Can I trust machine learning analytics? Downer’s approach to predictive maintenance

Machine learning models for predictive maintenance requires a high level of confidence pertaining to its predicted outputs to ensure user acceptance. However, the predictions of machine learning models are largely dependent on the contextual quality of the training data consisting of both normal and failure samples. The main challenge is that in reliability engineering, there is likely to be minimal failure data available in environments where good reliability management processes are in place to keep failures to a minimum. This often results in an imbalanced data set (i.e. abundance of data showing components functioning properly, but little data showing components failing), leading to a low level of confidence in predictions of failure cases and trust on machine learning models from the perspective of users. This paper presents Downer’s approach to predictive maintenance by developing a trust-focused methodology with the support of a human-machine teaming framework and optimal machine learning techniques in imbalanced data environments. This framework supports Downer’s integrated data and analytics platform known as TrainDNA into operations.

Media Info

  • Pagination: 8p. ; PDF
  • Monograph Title: AusRAIL PLUS 2019, Delivering growth; creating opportunity; embracing technology, 3-5 December 2019, Sydney, NSW, Australia

Subject/Index Terms

Filing Info

  • Accession Number: 01748342
  • Record Type: Publication
  • Source Agency: ARRB Group Limited
  • Files: ATRI
  • Created Date: Aug 20 2020 2:10PM