HF alkylation unit AI adversarial injection: how ±10 DN in the rendered acid settler interface level display suppresses HF acid carryover — and why API RP 751 has no adversarial robustness criterion for HF alkylation advanced process control AI classifying rendered acid circuit monitoring displays

HF alkylation chemistry, the acid circuit integrity concept, and API RP 751

The HF alkylation reaction occurs in a liquid-liquid contactor reactor where isobutane and olefin feed (C3–C4 olefins at 40–60 vol% olefin concentration in the combined feed) are dispersed as a fine emulsion in a continuous HF acid phase. The alkylation reaction — isobutane + olefin → isoalkane (alkylate) — proceeds via a carbocation mechanism at the acid-hydrocarbon interface, with H⁺ from the HF catalyst protonating the olefin to form a secondary carbocation that undergoes hydride transfer with isobutane, producing a tertiary carbocation that alkylates with additional olefin and ultimately terminates as a branched C7–C9 isoalkane (primarily 2,2,4-trimethylpentane (isooctane) from butylene alkylation, with smaller quantities of C8 isomers and minor C9 species). The alkylation reaction is exothermic (ΔH approximately −75 to −85 kJ/mol for butylene alkylation with isobutane at 20–35°C), requiring acid cooling to maintain the contactor at 20–40°C; higher temperatures increase unwanted side reactions including olefin polymerisation and acid-soluble oil (ASO) formation.

The acid settler vessel — the gravity-separation vessel downstream of the contactor — is the critical boundary between the acid circuit and the downstream hydrocarbon fractionation circuit. In the acid settler, the reaction product emulsion (alkylate, unreacted isobutane, HF acid, and ASOs) separates by density: the HF acid phase (density approximately 1.12–1.15 g/mL at 88–92 wt% HF, 25°C) settles to the bottom of the vessel; the hydrocarbon phase (density approximately 0.60–0.65 g/mL for the isobutane-alkylate-propane mixture at contactor conditions) rises to the top. The HF-hydrocarbon interface position in the acid settler — the level of the phase boundary between the dense lower HF acid layer and the lighter upper hydrocarbon layer — is the primary indicator of acid circuit containment integrity at the fractionation boundary: if the interface rises too high (HF phase too thick in the settler), HF acid entrains into the hydrocarbon overflow from the settler top, contaminating the downstream depropanizer, isobutane fractionator, and alkylate splitter with HF concentrations far above the 0.5 wt% corrosion threshold for carbon steel equipment that has not been designed for HF service. The consequence of HF carryover to non-acid-rated fractionation equipment includes severe HF corrosion of column internals and overhead piping; HF vapour accumulation in the fractionator vapour space (HF has significant vapour pressure at 40–80°C fractionator overhead conditions); and acute personnel exposure risk when contaminated equipment is subsequently opened for maintenance operations that operators and technicians believe to be safe based on the acid-free classification of the fractionation circuit. If the interface falls too low (HF phase too thin), hydrocarbon entrains into the HF acid recirculation to the contactor, reducing the acid-to-hydrocarbon (A/HC) ratio and degrading alkylate octane and acid catalyst activity.

API RP 751 defines the acid circuit as the boundary containing all HF-wetted equipment: the contactor, acid settler, acid coolers, acid recirculation piping, acid surge drum, and the HF storage and transfer system. All equipment within the acid circuit is specified to HF-compatible materials (plain carbon steel AISI 1010–1020, free of residual stress above 207 MPa) and is maintained under positive HF acid pressure. The acid circuit integrity concept in API RP 751 is that acid containment breaches — whether through equipment failure, relief device actuation, or process upsets that carry HF into the non-acid-rated fractionation circuit — constitute the primary HF release pathway requiring management. API RP 751’s monitoring specifications for acid settler interface level, acid strength, contactor temperature, and acid relief header pressure are all motivated by this containment integrity concept: these monitored parameters are the primary early indicators of acid circuit integrity degradation, upstream of any actual HF release. The AI systems deployed to classify rendered images of these monitoring displays sit directly at the acid containment monitoring boundary — and their adversarial robustness is the gap that neither API RP 751 nor OSHA PSM nor EPA RMP currently address.

ExxonMobil Torrance 2015: the community-scale consequence anchor for HF alkylation unit AI

The ExxonMobil Torrance California refinery is located in Torrance, California — the South Bay area of Los Angeles County, a densely urbanised coastal industrial corridor with residential communities in Torrance, Lomita, Redondo Beach, Hawthorne, and Carson directly adjacent to the refinery fence line. At the time of the 2015 explosion, the refinery processed approximately 155,000 barrels per day of crude oil and was one of the three largest petroleum refineries in California. The refinery operated a UOP-licensed HF alkylation unit processing approximately 24,000 barrels per day of FCC-derived C3/C4 olefin feed.

On 18 February 2015, maintenance activities were underway on the electrostatic precipitator (ESP) in the FCC unit’s regenerator flue gas treatment train — a high-temperature vessel that removes catalyst fines from the regenerator flue gas before the gas is routed to the CO boiler or released to the atmosphere through the stack. During the maintenance event, a pressure relief device in the ESP system actuated in an uncontrolled manner, producing a high-energy pressure release and explosion. The explosion scattered metal fragments and equipment debris across the FCC unit plot area; four refinery workers were injured; the FCC unit was emergency-shutdown and subsequently required extended inspection and repair. Community shelter-in-place advisories were issued for surrounding residential areas during the emergency response. Cal/OSHA subsequently cited ExxonMobil for multiple violations relating to permit-to-work procedures and mechanical integrity management at the FCC/ESP boundary.

The critical chemical safety finding from the CSB investigation of the Torrance 2015 event — documented in the CSB video case study released following the investigation — was the near-miss with the HF alkylation unit. The Torrance HF alkylation unit was located approximately 50 feet from the FCC ESP that exploded. The ESP explosion produced metal fragment trajectories across the refinery plot that passed in close proximity to the HF alkylation unit’s acid settler vessel, acid cooler bundles, and acid recirculation piping — all of which contain the principal HF acid inventory of the alkylation unit (estimated at approximately 800,000 pounds of HF for the Torrance unit at the time). The CSB investigation found that if the ESP explosion debris had struck and penetrated the HF acid containment — specifically, the acid settler shell, acid cooler shell-and-tube bundles, or acid circuit piping flanges — the resulting uncontrolled HF release could have created a toxic aerosol cloud potentially affecting up to 125,000 residents in the surrounding South Bay communities. The EPA RMP worst-case release scenario for HF alkylation facilities of comparable inventory produces toxic endpoint distances (to the IDLH concentration of 30 ppm) of 1–10 miles under Pasquill-Gifford F stability class (stable, low-wind) conditions typical of the Torrance coastal location; the CSB’s 125,000-resident consequence finding is consistent with the community population density within this toxic endpoint radius.

The Torrance 2015 event was not triggered by an HF alkylation unit monitoring AI failure — it arose from a maintenance boundary management failure at the FCC ESP, a unit adjacent to but mechanically independent of the HF alkylation unit. Its relevance to HF alkylation unit AI adversarial injection is as a consequence magnitude anchor: it established empirically, through CSB consequence analysis, that HF alkylation unit containment failures at a typical US refinery carry community-scale consequence at the 100,000-resident level. Any monitoring AI failure that suppresses early detection of an HF acid circuit integrity degradation — acid settler interface rising toward carryover, acid strength below operating floor, contactor temperature above normal, relief header pressure approaching pre-lift — increases the probability that the acid circuit integrity degradation progresses to an HF containment event at the community-consequence scale the CSB analysis quantified. The FCC regenerator afterburn AI adversarial injection blog documents the Torrance 2015 event from the FCC unit monitoring perspective — the same refinery near-miss that anchors the community consequence scale for both FCC and HF alkylation unit AI monitoring failures.

Four adversarial injection surfaces in HF alkylation unit AI

1. Acid settler HF-hydrocarbon interface level display AI (UOP HF Alkylation APC AI, Phillips 66 HF Alkylation process AI, Honeywell Experion PKS acid settler AI — rendered DCS level indicator AI classifying HF-HC interface position against carryover setpoints)

The acid settler HF-hydrocarbon interface level is the primary real-time indicator of acid circuit integrity at the fractionation boundary. The DCS display for the acid settler interface level presents one of three instrument types depending on the alkylation unit design: a differential pressure level transmitter display (showing the DP between the bottom drain nozzle and a point above the expected interface, calibrated to interface position via acid-hydrocarbon density differential); a float-type displacer level instrument display; or a nuclear density gauge level display (non-contacting, particularly favoured for HF service given the material compatibility constraints). The interface level is expressed as a percentage of the acid settler design operating range, with 0% indicating the minimum HF acid level above the settler drain and 100% indicating the maximum acid level below the hydrocarbon overflow weir; the design operating range is typically 50–70% of this span, with high-interface alarm at 75% and high-high trip (emergency acid pump-down to acid surge drum) at 85–90%. The display is updated every 2–10 seconds from the transmitter scan cycle and rendered on the HF alkylation operations console at the acid settler monitoring section.

The adversarial injection scenario for surface 1: the acid settler interface level is at 68% — within the normal operating range but approaching the high-interface alarm at 75%. The root cause is a partial blockage in the acid cooler return line reducing the acid circulation rate back to the contactor, allowing the acid phase in the settler to accumulate above the normal level; the correct classification is interface approaching high alarm — initiate investigation of acid circulation pump operation, check acid cooler outlet temperatures for fouling indicators, evaluate whether acid-to-hydrocarbon ratio in the contactor has shifted. The adversarial pixel perturbation: a ±10 DN downward shift applied to the pixel region encoding the interface level bar indicator or displacer float trace in the rendered DCS display image. On a standard acid settler level display with a 0–100% y-axis spanning approximately 200 pixels, the 68% reading renders at 136 px from the bottom; the 49% apparent reading after the downward shift renders at 98 px from the bottom. The 30–40 pixel downward shift (at ±10 DN) moves the interface indicator from the upper portion of the normal operating range (68%, approaching the high-alarm yellow zone beginning at 75%) to the mid-range of the normal operating band (49%), with no alarm zone proximity visible. The HF alkylation AI classifies normal acid settler operation — no acid circuit intervention initiated, no investigation of acid cooler circulation, no reduction in feed rate to reduce acid loading. The actual interface continues to rise as acid cooler circulation is reduced; the interface reaches 75% (high-alarm setpoint) within approximately 15–25 minutes and continues toward 85% (high-high trip) if the cooler blockage is not addressed. Above 80–85%, HF acid begins to entrain over the acid settler overflow weir into the hydrocarbon return line to the isobutane fractionator, contaminating the downstream fractionation circuit with HF.

The downstream consequence of undetected acid carryover: the isobutane fractionator overhead (primarily propane and isobutane vapour at 60–80°C and 8–15 bar) contacts HF-contaminated liquid from the fractionator feed; HF partitions preferentially to the liquid phase at these conditions but vapour-phase HF at ppm concentrations accumulates in the overhead condenser and reflux drum. Maintenance personnel opening flanges on fractionator overhead piping, reflux drum nozzles, or heat exchanger covers — on the assumption that the fractionation circuit contains only propane-isobutane-alkylate — encounter HF vapour and liquid HF at concentrations that cause immediate dermal and respiratory burns, with potential for fatal systemic HF toxicity from acid skin absorption exceeding the 1.5 g HF body burden threshold for cardiac arrhythmia from hypocalcaemia.

2. HF acid strength analyser display AI (upward-direction adversarial attack — 81.4 wt% degraded acid displayed as 87.8 wt% on-spec)

The HF acid strength analyser display AI processes a rendered image of the in-line potentiometric or conductometric acid strength analyser output — a continuous-flow analyser sampling the acid recirculation stream from the acid settler return to the contactor, reporting HF concentration (wt% HF) on a DCS trend display updated every 2–5 minutes. The analyser sample stream is drawn from the acid circuit, diluted and treated to bring the HF concentration within the potentiometric electrode measurement range, and the HF content determined by the voltage response of a fluoride-selective ion electrode (most modern HF alkylation acid strength analysers use a lanthanum fluoride membrane ion-selective electrode (ISE) calibrated against HF acid standards at 70–100 wt% HF). The DCS display renders the analyser output as a single-trace trend chart on a y-axis scale of 70–100 wt% HF (30 wt% total range), with the normal operating target at 88–92 wt%, the lower operating limit (floor) at 85 wt%, and a low-strength alarm at 85 wt% requiring acid management intervention.

The root cause of acid strength decline in HF alkylation is ASO (acid-soluble oil) accumulation. ASOs are formed from the polymerisation of olefins — particularly propylene and heavier C5+ olefins in the fresh feed — in the HF acid phase, producing heavy, dark-coloured oily compounds (polynuclear aliphatic and aromatic structures) that are fully miscible with the HF acid phase but act as diluents, reducing HF activity and increasing acid viscosity. ASOs are continuously formed in the alkylation reaction and continuously removed from the bottom of the acid settler as an acid-ASO-HF mixture; the ASO removal rate is managed by the alkylation unit acid management system (acid settler bottom drawoff flow to the acid regeneration unit or to spent acid disposal). If ASO formation rate exceeds ASO removal rate — from elevated feed olefin concentration, higher-than-design feed propylene content, or reduced acid regeneration throughput — ASO accumulates in the acid phase, diluting HF concentration below the operating floor.

The adversarial perturbation for surface 2 is an upward-direction pixel shift — the opposite direction from the majority of refinery AI adversarial surfaces documented in this blog series. A ±8 DN upward shift applied to the acid strength trend line in the rendered analyser DCS display moves the trace upward by approximately 15–25 pixels in a 200-pixel-tall trend chart. On the 70–100 wt% (30 wt% range) display, 81.4 wt% renders at (81.4–70)/(100–70) = 38.0% from the chart bottom (76 px in a 200-px chart); 87.8 wt% renders at (87.8–70)/(100–70) = 59.3% (119 px). The upward shift moves the apparent reading from 76 px to approximately 119 px — a 43-pixel upward displacement at ±8 DN — classifying the 81.4 wt% degraded acid as 87.8 wt% on-spec (2.2 wt% above the operating floor, no intervention required). This upward-direction attack is structurally equivalent to the urea passivation O2 injection upward attack (deficient O2 appears adequate) documented in the process-chemical hazards SEO series — both create false safety by showing a deficient-value parameter as safely within its operating window. The acid strength AI misclassification allows ASO-contaminated acid at 81.4 wt% to continue circulating; acid regeneration is not triggered; HF makeup is not initiated. The degraded acid has elevated free water content (approximately 3–5 wt% water vs below 1 wt% for on-spec acid) that increases corrosion rates in the acid circuit and, critically, increases HF volatility: at 81.4 wt% HF and 25°C, the HF vapour pressure above the acid phase is approximately 2.5 times higher than at 91 wt% HF, increasing HF losses to the acid cooler vapour space and acid settler vent.

3. Acid-hydrocarbon contactor temperature display AI (±8 DN downward shift — 44°C overtemperature approach suppressed to 36°C normal)

The acid-hydrocarbon contactor temperature display AI processes a rendered DCS temperature trend chart for the contactor reaction zone — typically 2–4 thermocouples at different elevations within the HF acid-HC contacting zone, with the highest temperature typically at the top of the contacting zone where the exothermic alkylation reaction has accumulated heat over the acid-HC contact time. The normal operating range for the contactor reaction zone temperature is 20–40°C for commercial HF alkylation (optimal alkylate quality at 25–35°C; higher temperatures shift selectivity toward heavier-than-target alkylate and increase ASO formation rate); the high-temperature alarm is typically set at 40–45°C, with high-high alarm at 50–55°C requiring reduction in feed rate and increase in acid cooling duty. The DCS display presents the thermocouple traces on a y-axis scale of 15–65°C (50°C range) spanning approximately 200 pixels in a standard HF unit operations console render.

The adversarial injection scenario for surface 3: the contactor top thermocouple reads 44°C — 4°C above the normal operating ceiling of 40°C, approaching the high-temperature alarm. The root cause is reduced acid cooling duty from partial fouling of the acid cooler tube bundle (calcium fluoride or iron fluoride deposits on the acid-side tube surfaces reducing heat transfer coefficient); with reduced cooling, the contactor reaction zone temperature rises above the operating ceiling as the exothermic alkylation reaction heat is not fully removed. The ±8 DN downward perturbation applied to the contactor thermocouple trace: on the 15–65°C display scale (50°C range, 200 pixels), 44°C renders at (44–15)/50 = 58.0% from the bottom (116 px); 36°C renders at (36–15)/50 = 42.0% from the bottom (84 px). The ±8 DN downward shift moves the trace from 116 px to approximately 84 px — a 32-pixel downward displacement — suppressing the approaching high-alarm reading at 44°C to appear as normal mid-range operation at 36°C. The HF alkylation AI does not flag contactor temperature concern; acid cooler fouling investigation is not initiated; feed rate is not reduced to compensate for reduced cooling capacity. At 44°C contactor temperature, HF vapour pressure above the acid-HC interface has increased relative to the 25–35°C operating target, increasing HF losses to the vapour spaces above the contactor acid level and increasing acid-soluble oil formation rate from the exothermic olefin polymerisation side reaction that accelerates above 40°C — the compounding factor that accelerates ASO accumulation and acid strength degradation (surface 2) when contactor overtemperature is not detected and corrected.

4. Acid relief header pressure display AI (±10 DN downward shift — 18 mbar pre-lift suppressed to 4 mbar normal)

The acid relief header display AI processes a rendered DCS pressure indicator image for the HF acid circuit pressure relief header — the common discharge header that collects any HF vapour releases from pressure safety valves (PSVs), thermal relief valves (TRVs), and pressure-vacuum (P-V) valves on acid circuit vessels (acid settler, acid cooler, contactor), routing them to the HF scrubber/absorber system for neutralisation before atmospheric release. The acid relief header is maintained at a slight positive pressure (typically 2–8 mbar) by a nitrogen blanket or by the sealing effect of the relief device trim — ensuring that atmospheric oxygen does not enter the acid circuit vapour spaces and that any small HF vapour releases from the acid phase surface are contained within the header system. The normal operating range for the acid relief header pressure display is 2–10 mbar; an elevated reading above 15–20 mbar indicates that a pressure relief device on the acid circuit is near its setpoint or actively relieving, generating HF vapour to the header at a rate above the scrubber absorption capacity — a condition requiring immediate investigation of the overpressure source in the acid circuit.

The adversarial injection scenario for surface 4: the acid relief header reads 18 mbar — elevated above the normal 2–10 mbar range, indicating that the acid settler vent P-V valve is near its opening setpoint from the accumulation of HF vapour in the acid settler vapour space, driven by the elevated contactor temperature (surface 3) increasing HF vapour pressure above the acid phase. A ±10 DN downward shift applied to the rendered acid relief header pressure indicator display (a dial-type pressure gauge or DCS bar indicator on the HF scrubber inlet monitor screen) moves the displayed reading from 18 mbar to approximately 4 mbar — well within the normal nitrogen-blanketing range, no pre-lift concern visible to the HF alkylation AI. The combined suppression of all four surfaces — acid settler interface approaching carryover (surface 1), acid strength below floor (surface 2), contactor overtemperature (surface 3), and relief header pre-lift (surface 4) — eliminates all four acid circuit integrity indicators from the HF alkylation AI classification layer simultaneously. The actual acid circuit state is: elevated interface (68%, rising), degraded acid (81.4 wt%), overtemperature contactor (44°C), and elevated relief header (18 mbar) — all four parameters consistent with an acid circuit state approaching an HF containment excursion. The AI presents all four as within-normal-range operation. See the full HF alkylation unit AI prompt injection technical specification for all four adversarial surface parameters and vendor coverage.

API RP 751, OSHA PSM, EPA RMP, and the adversarial robustness gap for HF alkylation APC AI

API RP 751 (Safe Operation of Hydrofluoric Acid Alkylation Units, 5th edition 2020) is the most comprehensive industry standard specifically addressing HF alkylation safety — the fact that a dedicated RP exists for this single unit type, where most refinery unit types are covered under general API refinery safety standards like API RP 505 or API RP 571, reflects the exceptional consequence severity of the HF hazard. API RP 751 Section 5 specifies monitoring requirements for HF alkylation operations: acid settler interface level monitoring at defined frequency intervals with DCS alarm setpoints; acid strength analyser calibration requirements (monthly analyser validation against laboratory titration, with maximum 1.0 wt% HF deviation before analyser recalibration or replacement); contactor temperature monitoring with high-temperature alarm and operator response requirements; and acid relief header pressure monitoring with scrubber capacity verification. Section 6 covers acid circuit integrity inspection: material verification for HF-compatible carbon steel (hardness below 200 HB Brinell; weld heat-affected zone hardness verification to prevent stress corrosion cracking in HF service); corrosion coupon monitoring at representative acid circuit locations; inspection intervals for acid settler shell, tube bundle, and piping systems. Section 7 covers emergency response, including HF vapour detector network design, RADS system design criteria, and API RP 751 Appendix A incident documentation.

Despite these comprehensive monitoring requirements, API RP 751 specifies no adversarial robustness provisions for AI systems classifying rendered acid circuit monitoring display images at the acid containment monitoring boundary. The standard specifies what must be monitored and how frequently — but it does not address whether the AI display classification systems that have been deployed to automate the real-time interpretation of acid settler interface level displays, acid strength analyser outputs, contactor temperature trend charts, and acid relief header pressure indicators are robust against adversarial pixel perturbation at the acid containment monitoring boundary. The 5th edition update (2020) incorporated revised acid inventory management guidance, updated RADS design criteria, and new MHF additive monitoring provisions — but did not add adversarial robustness requirements for AI display classification systems, because the API RP 751 committee’s scope covers physical process safety and chemical hazard management, not adversarial machine learning robustness for the AI systems operating at the monitoring boundaries the standard governs.

OSHA PSM 29 CFR 1910.119 governs HF alkylation units at US refineries under HF’s threshold quantity of 1,000 lbs — a threshold that virtually all commercial HF alkylation units exceed by factors of 100–1,000 (typical HF inventories of 100,000–1,000,000 lbs). PSM element (e) (Process Hazard Analysis) requires PHA studies covering HF alkylation hazard scenarios; at all US HF alkylation units, PHAs identify acid settler interface high-high (HF carryover to fractionation), acid strength below operating floor (increased corrosion, reduced alkylate quality), contactor temperature high-high (HF vapour generation and relief actuation), and acid relief header pressure excursion (scrubber capacity exceedance and HF atmospheric release) as major accident scenarios. The primary prevention safeguard identified in these PHAs for all four scenarios is the acid circuit monitoring system — specifically the DCS display indicators that AI systems now classify in real time. PSM element (e) does not specify adversarial robustness requirements for the AI classifying these monitoring displays. PSM element (j) (Mechanical Integrity) requires inspection and testing of HF alkylation acid circuit components — vessels, piping, heat exchangers, instrumentation — but does not address adversarial robustness of the AI interpreting the instrumentation outputs.

EPA RMP 40 CFR Part 68 requires HF alkylation facilities to complete worst-case release scenario analysis for HF — typically producing toxic endpoint distances of 1–10 miles under stable atmospheric conditions at US HF alkylation unit inventory levels — and to submit a Risk Management Plan describing accident prevention and emergency response programmes. The EPA RMP worst-case analyses for HF alkylation quantify the community consequence of HF containment failure at the same scale as the CSB Torrance 2015 consequence finding (125,000 residents). Like API RP 751 and OSHA PSM, EPA RMP does not specify adversarial robustness requirements for AI systems classifying rendered acid circuit monitoring display images. The pattern is structurally consistent with the regulatory gap documented for FCC regenerator afterburn AI under API RP 571 / OSHA PSM: comprehensive post-incident regulatory frameworks governing HF alkylation operations — updated after major HF alkylation incidents — that do not extend requirements to the adversarial robustness of the AI display classification layer now operating at the monitoring boundaries the regulations were designed to protect. OSHA PSM 29 CFR 1910.119 has the same adversarial robustness gap across all refinery AI monitoring contexts — the Texas City BP 2005 tragedy (15 killed, 180 injured) documents the refinery APC monitoring context and PSM regulatory framework.

Glyphward threshold 35 for HF alkylation unit AI

Glyphward’s adversarial detection API operates as a pre-classification gate at each rendered-image ingestion boundary in the HF alkylation APC AI pipeline: before the acid settler interface level display AI processes each rendered DCS level indicator image, before the HF acid strength analyser display AI processes each rendered analyser trend chart, before the contactor temperature display AI processes each rendered DCS temperature trend image, and before the acid relief header pressure display AI processes each rendered pressure indicator image. Each rendered display image receives a risk score (0–100) in 8–15 ms. At or above threshold 35, Glyphward gates the AI classification and generates an alert triggering manual verification against the underlying DCS process historian data — the raw transmitter and analyser records stored as engineering-unit time series that are not accessible to pixel-level adversarial perturbation.

Threshold 35 for HF alkylation unit AI reflects three factors that place this context at the highest detection sensitivity level in the refinery portfolio — consistent with FCC regenerator afterburn AI (also threshold 35) and CDU overhead HCl corrosion AI (also threshold 35) — and above the baseline for offshore well control AI and underground mine ventilation AI (threshold 30).

First, community-scale consequence at the 125,000-resident level — directly quantified by the CSB Torrance 2015 investigation for HF alkylation unit monitoring boundary failures. The toxic endpoint distance for a major HF alkylation unit HF release (EPA RMP worst-case: full inventory release, passive mitigation) encompasses residential communities at the 100,000–500,000 population scale for refinery locations typical of US HF alkylation operations (Gulf Coast, West Coast, and mid-continent refinery corridors with significant urban encroachment on refinery fence lines since original siting). The HF aerosol-forming characteristic — the partially ionised HF vapour and fine-mist condensate that produces a denser-than-air toxic cloud at temperatures and humidity conditions typical of US coastal and inland refinery sites — produces a toxic endpoint radius that is larger at equivalent mass release than any other refinery-scale toxic chemical hazard. HF alkylation unit AI monitoring failures therefore carry the highest per-event community consequence exposure of any refinery monitoring AI context in the Glyphward threshold calibration framework.

Second, the multi-surface compounding attack creates a failure pathway that bypasses the independent HF detection network. HF alkylation units are equipped with fixed-point HF vapour monitors at the acid circuit perimeter — electrochemical or ion-selective electrode detectors calibrated to the HF IDLH of 30 ppm, with audible and visual alarm on the HF unit operations console and in the surrounding work areas. These detectors provide a layer of protection independent of the acid circuit monitoring displays classified by the AI: if HF acid is actually releasing to atmosphere, the perimeter detectors will alert regardless of what the AI classifies. However, the acid settler interface AI misclassification (surface 1) produces HF carryover to the non-acid-rated fractionation circuit — an HF containment breach that is entirely within the liquid-phase fractionation system, below atmospheric release threshold, that the perimeter HF monitors would not detect until mechanical failure of fractionation equipment (flange leak, pump seal failure, or opening for maintenance) releases the accumulated HF from the contaminated fractionation circuit. The multi-surface suppression — surface 1 (interface) plus surface 2 (acid strength) plus surface 3 (contactor temperature) plus surface 4 (relief header) — eliminates all four early indicators of acid circuit integrity degradation, allowing the system to progress from early-warning state to HF carryover to fractionation without triggering either the AI-layer safeguard or the atmospheric HF detection network. Glyphward threshold 35 at the rendered-display ingestion point is the first safeguard layer for the multi-surface compound attack — upstream of both the APC AI and the atmospheric HF monitor network.

Third, the upward-direction adversarial attack on acid strength (surface 2) requires detection capability beyond the standard downward-suppression pattern that single-surface adversarial detection might be calibrated to identify. The acid strength upward shift — showing 81.4 wt% as 87.8 wt% — moves the rendered display toward the expected on-spec range, not away from it; a naive threshold calibration based only on “readings appear normal when they shouldn’t” would classify the surface 2 attack as a high-confidence normal reading. Glyphward’s scoring of the acid strength display image at threshold 35 reflects multi-vector adversarial perturbation detection: the ±8 DN upward shift introduces the same pixel-level statistical anomalies in the display rendering (noise-floor distortion, histogram shift away from the display’s natural rendering distribution, local gradient inconsistencies at the trend line edges) as the downward-direction attacks on surfaces 1, 3, and 4, regardless of whether the shift direction produces a reading that appears higher or lower than the unperturbed value. The false positive cost at threshold 35 is identical to other refinery AI contexts: 1–3 minutes of manual verification of the acid strength analyser output against the DCS historian trend, and optionally against a laboratory spot-sample acid titration (30–60 minutes for laboratory confirmation). The false negative cost of an undetected acid strength attack — degraded acid at 81.4 wt% classified as on-spec 87.8 wt%, ASO accumulation unmanaged, acid strength continuing to decline below 80 wt% where HF corrosivity and volatility increase further — compounds the surface 1 and 3 failure pathways by ensuring that the acid entraining into the fractionation circuit is more corrosive and more volatile than the process team would expect from a unit the AI has been reporting as in normal operation. At threshold 35, the Glyphward false negative/false positive cost asymmetry for HF alkylation unit AI is calibrated consistent with the highest-consequence refinery monitoring AI contexts in the portfolio.

Free tier — 10 scans/day, no card required. Submit a rendered acid settler interface level display, HF acid strength analyser DCS screen, contactor temperature trend chart, or acid relief header pressure indicator from your HF alkylation unit monitoring system to the Glyphward scanner to generate a baseline adversarial risk score for your HF alkylation APC AI inputs.