基于深度支持证据统计的轴承运行可靠性评估

Bearing operation reliability evaluation based on deep support evidence statistics

某台具体设备的运行可靠性评估是"个性"问题。基于设备运行数据大而丰富的特点,从证据论的观点出发,结合深度卷积自编码器,提出了一种深度支持证据统计方法,以实现无失效样本信息或缺乏分布函数等相关先验知识时的轴承运行可靠性评估。在传统的证据或特征获取过程中需要大量的人工干预或先验知识,利用深度卷积稀疏自编码器以实现证据的自动获取,借助支持向量数据描述的思想,通过统计比较标定与过程证据间的动态变化过程,最终完成某台具体设备的运行可靠性评估。随后通过航空轴承试验对提出的方法进行了验证,结果表明运用该方法得到的可靠度,能较好的反映轴承运行可靠性的"个性"。

The operation reliability evaluation of a specific equipment is a “personality” problem. Here, based on large and rich characteristics of equipment operation data, starting from the point of view of the evidence theory, and combining with deep convolution auto-encoder, a deep support evidence statistics method was proposed to realize bearing operation reliability evaluation without failure sample information or due to lack of relevant prior knowledge, such as, distribution function. In traditional evidence or feature acquisition process, a lot of manual interventions or prior knowledge were required. The deep convolution sparse auto-encoder was used to realize automatic acquisition of evidence. With help of the idea of support vector data description, the operation reliability evaluation of a specific equipment was finally completed by statistically comparing the dynamic change process between calibration and process evidence. Then, the proposed method was verified with aviation bearing tests. The results showed that the reliability obtained using the proposed method can better reflect the “personality” of bearing operation reliability.