There are a number of methods in RUL prediction for using incomplete data set. Similarity-based Method is one of them. You can see the paper below for reference Conference Paper A Novel Similarity-based Method for Remaining Useful Life Pr...

Also for Bearing PHM Dataset, you can use NASA Bearing Data Set for comparison. https://ti.arc.nasa.gov/tech/dash/groups/pcoe/prognostic-data-repository/

Thanks for your reply Jay Lee

The dataset mentioned above is for Rexnord ZA-2115 double row bearings. Will the degradation pattern be same for other bearings? Wont it change from application to application?

• Generally speaking, we can lifetime data or run-to-failure histories data for similar bearings then develop a reference pattern/model.
• We can further update/revise it using additional new data or peer data.
• Overtime, the data set can be more useful.

Good bearing remaining useful life RUL prediction depends on the use of right feature as well as the ability of using adaptive feature extraction. Once the baseline is established using one bearing, other machine learning algorithms can be used such as deep-learning-based and long short-term memory neural network methods for improved feature extraction.

Thanks for your time Jay Lee

To my experience, in case of incomplete dataset, the only reliable RUL prediction method is based on physical features and a pre- defined degradation model incorporated into a filtering framework, such as Kalman filtering or particle filtering.

You can refwr to the following

https://www.researchgate.net/publication/328405905_An_evidential_similarity-based_regression_method_for_the_prediction_of_equipment_remaining_useful_life_in_presence_of_incomplete_degradation_trajectories?_sg=qRq98fxnoDCCgoXahMFh9pTT66gBewawbbxJDfFth6edgWOc_bu4NApF6cXi_A9ZY6We9dfdV9bWmmChDBlLr05RmDoSP0avyOWbom-S.iWc7G4M6OuzJHC52sH03uza5wB_2LT7BxmSygm4M_q1IaneLR5eIuz5iPR7Yyw3ehXNVMMfIgt4OjbObw0-V-g

For any question do not hesitate to write me!