使用韋伯可靠性函數建立軸承剩餘壽命預測模型

Weibull Reliability Regression Model for Bearing Remaining Useful Life Prediction

賴駿傑(Jun-Jie Lai)

指導教授 ： 陳誠亮

國立臺灣大學/工學院/化學工程學研究所/碩士(2021年)

https://doi.org/10.6342/NTU202101707

摘要

剩餘使用壽命 (RUL) 是對機械元件壽命的預估，其被視為機器健康狀況的重要指標。正常來說，機器的健康狀況會隨著時間的推移而變差。因此在工廠中常常需要定期的維護及更換機械元件。然而，如何決定維護的週期事件重要的事。提早維護可以防止機器損壞，但要付出更多不必要的成本；若是維修太晚，可能會損壞機器並造成人員傷亡。若是能準確的預測RUL便可提供合適的時機，並為操作人員保持安全的環境。
在本研究中，數據來源來自IEEE在2012的挑戰。有一些論文使用數據驅動模型(data-based model)，例如人工神經網絡來解決這個問題。然而，數據驅動的模型通常缺乏對機械設備狀況的概述。此外，數據驅動模型中參數的物理意義知之甚少。
在我們的參考研究中，使用韋伯加速失效時間回歸（Weibull Accelerated Failure Time Regression, WAFTR）模型，其中用韋伯可靠性函數來敘述RUL。通過將可靠性函數中的參數η進行指數展開來修正韋伯可靠性函數。在這項研究中，我們很好奇是否可以將參數展開為不同的形式。此外，我們考慮了更多的可靠性參數進行展開修飾 (η、β)。此外，我們假設實際的 RUL 是線性衰減的。希望透過我們的模型可以使預測的RUL能趨近線性衰減。為了量化預測的表現，我們計算了預測 RUL 和實際 RUL 的均方誤差 (MSE)，MSE越小代表預測結果越好。最後，我們發現應用單一模型會使可靠性參數失去其物理意義。我們進一步介紹了多重模型，其誤差在 20% 以內，並展現了更有意義的可靠性參數。

關鍵字

韋伯方程式 ； 機械剩餘壽命 ； 可靠性函數

並列摘要

Remaining useful life (RUL) is an estimate of the time that an item or component can function. It is an important indicator of machine health condition. Normally, the health condition of a machine will be worse with time goes by. Thus, maintenance and replacement are needed. In a plant, although earlier maintenance can prevent machines from broken, the more unnecessary cost have to pay. However, if the maintenance is too late, the machine may be broken and cause casualties. The accurate RUL prediction provides a suitable moment for replacement actions and keeps a safe environment for workers.
In this research, the data source is from IEEE 2012 challenge. Some of the papers used data-driven models such as artificial neural networks to solve this problem. However, the data-driven model usually lacks an overview of the research condition. Moreover, little physical meaning is known in the data-driven model.
In the reference research1, Weibull Accelerated Failure Time Regression (WAFTR) model is used, which modifies the Weibull reliability function by exponentially expanding the parameter. In this study, we are curious that if we can expand the parameter in different forms. Moreover, we expend more reliability parameters. We assume the actual RUL is linear decayed. A good performance of predicted RUL needs to fit the actual RUL well. To check the performance of our model, we calculate the mean square error (MSE) for the predicted RUL and actual RUL. In the end, we find that applying a single model will let the reliability parameters lose their physical meaning. We further introduce the multiple model, of which the error is within 20% and meaningful reliability parameters are achieved.

並列關鍵字

Weibull distribution ； remaining useful life ； reliability function

參考文獻

Kundu, P.; Darpe, A.; Kulkarni, M., Weibull accelerated failure time regression model for remaining useful life prediction of bearing working under multiple operating conditions. Mechanical Systems and Signal Processing 2019, 134, 1-19.

Ben Ali, J.; Chebel-Morello, B.; Saidi, L.; Malinowski, S.; Fnaiech, F., Accurate bearing remaining useful life prediction based on Weibull distribution and artificial neural network. Mechanical Systems and Signal Processing 2015, 56.

Nectoux, P.; Gouriveau, R.; Medjaher, K.; Ramasso, E.; Chebel-Morello, B.; Zerhouni, N.; Varnier, C. In PRONOSTIA : An experimental platform for bearings accelerated degradation tests, IEEE International Conference on Prognostics and Health Management, PHM'12., Denver, Colorado, United States, 2012-06-18; IEEE Catalog Number : CPF12PHM-CDR: Denver, Colorado, United States, 2012; pp 1-8.

Wang, T., Bearing life prediction based on vibration signals: A case study and lessons learned. 2012; p 1-7.

Bucknam, J. S. In Data analysis and processing techniques for remaining useful life estimations, 2017.