Structural prototypes for diagnosis of combustion engines
Abstract
A method for structural of one-dimensional low-frequency quasiperiodic sensor signals is presented, based on the classification of strings representing the course of the signal within periods. Basic structural units, called primitives, are defined rather general patterned on geometric elements a human expert may choose for signal description, so that the proposed signal-to-symbol transformation is not restricted to a specific signal but applicable to various low-frequency signals. By means of weighted Levenshtein distance and nearest-neighbor classification the applicability of the presented structural approach to an actual task in diagnosis of combustion engines is shown, namely to the cylinder-specific detection of misfiring through analysis of the angular acceleration of the crankshaft.
