Özet:

Abstract Data-driven predictive maintenance commonly uses machine learning algorithms to conduct prognostics of an asset’s condition over its life cycle. Asset information and domain expert knowledge are essential in data-driven predictive maintenance to support maintenance-related decisions. Using a general feature selection approach in data-driven prognostics can cause misinterpretation, removal, or loss of domain-specific information of assets. The high dimensionality characteristics of asset data due to a large number of features sourced from various sensor measurements can affect the performance and reliability of machine learning algorithms. This paper presents a feature selection approach to overcome the challenges of retaining domain-specific asset data information by utilising the Safe Operating Limit of an asset. The asset information is combined with the filter method to reduce the high dimensional aspects of asset data for application in equipment’s remaining useful life prediction. The proposed feature selection approach is demonstrated on an oil and gas equipment dataset that contains multiple run-to-failure situations of a gas compressor.

Anahtar Kelimeler: