PaperlayerView exact revision on arXiv

Source-linked AI summary

XPhysICS: Cross-Physical-Domain Threat Grounding for Industrial Control Systems Security

Sangshin Park, Jainta Paul, Lawrence Ponce, Md Raihan Ahmed, Mu Zhang, Luis Garcia

arXiv:2609.30805v2cs.CRcs.AI

TL;DR

Industrial threat reports are tied to their source plants, while similar component names alone do not show that another target can support or evaluate the same physical effect. XPhysICS grounds source evidence against a target contract and produces validation slices; across evaluated settings, its studies distinguish structural grounding from slice execution, downstream compatibility, and realizability.

  • Problem

    Documented threats use source-plant evidence, and similar labels or components do not establish whether a specific target can support and evaluate the same physical effect.

    Source: §1 · p. 2

  • Method

    XPhysICS creates a provenance-linked threat abstraction from source evidence, checks it against a target contract, and materializes accepted groundings as validation slices.

    Source: §1 · p. 2

  • Results

    The evaluation covers 78 continuous-process abstractions, executable water-domain slices, nine additional Hydro and chemical-process cases, upstream GeCo compatibility, and a separate three-objective, one-target realizability study.

    Source: §8 · p. 27

  • Takeaways & Limitations

    Grounding acceptance, slice adequacy, dynamic realizability, and downstream analysis outcomes are distinct questions, even when they concern the same grounded threat.

    Source: §8 · p. 27

    The reported grounding evaluation covers five of the seven effect families available during source abstraction, leaving two outside its grounding surface.

    Source: §6.1 · p. 24

Abstract

from arXiv · show

Industrial control system attacks are usually documented in terms of the plant where they occurred: its sensors, actuators, process stages, and control logic. Yet many attacks express a more general physical pattern--such as suppressing flow, corrupting chemical dosing, or driving a vessel toward overflow--that may also matter in a different plant. The challenge is deciding when such a threat remains meaningful on a new system rather than relying on similar component names or broad semantic labels. We present XPhysICS, a methodology for grounding documented cyber-physical threats onto a specific target system. XPhysICS converts source evidence into a provenance-linked description of what is manipulated, what physical consequence is expected, and what observations the evidence calls for. Once this analyst-guided abstraction, its vocabulary and schema version, and a target contract are fixed, XPhysICS applies deterministic grounding checks. An accepted result can be represented as a validation slice that records the mapped roles, signals, dependencies, and context intended to support later evaluation. We study 83 threat abstractions across continuous-process and manufacturing sources using separate evaluation denominators. The continuous-process study evaluates 78 abstractions against target contracts spanning water treatment, water distribution, hydropower, and chemical processes. Selected cases are exercised through controlled perturbations of simulator-role signals. We also test compatibility with several analysis styles, including the released upstream GeCo implementation, and conduct a three-objective, one-target realizability study using a paper-derived search reproduction. Across these evaluated settings, the results support treating explicit target checks and traceable evidence as separate from semantic similarity alone.

1 Introduction

XPhysICS addresses the gap between documented source threats and evidence available for a specific target system. It combines analyst-guided threat abstraction with deterministic target checks and target-specific validation slices, then evaluates the approach across multiple settings and downstream analysis styles.

1 Introduction

  • XPhysICS targets a gap between semantic correspondence and evidence that a specific system can support and evaluate a documented physical effect.Existing classifications and knowledge representations organize threats and relationships, but similar labels alone do not establish target support or evaluability.Source: §1 · p. 2
  • XPhysICS separates analyst-guided source-threat abstraction from deterministic grounding against a fixed target contract, using five explicit eligibility checks.The checks cover component coverage, role/type compatibility, source-stage coherence, minimum-tag slice viability, and rule-surface intersection.Source: §1 · pp. 2–3
  • The evaluation spans continuous-process and manufacturing settings, with grounding coverage, executable slice studies, downstream compatibility tests, and a bounded realizability study reported separately.The studies use distinct denominators and answer different questions.Source: §1 · pp. 2–3
  • A validation slice records mapped manipulation and consequence paths, observable signals, dependencies, timing assumptions, and consumer-relevant context for later evaluation.This lets grounding acceptance, slice adequacy, consumer applicability, and consumer outcome be assessed separately.Source: §1 · p. 3
  • XPhysICS preserves provenance through schema-validated source abstractions and target contracts, versioned vocabularies, and an evidence ledger linking source evidence to grounding decisions.CrossPLC controller enrichment and temporal guardrails are optional.Source: §1 · pp. 2–3

2 Operational Context and Problem Statement

Recurring component names across industrial plants do not establish equivalent physical roles, process context, or observability. XPhysICS therefore checks whether source threat abstractions fit a target contract, separates grounding from downstream use, and requires traceable evidence and explicit scope.

2 Operational Context and Problem Statement

  • Similar pumps, valves, tanks, or interlocks across plants do not establish equivalent physical roles, process context, or observability.The implemented filters check component coverage, role/type, source-stage coherence, minimum-tag slice viability, and rule intersection; complete source-role coverage and slice adequacy require separate evidence.Source: §2 · pp. 3–4, Figure 1 · p. 4

2.1 Problem Statement

  • XPhysICS determines whether a source threat abstraction is structurally admissible under a selected system's target contract and constructs a target-specific validation slice for accepted groundings.The slice includes manipulated and consequence paths, observable signals, dependencies, timing assumptions, and consumer-relevant context; its adequacy is assessed separately from grounding acceptance.Source: §2.1 · p. 4

2.2 Use by Downstream Analyses

  • A validation slice can expose target context for different downstream analyses, while consumer applicability and outcome remain separate from grounding and slice adequacy.Analyses may need different combinations of ranges, correlations, time-series windows, trajectories, operating context, and provenance.Source: §2.2 · pp. 4–5

2.3 Design Requirements and Scope

  • The design requires target-specific decisions, explicit eligibility filters, independently assessed slice adequacy, and provenance linking evidence to mappings and outcomes.Controller artifacts are optional, and extension studies use separate evidence surfaces and denominators from the primary continuous-process evaluation.Source: §2.3 · pp. 5–6

3 XPhysICS Framework

XPhysICS connects source-side threat evidence to target-conditioned evaluation without requiring source and target plants to be equivalent. Its workflow separates analyst-guided abstraction from deterministic grounding, then constructs a validation slice whose adequacy is assessed separately.

3 XPhysICS Framework

  • XPhysICS uses four stages—source threat abstraction, target-conditioned grounding, validation-slice construction, and slice adequacy with downstream interfaces—to assess threats against a specified target.The objective is not plant equivalence, but whether documented source semantics satisfy explicit structural and evaluability requirements.Source: §3 · p. 6

3.1 Source Threat Abstraction

  • The source threat abstraction records an effect family, manipulated and consequence roles, required observations, and provenance linking fields to source evidence.It can draw on heterogeneous source documents, traces, or controller artifacts; ambiguous evidence may require analyst interpretation before grounding.Source: §3.1 · pp. 7–8

3.2 Analyst Workflow

  • Analysts select and interpret source evidence and construct target contracts, while normalization, candidate mapping, eligibility checks, and slice construction follow predefined procedures once inputs are fixed.A bounded review checks a subset of abstractions against source evidence, rather than providing an independent multi-analyst agreement study.Source: §3.2 · p. 9

3.3 Target-Conditioned Grounding

  • For fixed inputs, grounding requires five implemented filters, but acceptance does not establish complete role coverage, a target manipulation-to-consequence path, or slice adequacy.The reported grounding evaluation covers a five-family subset of the seven source abstraction effect families.Source: §3.3 · p. 10

3.4 Validation Slice Construction

  • An accepted grounding becomes a bounded, target-specific validation slice preserving the manipulation surface, expected effect path, required observations, and obligations for later analyses.The slice provides an intermediate representation between semantic correspondence and whole-plant equivalence.Source: §3.4 · p. 11

3.5 Slice Adequacy and Downstream Interfaces

  • Slice adequacy is assessed for the intended evaluation using signal coverage, dependency closure, and timing evaluability, with strong, partial, or insufficient classifications.Adequacy is distinct from consumer applicability and outcome; it does not establish whole-plant fidelity or dynamic realizability.Source: §3.5 · p. 11

4 Implementation

XPhysICS turns source evidence and target-system evidence into structured validation artifacts, then separates grounding acceptance from slice adequacy and downstream evaluation. Optional enrichment and consumer analyses add evidence when their required artifacts are available.

4.1 Evidence Records and Threat Abstractions

  • Source records preserve provenance, and threat abstractions capture manipulated roles, consequences, observation obligations, timing, confidence, and supporting evidence.The primary grounding results use structured or curated extraction, not exploratory LLM-assisted extraction.Source: §4.1 · p. 12

4.2 Target Contracts and Study Bundles

  • Target contracts are machine-validated descriptions of target roles, signals, dependencies, effect-family scope, rules, and evaluation context.The grounding vocabulary supports five of the seven extraction effect families; abstractions outside that scope are reported as outside the current grounding surface, not rejected threats.Source: §4.2 · pp. 12–13

4.3 Deterministic Grounding and Slice Generation

  • Grounding tests candidate mappings against five eligibility predicates, while retaining per-check diagnostics and provenance in the grounding record.Candidate selection follows a fixed precedence and contract order; determinism does not imply that alternative candidate orders converge.Source: §4.3 · p. 13
  • Validation slices preserve the signals, dependencies, timing assumptions, and context needed for target-side evaluation without claiming to be complete digital twins.Adequacy is assessed independently through signal coverage, dependency closure, and timing evaluability.Source: §4.3 · p. 14

4.4 Bounded Validation Consumers

  • Four bounded consumer families evaluate slices using declared invariants, predictive residuals, state-conditioned thresholds, or phase-conditioned scoring.The bounded GeCo-, SAIN-, and SCAPHY-style lanes test whether slices supply information for those analysis styles, not comparative performance against the published systems.Source: §4.4 · p. 14

4.5 Optional Enrichment Paths

  • Optional CrossPLC enrichment recovers controller-side routines, tags, dependencies, call relationships, and state when PLC or program artifacts are available.Optional STL guardrails provide slice-level temporal robustness assessments; in this evaluation they are bounded extension evidence, not a grounding criterion or full-system verification.Source: §4.5 · pp. 14–15

4.6 Assessment Outputs and Provenance

  • Assessment packages record grounding decisions, slice adequacy, consumer outputs, contextual limitations, and provenance for the selected target.An evidence ledger links source documents and field-level evidence to threat objects and grounding records, with JSON, CSV, and Cypher exports.Source: §4.6 · pp. 15–16

5 Evaluation

The evaluation uses separate denominators and evidence levels, from 83 source abstractions to selected executable slices and bounded extensions. Across these settings, structural grounding, slice adequacy, dynamic realizability, and consumer outcomes are treated as distinct measurements.

5.1 Source Semantics and Provenance

  • The source corpus contains 83 abstractions from 12 documents, with 78 continuous-process abstractions and five manufacturing abstractions evaluated separately.The continuous-process set comprises 51 SWaT, 17 WADI, and 10 OilTreatment abstractions; the broader ledger is not a count of independently validated or executable attacks.Source: §5.1 · p. 16

5.2 Continuous-Process Grounding Coverage

  • Grounding coverage is highest for water-domain sources evaluated against water-domain targets and lower for several adjacent process families.The observed differences reflect eligibility conditions in the target contracts, including role/type compatibility, slice viability, and rule-surface intersection.Source: §5.2 · p. 17, Figure 4 · p. 18

5.3 Testing Grounded Water-Domain Cases

  • All five selected water-domain slices meet adequacy requirements, but controlled perturbations reveal distinct consumer outcomes, including nominal-confounded violations and responses near decision boundaries.The cases are a selected subset with available execution evidence, not a random sample or execution of all 78 abstractions.Source: §5.3 · pp. 18–19

5.4 Consumer Compatibility over Shared Slices

  • Shared WT/WD slices support several bounded analysis styles, but their compatibility results do not establish fidelity to or comparative performance against the published systems.The lanes include invariant, predictive, state-aware, and phase-aware analyses; the bounded SCAPHY-style lane identifies the WD low-flow boundary case, raising observed WD coverage from 2/3 to 3/3.Source: §5.4 · pp. 19–20

5.5 Bounded-Support Validation Slices: Hydro and Chemical-Process Targets

  • Nine Hydro and chemical-process cases support slice executability and bounded consumer compatibility, but unavailable nominal-subtracted baselines limit outcome interpretation.Several Hydro cases share an execution bundle, and the results do not establish target-independent behavior or validation against independently executed attacks.Source: §5.5 · p. 20

5.6 Upstream GeCo as a Consumer

  • Across nine upstream GeCo executions, recall is 1.0 with tp = 21 and fn = 0 for each case, while false positives range from 2 to 3.Only two distinct eTaF1 values occur; shared Hydro target executions mean the nine cases are not nine independent target-side observations.Source: §5.6 · pp. 20–21

5.7 Dynamic Realizability: A Bounded Example

  • On the specified model and command surface, the pressure objective is reached while both level objectives are not; under the live OpenPLC-driven configuration, all three are reached.This bounded study covers three objectives on one target and does not establish a general impossibility result for the two model-specific negative outcomes.Source: §5.7 · pp. 21–22

5.8 Extension Evidence and Boundaries

  • Extension studies instantiate the representation over additional evidence types and mechanisms, but do not provide the same target-side evidence as the primary WT/WD studies.They use separate evidence scopes and denominators rather than being combined with the continuous-process grounding percentages or primary validation-slice results.Source: §5.8 · p. 22

6 Discussion

XPhysICS treats structural grounding as distinct from slice adequacy, dynamic realizability, and downstream analysis outcomes. Its evaluation is bounded by fixed representations and criteria, available evidence, selected research-environment cases, and the evaluated target domains.

6 Discussion

  • A grounding pass means an abstraction satisfies five implemented filters, not that its slice is adequate, its effect is dynamically realizable, or an exploit would succeed.Later evaluation can reveal insufficient evidence or nominal-confounded outcomes without changing the grounding decision.Source: §6 · pp. 22–24
  • The downstream studies assess consumer compatibility, not comparative detector performance; only the GeCo study uses a released upstream implementation.The SCADMAN- and ICSTracker-motivated adapters exercise interfaces without running those published systems.Source: §6 · p. 23
  • Output quality depends on complete, accurate source and target evidence, while deployment also requires alias reconciliation, normalization, timestamp alignment, and dependency maintenance.Incomplete or stale documentation can prevent plausible mappings or reduce slice adequacy.Source: §6 · pp. 23–24
  • Determinism applies only after the abstraction, vocabulary and schema version, and target contract are fixed; the current grounding vocabulary supports fewer effect families than extraction.Source interpretation and target-contract construction may involve analyst judgment or source-specific parsing.Source: §6 · p. 24

6.1 Threats to Validity

  • The construct depends on a study-developed vocabulary and five eligibility predicates that are not derived from an external standard; the reported matrix covers five of seven effect families.The implemented filters measure acceptance, not independently adjudicated mapping correctness or exploit success.Source: §6.1 · pp. 24–25
  • Abstraction evaluation used a bounded 20-record spot-check, not an independent multi-analyst study, so it establishes neither inter-annotator agreement nor identical independent abstractions.The deep validation cases were selected for available contracts, signals, dependencies, and perturbation surfaces rather than sampled randomly or exhaustively.Source: §6.1 · p. 25
  • Controlled synthetic perturbations and research-environment evaluations do not establish independently executed cross-domain attacks or generalization beyond the evaluated water, energy, chemical, and bounded manufacturing settings.Large offsets may be easier to detect than subtler manipulations; deep executable evidence remains concentrated in SWaT→WT and WADI→WD.Source: §6.1 · p. 25

7 Related Work

Related work offers structured threat representations, target-side analyses, and controller-level techniques, but these address different stages or questions. XPhysICS connects externally documented threat semantics to target contracts and validation slices, with optional evidence enrichment and separate downstream analyses.

7 Related Work

  • Systems-engineering approaches connect architecture, mission objectives, hazards, and resilience, whereas XPhysICS grounds externally documented threat evidence onto a selected target without requiring a complete mission-engineering model.Its validation slice records manipulated and consequence paths, observability requirements, and target-side evaluation context.Source: §7 · pp. 25–26
  • Threat ontologies and knowledge graphs organize threat, asset, and evidence information, while XPhysICS additionally tests candidate mappings against a target contract and preserves provenance through grounding.The implemented criteria include component coverage, role/type, source-stage, minimum-tag, and rule-intersection checks.Source: §7 · p. 26
  • Downstream systems analyze characterized targets; XPhysICS constructs target-conditioned context and validation slices that expose signals, dependencies, and consumer-relevant information for later analysis.Compatibility studies include native lanes and the released upstream GeCo pipeline, but assess compatibility rather than detector comparisons or cross-system fidelity.Source: §7 · p. 26
  • CrossPLC can optionally recover controller-side roles and dependencies from PLC artifacts, while STL/RTAMT can evaluate temporal robustness when a grounded slice exposes the required signals and thresholds.Neither is required for grounding acceptance.Source: §7 · p. 26
  • XPhysICS and ICSFlux address complementary questions: structural admissibility of documented threat semantics versus dynamic realizability under a model and command surface.The bounded study used a paper-derived ICSFlux search reproduction on three frozen groundings, not the published artifact or an implemented integrated workflow.Source: §7 · p. 27

8 Conclusion

XPhysICS grounds documented threat semantics against a target contract using deterministic checks once its inputs are fixed, and records accepted results as provenance-linked validation slices. Evaluation across 78 continuous-process abstractions supports explicit target checks and traceable evidence as distinct from semantic similarity alone.

8 Conclusion

  • XPhysICS makes grounding deterministic only after the source abstraction, vocabulary and schema version, and target contract are fixed, and records accepted mappings, signals, dependencies, timing assumptions, and provenance.Slice adequacy, dynamic reachability, and downstream analysis results remain separate questions.Source: §8 · p. 27
  • The grounding matrix evaluates 78 continuous-process abstractions under implemented target-contract filters, while the deepest executable evidence remains the water-domain SWaT→WT and WADI→WD studies.Nine Hydro and chemical-process cases extend controlled slice execution but are research-environment perturbations, not independently captured cross-domain attacks.Source: §8 · p. 27
  • Across evaluated contracts and cases, semantic similarity alone did not characterize target applicability; source interpretation needed traceable evidence and candidate mappings needed explicit target checks.The paper proposes broadening the target-contract vocabulary and extending deep executable evaluation to additional target families and control architectures.Source: §8 · p. 27

A Source Evidence and Extraction Details

The appendices provide source-evidence and extraction metadata, then expand the water-domain slice results with adequacy, perturbation, and consumer-outcome detail.

A Source Evidence and Extraction Details

  • Table 7 reports source- and domain-level evidence, manual spot-check results, and provenance metadata associated with the evaluated source abstractions.The evidence representation is informed by OTThreat, while XPhysICS implements the evidence ledger and records.Source: A Source Evidence and Extraction Details · p. 28, Table 7 · p. 28

B Water-Domain Validation-Slice Details

  • Table 8 separates raw consumer responses from nominal-subtracted outcomes to distinguish nominal-confounded observations and cases near the evaluated consumer’s decision boundary.Its metadata also covers slice adequacy and controlled perturbations; target signals come from the corresponding SPHERE contracts.Source: B Water-Domain Validation-Slice Details · p. 28, Table 8 · p. 28

C Bounded-Support Validation-Slice Details (Hydro/GRFICS)

All nine Hydro/GRFICS bounded-support cases satisfy the declared adequacy requirements, but they lack nominal-subtracted baselines. Their earliest recorded rule violations occur at perturbation onset, which supports slice executability rather than independent attack evidence.

C Bounded-Support Validation-Slice Details (Hydro/GRFICS)

  • All nine bounded-support cases satisfy the declared adequacy requirements for their respective studies.The cases lack nominal-subtracted baselines, so their rule responses do not receive the clean consumer-outcome classification used for flagship cases with corresponding nominal evidence.Source: C Bounded-Support Validation-Slice… · p. 29
  • The earliest recorded rule violation occurs at sample 20, the perturbation onset, in all nine cases.This timing is execution metadata, not a substitute for nominal-subtracted evidence or evidence of independently executed attacks.Source: Table 9 · p. 29, C Bounded-Support Validation-Slice… · p. 29
  • Observed rule responses characterize bounded consumer behavior without a nominal-subtracted baseline.The baseline is unavailable for these cases.Source: Table 9 · p. 29, C Bounded-Support Validation-Slice… · p. 29

D Bounded Consumer Implementations

The bounded consumer lanes use XPhysICS-native implementations of selected analysis patterns over a common validation-slice representation. They assess whether that representation supplies the signals and context those styles require, not the published systems themselves.

D Bounded Consumer Implementations

E Additional Downstream Adapter Evidence

The downstream-adapter results characterize XPhysICS-authored prototypes, not executions of the published SCADMAN or ICSTracker systems. The evidence includes a classification trade-off and provenance-workflow output under evaluated configurations.

E Additional Downstream Adapter Evidence

  • The reviewed SCADMAN-Light artifact used a weighted attack-signature classifier with a fixed threshold, not the control-flow-attestation mechanism considered by the adapter.This observation is limited to the reviewed artifact; no corresponding implementation assessment was performed for semantic-icstracker.Source: E Additional Downstream Adapter Evidence · p. 31
  • At 1 FA/hr, the semantic classification stage changes precision from 0.375 to 1.000 and F1 from 0.545 to 0.800, while recall changes from 1.000 to 0.667.These measurements characterize the evaluated XPhysICS prototype, not published SCADMAN or a comparison against it.Source: E Additional Downstream Adapter Evidence · p. 31, Table 11 · p. 31
  • The classification adapter consumes detector windows and emits typed events with robustness and provenance metadata after evaluating target-grounded temporal predicates.Source: E Additional Downstream Adapter Evidence · p. 31
  • The provenance-workflow prototype instantiates all 17 templates, produces 2,074 typed mappings, and records rerun agreement of 0.95 under the evaluated configuration.These measurements characterize the XPhysICS prototype workflow, not an execution or evaluation of published ICSTracker.Source: E Additional Downstream Adapter Evidence · p. 31, Table 11 · p. 31
  • The provenance adapter consumes grounded mappings, eligibility outcomes, projected edges, impact chains, and lineage records to instantiate rule templates over eligible mappings.Source: E Additional Downstream Adapter Evidence · p. 31

F Extension Evidence Ledger

The extension studies cover evidence surfaces beyond the primary continuous-process evaluation and report results under study-specific denominators. Figure 5 presents the separate TXT-to-S7 manufacturing/program-artifact grounding study.

F Extension Evidence Ledger

  • Extension studies cover source-semantic breadth, manufacturing/program-artifact grounding, controller-code enrichment, temporal guardrails, protocol evidence, and event/timing obligations.Their results are reported separately from continuous-process grounding percentages and primary WT/WD validation-slice studies.Source: F Extension Evidence Ledger · p. 32
  • Table 12 identifies each evaluated extension, its principal result, the XPhysICS capability exercised, and the result's evidentiary scope.Source: Table 12 · p. 32
  • Figure 5 shows Fischertechnik/TXT source abstractions grounded against a Siemens S7 target using available program, tag, and state evidence.This manufacturing/program-artifact study uses a denominator distinct from the 78-threat continuous-process corpus in Figure 4.Source: F Extension Evidence Ledger · p. 32, Figure 5 · p. 33
Loading 2609.30805v2…