Feature-based design analysis for automatic classification of simulated nonlinear responses using machine learning
Paper i proceeding, 2016
A practical concept design analysis approach for automatic processing and classification of simulated responses is presented. Deterministic and nonlinear dynamics are studied under ideal loading and low noise conditions to determine fundamental system properties, how they vary and possibly interact. Responses should be classified into characteristic types, such as periodic or nonperiodic, resonant or nonresonant, linear or nonlinear, and further into subcategories such as single or dual frequency responses, hardening or softening. For this, time-signals are evaluated using methods and metrics commonly used within structural dynamics and then possibly associated with qualitative features according to measures based on a set of rules-of-thumb criteria. A support vector machine is trained to determine whether a single feature, or combinations of features, applies or not. This paper describes elements of the analysis, report practical considerations and discuss the effectiveness of evaluated features using known few-degree-of-freedom examples.