Edward M. Huff

Analysis of Triaxial Vibration Data for Health Monitoring of Helicopter Gearboxes

Research on the nature of the vibration data collected from helicopter transmissions during flight experiments has led to several crucial observations believed to be responsible for the high rates of false alarms and missed detections in aircraft vibration monitoring systems. This work focuses on one such finding, namely, the need to consider additional sources of information about system vibrations. In this light, helicopter transmission vibration data, collected using triaxial accelerometers, are explored in three different directions, analyzed for content, and then combined using Principal Components Analysis (PCA) to analyze changes in directionality. The frequency content of the three different directions is compared and analyzed using time-synchronously averaged vibration data. To provide a method for analysis and monitoring purposes, the triaxial data are decorrelated using a mathematical transformation, and compared to the original axes to determine their differences. The benefits of using triaxial data for vibration monitoring and diagnostics are explored by analyzing the changes in the direction of the principal axis of vibration formed using all three axes of vibration. The statistical variation introduced due to the experimental variables is further analyzed using an Analysis of Variance approach to determine the effect of each variable on the overall signature. The results indicate that triaxial accelerometers can provide additional information about the frequency content of helicopter gearbox vibrations, and provide researchers and industry with a novel method of capturing and monitoring triaxial changes in the baseline vibration signatures.

An Analysis of Maneuvering Effects on Transmission Vibrations in an AH-1 Cobra Helicopter

During the past several years, an AH-1 Cobra aircraft at the NASA Ames Research Center has been instrumented with tri-axial accelerometers and a data acquisition system to support the experimental study of in-flight transmission vibration patterns. This paper describes the on-board HealthWatch system and presents important statistical analyses of the collected data sets. These analyses provide insight into how transmission vibration responds to several factors typically related to health and usage monitoring systems (HUMS), such as maneuver condition, order of execution, and pilot differences. Although a large database of flight recordings has been collected, these results focus on an overall analysis of planetary ring gear data that were recorded in two sets of flights. It is shown that variability due to torque is a major factor to be considered in real-time HUMS design and that certain steady state maneuvers yield a dramatically higher percentage of stationary recordings. Finally, it is conjectured that multi-axis recording may have previously unrecognized advantages for signal conditioning or analysis.

On the effects of production and maintenance variations on machinery performance

The variations introduced during the production and maintenance of rotating machinery components are correlated with the vibration and noise emanating from the final system during its operational lifetime. Vibration and noise are especially unacceptable elements in high‐risk systems such as helicopters and aircraft engines, resulting in premature component degradation and a potentially unsafe flying environment. In such applications, individual components often are subject to 100 per cent inspection following production and during operation through rigorous maintenance, resulting in increased product development cycles and high production and operation costs. In this work, the aim is to provide engineers with a technique to evaluate vibration modes and levels for each component or subsystem prior to putting them into operation. This paper presents a preliminary investigation of the correlation of manufacturing and assembly variations with vibrations, using an experimental test rig. A factorial design is used to study the effects of various factors. Challenges in developing a process monitoring and inspection methodology to predict performance quality are identified, followed by a discussion of future work.

Classification of aircraft maneuvers for fault detection

Ensemble classifiers tend to outperform their component base classifiers when the training data are subject to variability. This intuitively makes ensemble classifiers useful for application to the problem of aircraft fault detection. Automated fault detection is an increasingly important problem in aircraft maintenance and operation. Standard methods of fault detection assume the availability of data produced during all possible faulty operation modes or a clearly-defined means to determine whether the data represent proper operation. In the domain of fault detection in aircraft, the first assumption is unreasonable and the second is difficult to determine. Instead we propose a method where the mismatch between the actual flight maneuver being performed and the maneuver predicted by a classifier is a strong indicator that a fault is present. To develop this method, we use flight data collected under a controlled test environment, subject to many sources of variability. In this paper, we experimentally demonstrate the suitability of ensembles to this problem.

Helicopter transmission health monitoring using real-time neural computing methods

A real-time helicopter transmission health monitor is being developed and evaluated at the NASA Ames Research Center in conjunction with the U.S. Army Aeroflightdynamics Directorate. This system uses non-real-time neural computing techniques to first learn the vibration signatures of faulty gearbox modes. Real-time capability is then achieved by faithful replication of the neural network processing model in an air-worthy integrated electronics architecture. The latter is based upon seminal work done at Draper Laboratory on INCA (Integrated Neural Computing Architecture). Prior work done by a number of organizations has substantiated the utility of neural computation for this kind of application in static laboratory environments. The present effort extends that basic research into dynamic flight by use of the FLITE (Flying Laboratory for Integrated Test and Evaluation) vehicle, which is an instrumented Cobra helicopter located at Moffett Field, CA.<<ETX>>