Identifying emerging failure phenomena in complex systems through engineering data mapping

In the past, standard reliability and risk approaches have sufficed to identify the dominant causes of failure in forensic analyses, and the dominant risk contributors for proactive risk investigations. These techniques are particularly applicable when individual or even simple common failure events of a similar type dominate the analysis. However, nowadays due to increased understanding of the 'simple' mechanisms and the increasing complexity of the systems we build, failures in highly dependable systems arise from unexpected interactions between subsystems and the external and internal environment. Engineering data analysis is the process of data collection and investigation from a variety of perspectives, alternatively dissecting it into its underlying (yet often unknown) patterns; this process is becoming ever more necessary as systems become more complex. Some of the techniques employed are slicing the data sets according to known underlying variables, or overlaying data gathered from different perspectives, or imbedding data into previously established logical or phenomenological structures. This paper addresses the issues involved in visualizing patterns in data sets by providing examples of interesting maps from the past, indicating some of the maps currently in use, and speculating on how these visual maps might be developed further and used in the future to discover problems in complex systems before they lead to failure. Guidance is proposed as to how to explore and map data from different technical perspectives in order to evoke potentially significant patterns from reliability data. The techniques presented have been developed by combining approaches to common cause failure (CCF) classification with multidimensional scaling (MDS) to produce a new method for exploratory engineering data mapping