failures with regards to condition based maintenance Kris De Gussem - - PowerPoint PPT Presentation

Data based detection of pump failures with regards to condition based maintenance

31/03/2020 Kris De Gussem

Problem definition

3

Background

• Pumping station: pump between different levels in sewerage system, under roads, …
• Prevent discharge to clean surface water  backup equipment
• Pump failure: reactive maintenance

4

Background

• Pumping station: pump between different levels in sewerage system, under roads, …
• Prevent discharge to clean surface water  backup equipment
• Pump failure: reactive maintenance
• Reactive  condition based maintenance
• E.g. add new high frequency vibration measurement
• Ideal situation
• Data driven maintenance: fewer, when appropriate
• No weather independent nightly reactive maintenance

5

Goal

Existing data: any indicative value for maintenance? Hackathon challenge:

• Any algorithm to detect malfunctioning pumps
• Prefer simple, easy to explain over black box / neural nets
• Prefer general applicability over different algorithm per installation
• Any visualisation to show malfunctioning pumps

Future:

• Apply in big data platform (AWS)

Dataset

7

Dataset description

4 pumping stations (dataset completeness varies):

• Installation layout: screendump scada system
• Measurement logs:
• Description how/when values are stored (30s interval – threshhold based)
• Values include (varies per installation):
• Level
• Current / power consumption
• Flow
• Available maintenance data
• Format: parquet and csv

Access via: https://www.aquafin.be/en/hackathon-challenge

8

Wrap up up

Dataset challenge:

• Installation layout: differs, illustrates diversity and problem complexity
• Not all issues may pop up in maintence data

Result evaluation:

• Originality
• Simplicity
• Generally applicable?

Succes!

Contact: Kris De Gussem Hackathon dataset: https://www.aquafin.be/en/hackathon-challenge Tel.: +32 3 450 44 05 E-mail: Kris.DeGussem@aquafin.be