Submitted by admin on Tue, 08/02/2022 - 15:29

Unplanned downtime impacting production

Sequence Number 1 Industry Energy Banner Unplanned downtime impacting production How 5G enabled

Improve asset availability through forecasting early fault detection in critical components and subsequently conduct proactive replacements

Data Flows
Title Devices Icon Devices1 Description
  • Sensors deployed on all critical operational assets
  • Purpose: Measure pressure, temp, flow, etc.
Title Connectivity Icon Connectivity1 Description
  • Time series data transport
  • Asset data (maintenance records) access
Title Edge Compute Icon Edge Compute Description
  • Not time critical at present (predictions timeframe are in days), may change when real-time prediction is required from other equipment
Title Cloud Compute & Storage Icon Cloud Compute & Storage Description
  • All data from the edge sensors (both historical and real-time)
  • Enterprise storage is company owned
Title Applications & Services Icon Applications & Services Description
  • Predict future failure of critical assets
  • PaaS-based set up
Title Inform Decision Makers Icon Inform Decision Makers Description
  • Only reporting potential future failure so action can be taken in time
Title Support Decision Making Icon Support Decision Making Description
  • End of process
Application Logic
Description
  • Identify critical assets such as pumps, valves, compressors, etc.
  • Collect event and/or time series data from these critical assets, potentially using sensors
  • Collect as much data as possible from selected assets and surroundings 🡪 ML models will be strengthened with more data
  • If there is a lack of data, look at options of creating and using synthetic data
Description
  • Timeseries data will be stored (long-term) in the Enterprise storage
  • AI (ML) application requests all the data it needs for ML model development, from the Enterprise storage to copy in its own storage (cache)
  • Development of the ML model is done through an iterative process; a quality ML model (fully data-driven) will require multiple steps to detect anomalies/potential failures
  • SME involvement working with data scientists is required to develop the model
  • Model will be stored and maintained by AI application
Description
  • Process of data collection to execution of models is fully automated and runs unattended
  • Only alerts to operations in case of anomalous data behavior, resulting in expected asset failure in X days
  • Operations take action to replace suspected assets in time, therefore avoiding unplanned downtime
  • ML model management is important to handle large number of models
  • Various options for visualisation of entire operations (e.g., XR options on OpenXR platform)
Expected benefits Key value created