A methodology for cracks identification in large crankshafts

Abstract

Diesel engines used in power plants and marine propulsion are especially sensitive to outage events. Any advance in the early detection of failure will increase the reliability of the electricity supply and improve its productivity by reducing costly power outages. Fault detection and diagnosis is important technology in condition-based maintenance for diesel engines. This article presents a classifier based on neural networks for identifying failure risk level in crankshafts, the engine component of greatest cost concern. The authors have developed a finite element model for crack growth that fits well with fracture appearance and produces the evolution of crankshaft stiffness with crack depth. A lumped system model of the engine uses this evolution as input, giving the instantaneous speed at the engine flywheel as a function of crack depth. All the results shown in the paper come from outputs of the simulation models which have been built from real engine data. Measurements of the instantaneous flywheel speed were not available due to the crankshaft failure. All data are extracted from this speed and are then classified using a Radial Basis Function neural network.