Data-mining of materials-properties to implement ultra-precision process automation

Precise optics underpin diverse applications, including remote-sensing from space, astronomy, laser-physics, medical-diagnostics, security & defense. Laser-fusion potentially requires thousands of optics, and refurbishment due to laser-damage. Consumer optics are mass-produced using ultra-precise molds & dies, which require exquisite surfaces. Topically, autonomous vehicles require cameras, sensors and advanced lighting. Beyond optics, an aging population demands joint-implants with extended lifetimes, requiring superior surface-quality. In the case of turbine-blades, surface-quality drives energy-efficiency. Overall, we need to control surfaces from centimeters to meters in size, from microns down to nanometers in form and smoothness!

There are at least 150 optical glass-types, plus metals, crystals and ceramics, used for lenses, mirrors and other optical components. This diversity gives the lens designer the breadth of parameters needed for optimizing image formation in demanding applications. Different mirror materials are chosen on thermal and mechanical properties, stability and economic grounds. Semiconductor wafer-fabrication, the mold & die sector, prosthetic joint implants, turbines, and the whole gamut of additive-manufacturing, further extends the list of relevant materials that need highly-finished surfaces.

CNC machine-operators, when confronted by a new material selected by the designer, typically conduct test-runs to optimize detailed finishing approaches to achieve the required surface-quality in the minimum time. This empirical approach is extremely tedious, and sits ill with our overall ambition – fully autonomous manufacture of ultra-precision surfaces.

The proposed project will start by creating a data-base of fundamental physical, chemical and thermal properties of i) the widest practical range of relevant materials, encompassing ii) a manageable subset for which prior process-data is available, or for which new process trials will be conducted. The source of the materials data will be manufacturers’ data and publications, with some practical materials-tests to ‘in-fill gaps’. All these data will be collated in a common database format. Typical parameters related to material removal comprise ductility-index, fracture-toughness, Young’s modulus, density, thermal expansion-coefficient and chemical composition.

Multivariant analysis with pattern recognition on the materials’ database, joined with data on process strategies, will be applied to the dataset. The analytic results will be interpreted to give insight into how materials-properties affect process-results. Then, using quantitative materials characteristics in the data-base, correlative analysis will be used to predict optimum process parameters for materials for which process-data is not available. This should ultimately eliminate experimental process-optimization. The developed machine-based expertise will then be applied to materials not previously trialled, to establish the success of the predictive methods.