Predictive Maintenance AI Tools

Predictive Maintenance exposes AI capabilities through two layers:

• MCP tools under /mcp/pdm/ for external agents and enterprise AI clients.
• AI Engine endpoints under /ai/pdm/* for model training, detection, forecast, diagnosis, and payload processing inside the Predictive Maintenance runtime.

Use MCP tools when an Agent workflow needs governed read access to health, anomaly, and component intelligence. Use AI Engine endpoints when the product workflow needs to train, score, diagnose, forecast, or process maintenance signal payloads.

Tool layers

Layer	Primary users	Access boundary	Output type
/mcp/pdm/ MCP tools	Agent clients, copilots, workflow orchestrators	pdm.read scope	Health summaries, anomaly context, component intelligence, circular recovery outlook
/ai/pdm/* AI Engine endpoints	Predictive Maintenance backend workflows and model console	Product service authentication and workflow permissions	Model status, anomaly scores, forecasts, health score, diagnosis, remaining-life estimate
/api/v1/pdm/* module APIs	Predictive Maintenance UI and backend integration	pdm:read and pdm:write permissions	Equipment profiles, dashboards, advisories, templates, work-order feedback, reliability metrics

For user-facing explanations, write Predictive Maintenance. Use pdm only when showing API paths, MCP endpoints, scopes, or permission strings.

MCP endpoint and scope

Endpoint	Scope	Use
/mcp/pdm/	pdm.read	Read equipment health, fleet summary, anomaly list, filter component intelligence, and circular recovery outlook.
/mcp/base/	base.read	Add asset records, documents, work orders, inspections, or SOP context when a maintenance answer needs supporting evidence.
/mcp/base/	base.action.write	Draft follow-up actions after a responsible engineer approves the recommendation.

MCP tool map

Tool	Primary job	Typical question
get_pdm_summary	Summarize fleet health and alert state.	"Which equipment groups need maintenance attention this week?"
list_pdm_anomalies	List open or historical anomaly events by severity and status.	"Show critical anomalies that are still open."
get_equipment_health	Inspect health for one equipment item.	"Why is this pump graded C?"
get_filter_component_intelligence	Find filter components that need attention.	"Which components should maintenance review now?"
get_filter_circular_recovery	Review recovery, reuse, aftermarket, remanufacture, or disposal outlook for filter components.	"Which components have recovery value after replacement?"

get_pdm_summary

Use get_pdm_summary at the start of a maintenance review. It gives the Agent a fleet-level view before drilling into equipment or anomalies.

Field	Details
Scope	pdm.read
Inputs	Optional tenant_filter when the environment supports tenant or fleet filtering.
Returns	Equipment count, grade distribution, critical count, active alerts, and 7-day health trend.
Best used for	Daily triage, weekly maintenance review, fleet-level prioritization, and narrowing the next query.
Follow-up tools	list_pdm_anomalies for events, get_equipment_health for a specific asset, /mcp/base/ tools for work orders and asset documents.

Review checks:

• Confirm the tenant, site, equipment group, and time window with the requester.
• Treat grade distribution as a triage view; ask for equipment-level evidence before recommending field action.
• Include data freshness and missing source notes when the workflow has DFS quality context.

Example Agent flow:

1. Call get_pdm_summary.
2. Identify critical or degrading groups.
3. Call list_pdm_anomalies with severity=CRITICAL or HIGH.
4. Call get_equipment_health for the top affected equipment.
5. Ask a maintenance engineer to review the evidence before drafting an action.

list_pdm_anomalies

Use list_pdm_anomalies to retrieve anomaly events that need engineering review.

Field	Details
Scope	pdm.read
Inputs	severity (ALL, CRITICAL, HIGH, MEDIUM, LOW), status (OPEN, RESOLVED, ALL), optional limit.
Returns	Anomaly type, severity, affected equipment, and AI recommendation.
Best used for	Anomaly inbox review, event triage, escalation preparation, and maintenance meeting summaries.
Follow-up tools	get_equipment_health, /mcp/base/ asset lookup, Inspector work-order or inspection context.

Review checks:

• Separate open events from resolved events.
• Check whether the same equipment has repeated anomaly types.
• Compare anomaly output with recent inspections, operator notes, work orders, and operating mode.
• Include unresolved evidence gaps in the reviewed answer.

Example request:

{
  "severity": "HIGH",
  "status": "OPEN",
  "limit": 20
}

get_equipment_health

Use get_equipment_health when the Agent needs detail for one equipment item.

Field	Details
Scope	pdm.read
Inputs	Required equipment_name; optional equipment_id when the workflow already resolved a stable equipment UUID.
Returns	Health score from 0 to 100, grade from A to F, anomaly level, crest factor, and vibration RMS.
Best used for	One-equipment health explanation, next-check recommendation, and escalation context.
Follow-up tools	/mcp/base/ for work records and documents; DFS readiness pages for signal freshness and unit mapping.

Review checks:

• Prefer stable equipment_id when available; names and codes can have aliases.
• Treat health score, grade, crest factor, and vibration RMS as evidence fields in one engineering view.
• Ask for signal history and maintenance records before explaining root cause.
• Record whether the output supports a field check, a data-quality correction, or a work-order draft.

Example request:

{
  "equipment_name": "critical pump",
  "equipment_id": "optional-equipment-uuid"
}

Suggested reviewed output sections:

Section	Content
Health state	Score, grade, anomaly level, review window, and current trend.
Signal evidence	Vibration RMS, crest factor, temperature or other available signals, and source timestamps.
Maintenance context	Recent work orders, inspection findings, replaced parts, or operator notes.
Engineering checks	Field checks ranked by evidence strength.
Review boundary	Engineer approval state and missing evidence.

get_filter_component_intelligence

Use get_filter_component_intelligence for component-level maintenance priorities in equipment fleets.

Field	Details
Scope	pdm.read
Inputs	Optional equipment_id, optional equipment_code, optional top_n.
Returns	Components needing attention, risk band, predicted remaining life, recommended action, benchmark context, and aftermarket narrative.
Best used for	Component attention lists, filter maintenance planning, spare-parts planning, and fleet-level component review.
Follow-up tools	get_equipment_health, get_filter_circular_recovery, work-order context from /mcp/base/.

Review checks:

• Use equipment_id or equipment_code when the request is about one asset.
• Use top_n to keep outputs usable in daily triage.
• Pair component recommendations with source timestamps and maintenance history.
• Confirm whether the component recommendation should become an inspection item, a planned replacement, or a parts review.

Example request:

{
  "equipment_code": "EX-003",
  "top_n": 5
}

get_filter_circular_recovery

Use get_filter_circular_recovery when the maintenance workflow includes component recovery, reuse, remanufacture, disposal, sustainability, or aftermarket planning.

Field	Details
Scope	pdm.read
Inputs	Optional equipment_id, optional equipment_code, optional top_n.
Returns	Recovery candidates, risk band, remaining life, recommended actions, and aftermarket or disposal guidance.
Best used for	Circular recovery review, replacement planning, sustainability reporting input, and aftermarket decision support.
Follow-up tools	get_filter_component_intelligence for component priority and /mcp/base/ for asset or work-order evidence.

Review checks:

• Keep recovery guidance tied to component condition, remaining life, and maintenance history.
• Capture reviewer notes when a component is selected for reuse, remanufacture, disposal, or aftermarket handling.
• Use work-order closure feedback to improve future component recommendations.

Filter tool selection

Predictive Maintenance filter component tools focus on equipment and fleet components. Cleanroom HEPA or ULPA filter life analysis belongs to the SemiOps cleanroom workflow and uses the SemiOps tool surface.

Question type	Tool area
Fleet or equipment component needs attention	Predictive Maintenance get_filter_component_intelligence
Component reuse, recovery, remanufacture, or disposal	Predictive Maintenance get_filter_circular_recovery
Cleanroom HEPA or ULPA pressure-drop life prediction	SemiOps cleanroom filter life workflow

Recommended MCP tool sequences

Fleet triage

get_pdm_summary
  -> list_pdm_anomalies(severity=CRITICAL or HIGH, status=OPEN)
  -> get_equipment_health for top affected equipment
  -> /mcp/base/ asset, work-order, and document context
  -> reviewed engineering summary

Use this sequence for daily or weekly reliability meetings.

One-equipment health explanation

get_equipment_health(equipment_name or equipment_id)
  -> /mcp/base/ work history and inspection notes
  -> DFS signal-history freshness check
  -> ranked engineering checks

Use this sequence when a maintenance engineer asks why one asset changed health grade.

Component attention and recovery

get_filter_component_intelligence(equipment_id or fleet scope)
  -> get_filter_circular_recovery for selected candidates
  -> /mcp/base/ work-order or parts context
  -> reviewer decision

Use this sequence when component risk and recovery value should be reviewed together.

AI Engine analysis tools

AI Engine endpoints power the Predictive Maintenance runtime. They are usually called by product services, model workflows, or controlled backend jobs.

Endpoint	Purpose	Inputs to prepare	Output to review
/ai/pdm/train	Train a baseline model for equipment features.	Equipment ID or class, feature list, training window, cleaned time-series data.	Model ID, training status, data points used, readiness state.
/ai/pdm/baseline/{model_id}/status	Check model training status.	Model ID.	Status, readiness, errors, and next action.
/ai/pdm/detect	Run anomaly detection.	Sensor data window, optional model ID, feature set, asset context.	Anomaly score, threshold, severity, anomaly type, feature contributions, model version.
/ai/pdm/forecast	Forecast a feature trend and estimate threshold timing.	Feature history, horizon, confidence interval, optional threshold.	Forecast points, predicted threshold date, days to threshold, recommendation.
/ai/pdm/health-score	Calculate composite health score.	Dimension scores, weights, asset or equipment context.	Composite score, grade, dimension contribution, summary.
/ai/pdm/rul/predict	Estimate remaining useful life.	Health-index history, failure threshold, degradation history.	Remaining days, confidence, method, degradation rate, predicted failure date, trend data.
/ai/pdm/diagnose	Diagnose one asset from vibration and temperature evidence.	Equipment template, sensor position, latest reading, optional history.	Severity, diagnosis, likely fault modes, recommendation, ISO grade, evidence.
/ai/pdm/fleet-diagnose	Diagnose a group of similar equipment.	Equipment type, batch readings, comparable feature set.	Per-equipment diagnosis, fleet statistics, outlier detection.
/ai/pdm/templates	List diagnosis templates.	Optional equipment type or tenant context in the product workflow.	Available templates and supported equipment types.
/ai/pdm/templates/{equipment_type}	Retrieve one diagnosis template.	Equipment type.	Template configuration, expected signals, thresholds, and diagnosis logic.
/ai/pdm/ingest/xs770a	Process XS770A-style vibration payloads.	Three-axis velocity, acceleration, temperature, timestamp, equipment metadata.	Composite vibration RMS, peak vibration, crest factor, quick anomaly flag, health-score estimate.

AI Engine workflow guidance

Workflow	Recommended path	Review owner
New equipment model	Prepare signal history in DFS, run /ai/pdm/train, check status, assign model in the model console.	Reliability engineer or model owner.
Health review	Run health scoring or read latest product health snapshot, then compare with anomaly and maintenance context.	Maintenance engineer.
Anomaly investigation	Run detection, inspect feature contributions, compare with work history and operating mode.	Maintenance engineer.
Trend and threshold planning	Run forecast, confirm confidence interval and expected operating regime.	Reliability engineer.
Remaining-life review	Run remaining useful life prediction with validated health-index history.	Reliability engineer and maintenance planner.
Fleet diagnosis	Run batch diagnosis for comparable equipment only after units, templates, and operating conditions are aligned.	Fleet owner or reliability lead.

Data requirements

Predictive Maintenance AI tools are strongest when the data package includes:

• stable equipment ID and equipment type;
• sensor-to-equipment mapping;
• timestamped signal history;
• units, scale factors, axis definitions, and sampling interval;
• operating state or schedule when running-state filtering matters;
• maintenance records and work-order outcomes;
• approved equipment template or rule overlay;
• DFS quality notes for missing intervals, stale values, and outliers.

Use Prepare Signal History for Predictive Maintenance before training or scoring workflows that depend on source telemetry.

Output review checklist

• Equipment identity is stable across Predictive Maintenance, DFS, Inspector, CMMS, and work-order records.
• Signal timestamps and source freshness are visible.
• Units, axis definitions, and feature names are documented.
• Model or template version is visible when a score or diagnosis depends on it.
• Health score, anomaly severity, forecast, and remaining-life output include confidence or limitation notes where available.
• Recommendations are routed to a maintenance owner before field action.
• Work-order outcomes and false-positive labels are fed back into the Predictive Maintenance workflow.

Related docs

• Predictive Maintenance Operations
• Predictive Maintenance Workflow Guide
• Predictive Maintenance use case
• Prepare Signal History for Predictive Maintenance
• MCP Tool Reference
• MCP Scope Matrix