A similarity based methodology for machine prognostics by using kernel two sample test.

Area of Science:

• Mechanical Engineering
• Reliability Engineering
• Data Science

Background:

• Accurate Remaining Useful Life (RUL) prediction is crucial for effective machine prognostics and maintenance.
• Existing similarity-based methods often require additional health assessment procedures and lack robust uncertainty quantification.

Purpose of the Study:

• To propose a novel similarity-based algorithm for Remaining Useful Life (RUL) prediction.
• To develop a comprehensive methodology for machine prognostics.
• To enhance the accuracy and probabilistic interpretation of RUL predictions.

Main Methods:

• A Similarity Matching Procedure utilizing the Kernel Two Sample Test (KTST) to identify similar run-to-failure (R2F) profiles.
• Preliminary RUL predictions derived from identified similar R2F records.
• Weibull analysis to fuse preliminary predictions and determine the probability distribution of RUL.

Main Results:

• The proposed method directly measures similarities without extra health assessment.
• It provides robust probabilistic interpretations of prediction uncertainties.
• Statistical soundness of RUL distribution estimation is achieved through KTST prescreening.

Conclusions:

• The novel similarity-based algorithm offers a superior approach to RUL prediction and machine prognostics.
• The method demonstrates effectiveness and advantages over existing techniques, particularly in uncertainty handling and statistical rigor.
• Validation on a public aero-engine dataset confirms the proposed method's efficacy.