Liquid electrolytes are one of the key components of lithium ion batteries (LIBs). The state-of-the-art electrolytes are highly optimised and hard to vary, and suffer from volatility, flammability and limited electrochemical stability. Highly concentrated electrolytes (HCEs) have emerged as a viable strategy to open up a new design space and have shown promise w.r.t. both lower volatility and greater stability, in large part due to the elimination of free (uncoordinated) solvent. At the same time, the structure and ion transport mechanisms of HCEs are much more complicated than for conventional organic electrolytes, making rational design difficult. Molecular-scale methods are needed to gain a deeper understanding of this class of electrolytes. Ab initio molecular dynamics (AIMD) is routinely used to study battery electrolytes and gives a detailed view of the structure and dynamics of the system. However, conventional modes of analysis based on the notion of Li+ first solvation shells are unsatisfactory in explaining ion transport for electrolytes with complex dynamic structure and non-vehicular transport mechanisms, which are believed to be important in HCEs. We present a novel method for automated dynamic structure discovery (DSD), in which the bonds connecting the possibly changing structures are discovered dynamically and chemical species are uniquely identified by their bond graphs, enabling a detailed statistical analysis of each species in terms of topology, geometry, population, lifetime distribution, and transport properties. We give a proof-of-concept by analysing the mechanisms of lithium-ion transport in a HCE of relevance to LIBs, and outline future extensions to this work, through which we hope to automate the generation of classical interatomic potentials with quantum accuracy by combining DSD with a genetic algorithm. CHAMPION, the software implementing this method, will be available on Github under an open source MIT license towards the end of 2019.