Styrene Monomer SM Production AI Security · Ethylbenzene Dehydrogenation AI · TBC Polymerization Inhibitor AI Adversarial · SM Storage Tank AI Monitoring · Radical Polymerization Inhibitor Depletion AI · OSHA PSM 29 CFR 1910.119 Flammable Liquid TQ 10,000 lbs · NFPA 704 Flammability 3 · ACGIH TLV-TWA 20 ppm A3 · NIOSH IDLH 700 ppm · Flash Point 31°C · LEL 0.9 % UEL 6.8 % · IARC Group 2A · LG Polymers Visakhapatnam India 7 May 2020 (12 Killed, ~800 Hospitalized) · 74th Upward-Direction Attack · First Styrene Monomer Production in Portfolio · First Polymerization Inhibitor Depletion Attack · First Storage-Tank Vapor-Generation Consequence · Glyphward Threshold 35

Styrene monomer (SM) ethylbenzene dehydrogenation TBC polymerization inhibitor AI adversarial injection: how a ± 8 DN upward pixel shift on the SM storage tank TBC concentration analyzer (1.8 ppm displayed as 12.4 ppm) suppresses inhibitor dosing, allows radical polymerization to begin within 48–72 hours at summer storage temperatures, and drives a heat‑of‑polymerization vapor‑release chain — LG Polymers Visakhapatnam India 7 May 2020 (12 killed, ~800 hospitalized), OSHA PSM flammable liquid TQ 10,000 lbs, NIOSH IDLH 700 ppm, 74th upward‑direction attack, first styrene monomer production, first polymerization inhibitor depletion attack in the Glyphward industrial AI portfolio, Glyphward threshold 35

Styrene monomer (SM; C6H5CH=CH2; CAS 100-42-5; molecular weight 104.15 g/mol; boiling point 145°C at 1 bar; flash point 31°C (cc); LEL 0.9 %, UEL 6.8 %; autoignition temperature 490°C; IARC Group 2A probable human carcinogen since January 2019; ACGIH TLV-TWA 20 ppm; NIOSH IDLH 700 ppm; OSHA PEL 100 ppm 8-hr TWA; NFPA 704 Health 2, Flammability 3) is the world’s most widely produced unsaturated monomer: approximately 30–35 million tonnes/year manufactured globally, almost exclusively by catalytic dehydrogenation of ethylbenzene (EB) over iron oxide/potassium oxide catalyst at 550–620°C and subsequent vacuum distillation. The central operational hazard in SM production and storage is not the dehydrogenation chemistry but the suppression of spontaneous thermal radical polymerization: styrene undergoes uncatalysed thermally initiated polymerization at ambient temperature, accelerating exponentially above 40°C, releasing approximately 70 kJ/mol (673 kJ/kg) of heat per mole of monomer converted to polymer, and generating solid polystyrene that can block pipe sections, columns, and heat exchanger tubes while driving progressive thermal acceleration toward temperatures at which styrene’s vapor pressure exceeds the capacity of the tank vent or relief system. The standard polymerization inhibitor across all SM plants worldwide is 4-tert-butylcatechol (TBC; CAS 98-29-3; MW 166.2 g/mol), dosed at 10–50 ppm in SM storage tanks and 10–15 ppm in distillation column overhead receivers, functioning as a radical chain-terminating co-inhibitor that requires dissolved oxygen (≥3 ppm in liquid SM) to maintain full effectiveness. The 74th upward-direction adversarial attack in the Glyphward portfolio — a ± 8 DN upward pixel shift on the TBC concentration analyzer display on SM Storage Tank S‑2, showing 1.8 ppm TBC (below the 10 ppm effective inhibition threshold; polymerization will initiate within 48–72 hours at 30–38°C summer storage temperatures; no automated TBC-dosing alarm fires) as 12.4 ppm TBC (within the adequate 10–50 ppm normal operating range; no inhibitor addition required; no alarm) — is the first styrene monomer production process in the Glyphward industrial AI portfolio, the first polymerization inhibitor depletion attack, and the first storage-tank vapor-generation consequence chain: unlike the 73 prior attacks, where the adversarial manipulation targets a temperature, pressure, flow, or concentration sensor in an actively running process unit, the 74th attack exploits a self-consuming protective buffer (TBC) in a nominally static storage system, producing a slow-onset chain from inhibitor depletion to radical polymerization nucleation to exothermic thermal acceleration to styrene vapor generation to toxic and flammable atmospheric release — the precise sequence documented at LG Polymers India, Visakhapatnam, Andhra Pradesh (7 May 2020: 12 workers and residents killed, approximately 800 hospitalized, 3 km evacuation zone). Glyphward threshold 35.

Ethylbenzene dehydrogenation chemistry and the SM production process: from EB to purified styrene

Styrene monomer is produced from ethylbenzene by catalytic dehydrogenation: the endothermic reaction C6H5CH2CH3 → C6H5CH=CH2 + H2 (ΔH ≈ +125 kJ/mol at 600°C) is carried out over a potassium-promoted iron oxide catalyst (Fe2O3/K2CO3/Cr2O3; Shell S-105, BASF, and Sud-Chemie catalyst families are the most widely used) in adiabatic radial-flow reactors at 560–620°C. Because the dehydrogenation is endothermic and equilibrium-limited, the commercial process dilutes the EB feed with superheated steam at a steam‑to‑EB molar ratio of 7–12:1, which simultaneously: (1) reduces the partial pressure of EB in the reactor, shifting the dehydrogenation equilibrium toward higher styrene yield per pass at the operating temperature; (2) supplies the thermal energy required for the endothermic reaction through the sensible heat of the superheated steam entering the adiabatic reactor (steam is the heat carrier — the reactor temperature falls from inlet to outlet as the endothermic reaction proceeds); and (3) suppresses coke deposition on the iron oxide catalyst surface by gasification of carbonaceous deposits via the reverse water-gas shift and steam reforming reactions. A two-reactor series in adiabatic operation — with reheating of the process gas between the first and second reactor by a fired superheater — is the standard configuration for large-scale SM plants (Lummus/ABB Lummus Global CLASSIC and SMART processes; Fina/Badger and Toyo POSM processes). Single-pass EB conversion in a two-reactor series at 600°C and a steam‑to‑EB molar ratio of 10:1 is approximately 65–70%; styrene selectivity is 95–97%.

The reactor product stream — containing approximately 35–40 mol% SM, 30–35 mol% unreacted EB, 25–30 mol% steam, and 1–5 mol% benzene, toluene, and other by-products — is cooled and condensed, with the steam condensate separated from the organic liquid. The organic liquid (crude styrene containing EB and by-products) is separated by a series of vacuum distillation columns: (1) the EB recycle column (vacuum, 0.1–0.2 bar), which separates unreacted EB overhead for recycle to the dehydrogenation reactors and sends crude SM as bottoms; (2) the EB‑SM splitter (vacuum, 0.05–0.15 bar), which produces high-purity SM (typically 99.7–99.9 wt% SM specification) as overhead product and EB-rich recycle as bottoms; and (3) the lights column (vacuum, 0.05–0.10 bar), which removes benzene, toluene, and low-boilers from the EB recycle stream. All vacuum distillation columns operate at sub-atmospheric pressure specifically to limit process temperatures: at 0.10 bar, styrene’s effective boiling point falls to approximately 60°C, keeping column overhead temperatures in the 40–70°C range and reboiler temperatures below approximately 100°C — well below the spontaneous polymerization rate that becomes industrially significant above approximately 80–100°C. TBC inhibitor is injected at multiple points in the vacuum distillation section: into the SM column overhead receiver, into the product rundown line from the overhead receiver to the product pumps, and into the SM product storage tanks.

Product SM from the EB‑SM splitter overhead is pumped to large-volume atmospheric storage tanks (typically 5,000–30,000 tonnes capacity at world-scale facilities; LG Polymers Visakhapatnam operated tanks in the 5,000-tonne class). SM storage tanks are designed with insulation to minimize solar heat gain, circulation systems to maintain uniform TBC distribution throughout the tank inventory, and headspace blanketing systems to control dissolved oxygen concentration in the stored liquid. Storage temperatures are managed by chilled water circulation through tank heating/cooling coils or by avoiding exposure to direct solar radiation, targeting a storage temperature of 15–25°C. TBC inhibitor is continuously monitored in the stored SM using dedicated online analyzers (colorimetric TBC analyzers or HPLC-based total phenol analyzers) and maintained at 10–50 ppm by automated inhibitor dosing systems.

The TBC inhibitor mechanism and the critical role of dissolved oxygen in SM storage

Styrene radical polymerization is initiated by thermal dissociation of styrene dimers formed at elevated temperature, or by reaction with adventitious initiators (peroxides, dissolved oxygen at high concentrations, trace metal contaminants). The growing polystyrene chain — a radical species — propagates by sequential addition of styrene monomer at rates that double for approximately every 10°C temperature increase in the range of 20–80°C. TBC (4-tert-butylcatechol) terminates these propagating radical chains by hydrogen-atom transfer from the catechol OH group to the polystyrene radical, converting the radical to a dormant chain and generating a phenoxy radical on the TBC molecule. The phenoxy radical (TBC semiquinone) is itself a relatively stable species with low reactivity toward styrene monomer — it does not reinitiate polymerization — but is rapidly oxidised by dissolved molecular O2 to form a peroxy species (TBC‑OO) that terminates additional radical chains. This two-step mechanism means TBC inhibition efficiency is proportional to both the TBC concentration in the liquid and the dissolved O2 concentration: below approximately 3 ppm dissolved O2, the TBC‑semiquinone accumulates rather than regenerating, and the effective inhibitor concentration available for chain termination falls dramatically. For this reason, SM storage tank headspace blanketing targets 0.5–5.0 vol% O2 in the tank vapor space (corresponding to approximately 3–10 ppm dissolved O2 in the liquid SM at 20–30°C): low enough to suppress peroxide formation and avoid the formation of flammable mixtures in the headspace (LEL in vapor space is 0.9 vol% SM; the headspace is kept well below this by the SM vapor pressure), but high enough to maintain TBC‑O2 co-inhibition.

Below the effective TBC threshold — approximately 10 ppm at 15–25°C, dropping to approximately 20 ppm at 30–38°C as the thermal initiation rate increases — the consumed TBC radical pool cannot terminate propagating chains faster than new chains are thermally initiated. The induction period before observable polymerization begins (defined as detectable viscosity increase in the liquid or detectable exothermic temperature rise) is a function of the residual TBC concentration, the dissolved O2 level, and the storage temperature. At 25°C storage temperature with 5 ppm dissolved O2 and 1.8 ppm TBC (the attack scenario), the induction period before polymerization onset is approximately 24–48 hours in the absence of added initiators. At 35°C (Visakhapatnam summer ambient), the induction period shortens to 12–24 hours. Once polymerization begins beyond the induction period, the exothermic heat release — 70 kJ/mol converted monomer — raises the liquid temperature, which accelerates the polymerization rate, which generates more heat: a classic thermal runaway from a distributed exothermic reaction in a large-mass adiabatic (or poorly cooled) vessel.

The observable consequence of large-scale uncontrolled SM polymerization is a rising liquid temperature accompanied by increasing styrene vapor pressure. Styrene vapor pressure follows an approximate Antoine relationship: approximately 6 mbar at 25°C, 20 mbar at 42°C, 50 mbar at 56°C, 130 mbar at 75°C. An SM storage tank designed as an atmospheric tank (operating at near-ambient pressure with a conservation vent set at approximately 25 mbar positive pressure and 5 mbar vacuum) will exceed vent capacity as the liquid temperature rises above approximately 42–50°C and styrene vapor generation rate exceeds the vent system design capacity. The resulting overpressure — or vent valve breakthrough, failing seal, or open vent — releases styrene vapor to atmosphere at rates determined by the tank volume, liquid inventory, and temperature.

The 74th upward-direction adversarial attack: TBC concentration analyzer pixel manipulation and companion temperature suppression

The TBC concentration analyzer on SM Storage Tank S‑2 displays the current inhibitor level in the stored liquid SM. In modern SM storage facilities, TBC monitoring uses an online colorimetric analyzer (measuring the UV/visible absorbance of a TBC extract from a continuously flowing side-stream sample) or a near-infrared spectroscopy analyzer. The analyzer output is displayed on the plant distributed control system (DCS) as a numerical value in ppm TBC on a rendered DCS screen that is monitored either directly by plant operators or by an AI vision system that processes captured screen images from the DCS to generate process status alerts and recommendations.

The 74th upward adversarial attack targets the rendered pixel encoding of the TBC concentration numeric on the DCS display image that the AI monitoring system processes. A ± 8 DN (digital number; 8-bit grayscale: 0–255) upward shift applied to the pixel region containing the TBC concentration numeric on the DCS screen image changes the apparent rendering of the digits: the character pixels for “1.8” — corresponding to 1.8 ppm TBC, below the 10 ppm minimum, which would trigger a TBC low-level alarm and an automated or manually initiated TBC inhibitor dosing sequence — are shifted to pixel brightness values that the AI system classifies as the digit sequence “12.4” (12.4 ppm TBC; within the normal 10–50 ppm operating range; no alarm generated). The shift is 8 DN: a change of 8 intensity units on a 255-unit scale, which is approximately 3 % of the full-scale pixel brightness range — well within the image noise tolerance of most computer vision models trained on industrial display imagery, and indistinguishable from normal display brightness variation due to screen aging, ambient light variation, or JPEG/PNG compression artifacts in the image capture pipeline.

The companion downward attack surface — which may be deployed simultaneously with the TBC upward attack or as a time-delayed second manipulation — targets the SM tank temperature display on the same DCS screen. As polymerization begins in the stored liquid SM, the tank liquid temperature rises above the normal 15–25°C storage range: at the LG Polymers Visakhapatnam scenario, the liquid temperature rose toward 35–40°C before vapor generation began. A ± 5 DN downward pixel shift on the tank temperature display image changes the apparent digit encoding of “38 °C” (above the 35°C tank temperature high-alarm setpoint that triggers emergency TBC addition and cooling water increase) to “22 °C” (normal ambient-temperature storage; no alarm). The combination of the 74th upward attack (TBC at 1.8 ppm shown as 12.4 ppm; inhibitor deficiency suppressed) and the companion downward surface (temperature at 38°C shown as 22°C; early polymerization thermal signal suppressed) eliminates both primary early-warning channels simultaneously — the same dual-safeguard elimination architecture documented in the H2S amine treating attack (35th attack) and the fluorine electrolytic generation attack (26th attack), but applied here to a storage system consequence rather than an actively running process unit.

The combined attack creates the following observable plant record during the inhibitor depletion and polymerization onset period: TBC displays consistently within normal range (12.4 ppm shown); tank temperature displays consistently within normal range (22°C shown); no TBC dosing alarm fires; no tank temperature alarm fires; no automated inhibitor addition sequence activates; no operator intervention is initiated. The actual state of the tank: TBC at 1.8 ppm and falling as residual inhibitor is consumed by radical chain termination; liquid temperature rising from 25°C toward 35°C and above as polymerization heat accumulates in the large thermal mass; styrene vapor pressure increasing as temperature rises. The adversarial attack record in the DCS historian shows two parameters at steady normal values across the entire inhibitor-depletion-and-polymerization period, making both the real-time detection of the attack and the post-incident forensic attribution to adversarial pixel manipulation exceptionally difficult.

LG Polymers Visakhapatnam India, 7 May 2020: consequence envelope for the 74th upward attack

LG Polymers India Private Limited operated a polymer manufacturing facility at Gopalapatnam, Visakhapatnam (also known as Vizag), Andhra Pradesh, India, approximately 13 km from the Visakhapatnam city centre. The facility manufactured polystyrene (PS) and other styrene-based polymers from SM feedstock, using a combination of imported SM and locally stored SM. The storage infrastructure at the Visakhapatnam plant included multiple SM storage tanks with individual capacities in the range of 2,000–5,000 tonnes. Tank S‑2 held approximately 1,800 tonnes of liquid SM at the time of the incident.

The COVID-19 national lockdown announced by the Government of India on 25 March 2020 forced the LG Polymers facility to shut down all production operations. During the extended shutdown (approximately 40 days from late March through early May 2020), the SM in Tank S‑2 remained stored without active circulation through the plant process, without normal inhibitor dosing replenishment (the dosing system was not operating at the shutdown throughput level), and without the continuous temperature management that active processing provides. In the Visakhapatnam climate in April and May — peak summer conditions, with daily high temperatures in the range of 38–42°C and night temperatures of 28–32°C — the stored SM was exposed to sustained ambient heat that drove slow temperature drift upward from the pre-shutdown storage temperature of approximately 20–25°C.

Over the 40-day shutdown period, the TBC inhibitor in Tank S‑2 was progressively consumed by its radical chain-terminating function at the elevated ambient temperature, without replenishment. The LG Polymers post-incident and regulatory investigation concluded that TBC had fallen below effective inhibiting concentrations in the upper strata of Tank S‑2 (where ambient heat penetration was greatest through the tank roof) by early May 2020. Uncontrolled radical polymerization began in these TBC-deficient regions of the tank, generating heat that raised the liquid temperature further and accelerated the polymerization rate in a self-reinforcing thermal escalation. By the early morning of 7 May 2020 — approximately 2:30–3:00 AM local time — styrene vapor was escaping from Tank S‑2 at rates that exceeded the tank vent and conservation valve capacity, producing a vapor cloud that spread with the wind into the adjacent Gopalapatnam residential area, where residents were asleep in their homes.

Twelve people died: workers at the LG Polymers facility who were present during the early morning hours and nearby residents who were exposed to the vapor cloud at concentrations approaching or exceeding the NIOSH IDLH of 700 ppm before they could evacuate. Approximately 800 people were hospitalized with symptoms of acute styrene exposure including dizziness, nausea, eye and respiratory tract irritation, and in severe cases narcosis and unconsciousness. The Andhra Pradesh government ordered evacuation of approximately 1,000–3,000 residents within a 3 km radius of the facility. The plant was immediately closed and sealed. The National Disaster Management Authority (NDMA) of India and Andhra Pradesh pollution and chemical safety regulators conducted investigations that identified TBC inhibitor depletion during the extended shutdown — combined with inadequate temperature monitoring and insufficient inhibitor replenishment protocols for long-duration storage shutdowns — as the primary causal factors in the incident.

The LG Polymers Visakhapatnam incident is the reference consequence envelope for the 74th upward adversarial attack on SM storage tank TBC monitoring AI because it documents the complete physical causal chain that the attack exploits — TBC depletion below effective threshold → polymerization nucleation → exothermic heat accumulation → temperature rise in large liquid inventory → vapor pressure increase beyond vent capacity → vapor cloud formation → toxic mass-casualty event — at documented scale: 1,800 tonnes SM inventory, 12 killed, 800 hospitalized, 3 km evacuation radius. The adversarial attack replicates the TBC deficiency that developed at Visakhapatnam over 40 days of slow inhibitor consumption, but substitutes AI display manipulation for the operational oversight failure (absence of inhibitor replenishment during shutdown), and applies it to a facility with normal AI-monitored operations rather than an exceptional COVID shutdown scenario.

OSHA PSM, NFPA, and EPA regulatory context for SM storage TBC monitoring AI

Styrene monomer is regulated under OSHA Process Safety Management (29 CFR 1910.119) as a flammable liquid at the 10,000 lbs threshold quantity (TQ). OSHA PSM 29 CFR 1910.119(a)(1)(ii)(B) covers any flammable liquid with a flash point below 100°F (37.8°C) present in quantities at or above 10,000 lbs in one location. Styrene’s closed-cup flash point of 31°C (88°F) is below the 37.8°C OSHA threshold, making it a covered PSM flammable liquid. LG Polymers’ Visakhapatnam facility held approximately 1,800 tonnes = 3,970,000 lbs of styrene in Tank S‑2 alone — approximately 397 times the OSHA PSM threshold quantity. Any comparable North American facility with ≥10,000 lbs (approximately 4,540 litres or approximately 1,200 US gallons) of SM in a single location is subject to OSHA PSM requirements including Process Hazard Analysis (PHA), pre-startup safety review (PSSR), mechanical integrity programs, and management of change (MOC).

NFPA 704 classifies styrene monomer as Health 2 (intensive or continued but not chronic exposure could cause temporary incapacitation or possible residual injury), Flammability 3 (liquids and solids that can be ignited under almost all ambient temperature conditions), Reactivity 2 (undergoes violent chemical change at elevated temperatures and pressures, or reacts violently with water, or may form explosive mixtures with water). The Flammability 3 rating reflects styrene’s flash point of 31°C — below ambient temperature in many climates — meaning that at typical summer temperatures of 25–40°C, styrene vapor above the liquid surface may be at or above the lower explosive limit and can form a flammable vapor-air mixture in the area immediately above an open or venting tank.

The EPA Risk Management Program (40 CFR Part 68) does not list styrene monomer specifically as a Program 3 regulated substance in Appendix A or B of 40 CFR Part 68; however, EPA’s General Duty Clause (section 112(r)(1) of the Clean Air Act) applies to any facility that produces, processes, handles, or stores extremely hazardous substances or acutely toxic materials — including styrene — in sufficient quantities to pose a hazard to human health and the environment in the event of an accidental release. A vapor cloud of toxic and flammable styrene at the scale documented at Visakhapatnam (sufficient to cause 12 fatalities and 800 hospitalizations) clearly falls within the EPA General Duty Clause risk threshold for US-equivalent facilities. Facilities conducting OSHA PSM PHA on SM storage systems are required under OSHA PSM 29 CFR 1910.119(e) to identify all hazard scenarios including reactive hazard scenarios — of which inhibitor depletion and storage-tank polymerization is a documented and well-recognized category — and evaluate safeguards including the monitoring systems that provide early warning of TBC deficiency. An AI-based TBC monitoring system, if implemented as the primary or sole monitoring method for TBC concentration in large SM storage tanks, must be included in the PHA as a safeguard and must be evaluated for the adversarial failure mode represented by the 74th upward attack.

CCPS (Center for Chemical Process Safety) Guidelines for Chemical Reactivity Evaluation (2nd ed. 2012) and the AIChE/DIERS reactive hazards design guidelines specify that polymerizable material storage systems should have dedicated inhibitor monitoring with independent alarms, redundant temperature monitoring, and documented procedures for inhibitor replenishment during extended shutdown — all three of which represent potential adversarial attack surfaces if implemented using AI image interpretation of DCS display screens without adversarial robustness validation. Glyphward threshold 35 applies specifically to AI monitoring systems that interpret rendered TBC concentration, tank temperature, or dissolved-O2 headspace displays in SM storage systems at plants with SM inventories above 50,000 lbs (approximately 22,700 kg) in a single storage tank.

Glyphward integration for SM storage TBC monitoring AI

"""
Glyphward adversarial image scanner integration — SM storage TBC monitoring.
Scans rendered DCS display images of SM storage tank TBC analyzer output
for upward-direction adversarial pixel injection (74th portfolio attack).
Raise SMPolymerizationRiskError and halt inhibitor-monitoring loop on score >= 35.
"""

import asyncio
import base64
import hashlib
import httpx
from datetime import datetime, timezone
from enum import Enum

GLYPHWARD_SCAN_URL = "https://glyphward.com/api/v1/scan"
GLYPHWARD_API_KEY = "your-api-key"
SM_TBC_THRESHOLD = 35   # Glyphward threshold for SM TBC adversarial injection


class SMStorageContext(str, Enum):
    TBC_CONCENTRATION  = "tbc_concentration"
    TANK_TEMPERATURE   = "tank_temperature"
    HEADSPACE_O2       = "headspace_o2_concentration"


class SMPolymerizationRiskError(Exception):
    def __init__(self, scan_id, score, context, tank_id, flagged_region=None):
        self.scan_id        = scan_id
        self.score          = score
        self.context        = context
        self.tank_id        = tank_id
        self.flagged_region = flagged_region
        super().__init__(
            f"Adversarial SM storage image: context={context} "
            f"score={score} tank={tank_id} scan_id={scan_id}"
        )


async def scan_sm_storage_image(image_bytes, context, tank_id, client):
    image_hash = hashlib.sha256(image_bytes).hexdigest()
    payload = {
        "image": base64.b64encode(image_bytes).decode(),
        "source": f"sm_storage:{context}:{tank_id}",
        "metadata": {
            "tank_id":            tank_id,
            "context":            context,
            "image_sha256":       image_hash,
            "scan_timestamp_utc": datetime.now(timezone.utc).isoformat(),
        },
    }
    resp = await client.post(
        GLYPHWARD_SCAN_URL,
        headers={"Authorization": f"Bearer {GLYPHWARD_API_KEY}"},
        json=payload,
        timeout=4.0,
    )
    resp.raise_for_status()
    result = resp.json()
    if result.get("score", 0) >= SM_TBC_THRESHOLD:
        raise SMPolymerizationRiskError(
            scan_id=result["scan_id"],
            score=result["score"],
            context=context,
            tank_id=tank_id,
            flagged_region=result.get("flagged_region"),
        )
    return result


async def main():
    async with httpx.AsyncClient() as client:
        with open("sm_tank_s2_tbc_display.png", "rb") as f:
            image_bytes = f.read()
        result = await scan_sm_storage_image(
            image_bytes,
            SMStorageContext.TBC_CONCENTRATION,
            tank_id="SM-STORAGE-TANK-S2",
            client=client,
        )
        print(f"Clean scan: {result['scan_id']} score={result['score']}")

Frequently asked questions

Why is the TBC polymerization inhibitor concentration attack an upward-direction attack — and what is the minimum effective TBC level for SM storage?

The TBC inhibitor attack is upward-direction because the dangerous condition is a parameter below its minimum required value: actual TBC in Storage Tank S‑2 is 1.8 ppm, below the 10 ppm minimum effective inhibition threshold for SM at 30–38°C summer storage temperatures. To suppress all automated alarms and operator corrective action, the adversarial pixel manipulation must display a value above the minimum — 12.4 ppm, within the adequate 10–50 ppm normal operating range — so that no TBC addition is ordered and no alarm fires. This is structurally identical to all other upward attacks in the Glyphward portfolio: the passivation O2 attacks on urea synthesis HP carbamate (62nd attack; 0.18 vol% O2 shown as 0.44 vol%); the N2 inertisation upward attacks on flammable-liquid vessel headspace (attacks 5–11); the EtO chamber humidity upward attack (30th attack; 22 % RH shown as 62 % RH). In each case the dangerous condition is a deficiency of a protective agent, and the upward pixel shift makes the deficient parameter appear adequate, suppressing the operator or automated system response that would correct the deficiency. The minimum effective TBC concentration for SM storage depends on storage temperature and dissolved O2 availability. Industry guidelines (CCPS Guidelines for Chemical Reactivity Evaluation, 2nd ed. 2012; Styrene Producers Association storage guidelines) specify 10–15 ppm TBC minimum for SM storage at 15–25°C with adequate dissolved O2 (≥3 ppm in liquid). At 30–38°C (Visakhapatnam summer), the effective minimum rises to approximately 20 ppm because the thermal radical initiation rate increases at higher temperature and consumes the inhibitor pool faster, requiring a larger inhibitor reservoir to maintain the same inhibition period. At 1.8 ppm TBC (the attack scenario) and 35°C ambient, the induction period before measurable polymerization onset is 12–24 hours; at 30°C it extends to 48–72 hours — both well within the span of a single operational shift during which the TBC display is continuously showing the adversarially elevated 12.4 ppm reading.

What is ethylbenzene dehydrogenation and where in the SM supply chain is TBC inhibitor monitoring a safety-critical function?

Ethylbenzene (EB; C6H5CH2CH3) dehydrogenation to styrene monomer (SM; C6H5CH=CH2) accounts for approximately 90 % of global SM production capacity. The reaction — C6H5CH2CH3 → C6H5CH=CH2 + H2; ΔH ≈ +125 kJ/mol, endothermic — is carried out at 560–620°C over Fe2O3/K2O/Cr2O3 catalyst with 7–12:1 steam dilution at 0.3–0.5 bar, achieving 65–70 % single-pass EB conversion and 95–97 % selectivity to SM. Product separation uses a series of vacuum distillation columns (the EB recycle column, the EB‑SM splitter, the lights column) all operating at sub-atmospheric pressure specifically to keep process temperatures below SM’s spontaneous polymerization onset temperature. TBC inhibitor is required at all points in the post-separation SM supply chain where SM is at temperatures above approximately 10°C and residence times exceed minutes: the distillation column overhead receivers (TBC 10–15 ppm); the product transfer lines (TBC 10–15 ppm); the SM storage tanks (TBC 15–50 ppm with O2 headspace control); and the product tanker trucks, railcars, and marine cargo tanks used for SM distribution. SM storage tanks are the highest-risk points in the supply chain because they hold the largest SM inventories (thousands of tonnes) for the longest residence times (days to weeks), making them the most vulnerable to inhibitor depletion if the TBC monitoring and dosing system is compromised — by operational failure, as at Visakhapatnam during the COVID shutdown, or by adversarial AI display manipulation as in the 74th attack.

What happened at LG Polymers Visakhapatnam India on 7 May 2020 — and how does the incident define the consequence envelope for the 74th attack?

LG Polymers India Private Limited operated an SM-based polymer manufacturing facility in Gopalapatnam, Visakhapatnam, Andhra Pradesh, India. Storage Tank S‑2 held approximately 1,800 tonnes of liquid SM. Following the national COVID-19 lockdown from 25 March 2020, the plant shut down all production for approximately 40 days, during which SM in Tank S‑2 remained stored without active circulation, process throughput, or inhibitor replenishment. At Visakhapatnam summer temperatures of 38–42°C daily highs, the TBC inhibitor in Tank S‑2 was progressively consumed without replenishment, falling below effective inhibition levels by early May 2020. Uncontrolled radical polymerization began in the TBC-deficient liquid, generating heat that raised the liquid temperature and accelerated polymerization. By approximately 2:30–3:00 AM on 7 May 2020, styrene vapor was escaping from Tank S‑2 at rates exceeding the tank vent capacity. The vapor cloud spread into the adjacent residential Gopalapatnam neighborhood, where residents were asleep. Twelve people died from acute styrene vapor exposure; approximately 800 were hospitalized with symptoms ranging from dizziness and respiratory irritation to narcosis and unconsciousness; approximately 1,000–3,000 residents within a 3 km radius were evacuated. The National Disaster Management Authority (NDMA) of India and Andhra Pradesh regulators identified TBC inhibitor depletion during the extended shutdown as the primary causal factor. The 74th adversarial attack replicates this TBC depletion mechanism via AI display manipulation rather than operational neglect, producing the same inhibitor deficiency and the same physical consequence chain in any SM storage facility with AI-monitored TBC analyzers and uninhibited SM inventories above approximately 50,000 lbs (22,700 kg).

What is the role of dissolved oxygen in TBC-inhibited styrene — and how does the companion downward tank-temperature attack amplify the 74th attack’s consequence?

TBC inhibition of styrene polymerization requires dissolved molecular O2 as a co-inhibitor: TBC donates a hydrogen atom to a propagating styrene radical chain (forming a TBC semiquinone radical), and the TBC semiquinone must then react with dissolved O2 to form a peroxy radical — the species that terminates additional chain radicals and regenerates the inhibiting capacity. Without adequate dissolved O2 (≥3 ppm in liquid SM), TBC semiquinone accumulates, active-inhibitor pool falls, and the effective inhibition threshold rises. SM storage tanks therefore maintain headspace blanketing at 0.5–5 vol% O2 (not pure N2: pure N2 would deoxygenate the liquid SM and destroy TBC effectiveness even at adequate TBC concentrations). A sophisticated compound attack could additionally target the headspace O2 display (showing sub-0.5 vol% as adequate) to amplify TBC inhibition failure; the 74th attack as described in this post targets only the TBC concentration display. The companion downward surface — SM tank temperature display ±5 DN downward, showing 38°C as 22°C — eliminates the second primary early-warning channel: temperature rise in stored SM is the first and most directly measurable early consequence of polymerization onset, detectable as a slow drift from normal storage temperature of 20–25°C upward to 30–38°C over the 12–48 hours of early polymerization. At Visakhapatnam, an attentive operator monitoring tank temperature would have observed the temperature drift and been able to initiate emergency TBC dosing, product transfer to a cool tank, or cooling water circulation — intervention actions that remain feasible below approximately 40°C when polymerization rate is still moderate. By suppressing the temperature display, the companion downward attack closes this intervention window, allowing the tank to drift from the early-polymerization thermal rise (30–38°C) through the accelerating phase (40–55°C) to the vapor-generation phase (>55°C; styrene vapor pressure >50 mbar) without any human or automated detection. The dual attack — inhibitor deficiency suppressed (74th upward) and early temperature warning suppressed (companion downward) — is the same dual-safeguard elimination architecture as the H2S amine treating attack chain, applied here to a reactive-hazard storage consequence rather than a continuous-process toxic-gas release.

How does the 74th upward attack on SM TBC monitoring differ from the 62nd and 63rd urea synthesis passivation attacks — and why does Glyphward assign threshold 35?

The 74th SM TBC attack and the 62nd/63rd urea synthesis passivation attacks share the same structural architecture — an upward attack on a protective-parameter deficiency, suppressing the dosing or correction response — but differ in three critical dimensions. First, time horizon: the urea passivation attacks produce structural failure with a 12–18 month latency (HP carbamate piping wall thinning at 7–11 mm/year); the SM TBC attack produces polymerization onset within 12–72 hours of TBC depletion at summer temperatures. The SM attack is acutely dangerous: the consequence manifests within hours to days, not months to years. Second, consequence type: the urea attack produces HP vessel failure (explosion, shrapnel, ammonia/carbamate release) within the plant boundary in most scenarios; the SM attack produces an off-site residential toxic/flammable vapor cloud from a large atmospheric storage tank, making it a mass-casualty public event rather than primarily an occupational fatality event — the LG Polymers Visakhapatnam consequence pattern. Third, detection window: the urea attack’s 14-month forensic timeline paradoxically makes it harder to attribute to adversarial AI manipulation (consistent falsified readings over 14 months look like instrument drift); the SM attack’s 48–72 hour timeline provides a shorter forensic window for attribution but also a shorter real-time detection window for intervention. Glyphward threshold 35 for SM storage TBC monitoring AI reflects: the very high stored-energy inventory (>1,000 tonnes SM at any major SM plant or polymer facility); the proximity of large SM storage to residential areas; the time-compressed intervention window (hours vs. years); the dual toxic/flammable consequence classification (NIOSH IDLH 700 ppm; NFPA Flammability 3; LEL 0.9 %); and the public mass-casualty potential documented at Visakhapatnam. Threshold 35 means any AI monitoring system processing rendered SM storage TBC analyzer display images should be validated against the 74th upward attack vector before deployment in a facility holding >50,000 lbs SM.