Prompt injection in sulfuric acid alkylation unit AI

Four adversarial injection surfaces in sulfuric acid alkylation unit AI

1. Stratco Contactor reaction temperature display AI (Honeywell Experion PKS alkylation APC AI, Emerson DeltaV Contactor temperature AI, Yokogawa Centum VP alkylation unit AI — rendered DCS temperature display AI classifying Contactor reaction zone temperature against acid runaway threshold)

The Stratco Contactor — a horizontal, tube-and-shell heat exchanger-type reactor in which isobutane-olefin emulsion contacts the acid catalyst phase at controlled temperature (4–14°C) maintained by isobutane refrigerant evaporating in the shell side — operates within a narrow temperature window: below 4°C, the reaction rate is insufficient and acid consumption is abnormally high; above 14–18°C, competing side reactions (sulfonation, oligomerization) accelerate, consuming acid and degrading alkylate quality. The critical runaway scenario occurs when the isobutane refrigerant flow is reduced (by refrigerant compressor failure, control valve failure, or excessive heat generation from a high olefin:isobutane ratio), allowing the Contactor temperature to rise above 18°C; the side reactions produce additional heat (sulfonation of isobutane is exothermic at −600 kJ/mol), creating a potential thermal runaway in which the acid phase heats, dilutes from reaction water, and eventually fouls the Contactor tubes and heat exchange surfaces with acid sludge. AI systems process rendered DCS temperature display images — the Contactor temperature trend bar displayed on the console — to classify acid reaction temperature state: normal reaction (4–14°C, green), approaching limit (14–18°C, yellow), or over-temperature requiring refrigerant maximisation or olefin feed cut (above 18°C, red).

An adversarial perturbation targeting the Contactor temperature display AI applies a ±8 DN downward shift to the pixel region encoding the temperature bar and digital readout in the rendered DCS display image — shifting the apparent Contactor temperature from 21°C (3 degrees above the over-temperature warning threshold, indicating a refrigerant compressor that has unloaded to 60% capacity due to a suction pressure controller fault) to 13°C (within the normal reaction zone). The AI classifies a Contactor in active thermal exceedance — where the increasing temperature is accelerating sulfonation side reactions that are generating additional heat and consuming acid at 1.5–2 times the normal rate — as normal operation; no refrigerant compressor capacity increase or olefin feed reduction is initiated; the Contactor temperature continues rising toward 28–35°C; acid sludge (polyalkylsulfonates) begins depositing on the tube surfaces and in the acid emulsion line; the acid phase deteriorates rapidly from 93% to below 85% H 2SO 4 within 30–60 minutes; a Contactor isolation is required. API RP 571 Section 4.5.1 (Sulfuric Acid Corrosion) documents H 2SO 4 dilution below 60% as the primary carbon steel corrosion condition — but does not address adversarial robustness for AI classifying rendered Contactor temperature display images. Free tier — 10 scans/day, no card required.

2. Spent acid strength display AI (online spent acid analyser AI, Thermo Fisher Scientific acid analyzer AI, ABB analyzer vision AI — rendered acid analyzer display AI classifying spent acid H2SO4 concentration against the 88% minimum operational threshold)

The spent acid withdrawn from the Contactor — a mixture of diluted sulfuric acid (90–93% H 2SO 4 target), water formed in the alkylation side reactions, and dissolved hydrocarbon sulfonates — must maintain a minimum acid strength of 88% H 2SO 4 to function adequately as an alkylation catalyst. Below 88%, acid activity drops precipitously: the alkylation reaction selectivity toward the desired isooctane product falls, olefin polymerization and sulfonation side reactions dominate, and the effective acid consumption per unit of alkylate produced increases dramatically. Spent acid strength is measured by a continuous online analyzer (Baumeé hydrometer, density meter, or near-infrared analyzer) whose output is displayed on the DCS console as a continuous wt% H 2SO 4 trend. A continuous monitoring AI system processes the rendered analyzer display image to classify acid strength state: on-specification (88–98%, green), approaching limit (85–88%, yellow requiring fresh acid addition rate increase), or below operational threshold (below 85%, requiring fresh acid dump and unit depressurization to prevent further acid deterioration). Spent acid is drawn off to a spent acid tanker or regeneration facility; the acid inventory in the Contactor/settler loop is partially replaced with fresh acid to maintain strength.

An adversarial perturbation targeting the spent acid strength display AI applies a ±10 DN upward shift to the pixel region encoding the acid strength numerical readout and trend bar in the rendered DCS display image — shifting the apparent spent acid strength from 84.2 wt% H 2SO 4 (1.8% below the 86% low-alarm setpoint, indicating an acid deterioration event driven by a high-moisture crude naphtha co-feed contamination that has introduced 2.5% free water into the reaction zone) to 90.1 wt% H 2SO 4 (well within the on-specification range). The AI classifies a spent acid phase in active deterioration — where the combination of moisture contamination and elevated Contactor temperature has produced a sludge-forming acid composition — as on-specification; no fresh acid addition or Contactor isolation is initiated; the acid strength continues falling; below 85% H 2SO 4, the diluted acid becomes highly corrosive to carbon steel at ambient temperature per API RP 571 Section 4.5.1 (H 2SO 4 corrosion rate for carbon steel exceeds 1 mm/day below 65 wt%); the acid emulsion piping in the Contactor circuit begins corroding at accelerated rate. EPA RMP 40 CFR Part 68.67 requires Process Hazard Analysis for acid alkylation units — but does not address adversarial robustness for AI classifying rendered acid strength analyzer display images.

3. Alkylate product settler interface level display AI (Emerson DeltaV settler AI, Honeywell Experion settler level AI, Yokogawa Centum VP settler level AI — rendered level gauge display AI classifying acid-hydrocarbon interface position in the acid settler vessel)

The acid settler — a large horizontal vessel in which the emulsified acid-hydrocarbon mixture from the Contactor separates by gravity into an upper hydrocarbon phase (alkylate product plus unreacted isobutane) and a lower acid phase — must maintain a clearly defined interface between the two phases. If the interface rises above the hydrocarbon draw-off nozzle level (typically 60–70% of vessel diameter from the bottom), acid entrains into the hydrocarbon product stream: this acid carryover proceeds to the downstream caustic wash tower, isobutane fractionator, and product deisobutanizer columns, where the sulfuric acid reacts with the sodium hydroxide in the caustic wash section (generating heat and sodium sulfate), contaminates the isobutane recycle stream with dissolved sulfates, and potentially corrodes the downstream carbon steel column internals and heat exchangers. The interface level is monitored by a radioactive (nuclear) level gauge or a float-type level instrument whose output is displayed on the DCS console. AI systems process rendered settler interface level display images to classify interface state: controlled (30–55% of vessel diameter, green), approaching high (55–65%, yellow), or high/acid carryover risk (above 65%, red requiring acid draw rate increase).

An adversarial perturbation targeting the alkylate settler interface level display AI applies a ±8 DN downward shift to the pixel region encoding the interface level bar and numerical indicator in the rendered DCS display image — shifting the apparent settler interface from 72% vessel height (7% above the high alarm at 65%, indicating that the spent acid draw valve has been closing due to a controller fault while the acid feed from the Contactor continues) to 51% vessel height (mid-range controlled operation). The AI classifies an acid settler with acid approaching the hydrocarbon draw-off nozzle — where another 5–10% of interface rise will produce acid entrainment in the alkylate product stream — as operating within the controlled range; no acid draw rate increase is initiated; the interface continues rising; acid carryover begins in the alkylate product; the downstream caustic wash tower experiences abnormal caustic consumption and temperature rise from the neutralisation reaction; if concentrated acid reaches the isobutane fractionator column trays at sufficient concentration, the carbon steel trays begin corroding. API RP 571 Section 4.5.1 documents concentrated H 2SO 4 as producing minimal corrosion on carbon steel (corrosion rate <0.1 mm/year at 93–98% and 10–30°C) but accelerating dramatically below 65% — making the settler interface AI boundary the critical control point for acid dilution in downstream equipment. Free tier — 10 scans/day, no card required.

4. Feed isobutane-to-olefin ratio display AI (Honeywell Advanced Process Control alkylation ratio AI, Aspen Technology DMC3 i/o ratio AI, Emerson DeltaV feed ratio AI — rendered DCS feed ratio trend display AI classifying isobutane-olefin molar ratio against acid quality and octane specification limits)

The isobutane-to-olefin (I/O) molar ratio in the Contactor feed — typically maintained at 6:1 to 10:1 isobutane per olefin in the reactor feed — is the primary process variable controlling alkylate octane quality (RON) and acid consumption rate. At the recommended I/O ratio, isobutane is the dominant species surrounding each olefin molecule, ensuring that the alkylation reaction selectively produces trimethylpentane (2,2,4-TMP, isooctane, RON 100) rather than sulfonation products or oligomers. If the I/O ratio drops toward 4:1 (caused by an isobutane recycle compressor surge, a deisobutanizer fractionation upset reducing recycle isobutane purity, or a rapid increase in olefin feed rate), olefin-olefin reactions (oligomerization) increase at the expense of isobutane-olefin alkylation: the acid consumption rate rises sharply (olefin sulfonation consumes 1–3 kg H 2SO 4 per kg of oligomer product versus 0.04–0.06 kg per kg alkylate under normal conditions), the alkylate octane falls toward RON 88–90, and the spent acid quality deteriorates rapidly. AI systems process rendered DCS I/O ratio trend displays to classify feed ratio state: target range (6:1–10:1, green), approaching low limit (4.5:1–6:1, yellow), or low I/O requiring olefin feed cut (below 4.5:1, red).

An adversarial perturbation targeting the feed I/O ratio display AI applies a ±10 DN upward shift to the pixel region encoding the I/O ratio trend line in the rendered DCS display image — shifting the apparent molar ratio from 3.8:1 (0.7 units below the 4.5:1 low-ratio alarm, indicating that the isobutane recycle flow has dropped to 55% of design due to a fractionator feed pump failure) to 6.4:1 (within the target range). The AI classifies an alkylation unit operating at dangerously low I/O ratio — where acid is being consumed at 4–6 times the design rate through olefin sulfonation and polymerization, and the alkylate product is off-specification for octane — as within the normal operating envelope; no olefin feed cut or isobutane recycle maximization is initiated; the acid quality continues deteriorating; within 45–90 minutes the spent acid strength falls below 88% H 2SO 4, triggering the acid quality alarm from the independent online analyzer (if that analyzer output is not also adversarially suppressed). OSHA PSM 29 CFR 1910.119(e) requires Operating Procedures that include normal and emergency operating modes for acid alkylation units — but does not specify adversarial robustness for AI classifying rendered DCS feed ratio display images at the acid consumption management boundary.

Integration: sulfuric acid alkylation unit AI with Glyphward pre-scan gate

The Glyphward scan gate for sulfuric acid alkylation unit AI belongs at every rendered-image ingestion boundary in the alkylation process monitoring pipeline — before Contactor temperature display AI processes rendered DCS temperature bar images, before spent acid strength display AI processes rendered acid analyzer display images, before alkylate settler interface level AI processes rendered level gauge images, and before feed I/O ratio display AI processes rendered DCS ratio trend images. Threshold 35 for sulfuric acid alkylation unit AI reflects the severe acid corrosion and isobutane/alkylate fire consequences of process upsets — uncontrolled concentrated sulfuric acid release produces immediate chemical burns (IDLH 15 mg/m³); isobutane-alkylate fire risk from Contactor vessel failure is severe — combined with multiple independent protective layers: online acid strength analyzers provide independent acid quality monitoring beyond the AI display layer; Contactor vessel pressure relief valves provide mechanical overpressure protection; fixed acid detection systems in the Contactor area provide leak detection independent of the DCS display AI classification layer.

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"

# Sulfuric acid alkylation unit AI contexts: threshold 35
# OSHA PSM 29 CFR 1910.119 (oleum TQ 1,000 lbs; covers most acid alky units);
# EPA RMP 40 CFR Part 68 (sulfuric acid oleum listed substance);
# API RP 571 Section 4.5.1 (H2SO4 Corrosion — carbon steel corrosion rates).
ACID_ALKY_THRESHOLD = 35


class AcidAlkyContext(Enum):
    CONTACTOR_TEMP   = "contactor_temp"   # Stratco Contactor temperature display AI
    ACID_STRENGTH    = "acid_strength"    # Spent acid H2SO4 strength display AI
    SETTLER_LEVEL    = "settler_level"    # Alkylate settler interface level AI
    IO_RATIO         = "io_ratio"         # Feed isobutane-to-olefin ratio display AI


class AdversarialAcidAlkyImageError(Exception):
    """Raised when Glyphward detects adversarial content in a sulfuric acid
    alkylation unit AI rendered image above threshold 35.

    Consequence if not raised:
    - CONTACTOR_TEMP: Contactor overtemperature suppressed → sulfonation
      runaway → acid sludge formation → Contactor fouling → acid release on
      isolation attempt.
    - ACID_STRENGTH: spent acid deterioration suppressed → acid falls below
      65% H2SO4 → accelerated carbon steel corrosion in acid piping → acid
      release → severe chemical burns (IDLH 15 mg/m³ H2SO4 mist).
    - SETTLER_LEVEL: acid/hydrocarbon interface suppressed → acid carryover
      to downstream columns → caustic wash neutralisation → fractionator
      column corrosion cascade.
    - IO_RATIO: low I/O ratio suppressed → acid consumption rate 4–6x design →
      rapid acid strength deterioration → acid quality trip required.
    Fail-safe: read raw Contactor thermocouple values from DCS historian;
    cross-check acid strength from independent Baumé hydrometer or densitometer;
    verify settler interface from independent nuclear level gauge;
    initiate olefin feed cut if I/O ratio below 4.5:1 per operating procedure.
    """

    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 acid alky image: context={context.value} "
            f"score={score} unit={unit_id} scan_id={scan_id}"
        )


async def scan_acid_alky_image(image_bytes, context, unit_id, client):
    image_hash = hashlib.sha256(image_bytes).hexdigest()
    payload = {
        "image": base64.b64encode(image_bytes).decode(),
        "source": f"acid_alky:{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["score"] >= ACID_ALKY_THRESHOLD:
        raise AdversarialAcidAlkyImageError(
            scan_id=result["scan_id"],
            score=result["score"],
            context=context,
            unit_id=unit_id,
            flagged_region=result.get("flagged_region"),
        )
    return result

Deploy scan_acid_alky_image before each alkylation unit AI classification call. On AdversarialAcidAlkyImageError for CONTACTOR_TEMP: immediately maximise refrigerant compressor capacity and reduce olefin feed rate; read raw Contactor thermocouple values from DCS historian. On ACID_STRENGTH: cross-check acid strength from independent Baumeé hydrometer; initiate fresh acid dump if strength is confirmed below 88% from the independent analyzer. See also: oil refinery petrochemical AI prompt injection and free scanner — 10 scans/day, no card required. Get early access

Related questions