Value from free-text maintenance records: converting wind farm work orders into quantifiable, actionable information using text mining

Erik Salo, David McMillan, Richard Connor

Research output: Contribution to conferencePaper

Abstract

The aim of this project is to demonstrate how text mining can help wind farm operators extract unique, quantifiable maintenance information from historic work orders. A good overview of past maintenance efforts can help develop an reliability-centred maintenance strategy for the future in terms of labour intensity, budgeting and spare parts logistics [1, 2]. However, work orders - where significant information is entered by a human in the form of free text – do not provide any straightforward means for automated analysis [3, 4]. Our approach introduces a novel combination of machine learning techniques supported by expert judgement. Significant focus is on the vocabulary - spelling error correction, semantic matching of synonyms and abbreviations. This allows tasks to be grouped by their underlying meaning, not only the characters they contain. The principal output is a frequency distribution of all groups of equivalent tasks. Further categorical analysis allows to focus on specific plant systems or components, as well as failure modes. Data from an industrial partner’s major onshore wind farms in Scotland was used to test our approach against manual analysis. Potential savings were identified in weeks of effort, or £2-9k in labour cost per site, in addition to an improved maintenance strategy. The remaining challenges mainly lie in increasing accuracy and reducing operator input. These are being addressed by our continued research, but also highlight opportunities for collaboration and standardisation across the industry to maximise the value of data.

Conference

ConferenceAnalysis of Operating Wind Farms 2018
CountryLithuania
CityVilnius
Period15/05/1817/05/18
Internet address

Fingerprint

Onshore wind farms
Personnel
Budget control
Error correction
Standardization
Failure modes
Learning systems
Logistics
Semantics
Costs
Industry

Keywords

  • text mining
  • wind power
  • wind farm maintenance
  • machine learning

Cite this

@conference{fbb69ea069f6491bba4a8454215e8711,
title = "Value from free-text maintenance records: converting wind farm work orders into quantifiable, actionable information using text mining",
abstract = "The aim of this project is to demonstrate how text mining can help wind farm operators extract unique, quantifiable maintenance information from historic work orders. A good overview of past maintenance efforts can help develop an reliability-centred maintenance strategy for the future in terms of labour intensity, budgeting and spare parts logistics [1, 2]. However, work orders - where significant information is entered by a human in the form of free text – do not provide any straightforward means for automated analysis [3, 4]. Our approach introduces a novel combination of machine learning techniques supported by expert judgement. Significant focus is on the vocabulary - spelling error correction, semantic matching of synonyms and abbreviations. This allows tasks to be grouped by their underlying meaning, not only the characters they contain. The principal output is a frequency distribution of all groups of equivalent tasks. Further categorical analysis allows to focus on specific plant systems or components, as well as failure modes. Data from an industrial partner’s major onshore wind farms in Scotland was used to test our approach against manual analysis. Potential savings were identified in weeks of effort, or £2-9k in labour cost per site, in addition to an improved maintenance strategy. The remaining challenges mainly lie in increasing accuracy and reducing operator input. These are being addressed by our continued research, but also highlight opportunities for collaboration and standardisation across the industry to maximise the value of data.",
keywords = "text mining, wind power, wind farm maintenance , machine learning",
author = "Erik Salo and David McMillan and Richard Connor",
year = "2018",
month = "5",
day = "15",
language = "English",
note = "Analysis of Operating Wind Farms 2018 ; Conference date: 15-05-2018 Through 17-05-2018",
url = "https://windeurope.org/workshops/analysis-of-operating-wind-farms-2018/",

}

Salo, E, McMillan, D & Connor, R 2018, 'Value from free-text maintenance records: converting wind farm work orders into quantifiable, actionable information using text mining' Paper presented at Analysis of Operating Wind Farms 2018, Vilnius, Lithuania, 15/05/18 - 17/05/18, .

Value from free-text maintenance records : converting wind farm work orders into quantifiable, actionable information using text mining. / Salo, Erik; McMillan, David; Connor, Richard.

2018. Paper presented at Analysis of Operating Wind Farms 2018, Vilnius, Lithuania.

Research output: Contribution to conferencePaper

TY - CONF

T1 - Value from free-text maintenance records

T2 - converting wind farm work orders into quantifiable, actionable information using text mining

AU - Salo, Erik

AU - McMillan, David

AU - Connor, Richard

PY - 2018/5/15

Y1 - 2018/5/15

N2 - The aim of this project is to demonstrate how text mining can help wind farm operators extract unique, quantifiable maintenance information from historic work orders. A good overview of past maintenance efforts can help develop an reliability-centred maintenance strategy for the future in terms of labour intensity, budgeting and spare parts logistics [1, 2]. However, work orders - where significant information is entered by a human in the form of free text – do not provide any straightforward means for automated analysis [3, 4]. Our approach introduces a novel combination of machine learning techniques supported by expert judgement. Significant focus is on the vocabulary - spelling error correction, semantic matching of synonyms and abbreviations. This allows tasks to be grouped by their underlying meaning, not only the characters they contain. The principal output is a frequency distribution of all groups of equivalent tasks. Further categorical analysis allows to focus on specific plant systems or components, as well as failure modes. Data from an industrial partner’s major onshore wind farms in Scotland was used to test our approach against manual analysis. Potential savings were identified in weeks of effort, or £2-9k in labour cost per site, in addition to an improved maintenance strategy. The remaining challenges mainly lie in increasing accuracy and reducing operator input. These are being addressed by our continued research, but also highlight opportunities for collaboration and standardisation across the industry to maximise the value of data.

AB - The aim of this project is to demonstrate how text mining can help wind farm operators extract unique, quantifiable maintenance information from historic work orders. A good overview of past maintenance efforts can help develop an reliability-centred maintenance strategy for the future in terms of labour intensity, budgeting and spare parts logistics [1, 2]. However, work orders - where significant information is entered by a human in the form of free text – do not provide any straightforward means for automated analysis [3, 4]. Our approach introduces a novel combination of machine learning techniques supported by expert judgement. Significant focus is on the vocabulary - spelling error correction, semantic matching of synonyms and abbreviations. This allows tasks to be grouped by their underlying meaning, not only the characters they contain. The principal output is a frequency distribution of all groups of equivalent tasks. Further categorical analysis allows to focus on specific plant systems or components, as well as failure modes. Data from an industrial partner’s major onshore wind farms in Scotland was used to test our approach against manual analysis. Potential savings were identified in weeks of effort, or £2-9k in labour cost per site, in addition to an improved maintenance strategy. The remaining challenges mainly lie in increasing accuracy and reducing operator input. These are being addressed by our continued research, but also highlight opportunities for collaboration and standardisation across the industry to maximise the value of data.

KW - text mining

KW - wind power

KW - wind farm maintenance

KW - machine learning

M3 - Paper

ER -