A Novel Approach for Evaluating Datasets Similarities Based on Analytical Hierarchy Process in the Industrial PHM Context

In prognostics and health management (PHM), data-driven approaches are crucial for performing prognostics based on historical data, relying on the analysis of extensive datasets to identify patterns and relationships that contribute to predicting or optimizing variables. However, their efficiency is contingent upon the availability of large, high-quality datasets tailored to the specific task at hand.Yet, real-world applications frequently face challenges as data may not always be readily available due to limitations in data acquisition systems or confidentiality concerns. Paradoxically, the contemporary era witnesses an unprecedented surge in the availability of online databases across various fields. These databases offer a plethora of data that can be harnessed to develop, prototype, and test PHM solutions.This study endeavors to introduce an innovative approach for assessing the similarity between datasets, specifically tailored for prognostic and health management applications. The objective is to empower the development of PHM solutions for predefined systems without relying on data generated from the system itself, but rather by leveraging analogous datasets.To quantify the similarity between different datasets, we propose a set of criteria and sub-criteria based on the characteristics of datasets. Subsequently, the analytic hierarchy process (AHP), a well-established multi-criteria decision-making approach, is employed to systematically compare the importance of criteria and sub-criteria for each elementary process within the PHM cycle. This dynamic process considers the varying importance of criteria across different phases, acknowledging that a criterion may not be uniformly significant for all elementary processes. The evaluation of dataset similarity incorporates the proposed criteria and sub-criteria, utilizing a fundamental scale of importance intensity and weights assigned through AHP. This holistic approach yields a comprehensive similarity score, enabling a nuanced understanding of dataset compatibility.To exemplify the efficiency of our proposed approach, we applied it to a practical case study. The study involves assessing the similarity between a run-to-stop database of mechanical bearings and a set of online databases dedicated to the same application. Our solution facilitated the identification of criteria pertinent to the case study, the determination of criterion weights, and ultimately, the calculation of a similarity score for each database. This process proved instrumental in selecting the most similar database, showcasing the practical utility of our proposed approach in real-world PHM scenarios.