|Ph.D Student||Alkahe Jonathan|
|Subject||Helicopter Faults - Simulation, Detection and Identification|
|Department||Department of Aerospace Engineering||Supervisors||Professor Emeritus Omri Rand|
|Professor Yaakov Oshman|
Helicopter health and usage monitoring systems (HUMS) have been developed in an attempt to reduce operational costs and improve aircraft readiness and flight safety. Existing HUMS are based on automatic exceedance monitoring of vibration levels measured at selected points on the helicopter, based on data supplied by the manufacturer. These systems are quite limited with regard to their health monitoring capability, since the monitored helicopter has to be identical, both in structural configuration and in flight parameters, to the manufacturer's reference. The great importance of fault detection and identification (FDI) capability arises mainly from the fact that the helicopter rotor is a flight critical system, with no structural redundancy.
In this work a novel, model-based helicopter FDI methodology is presented. Since it is difficult to obtain flight test data for a damaged helicopter rotor, a physics based model offers the opportunity to study the simulated behavior of the damaged helicopter. A key advantage of model-based methods is the inherent flexibility due to the ability to perform adjustments in structural parameters reflecting configuration changes, as well as flight condition modifications. Their major drawback arises from the need to develop a detailed model of the helicopter, including all of the complex rotor structural components and the intricate aerodynamic effects. This still remains a very challenging area for research.
The detailed analytic model of the helicopter rotor system, considered in this work, includes: elastic blades, all of the rotating system components and a rigid fuselage. All dynamic and aerodynamic effects are included in the model, which is based on the solution of nonlinear equations of motion. This model was utilized to simulate several common faults occurring in helicopters. A set of sensors placed at selected points comprises the noisy measurement input vector for the identification stage. The identification procedure is basically a probability-based comparison between all damaged cases, including a non-damaged model of the helicopter. A multiple model adaptive estimation (MMAE) technique is adopted, which is based on a set of parallel Kalman filters corresponding to the set of faults modeled. In contrast to neural-nets-based techniques published in the literature, this method does not require any training stage, and filter out the process and measurement noises, which are inherent in the formulation. The statistical nature of the results using the proposed algorithm also alleviates the decision making process regarding the faulty system. The performance of the proposed method is evaluated by calculating the false alarm rate along with the detection and identification probabilities for each specific damage type and location, sensor type and location and (measurement and process) noise level. In cases where these probabilities attain unacceptable values, a specifically designed statistical experiment is utilized.
An extensive numerical study is performed to demonstrate the performance, applicability and viability of the proposed FDI method. In the first stage, the method is applied to a relatively simple and accurate finite element model of a single rotating flapping blade in vacuum. Although the blade's natural frequencies and modal shapes are hardly influenced by the local damage, the algorithm provides good detection and identification capabilities. A parametric study is also conducted, which gives better insight on the various phenomena affecting the FDI process. Next, a full-scale rotor is considered with several component faults. A fixed-shaft case is simulated where various sensor types are tested, including blade tip tracking and hub load measurements. The statistical experiment is conducted in the case of blade tip deflection measurements, in order to improve poor FDI performance caused by insufficient information. The case of hub load measurements proves to give high damage detectability. This is followed by trimmed forward flight simulation, where the sensors used are fuselage vibration levels. Here also, good performance is demonstrated.