Testing and comparing the application of new techniques to condition based maintenance

拟议的研究的目的是引进和测试新技术的监测，诊断和预测的机器健康状况的基础上，状态监测，比较这些技术的性能，并相应地优化维护措施。第一种技术是比例风险模型（PHM）和部分可观察马尔可夫决策模型（POMDM），这两个模型经常在文献中提到，但单独使用的组合。第二种技术是基于统计过程控制（SPC）图，霍特林的T2。第三种技术使用支持向量机（SVM）模型，该模型最初是为解决模式识别问题而开发的。第四种技术应用数据逻辑分析（LAD），这是一种基于组合学和优化的数据分析方法。这些技术代表了四种不同的CBM方法，分别基于建模退化、离群值检测、回归分析和组分类。科学的方法是建模，然后基于CBM的以下关键要素比较这四种技术的性能：1.检测监测指标中的趋势或模式; 2.根据剩余使用寿命或故障概率预测机器的未来健康状况;3.优化工具或维护决策过程与这些技术的接口。这项工作的新奇是双重的：1.四种方法中的三种从未用于预测或与基于状态的维护中的优化工具一起使用，尽管已报道成功地应用这些技术进行诊断。预计这些技术将在建立信任措施中发挥良好的作用，因此将向感兴趣的生产者提供新的技术。预计这样的研究将有助于创业者确定最适合他们需要的技术。

The objective of the proposed research is to introduce and test new techniques for monitoring, diagnosis and prognosis of machine health based on condition monitoring, to compare the performance of those techniques, and to optimize maintenance actions accordingly. The first technique is a combination of the Proportional Hazards Model (PHM) and a Partially Observable Markov Decision Model (POMDM), two models which are frequently mentioned in the literature, but which are used separately. The second technique is based on a Statistical Process Control (SPC) chart, the Hotelling's  T2 . The third technique uses the Support Vector Machines (SVM) model, which was originally developed for solving pattern recognition problems. The fourth technique applies the Logical Analysis of Data (LAD), a combinatorics and optimization-based data analysis method.These techniques represent four different approaches to CBM based on modeling degradation, outlier detection, regression analysis, and group classification, respectively.The scientific approach is to model and then compare the performance of those four techniques based on the following key elements of CBM:1.The detection of trends or patterns in the monitored indicators;2.The prediction of the machine's future health based on the remaining useful life or the probability of failure;3. The interface of an optimization tool or a maintenance decision-making process with those techniques.The novelty of this work is twofold:1.Three of the four approaches have never been used for prediction or with an optimization tool in condition based maintenance, although success in applying these techniques for diagnosis has been reported. It is expected that those techniques will perform well in CBM and thus new techniques will be available to interested practitionners.2.A comparative study of different techniques used in CBM has never been reported. It is expected that such a study will help practitionner identify the most suitable techniques for their needs.