Skip to main content

TimeSeries-Insight — AI Analytics Engine

Overview

PlantPulse TimeSeries-Insight is an AI anomaly detection, forecasting, and comprehensive analytics MCP server built specifically for manufacturing time series data.

Using the IBM Granite TSPulse (anomaly detection) and TinyTimeMixer TTM (forecasting) AI models, it delivers intelligent analysis of every time series data set in the plant — sensor data, OEE, RAM, EMS, alarms, work orders, and more.

With 33 analysis tools and a feature store, it covers everything from real-time analysis to pre-computed comprehensive insights.

Split across two services

insight-api(:8970) handles the tools and data, while model inference (TSPulse and TTM) is delegated to insight-server(:8980). Since a GPU is only required on the inference side, the two services can run on separate machines.


AI Anomaly Detection

Base model: IBM Granite TSPulse

TargetToolAnalysis
Equipment sensorsdetect_asset_anomalyDetects abnormal sensor patterns in temperature, pressure, vibration, and so on
System resourcesdetect_system_anomalyDetects CPU, memory, and disk anomalies
OEEdetect_oee_anomalyAbnormal patterns in availability, performance, and quality
RAMdetect_ram_anomalyAbnormal MTBF and MTTR patterns
EMSdetect_ems_anomalyAnomalies in energy consumption, power, and carbon emissions
Work ordersdetect_order_anomalyProduction quality anomalies (OEE/RAM/EMS/combined)
Alarmsdetect_alarm_anomalyAbnormal alarm frequency patterns

Key parameters:

  • threshold — Anomaly decision threshold (automatic when set to 0)
  • top_k — Number of top anomalous metrics
  • lookback_minutes — Analysis time range
  • target_columns — Analysis target (queries all sensors automatically when unspecified)

AI Forecasting

Base model: TinyTimeMixer (TTM)

TargetToolForecast
Equipment sensorsforecast_assetFuture sensor value trends
System resourcesforecast_systemCPU/memory/disk utilization
OEEforecast_oeeFuture trends in availability, performance, and quality
RAMforecast_ramFuture trends in equipment reliability indicators
EMSforecast_emsEnergy consumption and carbon emission outlook
Work ordersforecast_orderFuture production quality trends
Alarmsforecast_alarmAlarm frequency trends, failure prediction

Key parameters:

  • prediction_length — Number of forecast steps
  • enable_bounds — Output confidence interval (upper/lower bounds)

In-Depth Alarm Analysis

ToolDescription
get_alarm_statsAlarm statistics summary — frequency, priority distribution, top alarms, top tags
analyze_alarm_durationAlarm duration/MTTA analysis — degradation detection, mean acknowledgment time trend
analyze_alarm_correlationAlarm-sensor correlation — which sensor changes trigger alarms

Equipment Health Index

get_asset_health — Combines sensors, OEE, RAM, EMS, and alarms to produce a health score from 0 to 100


Feature Store

Delivers fast comprehensive analysis using pre-computed features and LLM context.

ToolDescription
get_feature_statusFeature store status — scheduler jobs, execution times, backfill status
get_asset_featuresEquipment feature query — health score, anomaly score, alarms, status, production metrics
get_asset_contextLLM context — daily comprehensive analysis summary text (pre-generated)
analyze_assetComprehensive equipment analysis — health score, sensors, alarms, OEE/RAM/EMS, cause hints
explain_asset_anomalyAnomaly cause analysis — cause estimation based on anomalous tags, alarms, and status changes
trigger_feature_pipelineManual execution of the feature pipeline

Feature store architecture:

  • Scheduler: Automatic hourly and daily feature computation
  • Embeddings: Converts equipment features into vectors to support similarity search
  • LLM context: Pre-generates daily comprehensive analysis text to maximize response speed

MCP Resources (6)

Resource URIDescription
plantpulse://system/infoSystem information
plantpulse://domainsDomain list and API endpoints
plantpulse://metrics/oeeOEE metric descriptions
plantpulse://metrics/ramRAM metric descriptions
plantpulse://metrics/emsEMS metric descriptions
plantpulse://metrics/alarmAlarm metrics and analysis capabilities

Understanding Equipment and Tags

When a piece of equipment carries thousands of tags, no one can explain what each of them is. The tools below automatically characterize tags using real data statistics and platform metadata.

ToolDescription
list_asset_tagsTag list for the equipment
get_asset_profileEquipment profile — characterizes which tag is what
get_asset_tag_relationsCorrelation between tags — groups tags that move together

Case Search and Comparison

ToolDescription
find_similar_assetsSimilar equipment search — based on cosine similarity of vector embeddings
find_similar_anomaliesSimilar anomaly case search — whether an anomaly of the same shape occurred before
find_fleet_outlierFleet outliers — units of the same equipment type that behave differently from the rest
search_patternPattern search — finds intervals resembling a specified waveform
check_failure_precursorFailure precursor inspection — whether signals that preceded past failures are visible now

Usage Scenarios

ScenarioTools Used
Early detection of equipment anomaliesdetect_asset_anomalyexplain_asset_anomaly
Predictive maintenance planningget_asset_healthforecast_assetforecast_ram
Daily equipment briefingget_asset_contextget_asset_features
Root cause analysis of alarm floodsget_alarm_statsanalyze_alarm_correlationanalyze_alarm_duration
Comparative analysis of similar equipmentanalyze_assetfind_similar_assets
Has this anomaly happened before?detect_asset_anomalyfind_similar_anomalies
Catching failures before they occurcheck_failure_precursorforecast_ram
Same equipment, but only one is abnormalfind_fleet_outlieranalyze_asset
Energy anomaly detectiondetect_ems_anomalyforecast_ems
Production quality forecastingdetect_order_anomalyforecast_order