Design Mining is the use of computational intelligence techniques to iteratively search and model the attribute space of physical objects evaluated directly through rapid prototyping technology to meet given objectives. It enables the exploitation of novel materials and processes without formal models or complex simulation, whilst harnessing the creativity of both computational and human design methods. The traditional engineering design process and the data mining process share many similarities, and the proposed project will seek to exploit this fact and embed data mining within design. Models which enable what-if testing of the characteristics of the object design space are created throughout. A sample-model-search-sample loop creates an agile/flexible approach, ie, primarily test-driven, enabling a continuing process of prototype design consideration and criteria refinement by both producers and users. Parallel/sub-design scenarios will also be explored, considering the effects of the degree of prototype and data/model synchronisation in the concurrent tasks upon the utility of the approach. In particular, machine learning techniques will be used to iteratively search and model the object design space informed by the performance metrics of microbial fuel cells whose electrodes are fabricated using 3D printing, both as individual units and as collectives in cascades.

Potential Impact:
The proposed project will combine and extend techniques from machine learning with emerging 3D printing technology to develop a novel way to design 3D objects in domains which are difficult to model either formally or in simulation. As such, it will have the potential for impact in a wide range of areas. Academics and commercial organisations using optimization and modelling algorithms in complex domains will benefit from the new approaches to problem space approximation via direct sampling developed. Moreover, Microbial Fuel Cells have received increased attention over the last decade in particular, due to the unique advantage of generating electricity directly from the breakdown and treatment of organic waste, which renders the technology - at the very least - energy neutral. Some of the main challenges lie with the core materials, system configuration and design (eg, surface area-to-volume ratio), which are elements that design mining can help address. This is likely to lead in improvements, which can be directly implemented and tested in field trials that are already happening as part of the applicants' EPSRC New Directions as well as Gates Foundation projects, thus maximising the potential of the technology.

The design of novel microbial fuel cells has been chosen for a number of reasons, not least the pressing climate change agenda. The proposers will make use of the EU-funded Environmental Technologies Innovation Network (led by UWE) and other project networks (eg, EPSRC Supergen) to pursue impact. It can also be noted that the approach is directly applicable to the design of wind turbines, wave energy systems, 3D solar cells, etc. and so should be of benefit to the wider renewable energy research and development community.

Machine learning has also been used as part of human-centred design systems, including those intended for fabrication (e.g., see Cornell University's "EndlessForms"). In future, the design of green products such as 3D printed, wind-powered lights (e.g., see Ponoko Ltd) may exploit the techniques developed to both increase performance and aesthetic measures.