> For the complete documentation index, see [llms.txt](https://docs.layerlens.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.layerlens.ai/more-in-this-section-2/scenarios-7.md). # Manufacturing and industrial scenarios Five canonical scenarios where manufacturers, industrial operators, and field-service teams use Stratix evaluation to gate AI features. ## 1. Predictive maintenance and asset health **Setup:** AI ingests sensor telemetry, work-order history, failure-mode databases; produces remaining-useful-life estimates, recommended interventions, and parts/labor pre-pulls. **Why it's hard:** * Failures are rare events — most data is "normal" * Class imbalance crushes naive models * Wrong "all clear" leads to unplanned downtime; wrong "imminent failure" leads to wasted maintenance **Stakes:** * Unplanned downtime in process industries can cost \~$260K/hour or more (industry estimates) * Safety failures can be catastrophic * OEE (Overall Equipment Effectiveness) targets directly tie to plant P\&L **Stratix evaluation criteria:** | Aspect | What you measure | Pattern | | ------------------------ | ------------------------------------------------ | -------------------------------- | | RUL calibration | Predicted vs. actual time-to-failure | Deterministic on labeled history | | False-positive rate | Bad calls that triggered unneeded interventions | Code assertion | | False-negative rate | Missed failures | Tracked as drift signal | | Critical-asset rule | High-criticality assets always escalate to human | Hard rule | | Recommendation grounding | Cited failure-mode supports recommendation | Citation faithfulness judge | ## 2. Quality and visual-inspection AI **Setup:** Computer vision plus text reasoning identifies defects on production lines; explanations route to QA engineers. **Why it's hard:** * Defect taxonomy varies per line, per product, per shift * Lighting and substrate variation defeat naive models * Some defects matter for safety, others for cosmetics — categorization drives wildly different actions **Stakes:** * Recalls in regulated industries (auto, medical device, food) can run into hundreds of millions * Customer-experienced defects drive reputation cost beyond direct recall expense * Process-control records support FDA, FAA, NHTSA inquiries **Stratix evaluation criteria:** * **Per-class precision/recall** — Defect class-by-class, not aggregate * **Critical-defect rule** — Hard rule: safety-critical defects route to immediate stop-and-hold * **Operator-explanation quality** — Judge: would a line operator know what to do next? * **Drift detection** — Daily comparison vs. golden image set * **False-pass rate floor** — Hard threshold; any breach triggers line audit ## 3. Supply chain and procurement assistance **Setup:** AI surfaces supplier-risk signals, recommends substitutions for shortage-affected parts, drafts RFx documents. **Why it's hard:** * ITAR / EAR / OFAC compliance — wrong supplier can be a sanctions violation * Substitution may not meet safety / regulatory specs * Tier-N supplier visibility is poor **Stakes:** * OFAC violations carry per-incident fines + criminal exposure for willful violations * Substitution that breaks regulatory spec triggers recall + liability * Single-source dependencies drive multi-billion-dollar industry shocks **Stratix evaluation criteria:** * **Sanctions-list rule** — Hard rule: every supplier checked against current OFAC, BIS, State Dept lists * **Spec-match rule** — Hard rule: substitution must match regulatory spec (e.g., RoHS, REACH, Mil-Spec) * **Risk-citation rule** — Risk flags cite the specific signal (sanctions, financial distress, geographic risk) * **Country-of-origin accuracy** — Code assertion against verified database * **Audit-trail completeness** — Every recommendation logs evidence chain ## 4. Technical Q\&A and field-service assistance **Setup:** Field technician asks questions about a specific machine: troubleshooting steps, parts identification, safety procedures. AI grounds answers in maintenance manuals, prior tickets, OEM bulletins. **Why it's hard:** * Manual versions vary across product generations * Safety procedures must reflect current OSHA / OEM bulletins * A confident wrong answer can injure a tech **Stakes:** * Worker injury or death directly tied to incorrect AI guidance * OSHA recordable incidents drive plant-level scrutiny * Bad-faith failure-to-train allegations follow workplace incidents **Stratix evaluation criteria:** * **Manual-version match rule** — Hard rule: cited manual version applies to the actual machine * **Safety-procedure rule** — Hard rule: any answer involving energized/pressurized/hot work cites LOTO and PPE requirements * **Faithfulness judge** — Output grounded in cited manual section * **Out-of-scope refusal** — Hard rule: questions outside the machine's documented scope must be refused ## 5. Operator copilot and process safety **Setup:** Process operators ask the AI about set points, alarms, batch records. AI helps interpret state, surface relevant SOPs, draft shift-handover notes. **Why it's hard:** * Process-safety regulations (PSM, EPA RMP) are unforgiving * Cyber-physical exposure — wrong recommendation can crash a process * Shift-handover quality directly drives next-shift incident rates **Stakes:** * Process-safety incidents can be catastrophic (Bhopal-tier worst case; commonly multi-fatality if the chemistry is severe) * OSHA PSM and EPA RMP carry citations and criminal exposure * Insurance carriers price plant premiums on process-safety performance **Stratix evaluation criteria:** * **Set-point-recommendation rule** — Hard rule: AI never auto-applies; advisory only * **SOP-citation rule** — Output cites the specific SOP version * **Hazard-disclosure rule** — Any procedure involving energetic materials carries the required hazard disclosure * **Handover-completeness scorer** — Required fields present (alarms, deviations, equipment status) * **Plain-language scorer** — Reading level appropriate for the operator population ## See also * [Manufacturing eval patterns](/more-in-this-section-2/eval-patterns-7.md) * [Manufacturing compliance](/more-in-this-section-2/compliance-7.md) --- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter: ``` GET https://docs.layerlens.ai/more-in-this-section-2/scenarios-7.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.