Incremental Latin Hypercube Additive Design for LOLIMOT

Estimating data based models of industrial processes is an essential step in making existing production lines intelligent and increasing their flexibility and efficiency. The collection of data during the normal operation often does not provide sufficient information to achieve a high model quality, so that design of experiments methods are needed. In this paper a step-wise methodology for the design of experiments and modeling is proposed, which makes it more applicable for industrial processes. In contrast to commonly used experimental designs, in which all available inputs are varied at once, the proposed method starts with only one input and successively adds another input to the design in each step. The experimental design is combined with an incremental modeling strategy using LOLIMOT-trained models. Experimental results are presented based on a computer experiment and a heating system. In both experiments a model quality comparable to a standard approach is achieved.