See It Work
Documentation Support Trust Center Partners
See It Work
SYSTEM: OPERATIONAL OT/IT CONNECTORS: 150+ AUTONOMOUS OPERATION: 15+ DAYS GOVERNED AUTONOMY: ENFORCED AUDIT TRAIL: IMMUTABLE INDUSTRIES: ASSET-INTENSIVE & MISSION-CRITICAL DEPLOYMENT: 3-6 MONTHS VIA APEX CONTROL LOOPS: 3,400+ SYSTEM: OPERATIONAL OT/IT CONNECTORS: 150+ AUTONOMOUS OPERATION: 15+ DAYS GOVERNED AUTONOMY: ENFORCED AUDIT TRAIL: IMMUTABLE INDUSTRIES: ASSET-INTENSIVE & MISSION-CRITICAL DEPLOYMENT: 3-6 MONTHS VIA APEX CONTROL LOOPS: 3,400+
Available HARNESS-OIL-SAMPLE-CLASSIFIER-001 AI Workflow Harness

Oil Sample Classifier

Classifies oil sample results against known failure modes — structured output, confidence, and a recommended review path on the Data Stream Designer canvas.

MiningManufacturingOil & GasEnergy & UtilitiesIron & Steel Reliability & Maintenance

Target outcome · Turn lab oil samples into structured classifications the reliability engineer can act on without re-reading every record.

Business problem

Lab oil sample results arrive without consistent classification against known failure modes. Reliability engineers spend interpretation time on patterns they have already seen — sediment, water ingress, additive depletion, oxidation, wear-metal signatures — instead of focusing on the cases that warrant deeper investigation.

Read more Show less

The bottleneck is not the lab. It is the manual reading-against-context step between lab result and intervention decision.

What it does

Reads each new oil sample result against the asset’s history, the failure mode library, and the operating context, then produces a structured classification (which failure mode the sample matches, if any), a confidence score, and a recommended review path for the engineer to follow.

Read more Show less

Context assembly, output validation, and routing decisions all remain inspectable on the Data Stream Designer canvas — the harness adds governed reasoning, not autonomy.

Agent structure

  • Read the new oil sample result, asset history, and failure mode library from operational context
  • Classify the result against known degradation patterns
  • Produce a structured output with classification + confidence score
  • Recommend a review path (no-action / re-sample / engineer review / immediate escalation)
  • Log input sample, classification logic, output validation, and routing decision for audit

Operational profile

Classify lab result against known failure modes.

Control mode: Human-Controlled

User it helps
Reliability engineer, maintenance planner.
Context it uses
Oil sample result, asset history, failure mode library, operating context.
Decision supported
Does this result indicate abnormal degradation or a known failure pattern?
Action or output
Structured classification, confidence, recommended review path.
Evidence captured
Input sample, classification logic, output validation, routing decision.

What the team handles

Handles

Pattern-match each new sample against the failure mode library; produce structured classification, confidence, and recommended review path; capture full audit trail of input + logic + output + routing.

Does not handle

Final intervention decisions, work-order creation, asset removal from service, or trend-over-time degradation modelling — those belong to downstream advisors or decision teams.

Humans retain authority over

All intervention and capital decisions. The classifier output is reviewed by the reliability engineer or maintenance planner before action.

Current process vs. with AI Workflow Harness

TODAY · RELIABILITY & MAINTENANCEREACTIVE
×
Reading a new oil sample resultEngineer manually compares sample numbers against historical patterns and failure mode reference data.
×
Triaging which samples need deeper reviewSamples reviewed in turn or triaged by gut feel based on queue size.
×
Auditing the classification logicEngineer’s reasoning lives in notes or memory; difficult to trace later.

Outcomes and measurement

Engineer time per routine oil sample triage

Baseline Full manual read against history and failure mode library
With agent Pre-classified read with structured output to confirm or override

Auditability of classification logic

Baseline Engineer notes or memory
With agent Full per-sample trace of input, logic, validation, and routing

Consistency across engineers and shifts

Baseline Varies with individual experience and queue pressure
With agent Same logic applied to every sample, same evidence captured every time

*All figures are typical ranges. Achievable range depends on existing control maturity, data quality, and site-specific conditions.

Data inputs

Lab / LIMS

oil sample result fields (viscosity, particle count, water content, additive levels, wear metals)

CMMS / asset history

prior samples for the assetwork order and failure history

Failure mode library

known patterns and reference signatures

Operating context

asset duty cycleoperating envelope

*Categories only — no tag names or system-specific field references. Exact data mapping is scoped per site.

Scoping questions

Expect these questions in a first scoping conversation. They signal engineering discipline and help narrow the template to your specific site context.

  1. Which oil sample fields are captured today, and in which system (LIMS, CMMS, spreadsheet, lab portal)?
  2. How comprehensive is the existing failure mode library — coverage of degradation patterns, water ingress, contamination, additive depletion, wear metals?
  3. What is the routing today after a sample is read — direct to engineer review, queued by criticality, or fixed schedule?
  4. What confidence threshold and review paths do you want for routine, unclear, and high-risk classifications?
  5. Where should classification audit records live — Data Stream Designer trace, CMMS, or both?

Want our AI to walk you through these scoping questions?

SPEAK WITH OUR TEAM

Get specialist advice on scoping this for your site.

Our specialists will help you understand how the Oil Sample Classifier fits your operations, what data you'd need, and what a scoping engagement typically looks like.

← Browse all templates