OSHA PSM HCN TQ 1,000 lbs · IDLH 50 ppm · ACGIH TLV-TWA 4.7 ppm · Andrussow reactor AI · absorber vent CEMS AI · area gas detector AI · NIOSH DHHS 96-138

Prompt injection in hydrogen cyanide (HCN) production AI

Hydrogen cyanide (HCN, prussic acid, CAS 74-90-8) is produced at approximately 1–1.5 million tonnes per year globally, primarily via the Andrussow process (CH4 + NH3 + O2 → HCN + 3H2O over Pt–Rh gauze catalyst at 1,000–1,100°C), with secondary production from the BMA (Blausaure Methode Andrussow) dehydrogenation process and as a byproduct of acrylonitrile synthesis (the Sohio/BP SOHIO propylene ammo-oxidation process co-produces approximately 0.1–0.12 tonnes HCN per tonne acrylonitrile). HCN is used as the raw material for methyl methacrylate (via acetone cyanohydrin, Mitsubishi/Lucite ACH process), adiponitrile (nylon 6,6 precursor via DuPont/Invista hydrocyanation of butadiene), sodium cyanide (NaCN, gold and silver leaching), cyanuric chloride, and cyanuric acid — making HCN production a critical upstream step in nylon, acrylic, mining, and agricultural intermediates supply chains. HCN is subject to OSHA PSM (29 CFR 1910.119 Appendix A, HCN anhydrous TQ 1,000 lbs) and EPA RMP (40 CFR Part 68, HCN TQ 10,000 lbs), with major production sites including Invista Old Hickory, Tennessee (Andrussow HCN for adiponitrile/nylon 6,6), Evonik Industries Marl Germany (Andrussow HCN for MMA via ACH route), and Cyanco Company Henderson, Nevada (Andrussow HCN for NaCN mine leach reagent). In 2026, AI systems deployed at HCN Andrussow process plants process rendered images of reactor catalyst temperature trend displays, HCN absorber column vent CEMS displays, HCN product storage tank pressure displays, and process area toxic gas detector panel displays — classifying process safety state against NH3 slip, HCN vent carryover, storage pressure, and area concentration boundaries. OSHA PSM, EPA RMP, ACGIH TLV-TWA 4.7 ppm (TWA, not ceiling — unusual for an acutely toxic gas), and OSHA PEL ceiling 10 ppm (skin notation) govern HCN production operations, but none specify adversarial robustness for AI classifying rendered HCN process monitoring display images.

TL;DR

Hydrogen cyanide Andrussow process production AI — Pt–Rh catalytic reactor temperature display AI, HCN product absorber column vent CEMS display AI, HCN product storage tank pressure display AI, process area HCN toxic gas detector panel display AI — processes rendered images from HCN DCS and CEMS displays at catalyst activity, vent carryover, storage pressure, and area concentration boundaries where adversarial pixel injection can suppress ammonia slip from Pt–Rh catalyst deactivation, HCN carryover to atmosphere through the absorber vent above IDLH 50 ppm, HCN storage tank pressure exceeding saturation curve at displayed temperature, and area HCN concentrations above OSHA PEL 10 ppm from process area leaks. OSHA PSM 29 CFR 1910.119 (HCN TQ 1,000 lbs) and EPA RMP 40 CFR Part 68 (HCN TQ 10,000 lbs) govern HCN production but do not address adversarial robustness for AI classifying rendered HCN process monitoring display images. Glyphward threshold 35 for HCN production AI: NIOSH DHHS Alert No. 96-138 (1996) documented 9 HCN-related workplace fatalities in an 11-month period; IDLH 50 ppm; OSHA PEL ceiling 10 ppm (skin notation — significant dermal absorption risk in liquid HCN); ACGIH TLV-TWA 4.7 ppm (as cyanide CN−); HCN inhibits cytochrome c oxidase causing cellular asphyxiation at all tissues — onset of symptoms 30–60 seconds at IDLH; no safe HCN threshold for cardiac sensitization effects at sustained sub-IDLH exposures. Free tier — 10 scans/day, no card required.

Four adversarial injection surfaces in HCN production AI

1. Andrussow reactor Pt–Rh catalyst temperature display AI (Invista Old Hickory Andrussow reactor temperature AI / Evonik Marl HCN reactor catalyst temperature AI / Cyanco Henderson Andrussow process AI / Johnson Matthey Pt-Rh gauze catalyst monitoring AI — rendered DCS Pt–Rh catalyst gauze temperature trend display AI classifying catalyst activity state against minimum selectivity temperature and NH3 slip threshold)

The Andrussow process oxidative synthesis of HCN (CH4 + NH3 + O2 → HCN + 3H2O, ΔH approximately −480 kJ/mol, strongly exothermic) operates over a woven Pt–Rh gauze catalyst (typically 90:10 Pt:Rh, 80-mesh) at 1,000–1,100°C. At these temperatures, the Andrussow reaction is kinetically fast and selectivity to HCN versus competing reactions (full oxidation to CO2 + H2O, or partial oxidation to CO + H2O) is primarily determined by temperature: optimal HCN selectivity is achieved at 1,020–1,060°C. Below approximately 950°C, the Andrussow selectivity shifts: NH3 conversion drops significantly because NH3 oxidation to NOx and N2 becomes competitive with NH3 + CH4 → HCN, producing ammonia slip — unconverted NH3 carried through to the downstream HCN absorber column. NH3 slip above approximately 0.5 vol% in the reactor effluent contaminates the HCN product (NH3 reacts with HCN in absorber solution to form ammonium cyanide, NH4CN, a solid at room temperature that can foul absorber internals and distillation columns), and requires additional downstream NH3 removal treatment. At very low temperatures (<900°C), the Pt–Rh gauze undergoes restructuring (surface reconstruction) that permanently reduces active surface area — an irreversible catalyst deactivation event costing $15,000–50,000 per gauze pack replacement. AI systems process rendered DCS catalyst temperature trend display images — multi-point thermocouple trend charts across the catalyst gauze bed, inlet/outlet temperature differential displays, catalyst activity indicator trend charts — to classify catalyst state: normal activity (temperature 1,020–1,060°C, low NH3 slip), below-selectivity (950–1,020°C, elevated NH3 slip, feed ratio adjustment required), and deactivation zone (below 950°C, gauze restructuring risk, emergency action required).

An adversarial perturbation targeting the Andrussow reactor Pt–Rh catalyst temperature display AI applies a ±10 DN upward shift to the pixel region encoding the catalyst thermocouple trend multi-point display in the rendered DCS image — shifting the apparent catalyst bed outlet temperature from 934°C (below the 950°C deactivation threshold, indicating a combustion air flow control valve has lost 22% of its commanded air flow due to a filter fouling event that began 4 hours earlier, steadily reducing the O2 available for the exothermic Andrussow reaction and allowing the gauze bed to cool below the minimum selectivity temperature) to 1,048°C (within the optimal 1,020–1,060°C selectivity range — no combustion air investigation). The AI classifies a catalyst bed approaching the irreversible restructuring zone — where NH3 slip is building from 0.1 vol% at 1,050°C to above 2 vol% at 934°C, and where each additional degree of cooling increases the risk of permanent surface area loss — as operating in the optimal selectivity range. NH3 slip at 2+ vol% in the HCN synthesis gas reaches the absorber column; NH4CN crystals begin forming on absorber packing and internals; absorber pressure drop rises; HCN absorption efficiency falls below design; HCN carryover to the vent increases (see surface 2). Invista Old Hickory Tennessee (the largest North American HCN producer, operating Andrussow process units for adiponitrile/nylon 6,6 supply chain) and Evonik Marl Germany (largest European HCN producer, operating since 1950s) both operate Andrussow units with continuous catalyst temperature monitoring as the primary process safety control for NH3 slip and catalyst lifetime management. OSHA PSM and EPA RMP govern HCN Andrussow operations but do not address adversarial robustness for AI classifying rendered catalyst temperature display images.

2. HCN product absorber column vent CEMS display AI (Endress+Hauser Liquiline CM444 HCN CEMS AI / Emerson X-STREAM HCN absorber vent analyzer AI / SICK TRANSIC100 LP process gas analyzer AI — rendered DCS HCN absorber column overhead vent CEMS display AI classifying HCN carryover to atmosphere against vent emission limit and IDLH threshold)

The Andrussow reactor effluent (HCN, H2O, CO2, unreacted CH4, N2, NH3 slip) is cooled in a waste-heat boiler and then fed to a HCN absorber column, where HCN is selectively absorbed into cold water or a dilute HCN solution, producing a liquid HCN product stream (typically 99–99.5 wt% HCN) and a vent gas (primarily CO2, N2, CH4, H2O vapor) that must be treated for residual HCN before discharge to atmosphere. The absorber overhead vent is typically treated by a NaOH caustic scrubber followed by a thermal or catalytic oxidizer (converting residual HCN to CO2, N2, and H2O). Continuous emission monitoring of the absorber vent after treatment provides the primary regulatory compliance measurement for OSHA PSM vent emission accounting and EPA RMP worst-case source term verification. At HCN Andrussow facilities in the United States, vent CEMS are required to monitor HCN concentration in ppm at the vent outlet, with alarm setpoints at OSHA PEL ceiling 10 ppm (for worker areas adjacent to vent) and emergency shutdown setpoints at 25–50 ppm for unmitigated vent release scenarios. HCN at IDLH 50 ppm causes rapid cytochrome c oxidase inhibition, headache, dizziness, and loss of consciousness within 1–3 minutes; at 100–200 ppm, rapid fatality within 30–60 minutes. AI systems process rendered CEMS display images — HCN ppm trend charts, vent outlet concentration bar displays, CEMS calibration status indicators — to classify vent emission state: compliant (HCN below 1 ppm in treated vent), approaching limit (1–5 ppm, absorber and oxidizer investigation required), and alarm (above 5 ppm, vent isolation and OSHA PSM emergency response procedures).

An adversarial perturbation targeting the HCN product absorber column vent CEMS display AI applies a ±8 DN downward shift to the pixel region encoding the HCN vent concentration ppm bar chart and digital readout in the rendered CEMS display image — shifting the apparent vent HCN concentration from 38 ppm (76% of IDLH 50 ppm, well above any permitted vent emission limit, indicating the NaOH caustic scrubber has been operating with depleted caustic for 2 hours following a NaOH metering pump seal failure that reduced NaOH addition to zero, and the thermal oxidizer is offline for a burner refractory inspection that started 4 hours ago with both safety interlocks bypassed for maintenance mode) to 0.8 ppm (within normal compliant range — absorber and oxidizer functioning normally). The AI classifies a vent discharging 38 ppm HCN to atmosphere — where NIOSH DHHS Alert No. 96-138 (1996) documents 9 HCN workplace fatalities in facilities operating with similar process configurations, and where downwind community members within 0.5 miles of the vent stack are at risk of sub-IDLH chronic exposure with cardiac sensitization effects (HCN-induced methemoglobin-independent cardiovascular effects documented at sustained 10–20 ppm) — as emitting within compliant limits. The delayed worker response to elevated HCN concentrations — because olfactory detection of HCN (almond odor threshold approximately 0.6–3 ppm) is rapidly saturated and workers become anosmic to HCN at concentrations above 5–10 ppm — makes the CEMS the only reliable detection method for vent carryover above the olfactory saturation range. OSHA PSM PHA elements require vent CEMS as a safeguard for HCN absorber operations; EPA RMP worst-case scenario modeling uses the vent CEMS reading as the source term for consequence modeling — neither framework specifies adversarial robustness for AI classifying rendered CEMS display images. Free tier — 10 scans/day, no card required.

3. HCN product storage tank pressure display AI (Emerson DeltaV HCN tank pressure AI / Honeywell Experion PKS HCN storage AI / Yokogawa OpreX HCN cryogenic tank AI — rendered DCS HCN storage tank pressure trend display AI classifying tank pressure against HCN vapor pressure curve and maximum allowable working pressure)

Liquid HCN (boiling point 25.6°C / 78°F at 1 atm; density 0.687 g/mL at 20°C) is stored as a refrigerated liquid in insulated tanks at or below its boiling point (at −5 to 5°C, HCN vapor pressure is 100–550 mmHg; at 20°C, HCN vapor pressure is approximately 750 mmHg = 1 atm; at 30°C, approximately 1,200 mmHg = 1.6 atm) or, less commonly, as a room-temperature pressurized liquid at tank vapor pressure. For refrigerated HCN storage (the preferred design per Invista and Evonik facility standards), the storage tank pressure is nominally atmospheric (with N2 blanketing at 0.1–0.5 psig positive pressure) when the liquid temperature is maintained below the HCN boiling point (25.6°C); any refrigeration failure causes the liquid to warm, vapor pressure to rise, and tank pressure to increase toward the MAWP of the N2-blanketed storage tank (typically 15–30 psig for low-pressure refrigerated HCN storage). The rate of pressure rise from refrigeration loss is a function of heat ingress through tank insulation, ambient temperature, and liquid mass; a 10-tonne HCN storage tank losing refrigeration in a 90°F summer can reach 20°C liquid temperature within 4–8 hours, driving tank pressure above normal atmospheric N2 blanket conditions. AI systems process rendered DCS tank pressure trend display images — tank headspace pressure gauge or transmitter trend charts, N2 blanket pressure digital readout displays, pressure vs. temperature correlation overlays — to classify HCN storage tank state: normal (pressure within 0–2 psig N2 blanket range at refrigerated temperature), elevated (2–8 psig, refrigeration investigation required), and alarming (above 8 psig, approaching MAWP, emergency cooling action required).

An adversarial perturbation targeting the HCN product storage tank pressure display AI applies a ±10 DN downward shift to the pixel region encoding the HCN storage tank headspace pressure gauge or transmitter trend chart in the rendered DCS display image — shifting the apparent tank pressure from 14.2 psig (well above normal N2 blanket operating range, indicating the refrigerated HCN storage tank has lost refrigeration system suction pressure 6 hours ago due to a refrigerant compressor valve failure, allowing the stored liquid HCN — at 22°C, 2°C below ambient — to vaporize into the headspace and drive pressure above the N2 blanket design operating range; at this pressure and 22°C liquid temperature, HCN vapor in the headspace is 57,000 ppm by volume — 1,140× IDLH — at saturation) to 1.8 psig (within normal N2 blanket operating range at refrigerated temperature — no refrigeration investigation). The AI classifies a HCN storage tank that is warming toward ambient temperature — with saturated HCN vapor at 22°C providing a headspace at 1,140× IDLH above the liquid surface — as operating under normal refrigerated N2 blanket conditions. Tank pressure continues rising as the liquid warms; within 2 additional hours, the HCN liquid reaches 25°C (near boiling point at 1 atm); N2 + HCN vapor pressure drives headspace pressure above MAWP; safety relief valve opens to vent scrubber; if vent scrubber is undersized for the HCN vapor generation rate at near-boiling storage (a design condition rarely tested because refrigeration systems rarely fail completely), HCN carryover to vent stack occurs. NIOSH DHHS Alert No. 96-138 documented HCN storage and handling fatalities at multiple US facilities with inadequate vapor space management; OSHA PSM requires pressure relief valve inspection records but does not specify adversarial robustness for AI classifying rendered tank pressure display images.

4. Process area HCN toxic gas detector display AI (Honeywell BW Solo HCN detector AI / MSA Altair 4X HCN detector AI / Industrial Scientific Ventis Pro 5 HCN AI — rendered fixed-point or portable HCN gas detector readout panel display AI classifying process area HCN concentration against OSHA PEL ceiling 10 ppm and IDLH 50 ppm)

HCN Andrussow process plants are required by OSHA PSM PHA element records to maintain continuous area monitoring for HCN at fixed detector locations throughout the HCN production, absorber, and storage areas, supplemented by portable electrochemical HCN detectors carried by all workers performing tasks in HCN-present areas. Fixed electrochemical HCN sensors — requiring annual calibration and quarterly bump testing per ISA 84.01 functional safety standards referenced in OSHA PSM elements — provide continuous ppm-level HCN monitoring at valve manifolds, absorber column base, tank vent connection areas, and transfer pump rooms. Portable HCN detectors (colorimetric Dräger tubes, electrochemical sensors, or photoionization detector units calibrated for HCN) supplement fixed monitoring. HCN is particularly dangerous in occupational settings because: (a) HCN olfactory detection threshold varies widely between individuals (0.6–3 ppm; well below OSHA PEL 10 ppm for acute detectors, but rapidly saturating the olfactory receptor so that continued exposure causes anosmia within minutes); (b) HCN inhibits cytochrome c oxidase at the mitochondria, causing cellular asphyxiation that is nearly instantaneous at high concentrations — workers cannot self-rescue once they develop symptoms because cognitive impairment (confusion, dizziness) begins before the worker recognizes the severity of their condition; (c) HCN penetrates rubber gloves and skin (OSHA PEL skin notation; ACGIH TLV skin notation), making dermal absorption a significant supplementary exposure route in liquid HCN handling. AI systems process rendered fixed-detector panel or SCADA displays showing multi-zone HCN concentration ppm bars, zone alarm status LEDs, and concentration trend charts — to classify area safety state: all clear (all zones below OSHA PEL 10 ppm), OSHA PEL approach (one or more zones 5–10 ppm, source investigation required), and IDLH approach (one or more zones above 10 ppm, evacuation of affected zones required).

An adversarial perturbation targeting the process area HCN toxic gas detector display AI applies a ±8 DN downward shift to the pixel region encoding the HCN concentration bar chart and digital readout for the absorber base zone in the rendered detector panel display image — shifting the apparent area HCN concentration in the absorber base zone from 42 ppm (84% of IDLH 50 ppm, well above OSHA PEL ceiling 10 ppm, indicating a liquid HCN drip from a failing pump mechanical seal at the HCN product recirculation pump — the seal has been seeping for 2 hours following the same refrigeration failure that elevated the storage tank pressure in surface 3, with the slightly warmer HCN product liquid creating marginally elevated vapor pressure at the seal face) to 1.2 ppm (within all-clear range — no source investigation, no evacuation). The AI classifies a process area with HCN at 84% of IDLH — where NIOSH DHHS Alert No. 96-138 documents that 7 of the 9 documented fatalities involved concentrations in the 40–100 ppm range that workers entered before symptom onset, and where the anosmic effect of prior low-level HCN exposure in the facility means that workers present in the area may not detect the odor at 42 ppm — as a normal area requiring no protective action. Workers continue operating in the 42 ppm zone; within 5–10 minutes of sustained 42 ppm exposure, dizziness and headache begin; within 15–20 minutes, loss of muscular control; workers in the process area cannot self-rescue without immediate colleague assistance and antidote administration (hydroxocobalamin injection or amyl nitrite inhalation per NIOSH cyanide poisoning protocol). OSHA PSM area monitoring requirements for HCN facilities do not specify adversarial robustness for AI classifying rendered gas detector panel display images. Free tier — 10 scans/day, no card required.

Integration: HCN production AI with Glyphward pre-scan gate

The Glyphward scan gate for HCN Andrussow production AI belongs at every rendered-image ingestion boundary in the HCN production safety pipeline — before Pt–Rh catalyst temperature display AI processes rendered catalyst temperature images, before absorber vent CEMS display AI processes rendered HCN vent concentration images, before HCN storage tank pressure display AI processes rendered tank pressure images, and before process area gas detector display AI processes rendered detector panel images. The HCN production AI adversarial surface set illustrates a compound multi-display attack where refrigeration system failure propagates across surfaces 3 and 4: the storage tank pressure display (surface 3) is suppressed from 14.2 psig to 1.8 psig (masking refrigeration loss); simultaneously, the area gas detector (surface 4) is suppressed from 42 ppm to 1.2 ppm (masking the consequent pump seal vapor release). An operator cross-checking storage pressure against area concentration would see: apparently normal tank pressure at 1.8 psig + apparently clean area at 1.2 ppm — a mutually consistent false picture with no refrigeration investigation triggered. Multi-surface simultaneous Glyphward scanning detects the compound pattern. Threshold 35 for HCN production AI reflects NIOSH DHHS Alert No. 96-138 (1996) documented 9 HCN fatalities in 11 months; OSHA PSM TQ 1,000 lbs; IDLH 50 ppm; cytochrome c oxidase inhibition mechanism producing cellular asphyxiation at all tissues with rapid cognitive incapacitation before self-rescue is possible; and the anosmia effect that eliminates olfactory early warning at sustained sub-IDLH concentrations.

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

import httpx

GLYPHWARD_API_KEY = "YOUR_GLYPHWARD_API_KEY"
GLYPHWARD_SCAN_URL = "https://glyphward.com/v1/scan"

# HCN Andrussow production AI contexts: threshold 35
# OSHA PSM 29 CFR 1910.119: HCN (anhydrous) TQ 1,000 lbs.
# EPA RMP 40 CFR Part 68: HCN TQ 10,000 lbs (toxic).
# ACGIH TLV-TWA 4.7 ppm (as CN-); OSHA PEL 10 ppm (ceiling, skin).
# IDLH 50 ppm; cytochrome c oxidase inhibitor — cellular asphyxiation.
# NIOSH DHHS Alert No. 96-138 (1996): 9 HCN fatalities in 11 months.
HCN_PRODUCTION_THRESHOLD = 35


class HCNProductionContext(Enum):
    REACTOR_CATALYST_TEMPERATURE = "reactor_catalyst_temperature"  # Pt-Rh gauze temp AI
    ABSORBER_VENT_CEMS           = "absorber_vent_cems"            # Vent HCN CEMS AI
    STORAGE_TANK_PRESSURE        = "storage_tank_pressure"         # HCN tank pressure AI
    AREA_GAS_DETECTOR            = "area_gas_detector"             # Process area HCN AI


class AdversarialHCNProductionImageError(Exception):
    """Raised when Glyphward detects adversarial content in an HCN production AI
    rendered image above threshold 35.

    Consequence if not raised:
    - REACTOR_CATALYST_TEMPERATURE: UPWARD SHIFT — 934°C (deactivation zone)
      shown as 1,048°C (optimal) → NH3 slip building → absorber fouling +
      vent HCN carryover → irreversible Pt-Rh gauze restructuring at <950°C.
    - ABSORBER_VENT_CEMS: 38 ppm HCN in vent (76% IDLH) classified as 0.8 ppm
      compliant → community sub-IDLH chronic exposure + worker acute risk.
    - STORAGE_TANK_PRESSURE: 14.2 psig (refrigeration failure) shown as 1.8 psig
      (normal N2 blanket) → tank warming toward boiling point → vent scrubber
      overwhelmed → HCN release; compound with AREA_GAS_DETECTOR.
    - AREA_GAS_DETECTOR: 42 ppm (84% IDLH) shown as 1.2 ppm → workers in
      42 ppm zone → cognitive incapacitation before self-rescue possible.
    Fail-safe: verify catalyst temperature from independent secondary thermocouple;
    confirm vent HCN from portable electrochemical detector at stack sample point;
    verify tank pressure from independent mechanical gauge; confirm area HCN
    via portable detector carried by entering worker.
    """

    def __init__(self, scan_id, score, context, unit_id, flagged_region=None):
        self.scan_id = scan_id
        self.score = score
        self.context = context
        self.unit_id = unit_id
        self.flagged_region = flagged_region
        super().__init__(
            f"Adversarial HCN production image: context={context.value} "
            f"score={score} unit={unit_id} scan_id={scan_id}"
        )


async def scan_hcn_production_image(image_bytes, context, unit_id, client):
    image_hash = hashlib.sha256(image_bytes).hexdigest()
    payload = {
        "image": base64.b64encode(image_bytes).decode(),
        "source": f"hcn_production:{context.value}:{unit_id}",
        "metadata": {
            "unit_id": unit_id,
            "context": context.value,
            "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) >= HCN_PRODUCTION_THRESHOLD:
        raise AdversarialHCNProductionImageError(
            scan_id=result["scan_id"],
            score=result["score"],
            context=context,
            unit_id=unit_id,
            flagged_region=result.get("flagged_region"),
        )
    return result


async def main():
    async with httpx.AsyncClient() as client:
        with open("hcn_reactor_temperature_screenshot.png", "rb") as f:
            image_bytes = f.read()
        result = await scan_hcn_production_image(
            image_bytes,
            HCNProductionContext.REACTOR_CATALYST_TEMPERATURE,
            unit_id="HCN-ANDRUSSOW-REACTOR-1",
            client=client,
        )
        print(f"Clean scan: {result['scan_id']} score={result['score']}")


asyncio.run(main())

Frequently asked questions

What is the Andrussow process and why is Pt–Rh catalyst temperature the critical safety parameter?
The Andrussow process (CH4 + NH3 + O2 → HCN + 3H2O over 90:10 Pt–Rh gauze at 1,000–1,100°C) accounts for ~70% of global HCN production. Catalyst temperature is critical because: below 950°C, NH3 slip builds and HCN selectivity falls; below 900°C, Pt–Rh surface reconstruction is irreversible ($15k–50k gauze replacement). Continuous temperature monitoring is the primary catalyst lifetime and NH3-slip safety control.
What does NIOSH DHHS Alert No. 96-138 document about HCN facility fatalities?
NIOSH Alert 96-138 (1996) documented 9 HCN workplace fatalities in 11 months across electroplating, production, and confined-space scenarios. Most fatalities occurred at 40–100 ppm — where HCN olfactory anosmia (smell-saturation at 5–10 ppm) prevented odor detection while cytochrome c oxidase inhibition caused cognitive incapacitation before self-rescue. The Alert established the four HCN controls now standard in US industry: continuous area monitoring, supplied-air respirators, antidote kits (hydroxocobalamin), and buddy systems.
Why does HCN olfactory adaptation make area detector AI the only reliable warning?
Workers with chronic low-level HCN background exposure (0.5–2 ppm from normal process fugitives) become anosmic to HCN within minutes of exposure above 5 ppm. A release escalating area concentration from 1 ppm to 30–40 ppm will not be detected by smell until neurological symptoms appear simultaneously with cognitive impairment — eliminating self-rescue. Area gas detectors are therefore the sole reliable early warning layer; adversarial robustness of their AI classifiers is a primary process safety gap.
Why does the reactor temperature attack use an upward shift rather than the standard downward shift?
The dangerous condition in Andrussow is too-low temperature (NH3 slip, catalyst deactivation), not too-high. The upward shift makes the actual 934°C (below 950°C deactivation threshold) appear as 1,048°C (optimal selectivity) — preventing combustion air investigation and catalyst recovery. This is the third upward-direction attack in the Glyphward portfolio, confirming that both upward and downward pixel shifts must be scanned at every adversarial surface.
Why is threshold 35 for HCN production AI?
OSHA PSM TQ 1,000 lbs; EPA RMP significant community consequence radii; NIOSH 9 fatalities in 11 months; IDLH 50 ppm (cytochrome c oxidase inhibition — all-tissue cellular asphyxiation, not just respiratory); anosmia eliminating olfactory warning; and compound multi-display attack (tank pressure + area gas detector suppressed simultaneously) requiring multi-surface scanning. Threshold 35 rather than 40 reflects primarily facility-scale rather than multi-kilometer community-scale acute consequence at typical HCN production volumes.