Probabilistic Machine Learning Could Eliminate No Fault Found

Many Condition Indicators have been implemented, yet success has been limited owing to their sensitivity to artifacts that invariably corrupt vibration measurements under real-life operations. Here we report a novel approach based on a stochastic non-linear fault evolution model. This probabilistic machine learning algorithm estimates fault magnitudes and probabilities, which were compared to component removals validated by tear down analyses, and achieved a 94% consistency rate over all available data thanks to excellent artifact rejection. This novel maintenance support tool can detect hidden conditions early while virtually eliminating NFF (false positives).