Ensemble Learning for Remaining Useful Life Prediction Prof. Chao HU and Zhixiong LI (Iowa State Univ.)
Remaining useful life (RUL) prediction is critical to implement predictive maintenance. While significant research has been conducted in model-based and data-driven prognostics, very limited research has been done to investigate the prediction of RUL using an ensemble learning method that combines prediction results from multiple learning algorithms. The objective of this research is to introduce a new ensemble prognostics method with degradation-dependent weights. Specially, this method assigns an optimized, degradation-dependent weight to each learner (i.e., learning algorithm) such that the weighted sum of the prediction results from all the learners predicts the RUL of mechanical components with better accuracy. The ensemble prognostic algorithm is demonstrated using an experimental data collected from an engine simulator. Experimental results have shown that the predictive model trained by the ensemble learning algorithm outperform the existing methods.
Prediction of Bearing Life by Exponential Curve-fitting Dr. Yun-Ho SEO and SangRyul KIM (KIMM)
Bearings are one of the most important parts to determine the operation of a machine. Then, the maintenance and replacement of bearings prevent from operation of a machine. Therefore, the prediction of bearing life is critical to manage the operation of a machine. In this paper, several degradation tests of bearings for the constant forces are carried out to gather various physical parameters with the lapse of time. An efficient parameter which consists of a linear combination of measured parameters is made to represent definite degradation process. The efficient parameter increases monotonically as time goes by. Then, a simple curve fitting method by using exponential functions is proposed in order to predict the life of the bearing based on real-time measurement results. The simple curve fitting method is demonstrated by various samples to represent feasibility of the proposed method.
A health monitoring method for wind power generators with hidden Markov and probabilistic principal components analysis models Mr. Riku SASAKI, Naoya TAKEISHI, Takehisa YAIRI, Koichi HORI, Kazunari IDE and Hiroyoshi KUBO (The University of Tokyo, Mitsubishi Heavy Industries, Ltd.)
In this work, we propose a data-driven health monitoring method for wind power generators, which learns an empirical model from the time-series sensor data and detects irregularities or faults in the turbines and blades. Our main objective is to predict any symptoms of faults as early as possible before the generators fall into malfunction. The data obtained from the wind power generators are strongly correlated multidimensional time-series with multiple states. In this study, we take the features into account and develop the probabilistic model for them, namely, hidden Markov and probabilistic principal component analysis. Once the model is learned with the data that contain no faulty events, it can be used to detect faults in new data by comparing the original sensor values and reconstructed one. In this research, we apply this method to synthetic data and real-world wind turbine data, and show the results of experiments to confirm the availability of the proposed method.