This paper presents an approach and the presented solution of the questions raised in the IEEE PHM 2012 Conference Challenge Competition. What was given (known) is the real run-to-failure data of 6 bearings only from the three groups exposed to different operating conditions. One should use this data to estimate the Remaining Useful Life (RUL) of the given set of 11 test bearings. The main feature of the presented data is significant loss of trendability (i.e. "non-trendability") of the defined significant parameters' behavior (horizontal and vertical vibration), thus avoiding the use of well-known supervised learning RUL prediction models. New models have been developed and used; further the Cross-Entropy method has been used for control parameter optimization based on the Cross-Validation procedure. The presented solution has been recognized as a "Winner from Industry" in the above mentioned Competition. The achieved results demonstrate the effectiveness of the approach for the RUL estimation for systems parameters with the non-trendability behavior.
