Document Type


Source Publication Title

Proceedings of CAMX 2018 Conference,Dallas, TX


The concept of “data driven” grew out of the general subject of data analytics, especially in the business world. The classical sequence proposed by Dykes (2010) shown below illustrates the general contents of the concept. Indeed, the idea of “predictive modeling” of the future behavior of a system has strong roots in the business community. In general, for engineering and other fields, the data driven challenge is to use the past with information from the present to predict the future behavior of a system. In turn, the user must pay for collection and processing, hosting and maintenance of the data, and for the cost of analysis, etc. and address the risk of breach of the system. The first and perhaps the most important question in this venture is the nature of the data itself, i.e., the “source of truth.” We have many measurable quantities for composite materials; which of those should we use for the objectives of our analysis? Traditional data such as “failure rates” result in sparse data sets that may be impossible to analyze, and abundant data from “health monitoring” systems may be distantly related to the physics of our objective function. Data interpretation and analysis for composites is also a challenge; there may be “missing physics” that motivates the use of machine learning or more general artificial intelligence / neural network systems for interpretation. The present paper discusses these general questions and discusses several paths forward, including the implementation of mixed physics / neural network analysis approaches, machine learning, material state variable definitions for composite materials, and recent experience with these and other concepts.


Engineering | Materials Science and Engineering

Publication Date





Contact if you are the author.

Available for download on Wednesday, January 01, 3000