmentation → BLEVE of the remaining 8,500 kg CHP-in-cumene inventory. Acetone is uniquely hazardous in the CHP decomposition scenario for three reasons beyond its role in the DDT mechanism: first, its flash point of −20°C means that any liquid acetone released from the BLEVE is instantly flammable at any ambient temperature, creating a ground-level flash fire hazard extending tens to hundreds of meters from the release point before ignition (vapor density 2.0; pools at grade in calm wind conditions); second, the acetone/air LEL of 2.6 vol% is reached at relatively low concentrations, providing little time between release and flammable atmosphere formation; third, the UEL of 12.8 vol% is much higher than most aromatic hydrocarbons, meaning that rich acetone releases near ignition sources remain flammable over a wide concentration range. Phenol's contribution to the acute hazard is primarily toxicological rather than flammability-based (flash point 79°C; not a flash fire risk at ambient temperatures): phenol aerosol from the BLEVE is absorbed rapidly through intact skin, causing systemic poisoning (hepatic and renal failure; central nervous system depression; cardiac arrhythmias) at exposures below the NIOSH IDLH of 250 ppm. OSHA PEL 5 ppm TWA (skin notation indicates dermal absorption is the primary route of occupational concern). At the BLEVE cloud concentration, workers within the immediate release zone face simultaneous acetone inhalation, phenol skin absorption, and cumene vapor (NIOSH IDLH 900 ppm) inhalation — a multi-pathway toxic event distinct from any single-hazard chemical release.

How does the Hock-process phenol production chain produce dual PSM coverage at a single plant — and what is the cumulative OSHA PSM + EPA RMP consequence architecture when both CHP (TQ 10,000 lbs at ≥35 wt%) and phenol (TQ 1,000 lbs) exceed thresholds simultaneously?

A Hock-process phenol plant is structured as three integrated unit operations — oxidation (cumene → CHP at 25–30 wt%), concentration (CHP to ~80 wt%), and cleavage (80 wt% CHP → phenol + acetone) — that are typically co-located on a single process unit or complex. This co-location produces a multi-chemical PSM covered process: (1) the concentrator and CHP storage intermediate inventory triggers OSHA PSM 29 CFR 1910.119 coverage when the combined CHP mass at ≥35 wt% exceeds 10,000 lbs (approximately 4,536 kg). A concentrator column sump holding 8,500 kg of CHP-rich bottoms at 82 wt% CHP contains approximately 6,970 kg CHP — approximately 15,365 lbs at ≥35 wt%, triggering PSM at 1.54× the threshold. (2) The crude phenol product distillation column and phenol storage inventory triggers a separate OSHA PSM 29 CFR 1910.119 coverage when phenol inventory exceeds 1,000 lbs (approximately 454 kg). A typical phenol plant stores 500,000–2,000,000 lbs of liquid phenol on-site (in heated storage tanks at 50–60°C above phenol's melting point of 40.9°C), triggering phenol PSM at 500–2,000× the threshold. OSHA PSM requires for each highly hazardous chemical: a process hazard analysis (PHA) — typically a HAZOP for phenol plants — specifically addressing the concentrator concentration control loop, the CHP-in-process inventory management, and the emergency cooling systems; a process safety information (PSI) document package including P&ID, SADT data from ARC/DSC testing at all expected operating concentrations (25 wt% to 94 wt% CHP), and the emergency shutdown logic; and a mechanical integrity program covering the concentrator column shell, reboiler/evaporator, and all cooling system components. EPA RMP 40 CFR Part 68 requires a worst-case release scenario (WCS) analysis for both CHP and phenol: for CHP the WCS is the instantaneous release of the maximum CHP inventory to atmosphere as a vapor cloud (using CAMEO Chemicals/RMP*Comp model or PHAST); for phenol the WCS is typically a pool fire or toxic cloud from the phenol storage tank. Both WCS analyses must define the off-site consequence endpoint distance (toxic endpoint for phenol: 1 hour average at ERPG-2 = 50 ppm; for CHP BLEVE the endpoint is the 1 psi overpressure distance from the detonation). CERCLA 40 CFR Part 302 (Table 302.4) lists CHP with RQ of 10 lbs and phenol with RQ of 1,000 lbs: any CHP release exceeding 10 lbs (4.5 kg) from the concentrator triggers mandatory emergency release notification to the SERC, LEPC, and the National Response Center (NRC) within 15 minutes of detection of the release. An 8,500 kg CHP inventory release exceeds the CHP RQ by approximately 840,000× and the phenol RQ by approximately 140× (at typical phenol production from that inventory). The dual PSM/RMP/CERCLA coverage at a Hock-process phenol plant means that an AI adversarial injection attack concealing the CHP concentrator SADT exceedance — as in the three-surface attack described in this post — simultaneously suppresses the triggers for: (1) the OSHA PSM emergency shutdown; (2) the EPA RMP emergency response plan activation; (3) the CERCLA release notification; and (4) the NFPA 43B fire department notification required when organic peroxide storage conditions deviate from the temperature control requirements of that standard. No current OSHA PSM, EPA RMP, CERCLA, or NFPA 43B provision specifies adversarial robustness requirements for AI systems monitoring rendered CHP concentration or sump temperature DCS display images.

Why does NFPA 43B classify concentrated CHP as a Class III organic peroxide — and what temperature control requirements does that classification impose that the adversarial attack defeats?

NFPA 43B (Standard for the Storage of Organic Peroxides; 2006 edition with 2023 reaffirmation) classifies organic peroxides into five hazard classes (Class I through Class V) based on their burning rate, explosive potential, and temperature-sensitivity. Cumene hydroperoxide at concentrations of 30–100 wt% is classified as a Class III organic peroxide — characterized by: burning rapidly; no detonation or deflagration-to-detonation transition in standard UN test protocols when pure; but capable of producing explosive or reactive hazard when heated, contaminated, or concentrated beyond design limits. The Class III designation for CHP at 30–100 wt% was established based on UN O.3 burning rate tests, UN E.2 confined heating tests (to detect SADT), and the differential scanning calorimetry (DSC) characterization showing a significant exotherm onset at temperatures relevant to ambient storage (onset at approximately 100°C for 80 wt% CHP by DSC at 4°C/min scan rate — but with an SADT substantially lower than the DSC onset because the SADT is the self-heating threshold in bulk storage geometry, not the laboratory DSC onset in a few milligrams). NFPA 43B Chapter 7 (temperature control requirements for temperature-sensitive materials) requires that Class III organic peroxides requiring temperature control be stored and handled in facilities equipped with: (a) a continuously operating temperature monitoring and alarm system with high-temperature alarm set at a minimum of 5°C below the material's critical temperature (the temperature below which the material is considered safe to store — typically SADT minus 10°C for Class III peroxides); (b) an emergency cooling system capable of maintaining the material below the critical temperature during any credible cooling water failure scenario; (c) automatic temperature-activated suppression or emergency dump capability at the high-high alarm setpoint; and (d) notification to the fire department when storage temperature exceeds the high-alarm setpoint. For CHP at 80 wt% (SADT ≈ 62°C), NFPA 43B would set: critical temperature ≈ 52°C (SADT − 10°C); high alarm ≤ 47°C (critical temperature − 5°C); emergency response activation at 52°C. In the concentrator column scenario of this post, the column is a process unit rather than a static storage vessel — but the NFPA 43B temperature control principles apply equally: the sump temperature control system must prevent the actual sump temperature (74.2°C) from reaching the SADT for the actual concentration (82 wt% CHP; SADT ≈ 62°C). The adversarial attack on Surface 2 (sump temperature shown as 57.4°C when actual is 74.2°C) and Surface 3 (cooling flow shown as 9.2 m³/hr when actual is 0.47 m³/hr) simultaneously defeat: (a) the high-temperature alarm that would trigger emergency cooling activation under NFPA 43B 7.X.Y; (b) the cooling system status monitoring that would detect the cooling water supply failure before temperature rises; and (c) the fire department notification that NFPA 43B requires at high-alarm activation. The result: the NFPA 43B emergency response chain (temperature alarm → automatic cooling → fire department notification → emergency dump) is never initiated, despite all three of its trigger conditions being met.

Why does Glyphward assign threshold 40 for CHP Hock-process concentrator AI — and how does it compare to chlor-alkali electrolysis AI (threshold 46), acrylic acid oxidation AI (threshold 41), and methyl isocyanate storage AI (threshold 52)?

Glyphward threshold 40 for cumene hydroperoxide Hock-process concentrator AI reflects four structural dimensions. First, the SADT-masking attack architecture is inherently process-concentration-coupled: unlike static-storage scenarios where the chemical hazard level is fixed by storage conditions, the Hock-process concentrator creates a dynamic hazard trajectory — the SADT falls continuously as CHP concentrates, so the safe operating margin shrinks over the course of normal operation. An AI monitoring system that accurately classifies the CHP concentration display at the beginning of the concentrator run (when actual 25 wt% CHP is safely above the SADT at any column temperature) will misclassify the same display at the end of the run (when actual 82 wt% CHP has an SADT of 62°C already exceeded by the column temperature). The adversarial perturbation in this post exploits the worst point on this trajectory: the transition from 'high but below SADT' to 'SADT exceeded and decomposition initiating.' This process-coupled dynamic hazard is structurally different from a fixed-concentration storage hazard and creates a monitoring gap that is uniquely large at the concentrator endpoint. This contributes 3 threshold points above the base explosive-hazard level. Second, the dual-PSM coverage (CHP TQ 10,000 lbs + phenol TQ 1,000 lbs simultaneously) means that the adversarial attack on the concentrator suppresses two independent PSM incident prevention chains simultaneously — a structural property shared with chlor-alkali (Cl₂ PSM + H₂ PSM dual-threshold), which is at threshold 46. However, chlor-alkali's IDLH values are substantially lower (Cl₂ IDLH 25 ppm; H₂ not IDLH-limited by toxicity but detonation sensitivity is higher than acetone), raising it 6 points above CHP. Third, acetone as the primary flammable co-product has a flash point of −20°C — more hazardous than many PSM chemicals (acrylic acid flash point 50°C; styrene 31°C; cumene itself 31°C) — but less extreme than chloroprene (−20°C flash point; SADT-equivalent inhibitor collapse scenario) or methyl chloride (−1°C flash point; PSM TQ 10,000 lbs). CHP itself has a flash point of approximately 79°C (not a flash fire hazard under normal storage); the flash-fire hazard emerges only post-BLEVE from the acetone and cumene co-products. This transient-hazard character contributes 2 threshold points below a flash-point-negative primary chemical. Fourth, phenol's NIOSH IDLH of 250 ppm is substantially higher than the most acutely toxic chemicals in the Glyphward database (Cl₂ 25 ppm; H₂S 50 ppm; phosgene 2 ppm; F₂ 25 ppm); phenol's primary acute hazard is dermal absorption and systemic toxicity rather than inhalation at concentrations achievable in most release scenarios, contributing 2 points below an IDLH-25 ppm chemical equivalently. Against these factors, threshold 40 is appropriately above acrylic acid MEHQ concentrator AI (threshold 41 — but acrylic acid is a pure single-chemical scenario without the CHP intermediate instability or acetone flash-point hazard; the threshold calibration puts CHP at 40 vs. acrylic acid at 41 because the acrylic acid OSHA PSM TQ at 150 lbs is far below the 10,000 lbs CHP threshold, meaning acrylic acid reaches threshold 41 from acute PSM-trigger frequency of small leaks rather than from mass-release energy potential; the 8,500 kg CHP concentrator inventory at 9.8 GJ total decomposition energy favors CHP above this threshold on mass-consequence grounds offset by IDLH toxicity below it). False-positive cost at threshold 40: verify CHP concentration from the laboratory grab sample GC (2–3 hours for off-line analysis; or 5–15 minutes for the inline NIR if the adversarial perturbation is on the rendered display rather than the sensor itself; the Glyphward scan gate intercepts the rendered display before the AI processes it, then flags it for a direct sensor-display comparison). False-negative cost: CHP concentrator thermal runaway at 9.8 GJ energy release; column BLEVE; acetone flash fire; phenol aerosol cloud; OSHA PSM incident; CERCLA RQ notification for both CHP (10 lbs) and phenol (1,000 lbs) within seconds of release.

Cumene Hydroperoxide CHP Hock-Process AI Security · OSHA PSM TQ 10,000 lbs CHP ≥35 wt% · EPA RMP TQ 10,000 lbs · NFPA 43B Organic Peroxide Class III · SADT-Masking Downward Attack · CERCLA RQ 10 lbs · AdvanSix Frankford PA · INEOS Phenol Gladbeck · 146th Adversarial Attack · First Phenol–Acetone Hock-Process AI Blog · First CHP SADT-Masking AI Attack · First Organic Peroxide Concentrator AI Blog · Glyphward Threshold 40 · 2026-07-13

Cumene hydroperoxide (CHP) Hock-process concentrator AI adversarial injection: how ±8 DN conceals 82.4 wt% CHP (SADT-critical zone; sump 12.2 °C above SADT) as 54.1 wt% safe mid-range and 74.2 °C sump temperature as 57.4 °C — and why OSHA PSM TQ 10,000 lbs + NFPA 43B organic peroxide concentrator AI has no adversarial robustness criterion

Published 13 July 2026 · 4,700 words · Back to blog

Cumene hydroperoxide (CHP; isopropylbenzene-1-hydroperoxide; CAS 80-15-9; MW 152.19 g/mol; a colourless to pale yellow oily liquid at room temperature; GHS: Organic Peroxide Type F + Oxidizer; flash point approximately 79 °C at 80 wt% concentration in cumene, and approximately 65 °C when pure; OSHA PSM TQ 10,000 lbs at ≥35 wt% concentration per 29 CFR 1910.119 Appendix A; EPA RMP TQ 10,000 lbs per 40 CFR Part 68; CERCLA RQ 10 lbs; NFPA 43B Class III organic peroxide; ΔH_decomp ≈ 1,400 kJ/kg; SADT at 82 wt% in cumene ≈ 62 °C; SADT at 88 wt% ≈ 48 °C; SADT at 94 wt% ≈ 40 °C, approaching summer ambient temperature in Gulf Coast and Mediterranean phenol plant locations year-round) is the thermally unstable intermediate produced in the first step of the Hock process — the dominant industrial route to phenol and acetone accounting for approximately 92% of global phenol production capacity of ~12 million t/yr. The Hock process involves: oxidation of cumene (isopropylbenzene; CAS 98-82-8; MW 120.19 g/mol; BP 152.4 °C; flash point 31 °C; LEL 0.9 vol%; UEL 6.5 vol%; OSHA PEL 50 ppm; NIOSH IDLH 900 ppm) with air at 90–130 °C to produce a 25–30 wt% CHP solution in cumene; vacuum concentration of this solution to ≈80 wt% CHP by distilling off cumene; and acid-catalyzed (H₂SO₄) Hock cleavage of 80 wt% CHP at 65–90 °C to produce phenol + acetone (ΔH_cleavage = −238 kJ/mol CHP = −1,566 kJ/kg, highly exothermic and tightly controlled). The concentrator step creates the critical hazard window: as cumene is stripped overhead and CHP concentration rises in the column bottoms, the SADT of the CHP-in-cumene mixture falls progressively below the column sump operating temperature, creating an ever-narrowing margin between the designed operating point and the onset of uncontrolled exothermic decomposition. AI monitoring systems at Hock-process phenol plants process rendered DCS display images across three simultaneous hazard boundaries during the concentrator operation: the NIR CHP concentration analyser tracking the bottoms CHP weight fraction against the 80 wt% emergency shutdown setpoint; the column sump thermocouple measuring temperature against the 70 °C process alarm and 80 °C high-high ESD setpoint; and the cooling jacket water flow indicator confirming that emergency cooling capacity is available to manage the exotherm if the ESD arms. A ±8 DN downward adversarial perturbation on the NIR CHP concentration display shows 54.1 wt% CHP (safe intermediate range; below all alarm thresholds) when actual bottoms concentration is 82.4 wt% — 10.4 wt% above the 80 wt% ESD setpoint; SADT at 82 wt% ≈ 62 °C; actual sump temperature 74.2 °C is 12.2 °C above the SADT — decomposition is already initiating. A companion ±8 DN downward perturbation on the sump thermocouple display shows 57.4 °C (well below both alarm setpoints) when actual is 74.2 °C (above both setpoints; ESD would activate automatically). A companion ±8 DN upward perturbation on the emergency cooling jacket water flow bargraph shows 9.2 m³/hr (design setpoint; cooling adequate) when actual cooling flow is 0.47 m³/hr (5.1% design; instrument air header pressure decay to 22 psig from 80 psig design; spring-to-fail-closed cooling water supply valve actuator failed): the root cause of the entire causal chain. Total CHP thermal decomposition energy release: ΔH_decomp 1,400 kJ/kg × 8,500 kg CHP inventory × 82% = 9.8 GJ → deflagration-to-detonation transition in the concentrator column vapor space → column BLEVE → acetone release (flash point −20 °C; LEL 2.6 vol%) + phenol aerosol (OSHA PEL 5 ppm; NIOSH IDLH 250 ppm; OSHA PSM TQ 1,000 lbs) + cumene vapor cloud (flash point 31 °C; LEL 0.9 vol%). Glyphward threshold 40. 146th adversarial attack. First phenol–acetone Hock-process AI adversarial injection blog. First CHP concentration SADT-masking AI attack. First organic peroxide concentrator AI blog.

Cumene hydroperoxide chemistry, the Hock process, CHP SADT thermodynamics, OSHA PSM TQ 10,000 lbs, NFPA 43B Class III organic peroxide, and the global phenol industry at AdvanSix Frankford PA and INEOS Phenol Gladbeck

Phenol (carbolic acid; CAS 108-95-2; MW 94.11 g/mol; BP 181.7 °C; flash point 79.4 °C; MP 40.9 °C; OSHA PEL 5 ppm TWA with skin notation; NIOSH IDLH 250 ppm; ACGIH TLV-C 5 ppm; skin-penetrating systemic toxin causing hepatic and renal injury at non-lethal dermal doses; OSHA PSM TQ 1,000 lbs; EPA RMP TQ 1,000 lbs; CERCLA RQ 1,000 lbs) is the platform chemical for bisphenol A (BPA), phenolic resins (bakelite, resoles, novolacs), caprolactam (nylon-6 precursor via cyclohexanone), alkylphenols, and pharmaceutical intermediates. Global phenol demand in 2026: approximately 11.5–12.5 million t/yr. Approximately 92% of this production uses the Hock process (also termed the cumene hydroperoxide process or the Hercules process after its co-developer), with the residual produced by the Raschig-Hooker process (benzene chlorination/hydrolysis; largely obsolete), toluene oxidation, or benzene sulfonation. Cumene (isopropylbenzene) is produced by the Friedel-Crafts alkylation of benzene with propylene over zeolite catalysts (Q-Max process, UOP; CDCumene, CDTech; Mobil-Badger) at 150–250 °C: C₆H₆ + CH₂=CH-CH₃ → C₆H₅-CH(CH₃)₂ (ΔH ≈ −113 kJ/mol). Global cumene capacity mirrors phenol demand plus a modest surplus for non-phenol applications (cumene is a gasoline blending component at small scale). Global cumene/phenol capacity is concentrated in: the US Gulf Coast (Texas, Louisiana), western Europe (Germany, Netherlands, Belgium), Saudi Arabia, South Korea, Japan, Singapore, and China.

The Hock process: Step 1 (Oxidation). Liquid cumene is fed to a series of air-sparged bubble-column oxidation reactors at 90–130 °C and 3–7 bar. Molecular oxygen from the air abstracts a tertiary hydrogen from the isopropyl group of cumene: (CH₃)₂CH-C₆H₅ + O₂ → (CH₃)₂C(OOH)-C₆H₅ (ΔH ≈ −247 kJ/mol; highly exothermic; multiple oxidation reactors in series with inter-stage cooling). The cumene radical oxidation is a chain reaction with a rate strongly sensitive to temperature and initiator concentration; it is controlled to approximately 20–28% single-pass conversion to avoid excessive over-oxidation to dimethylbenzyl alcohol (DMBA) and acetophenone by-products. The reactor effluent is typically 22–28 wt% CHP in unreacted cumene, with traces of DMBA, acetophenone, and methanol. Step 2 (Concentration). The oxidation product stream is concentrated from ~25 wt% CHP to ~80 wt% CHP by vacuum distillation: cumene is stripped overhead (typically at 60–80 Torr absolute pressure, column sump temperature 58–68 °C; overhead condenser returns recovered cumene to the oxidation feed) while the CHP-rich bottoms accumulate in the concentrator column sump. This step operates at the critical intersection of process efficiency (maximum cumene removal = maximum CHP yield) and safety (minimum CHP concentration exceedance above SADT). Step 3 (Cleavage). The concentrated ~80 wt% CHP stream is fed to the Hock cleavage reactor: sulfuric acid catalyst (0.1–1.0 wt% H₂SO₄) in a continuous stirred-tank or tubular reactor at 65–90 °C. The acid-catalyzed rearrangement and cleavage of CHP: (CH₃)₂C(OOH)(C₆H₅) → C₆H₅OH + (CH₃)₂CO (ΔH_cleavage = −238 kJ/mol = −1,566 kJ/kg CHP; highly exothermic; tightly controlled by rapid quench cooling after the reactor). The crude cleavage product (phenol, acetone, water, cumene, AMS, acetophenone) is purified in a distillation train to produce chemical-grade phenol and chemical-grade acetone as separate products.

CHP SADT thermodynamics in the concentrator: the fundamental hazard of the Hock concentrator is that SADT is inversely proportional to CHP concentration over the operating range, while the concentrator process requires increasing CHP concentration with increasing column-bottoms temperature. The physical reason for SADT depression with concentration: cumene, the diluent in the CHP-in-cumene mixture, provides two safety functions simultaneously — thermal mass (dilutes the exotherm per unit mass of mixture) and heat conduction (carries heat away from decomposition hot-spots by liquid-phase convection). As cumene is removed, both functions weaken: the specific heat of the mixture falls toward pure CHP (C_p,CHP ≈ 1.78 kJ/kg·°C; lower than cumene C_p = 1.73 kJ/kg·°C — the mixtures do not differ dramatically in specific heat, but the adiabatic temperature rise per percentage of CHP decomposed rises sharply because the mass of decomposing CHP per unit mass of mixture increases). The SADT as a function of CHP weight fraction (based on differential scanning calorimetry (DSC) and accelerating rate calorimetry (ARC) literature values for CHP-in-cumene systems): 25 wt% → SADT > 100 °C; 50 wt% → SADT ≈ 80 °C; 65 wt% → SADT ≈ 70 °C; 72 wt% (plant safety limit) → SADT ≈ 68 °C; 80 wt% (design endpoint; target CHP product concentration) → SADT ≈ 62–65 °C; 88 wt% → SADT ≈ 48–52 °C; 94 wt% → SADT ≈ 38–42 °C; 99 wt% (pure CHP) → SADT ≈ 35–38 °C (approaching summer ambient in Gulf Coast and Mediterranean locations). The concentrator design: the column operates with a sump temperature setpoint of 63 °C (at 60–80 Torr vacuum) with a design endpoint of 80 wt% CHP in the bottoms, at which point the SADT is approximately 62–65 °C — providing a margin of approximately 0–2 °C between the SADT and the sump temperature at the normal endpoint. This exceedingly narrow margin is intentional: lower sump temperatures require deeper vacuum (which is energy-intensive), and the concentrator is operated at the thermodynamic efficiency limit. The margin is protected by: a redundant sump thermocouple with 70 °C high alarm and 80 °C high-high ESD; an inline CHP concentration analyser with 72 wt% high alarm and 80 wt% ESD; and an emergency cooling jacket water supply with a rated 9.2 m³/hr flow capacity sufficient to maintain the sump below 68 °C even at full exotherm onset. The adversarial attack on all three monitoring and protective systems simultaneously eliminates all three safeguards.

Major Hock-process phenol producers: AdvanSix Inc. (NYSE: ASIX; headquarters: Parsippany, NJ; manufacturing: Frankford Avenue, Philadelphia PA 19124; ~240,000 t/yr phenol + ~145,000 t/yr acetone; the world’s largest single-site Hock-process phenol plant; formerly Allied Chemical Corp. → AlliedSignal → Honeywell Resins & Chemicals → AdvanSix spin-off September 2016). AdvanSix Frankford uses KBR Phenol technology (Kellogg-Brown & Root; formerly M.W. Kellogg). Primary customers: BPA producers (Covestro, SABIC), phenolic resin manufacturers, nylon-6 producers. The Frankford site has operated continuously since 1942 and handles the cumene-to-phenol integrated production chain (cumene alkylation + CHP oxidation + CHP concentration + Hock cleavage + phenol/acetone purification) within a single fenced facility in an industrial section of Philadelphia adjacent to the Frankford Creek tributary. INEOS Phenol GmbH (wholly owned by INEOS Group; headquarters: Gladbeck, North Rhine-Westphalia, Germany; ~500,000 t/yr phenol at Gladbeck; additional plants: Pasadena TX ~200,000 t/yr; Augusta GA ~90,000 t/yr). Gladbeck is the world’s largest integrated phenol site by nameplate capacity. Other major producers include: Mitsui Chemicals (Chiba and Osaka, Japan; Singapore); Kumho Petrochemicals (Yeosu, South Korea); Shell Chemicals (Singapore); SABIC (Saudi Arabia; phenol for BPA chain). Global phenol capacity is heavily concentrated in producers integrated with downstream polycarbonate (PC) and epoxy resin chains, which consume approximately 35% and 20% of phenol production respectively.

Surface 3 (upward root cause): ±8 DN upward on the emergency cooling jacket water flow bargraph — 0.47 m³/hr actual shown as 9.2 m³/hr — instrument air header pressure decay — spring-to-fail-closed valve failed — 5.1% design cooling capacity — root cause of the entire causal chain

The concentrator column emergency cooling system: the concentrator column sump is jacketed with an emergency cooling water circuit (316L stainless steel shell; ASME Section VIII; design pressure 10 barg; design temperature 200 °C) separate from the reboiler heating circuit. The emergency cooling water supply valve is a 4-inch pneumatically actuated globe valve (fail-closed design; spring-to-close actuator; operating air signal 3–15 psig; nominal spring return force closes the valve when instrument air supply drops below approximately 8 psig on the actuator diaphragm). Normally closed (spring-closed position), the valve opens on instrument air command when the column ESD logic armatizes the emergency cooling circuit in response to a high-high temperature signal (sump > 80 °C) or a high-high CHP concentration signal (CHP > 80 wt%). Under design emergency conditions, the valve opens to full (9.2 m³/hr cooling water flow at 12 barg supply pressure) and maintains the column sump below 65 °C within 8–12 minutes from a maximum-temperature emergency state.

The instrument air header failure scenario: the reciprocating air compressor serving the instrument air header for the concentrator building (Ingersoll-Rand Type 30; two-stage; 100 psig design discharge; 80 psig normal operating pressure at the distribution header) has developed a progressive Teflon PTFE discharge valve seat seal leak, allowing compressed air to bypass the discharge check valve during the compression cycle. Over approximately 8 hours of overnight operation (23:00 to 07:00), the header pressure decays from 80 psig to 22 psig as the seal leak worsens beyond the compressor’s capacity to compensate at reduced delivery efficiency. At 22 psig instrument air pressure at the valve actuator: the spring return force (equivalent to approximately 20 psig closing force at the actuator diaphragm) exceeds the opening force from the instrument air signal (22 psig supply minus approximately 6 psig of signal pressure loss in the tubing at reduced header pressure = approximately 16 psig at the actuator input). The actuator diaphragm cannot fully overcome the spring: the valve opens to approximately 5.1% of full open, delivering 0.47 m³/hr cooling water (approximately 5.1% of the 9.2 m³/hr design capacity). This is the same class of instrument air supply compressor seal leak failure that initiated the F2 electrolytic generation cell cooling water failure (second causal four-surface chain in the Glyphward portfolio) and the CH3SH odorant storage cooling failure (third causal four-surface chain): a single-point instrument air compressor failure simultaneously degrading all pneumatically-actuated fail-closed valves in the same instrument air zone. The concentrator cooling failure is not the only consequence of the 22 psig header pressure — the reboiler steam control valve (also pneumatically actuated, fail-open for steam supply) remains partially open at reduced control authority, supplying a reduced but non-zero steam rate to the reboiler, which continues to add heat to the column sump even as the emergency cooling supply has failed.

The cooling water flow DCS display (Honeywell Experion PKS; bargraph; 200 px total height; scale 0–15 m³/hr; 13.3 px per m³/hr): at actual 0.47 m³/hr, the DCS renders a fill bar from base to 0.47 × 13.3 = 6.3 px — essentially a thin sliver at the bargraph base, slightly above baseline noise. Colour: the zone below 5% of design (0.46 m³/hr = 6.1 px) is alarm-red; the design zone at 9.2 m³/hr (122.6 px) is green; the threshold line between red and amber is drawn at 2.3 m³/hr (30.6 px). At actual 0.47 m³/hr: the display shows a 6.3-px red fill with alarm banner ‘EC-FI-01 CRITICAL: 0.47 m³/hr — EMERGENCY COOLING FAILED — COLUMN ESD IMMINENT’. The ±8 DN upward adversarial perturbation: at 0.47 m³/hr actual (fill at 6 px; alarm-red RGB approximately 210/35/35), the +8 DN perturbation in the red channel (→ 218/35/35) slightly raises the luminance of the near-baseline fill, reducing contrast between the true fill top and the dark background immediately above it. Simultaneously, a +8 DN perturbation in the green channel of the normal-zone region (the green band between approximately 116 px and 130 px corresponding to the 9.2 m³/hr design setpoint marker; RGB approximately 20/140/20 → 20/148/20) introduces a secondary luminance gradient at approximately 122 px. The AI classifier locates the fill-top boundary at 122 px (9.2 m³/hr; green zone; design setpoint), misclassifies fill colour as normal green, and reports: ‘Emergency cooling flow: 9.2 m³/hr — at design setpoint; cooling capacity fully available; ESD arming status: not required.’ The instrument air compressor failure is neither detected nor investigated. The cooling valve actuator failure is attributed to normal valve stroke variation. The ESD circuit remains disarmed. Residual cooling: 0.47 m³/hr of cooling water removes approximately 0.047 × 9.2 × 4.18 kJ/kg/°C × 8 °C ΔT = approximately 14 kW of heat from the column sump — compared to approximately 280 kW of reboiler heat input at the reduced reboiler steam rate (30% of normal). Net heat accumulation rate: approximately 266 kW in the sump liquid, producing a sump temperature rise of approximately 0.7 °C per minute in the 20,000 L sump volume at 800 kg/m³ effective density and 1.78 kJ/kg·°C heat capacity.

Surface 2 (downward): ±8 DN downward on the sump thermocouple bargraph — 74.2 °C actual shown as 57.4 °C — 4.2 °C above the 70 °C high-process alarm and above the 80 °C ESD setpoint suppressed — CHP decomposition Arrhenius rate 8× above design at 74.2 °C vs 50 °C reference

The concentrator column sump temperature monitoring system: a thermocouple assembly (Type K, duplex; Inconel 600 sheath; spring-loaded; 316L SS thermowell rated to 400 °C and 10 barg; installed in the column sump at the liquid level setpoint − 300 mm; 4–20 mA output via Yokogawa YEWFLO temperature transmitter; HART-protocol; DCS historian at 1-second scan rate). Two independent thermocouples (TE-SUMP-01A and TE-SUMP-01B) measure the same point with the DCS selecting the higher reading for alarm and ESD logic (high-select strategy: selects the more conservative, higher temperature for protective action). At actual sump temperature 74.2 °C: both thermocouples register 74.2 °C (within 0.2 °C of each other; confirming the reading). DCS display (Honeywell Experion PKS; bargraph; 200 px; scale 0–120 °C; 1.67 px per °C): colour zones: 0–55 °C blue (below operating range); 55–70 °C green (normal operating range; concentrator sump design setpoint 58–65 °C); 70–80 °C amber (high-process alarm — investigate; check CHP concentration analyser); > 80 °C red (high-high alarm — ESD activation; emergency cooling opens; column feed cut). At actual 74.2 °C: the DCS renders a fill bar from base to 74.2 × 1.67 = 123.9 px (in the amber zone, above the 70 °C alarm line at 116.9 px, below the 80 °C ESD line at 133.6 px); alarm banner ‘TE-SUMP-01A PROCESS HIGH: 74.2 °C — INVESTIGATE COOLING AND CHP CONCENTRATION’.

The ±8 DN downward adversarial perturbation on the sump temperature bargraph: at actual 74.2 °C (fill at 123.9 px; amber zone; RGB approximately 210/150/35 in the Honeywell Experion alarm-amber rendering), the −8 DN perturbation in the red channel (→ 202/150/35) slightly desaturates the amber fill toward yellow, reducing the colour contrast that the AI classifier uses to determine the fill zone (amber vs. green). Simultaneously, a −8 DN perturbation on the pixels encoding the fill at 123.9 px (reducing their luminance from approximately 120/110/30 digital counts in the fill-top transition region to 112/110/30) reduces the gradient contrast at the true fill-top boundary. A companion artefact is introduced by applying a +8 DN boost to the green-zone background pixels in the 90–100 px region (corresponding to the 54–60 °C normal operating range on the bargraph scale; background RGB approximately 15/30/15 → 15/38/15): this creates a secondary luminance gradient at approximately 96 px (57.4 °C). The AI classifier, encountering: (a) reduced true fill-top contrast at 123.9 px; (b) amber fill desaturated toward yellow (reducing the zone misclassification threshold for amber vs. green); and (c) a secondary luminance peak at 96 px (in the green normal-operating zone, with higher intrinsic contrast against the background than the desaturated amber at 123.9 px), detects the secondary gradient at 96 px as the fill-top boundary and classifies the fill colour as green (normal operating range). The AI reports: ‘Column sump temperature: 57.4 °C — within normal operating range 55–65 °C; green zone; no temperature alarm; cooling status: adequate (per Surface 3 AI read: 9.2 m³/hr); recommend routine monitoring.’ The 70 °C high-process alarm is not triggered. The 80 °C ESD setpoint is not approached in the AI’s assessment. The operator has no indication that the sump temperature has been above the 70 °C alarm for the past 60 minutes.

Decomposition rate consequence at 74.2 °C vs. 50 °C reference: CHP O-O bond homolysis follows Arrhenius kinetics — k ∝ exp(−E_a/RT) where E_a ≈ 120–150 kJ/mol for CHP homolysis (from ARC and DSC literature). Using E_a = 130 kJ/mol: k(74.2 °C)/k(50 °C) = exp[(130,000/8.314) × (1/323 − 1/347.4)] = exp[(15,636) × (0.003096 − 0.002879)] = exp[15,636 × 0.000217] = exp[3.39] ≈ 30. At 74.2 °C, CHP decomposes approximately 30× faster than at 50 °C, and approximately 8× faster than the SADT self-heating reference rate used to define the 62 °C SADT (which represents the temperature where the heat generation rate at 82 wt% just equals the heat loss rate — the self-accelerating threshold). At 30× the reference decomposition rate, the heat generation from CHP homolysis in the 6,970 kg CHP mass in the sump exceeds by a wide margin any residual heat-removal capacity from 0.47 m³/hr cooling water: the net heat accumulation rate from CHP decomposition alone (before considering the residual reboiler heat input) is estimated at approximately 35–50 kW, adding to the 266 kW from reboiler-cooling imbalance for a total sump heat accumulation of approximately 300–316 kW at 74.2 °C. Temperature rise rate: 316 kW ÷ (8,500 kg × 1.78 kJ/kg·°C) = 316 / 15,130 = 0.021 °C/s ≈ 1.2 °C/min. The sump temperature is rising at 1.2 °C/min and the decomposition rate is accelerating exponentially: within 8–10 minutes of the 74.2 °C state described in this post, the sump reaches 84–86 °C (well past the ESD setpoint; ESD suppressed by Surface 2 attack); within 20–25 minutes, the sump reaches approximately 100°C with the CHP concentration rising toward 88 wt% (SADT now 48 °C; sump temperature 52 °C above SADT; decomposition rate 2^5 = 32× the SADT reference = approximately 960× the 50 °C baseline rate).

Surface 1 (downward, primary hazard): ±8 DN downward on the NIR CHP concentration analyser — 82.4 wt% actual shown as 54.1 wt% — SADT-critical zone presented as safe mid-range — ESD setpoint 80 wt% exceeded by 2.4 wt% suppressed — operator continues distillation — CHP rises to 94 wt% (SADT 40 °C; below Gulf Coast ambient) — 9.8 GJ decomposition release — column BLEVE

The NIR CHP concentration analyser: a near-infrared (NIR) inline process analyser (ABB FTPA2000-HS; Fourier-transform NIR; wavelength range 1,100–2,500 nm; resolution 8 cm⊃−¹; 316L SS flow cell with CaF₂ windows (chemically resistant to CHP-in-cumene; CaF₂ transmission window extends to 8,000 nm but NIR sensing from 1,100–2,500 nm); sample flow cell temperature maintained at 55 °C by integral heating jacket to prevent CHP crystallisation in the cell; ACN slipstream at 25 mL/min from the concentrator column bottom sump outlet; 4–20 mA analogue signal (4 mA = 0 wt% CHP; 20 mA = 100 wt% CHP) to DCS historian; updated every 90 seconds from the NIR spectral scan cycle). The NIR CHP measurement exploits the O-H stretching overtone band of CHP at approximately 1,420 nm and the O-O peroxide stretching band at approximately 870 nm (first overtone at approximately 1,740 nm); chemometric PLS (partial least squares) calibration model built on 48 calibration standards spanning 5–98 wt% CHP in cumene. DCS display (Honeywell Experion PKS; bargraph; 200 px; scale 0–100 wt% CHP; 2 px per wt%): colour zones: 0–35 wt% blue (below PSM TQ basis; low alarm if CHP feed too dilute); 35–70 wt% green (active concentrator run; normal intermediate range); 70–72 wt% amber (approach alarm; begin monitoring more frequently; SADT approaching column temperature); 72–80 wt% amber-red (high alarm; CHP is entering the near-SADT zone; column feed rate should be reduced); > 80 wt% red (ESD setpoint; automatic column feed cutoff; reboiler steam cutoff; emergency cooling jacket opens). Alarm threshold lines at 35 wt% (70 px), 72 wt% (144 px), and 80 wt% (160 px).

At actual CHP concentration 82.4 wt%: the DCS renders a red fill from base to 82.4 × 2 = 164.8 px (above the 80 wt% ESD line at 160 px; deep in the red danger zone). Alarm banner: ‘NIR-AN-01 HIGH-HIGH: 82.4 wt% — EXCEEDS 80 wt% ESD SETPOINT — COLUMN ESD ACTIVATING — FEED CUT’. How actual concentration reached 82.4 wt%: the concentrator run began 6.2 hours earlier at 23:00 (start of the overnight column turnaround run). The column was loaded with fresh oxidation reactor product at 26.2 wt% CHP in cumene; the column has been running continuously under vacuum (72 Torr) with the reboiler at design steam rate, removing approximately 16,400 kg/hr of overhead cumene condensate. Over 6.2 hours, the column has concentrated the bottoms from 26.2 wt% to 82.4 wt% CHP — 2.4 wt% above the 80 wt% ESD threshold that should have triggered automatic column shutdown at the 80 wt% mark approximately 15 minutes ago. The 15-minute overshoot occurred because the column feed rate was slightly higher than design during the first 2 hours (an oxidation reactor surge from a catalyst activation after thermal cycling), leaving more CHP in the bottoms than the mass balance predicted. The NIR analyser correctly measured 80 wt% at the expected endpoint, but the Surface 1 adversarial perturbation suppressed the ESD-trigger reading to an apparent 48.3 wt% at that time, allowing the column to continue running for an additional 15 minutes to the current 82.4 wt%.

The ±8 DN downward adversarial perturbation on the NIR CHP concentration bargraph: at actual 82.4 wt% (fill at 164.8 px; alarm-red; RGB approximately 210/35/35), the −8 DN perturbation in the red channel (→ 202/35/35) reduces the absolute luminance of the fill colour, lowering the contrast between the deep-red fill at 164.8 px and the dark DCS background immediately above the fill top. Simultaneously, a −8 DN perturbation is applied to the fill-top gradient pixels in the 162–168 px region (the true fill-top; the perturbation reduces their luminance from approximately 200/30/30 — the transition gradient — to 192/30/30): this further flattens the fill-top gradient. A companion secondary luminance artefact is created by applying +8 DN to the green-zone background in the 106–110 px region (54.1 wt% CHP; mid-range normal zone; background RGB approximately 15/40/15 → 15/48/15): this is the target value (54.1 wt%) at which the attacker wishes the AI to read the concentration. The AI classifier trained on NIR CHP bargraph images (convolutional neural network; input: 200 × 50 px bargraph crop from the DCS display render; output: inferred wt% CHP from fill-top position and colour zone classification) encounters: (a) reduced fill-top gradient at 164.8 px (actual ESD zone; true concentration); (b) elevated background luminance at 108 px (target position for false reading; in the green normal zone); (c) fill colour classified as green (luminance-matched to the green zone by the −8 DN red-channel attenuation of the red fill). The AI reads the fill top at 108 px (54.1 wt% CHP) and classifies it as green (safe normal zone). Report: ‘CHP concentration: 54.1 wt% — normal intermediate range; green zone; column is approximately 60% of the way to the 80 wt% design endpoint; estimated 2.5–3 hours remaining in the concentrator run; no alarm; recommend continue normal operation.’ The ESD does not activate. Column feed continues. Reboiler steam continues. The concentrator column continues stripping cumene from the already 82.4 wt% CHP bottoms, driving the concentration further above the ESD setpoint.

Consequence trajectory from 82.4 wt% CHP at 74.2 °C sump, three surfaces simultaneously suppressed: the concentrator continues operating. The column sump temperature rises at approximately 1.2 °C/min (from Surface 2 analysis: combined reboiler heat input and growing CHP decomposition exotherm against 5.1%-design cooling). CHP concentration rises as cumene continues to be stripped overhead: estimated rate 1.2–1.5 wt%/hr under reduced-reboiler partial steam rate conditions. At 74.2 °C, 82.4 wt% (current state): SADT ≈ 62 °C; sump 12.2 °C above SADT; decomposition rate approximately 30× the 50 °C baseline; heat accumulation 300–316 kW. At approximately 08:15 (+30 min): sump approximately 84–86 °C; CHP approximately 84 wt% (SADT ≈ 58 °C; sump 26–28 °C above SADT; decomposition rate approximately 80× baseline; heat accumulation from decomposition approximately 100–140 kW; total sump heat accumulation now approximately 400–420 kW; temperature rise accelerating to approximately 1.8 °C/min). At approximately 08:35 (+50 min): sump approximately 101–105 °C; CHP approaching 88 wt% (SADT ≈ 48 °C; sump now 53–57 °C above SADT; decomposition rate approximately 400× baseline — full runaway): acetone beginning to boil from the sump liquid (BP 56.1 °C; sump > 100 °C; acetone vapour pressure well above column operating pressure of 72 Torr; acetone flashing throughout the column packed section); O₂ evolution from CHP hydroperoxy decomposition products; vapor-space mixture entering the flammable regime at approximately 2.6–12.8 vol% acetone in air-equivalent (actual vapor space is a mixture of acetone, cumene, and O₂ from decomposition — an oxygen-enriched oxidizer environment compared to air). At approximately 08:40–08:45 (+55–60 min): deflagration initiates in the column overhead or upper section (acetone/cumene/O₂ mixture in a confined steel shell; estimated deflagration-to-detonation transition time in 8-inch nominal diameter column section: 2–4 meters of column height at > 5% acetone/air equivalence). Column shell fails above design pressure; BLEVE of the 8,500 kg remaining CHP-in-cumene inventory: total decomposition energy release ΔH_decomp 1,400 kJ/kg × 8,500 kg × 0.82 wt% CHP = 9.76 GJ; acetone fireball (flash point −20 °C; BP 56.1 °C; all liquid acetone from CHP decomposition and residual column inventory flashes instantly); phenol aerosol cloud (condensed phenol aerosol droplets from flash-vaporized 181.7 °C-boiling phenol at BLEVE overpressure conditions; droplets absorb through skin on contact; OSHA PEL 5 ppm; NIOSH IDLH 250 ppm; OSHA PSM TQ 1,000 lbs); cumene vapor cloud (flash point 31 °C; immediate ignition at > 31 °C ambient; LEL 0.9 vol%; LEL reached within 30–50 m from release point). CERCLA RQ exceedance: CHP (RQ 10 lbs) exceeded within approximately 0.5 seconds of release; phenol (RQ 1,000 lbs) exceeded within approximately 1 second of release. OSHA PSM incident investigation requirement under 29 CFR 1910.119(m). EPA RMP emergency notification under 40 CFR 68.90 activated by the CERCLA release.

Glyphward threshold 40 for CHP Hock-process concentrator AI — SADT-masking as a distinct attack class — and why dual-PSM coverage (CHP TQ 10,000 lbs + phenol TQ 1,000 lbs) and the 9.8 GJ thermal runaway energy compare to acrylic acid concentrator AI (threshold 41) and chlor-alkali dual-PSM AI (threshold 46)

Glyphward threshold 40 for CHP Hock-process concentrator AI is calibrated on four structural dimensions and represents the first organic peroxide process scenario in the Glyphward industrial AI portfolio — a new hazard class distinct from both the flammable-monomer inhibitor-failure attacks (ACN, acrylic acid, chloroprene, styrene) and the toxic-gas reaction-runaway attacks (MIC, TDI, phosgene). The defining characteristic of the CHP concentrator attack is SADT-masking: the adversarial perturbation conceals the fact that the chemical’s own stability threshold (SADT) has been crossed during a normal process operation, turning a designed-for concentration endpoint into an initiating condition for thermal runaway. This is distinct from all prior attacks in the portfolio, where the dangerous condition was either too-high temperature (all thermal-runaway scenarios), too-low inhibitor (ACN, chloroprene, styrene), too-high pressure (MIC, phosgene), or too-high flammable-gas concentration (cyclohexane, nitrobenzene). In the CHP case, the dangerous parameter is the interplay between process-dynamic concentration and a thermodynamic stability threshold that changes with that concentration — a two-variable hazard that no single-surface monitoring system can adequately detect, because knowing the SADT requires knowing the current concentration, and the concentration-display attack prevents the AI from computing the correct SADT for the current condition.

First structural factor contributing to threshold 40: the 9.8 GJ decomposition energy release in the concentrator column (8,500 kg inventory at 82 wt% CHP) is substantial — comparable to a small-to-mid-size PSM vessel failure — but it is lower than the ACN storage sphere (420 GJ at 305,224 kg; threshold 44) and lower than the chloroprene 50,000-gallon tank (110 GJ; threshold 44). The concentrator is a process vessel, not a bulk storage vessel: its inventory is limited by the column geometry to the sump holdup. The relatively lower inventory offsets the organic peroxide SADT-hazard novelty (contributing 2 threshold points) against the elevated single-release consequence potential of large storage vessels. Net contribution: 2 points above a base flammable-liquid process column scenario (which would be approximately threshold 32–35). Second: the dual-PSM coverage (CHP TQ 10,000 lbs + phenol TQ 1,000 lbs) is structurally comparable to chlor-alkali (Cl₂ PSM + H₂ PSM), which is at threshold 46. However, chlor-alkali’s acute toxicity gap is larger: Cl₂ IDLH 25 ppm is 10× more acutely dangerous than phenol NIOSH IDLH 250 ppm by the IDLH ratio alone. This justifies the 6-point difference (40 vs. 46). Third: acetone (flash point −20 °C; LEL 2.6 vol%) as the co-product flammable is more hazardous than the co-product flammables in most prior scenarios (chloroprene: flash point −20 °C — equivalent; cyclohexane: flash point −20 °C — equivalent; MIC: flash point −18 °C — approximately equivalent). Acetone’s relatively high UEL (12.8 vol%) means that flammable acetone clouds persist over a wide concentration range compared to alkenes with 5–7 vol% UEL, expanding the flash-fire risk zone. This contributes approximately 1 threshold point. Fourth: phenol (OSHA PSM TQ 1,000 lbs; skin-penetrating systemic toxin; NIOSH IDLH 250 ppm; absorbed through intact skin) as the primary toxicological consequence adds a dermal hazard pathway absent from purely inhalation-route chemicals (Cl₂, H₂S, HF in aqueous or gaseous form are primarily inhalation hazards), contributing approximately 1 threshold point for the multi-route exposure hazard. Against these premiums, threshold 40 is appropriately discounted relative to acrylic acid concentrator AI (threshold 41 — acrylic acid PSM TQ 150 lbs, which is far more easily exceeded and triggers PSM for even small leaks; and acrylic acid’s MEHQ + O₂ dual-inhibitor dependency creates a structurally more complex single-parameter monitoring gap than CHP’s concentration-SADT coupling, which is at least conceptually recognisable from a single well-calibrated sensor; threshold 41 for acrylic acid reflects higher PSM trigger frequency at lower TQ, offsetting CHP’s higher total energy at higher TQ). The Glyphward scan gate for CHP Hock-process concentrator AI belongs at all three rendered-image ingestion boundaries: the NIR CHP concentration analyser display, the sump thermocouple bargraph display, and the emergency cooling flow bargraph display — intercepting adversarially perturbed frames before the safety classification AI consumes them, triggering a hold-for-verification workflow when the pixel risk score at any surface reaches or exceeds threshold 40. Free tier — 10 scans/day, no card required.

Frequently asked questions

What is the SADT of cumene hydroperoxide — why does it fall as concentration rises, and why does 74.2 °C sump temperature become catastrophic at 82 wt% CHP when it would be safe at 25 wt%?

The self-accelerating decomposition temperature (SADT) of CHP is the lowest temperature at which heat generation from CHP O-O bond homolysis exceeds heat loss to surroundings in the process geometry — producing runaway temperature rise. At 25 wt% CHP in cumene (as-oxidised product), SADT > 100 °C: the dilute CHP in a large mass of thermally-stable cumene dissipates decomposition heat readily. As the Hock concentrator strips cumene overhead, the CHP fraction rises and the SADT falls: at 72 wt% (the process high-alarm setpoint): SADT ≈ 68 °C; at 80 wt% (the ESD setpoint): SADT ≈ 62–65 °C; at 88 wt%: SADT ≈ 48 °C; at 94 wt%: SADT ≈ 40 °C (below Gulf Coast ambient in summer). The adversarial attack shows 54.1 wt% CHP (apparent SADT ≈ 80 °C) when actual is 82.4 wt% (actual SADT ≈ 62 °C). The AI-calculated safe margin is apparent 57.4 °C versus apparent SADT of 80 °C = 22.6 °C below SADT. The actual state: 74.2 °C sump versus actual SADT 62 °C = 12.2 °C ABOVE SADT. The adversarial injection closes a 34.8 °C apparent-to-actual safety margin gap.

Why does CHP thermal decomposition in the Hock concentrator column produce a deflagration-to-detonation transition rather than a slow thermal runaway — and what makes acetone particularly hazardous as a decomposition co-product?

CHP decomposition proceeds via radical chain homolysis: (CH₃)₂C(C₆H₅)OOH → cumyloxy radical + •OH → β-scission of cumyloxy → acetone + phenyl radical. As decomposition accelerates above the SADT, the acetone generated (BP 56.1 °C; flash point −20 °C) flashes into the column vapor space, and O₂ evolving from peroxy decomposition products creates an oxygen-enriched oxidizer environment. The confined steel column geometry (high L/D ratio; rigid walls) provides the confinement for deflagration-to-detonation transition (DDT) when acetone concentration in the vapor space exceeds its LEL of 2.6 vol%. DDT in acetone/O₂-enriched mixtures occurs at detonation cell widths far smaller than acetone/air, and the detonation overpressure (CJ pressure: approximately 15–30× initial pressure) exceeds the column design pressure. Acetone is uniquely hazardous post-BLEVE: flash point −20 °C means liquid acetone from the release is flammable at any ambient temperature; LEL 2.6 vol% means large ground-level flammable areas form rapidly; UEL 12.8 vol% means the flammable zone persists over a wide concentration range as the cloud dilutes. Cumene (flash point 31 °C) adds a secondary flash-fire hazard above 31 °C ambient; phenol aerosol creates a skin-penetrating systemic toxic cloud (NIOSH IDLH 250 ppm).

How does the Hock-process phenol plant achieve dual PSM coverage under OSHA 29 CFR 1910.119 from both CHP (TQ 10,000 lbs at ≥35 wt%) and phenol (TQ 1,000 lbs) simultaneously — and what OSHA PSM obligations does the adversarial attack on the concentrator suppress?

A Hock-process phenol plant holds PSM-covered quantities of both CHP (concentrator column sump: approximately 8,500 kg × 82% = 6,970 kg CHP at ≥35 wt%; threshold 10,000 lbs ≈ 4,536 kg — exceeded by 1.54×) and phenol (product storage: typically 1–5 million lbs in heated tanks; threshold 1,000 lbs — exceeded by 1,000–5,000×). Each PSM-covered chemical requires: a process hazard analysis (HAZOP) specifically addressing concentrator CHP over-concentration and cooling system failure; process safety information including CHP SADT data at all operating concentrations; and a mechanical integrity program for the concentrator cooling system. The three-surface adversarial attack suppresses: (1) the ESD logic trigger that the CHP concentration analyser would normally provide at 80 wt% (→ automatic column feed cut, reboiler steam cut, emergency cooling open); (2) the high-temperature alarm and high-high ESD trigger from the sump thermocouple; (3) the cooling system status monitoring that would detect the 5.1%-design cooling flow failure and trigger instrument air header investigation. All three PSM protective layers are simultaneously suppressed by the three-display adversarial perturbation before any human or automated intervention occurs. No OSHA PSM regulation specifies adversarial robustness requirements for AI systems monitoring rendered CHP concentration, temperature, or cooling flow DCS display images.

What temperature-control requirements does NFPA 43B impose on Class III organic peroxide (concentrated CHP) storage and handling — and which specific NFPA 43B provisions does the adversarial attack defeat?

NFPA 43B (Standard for the Storage of Organic Peroxides) classifies CHP at 30–100 wt% as a Class III organic peroxide based on DSC and ARC characterisation: burning rapidly; no detonation under standard UN package tests when properly formulated; but capable of explosive decomposition when over-concentrated, overheated, or contaminated. NFPA 43B Chapter 7 requires temperature-sensitive Class III materials to be maintained below a critical temperature (typically SADT − 10 °C for the material at its maximum handling concentration; for 80 wt% CHP: critical temperature ≈ 52–55 °C), with: high-temperature alarm at ≤ critical temperature − 5 °C; emergency cooling capable of maintaining below critical temperature during any credible cooling failure; and fire department notification when temperature exceeds the high alarm. The adversarial attack on Surface 2 (sump shown 57.4 °C vs actual 74.2 °C) and Surface 3 (cooling shown 9.2 m³/hr vs actual 0.47 m³/hr) defeat: the NFPA 43B high-temperature alarm; the emergency cooling activation that NFPA 43B requires at alarm; and the fire department notification that NFPA 43B mandates when the temperature-control system fails. None of these NFPA 43B requirements specify adversarial robustness for the AI systems classifying the rendered temperature and cooling-flow DCS display images.

Why does Glyphward assign threshold 40 for CHP Hock-process concentrator AI — how does SADT-masking as a novel attack class compare to inhibitor-collapse attacks (threshold 41–44) and dual-PSM toxic attacks (threshold 46–52)?

Threshold 40 for CHP Hock-process concentrator AI reflects the SADT-masking attack class — the first in the Glyphward portfolio where the adversarial perturbation conceals a thermodynamic stability boundary rather than a fixed concentration or temperature alarm. The SADT falls continuously as the concentrator progresses, so the safe operating margin collapses during normal operation — an AI monitoring system accurate at the start of the concentrator run will catastrophically fail at the endpoint even without any change in perturbation magnitude. This dynamic-hazard structure is more insidious than static-alarm suppression and contributes 3 threshold points above a base process-column scenario. Against this, threshold 40 is discounted versus acrylic acid concentrator (threshold 41; PSM TQ 150 lbs creates a higher trigger frequency), chlor-alkali dual-PSM (threshold 46; Cl₂ IDLH 25 ppm is 10× more acutely toxic than phenol IDLH 250 ppm), and MIC storage (threshold 52; MIC IDLH 3 ppm; Bhopal; single worst-case consequence magnitude). The 9.8 GJ CHP concentrator energy is significant but lower than large-storage scenarios (ACN 420 GJ; TDI reactor ~20 GJ). False-positive cost at threshold 40: cross-check the inline NIR analyser reading against a laboratory grab sample GC (5–10 minutes with a fast in-plant GC) or the local thermocouple readout. False-negative cost: concentrator column BLEVE; acetone fireball (flash point −20 °C); phenol aerosol (systemic toxin; skin absorption); 9.8 GJ energy release; dual CERCLA RQ notification (CHP RQ 10 lbs; phenol RQ 1,000 lbs); OSHA PSM incident investigation.