On-line anomaly detection and resilience in classifier ensembles
Pattern Recognition Letters
| Hi-index | 0.00 |
One of the most important properties of autonomous ve- hicle is the reliability which means to detect the fault by itself and then isolate the fault. This paper combined the neural-fuzzy model with the fault hypothesis test, and put forward a Neuro-Fuzzy model-based Cumulative-Sum (NF- CUSUM) algorithm. It gave the assumptions aiming at the faults and set the alarm when the probability of the fault case was greater than the probability of the normal case. Under the fault case the system is called to have a fault, oth- erwise it is normal. The core of the NFCUSUM algorithm is to find a logic fault detector (decision function) which ex- presses whether the fault occurs at one sample time. The design idea of the decision function is that the system is suf- fered a fault and gives alarm when the value of the decision function is over the preset threshold; otherwise the system is in normal mode. The simulation results in Matlab show that the logic fault detector designed by the NFCUSUM al- gorithm in this paper is practical, efficient and robust.