Increasingly, extracting knowledge from data has become an important task in organizations for performance improvements. To accomplish this task, data-driven Prognostics and Health Management (PHM) is introduced as an asset performance management framework for data management and knowledge extraction. However, acquired data come generally with quality issues that affect the PHM process. In this context, data quality problems in the PHM context still an understudied domain. Indeed, the quality of the used data, their quantification, their improvement techniques and their adequacy to the desired PHM tasks are marginalized in the majority of studies. Moreover, many PHM applications are based on the development of very sophisticated data analysis algorithms without taking into account the adaptability of the used data to the fixed objectives. This paper aims to propose a set of data quality requirements for PHM applications and in particular for the fault detection task. The conducted developments in this study are applied to Scoder enterprise, which is a French SME. The feedback on the first results is reported and discussed.

从数据中提取知识已越来越成为组织提高性能的重要任务。为了完成此任务，引入了数据驱动的预测和健康管理（PHM）作为用于数据管理和知识提取的资产绩效管理框架。但是，获取的数据通常伴随有影响PHM过程的质量问题。在这种情况下，PHM上下文中的数据质量问题仍然是一个未被充分研究的领域。实际上，在大多数研究中，所用数据的质量，其量化，其改进技术以及对所需PHM任务的适用性都处于边缘地位。此外，许多PHM应用程序都是基于非常复杂的数据分析算法的开发而没有考虑所使用数据对固定目标的适应性。本文旨在为PHM应用程序提出一套数据质量要求，尤其是对于故障检测任务。在这项研究中进行的开发应用于法国的SME编码器企业。报告并讨论了对第一个结果的反馈。