Fault Detection in a 3DOF Helicopter System


This paper presents a fault detection scheme based on k-means clustering. In the proposed approach, the k- means algorithm is applied in unsupervised learning of patterns from system input-output data acquired under normal operating conditions. This method does not require an accurate model of the system, additional equipment or historical records of faults on the system. The pattern classification technique will be used to determine regions that reflect adequately the normal operating signature. The normal behavior is verified by calculating the Euclidean distance from each set of measurements to the centroids of the learned clusters. The clusters represent regions in the n-dimensional space composed by system variables. These regions contains the variables joint trajectories during normal operating conditions and are used to verify if the relation among system variables remains the same. Otherwise, a fault is declared. Additionally, an analytical redundancy using a wavelet filter bank scheme for fault detection based on the monitoring of the innovations of a Kalman filter is used for comparison.

For experimental validation of the proposed scheme, a pilot plant in the form of a three-degree-of-freedom helicopter is employed. The system has two DC (Direct Current) motors, each one coupled to a propeller. The actuation signals consist of voltages applied to the motors. Three types of movements are possible: elevation, pitch and travel. The fault under consideration consists of a 10% reduction in the gain of one of the motors during a landing procedure. The results in terms of detection delay and ROC (Receiver Operating Characteristic) curve indicate that the method has good potential when compared with analytical redundancy, with the advantage of not requiring an analytical model of the system and without necessarily being complex or costly to implement.

Autores: Elen Collaço de Oliveira e Guilherme Monteiro Garcia

Jackson Paul Matsuura (MATSUURA, Jackson Paul)

Roberto Kawakami Harrop Galvão (GALVÃO, ROBERTO K. H)

Produto: Helicóptero 3DOF

Titulo: Fault Detection in a 3DOF Helicopter System

Ano de Publicação: 2010.

Link: Download