基于信息熵的机械传动油液光谱监测数据选择方法

机械传动装置磨损产生的金属微粒在润滑油中均匀混合并不断积累，是一个缓慢退化过程，可通过油液光谱分析监测。MOA Ⅱ型原子发射光谱仪能够分析得到多达15种元素浓度数据，应用分析得到的油液光谱数据，便能够实现机械传动装置健康状态的监测与评估。然而，并不是所有的油液光谱数据都能够表征装备的健康状态，只有部分油液光谱数据能够提供有用的退化表征信息。应用全部油液光谱数据进行机械传动装置的健康状态监测会增加退化模型的复杂性。鉴于此，为实现机械传动装置健康状态的准确表征，提出了基于信息熵的油液光谱监测数据的选择方法，旨在为机械传动装置的健康状态监测与剩余寿命预测提供有效的退化数据。与传统的油液光谱监测数据选择方法相比，该方法使用信息熵表征各监测数据中蕴含退化信息量的大小，并以此为指标定量选择机械传动装置的退化数据。通过对综合传动装置可靠性试验油液光谱监测数据的实例分析证明了该方法的有效性，能够实现油液光谱数据的定量选择，提高了综合传动装置寿命预测的准确性，也为其他装备监测数据的选择提供了指导。

Spectral Oil Condition Monitoring Data Selection Method for Mechanical Transmission Based on Information Entropy

In mechanical transmission, the wear debris produced from different friction couplings is uniformly mixed in lubrication oil, which is a slow degradation process that can be observed by oil spectral analysis. The wear debris in a sample can be categorized into 15 groups of concentration (e.g., Fe, Cu and Mo) in parts per thousand using MOA II (atomic emission spectroscopy) during the sampling epochs. Its level is one of the most common data types used to monitor and evaluate the underlying health state. However, not all the oil spectral data can show the same degradation pattern. Only parts of the spectral oil data can provide useful information for degradation degree characterization. Using all the spectral oil data for condition monitoring will result in unreasonable degradation modeling for condition monitoring and unscheduled maintenance afterwards. Therefore, this article proposes a selection of degradation data based on information entropy to determine the appropriate degradation data for degradation modeling and remaining useful life prediction. Compared with the experiential selection method, the proposed method can characterize the degradation information contained in the multiple spectral oil dataset, leading to a quantitatively selecting the degradation data. The proposed method was verified through a case study involving a degradation dataset of multiple spectral oil data sampled from a power-shift steering transmission (PSST). The result shows that the proposed method can better characterize the degradation degree, which leads to an accurate estimation of the failure time when the transmission no longer fulfills its function.