Railway suspension fault detection under variable operating conditions via random vibration signals and the stochastic Functional Model based method

The problem of fault detection in railway vehicle suspension systems via random vibration signals under varying operating conditions, presently payload, is considered. The focus is on exploring the performance limits for small, incipient, faults, while maintaining low false alarm rates despite significant variations in the payload. The study is based on a recently introduced Functional Model (FM) method for robust fault detection and Monte Carlo simulations. Comparisons with an alternative state-of-the-art Principal Component Analysis based method are also made. The results indicate that the FM method, based on just two car body sensors and a narrow (0-40 Hz) frequency bandwidth, achieves very high to excellent detection performance for primary and certain secondary suspension faults, yet inadequate for certain small magnitude (5% to 20% reduction in suspension element properties) faults in the secondary suspension. The detection performance is significantly degraded when the PCA based method is employed.