Neural Networks and Information in Materials Science

Abstract

Abstract:Neural networks have pervaded all aspects of materials science resulting in the discovery of new phenomena and have been used in quantitative design and control. At the same time, they have introduced a culture in which both noise and modeling uncertainties are considered in order to realize the value of empirical modeling. This review deals with all of these aspects using concrete examples to highlight the progress made, whilst at the same time emphasizing the limitations of the method. 

A Constitutive Equation Relating Composition and Microstructure to Properties in Ti-6Al-4V: As Derived Using a Novel Integrated Computational Approach

“…In particular, the work of Bhadeshia [14][15][16] and his colleagues have, over the years, helped to bring ANNs into the main for materials science research. [19,[22][23][24]30,31] While much of the published work has focused on the effects of composition and processing on the resulting material properties, [32][33][34][35] his papers have inspired other researchers to assess other systems and, in particular, explore the influence of not only composition but also microstructure on the resulting properties.…”

“…The challenge associated with the problem is illustrated nicely in the work of Kar et al [10] who point out the natural scatter in the data when only considering one variable, and note the great difficulty that one would face attempting to extract a direct relationship via traditional regression-based approaches. Their solution to the problem, and one adopted in other similar studies [11][12][13][14][15] was to use artificial neural networks (ANNs), such as those based upon Bayesian statistics, [16][17][18][19] to formulate the interrelationships that exist among the various input variables. The details of the Bayesian neural network are described nicely in these papers, and are not reproduced here.…”

Critical assessment: Martensite-start temperature for the γ→ε transformation

“…Comprehensive details of this method have been described elsewhere [88][89][90][91] and hence are not reproduced here except when the information is necessary to reproduce the work.…”

Fatigue crack growth rate model for metallic alloys

“…The theory behind practical Bayesian networks has been described in [13,14] and the background information theory is available in a seminal textbook on the subject [15]. In addition, this method been reviewed thoroughly [11], as have been its applications [16]. Indeed, there have been diverse applications which lead to useful and verifiable predictions in the context of low-cycle fatigue [17], the estimation of bainite plate thickness [18], the calculation of ferrite number in stainless steels [19], the estimation of tensile strength [20,27], impact strength [21,26], the effect of processing parameters on marageing steels [22], the modelling of strain induced martensitic transformation [23], and the reduction in steel varieties [28], to name but a few.…”