An algorithm for data-driven prognostics based on statistical analysis of condition monitoring data on a fleet level

The availability of condition monitoring data for large sets of homogeneous products (in the following referred as a fleet) motivates the development of new data-driven prognostic algorithms. In this paper, an intuitive and an innovative data-driven algorithm to predict the health and, consequently, the Residual Useful Lifetime (RUL) of a product are proposed. The algorithm is based on the extraction and exploitation of knowledge at a fleet level. The fleet-specific usage and the degradation profile are extracted by statistically analyzing the condition monitoring data of all the products that's belongs to the fleet. The extracted knowledge, in terms of statistical distribution of health condition and sampling time, is then exploited to predict the health and RUL of a product in the fleet. The algorithm described in this paper is able to predict the RUL of a product with a good credibility even for observation window lengths that are smaller compared to the lifetime of the product.