摘要：AbstractCollecting useful and informative data play an essential role in ensuring the performance of data-driven solutions for intelligent maintenance. However, there is still a lack of methodology to systematically assess the data usefulness (or data suitability) for modeling. This lack of data suitability assessment becomes a more pressing issue in the big data environment where a large volume of machine data is generated at a high velocity. Therefore, there are imperative needs for standardized procedures and systematic solutions that can scan through a large amount of data to quantify the data suitability and locate the useful datasets for model development. To fill in this gap, this paper proposes a novel methodology to evaluate the data suitability for PHM modeling from the aspects of detectability assessment, diagnosability assessment, and prognosability assessment. In the discussion, new assessment procedures and algorithms are proposed by using a series of similarity metrics between data vectors or data distribution. Also, the proposed methods provide both visualization tools and quantitative metrics to assess the data suitability. The effectiveness of the methodology is demonstrated by using real-world examples about the ball screw degradation and boring tool degradation. The results successfully demonstrate the effectiveness and practicality of the proposed methodology and analytics.
