For detecting faults in rotating machinery, different fault types generally require different techniques for the effective detection of the faults; hence, many different techniques have been developed. However, we often find that the existing detection techniques are either incapable or ineffective for many new fault types. Therefore, we will need to develop brand new methods after the fault event. This can significantly constrain the usefulness and effectiveness of machine health monitoring systems. In this article, we propose a unified signal processing approach to detecting and trending changes caused by various types of faults in rotating machinery. The theoretical foundation of the proposed technique is based on extracting the differences between the synchronously averaged (angularly resampled) signals acquired under the changing health conditions after aligning their phases. The aligning of the phases is achieved through a novel and efficient approach, which is performed in the frequency domain. The alignment enables a direct comparison between the amplitudes of the synchronously averaged signals and provides a trustworthy evaluation using the same reference datum. The proposed method is novel and conceptually unsophisticated, and its effectiveness is demonstrated using vibration data from rotating machines with several different types of faults. The results have shown that this single unified change detection approach can be very effective in detecting and trending changes caused by many different types of faults in rotating machines. This is an ideal technique to be implemented for testing on aircraft engine onboard health monitoring systems.

A comparison was made between computer model predictions of gear dynamic behaviour and experimental results. The experimental data were derived from the NASA gear noise rig, which was used to record dynamic tooth loads and vibration. The experimental results were compared with predictions from the Australian Defence Science and Technology Organisation Aeronautical Research Laboratory’s gear dynamics code, for a matrix of 28 load-speed points. At high torque the peak dynamic load predictions agree with experimental results with an average error of 5 percent in the speed range 800 to 6000 rpm. Tooth separation (or bounce), which was observed in the experimental data for light-torque, high-speed conditions, was simulated by the computer model. The model was also successful in simulating the degree of load sharing between gear teeth in the multiple-tooth-contact region.

A Bell 206B main rotor gearbox was run at high load under test conditions in the Helicopter Transmission Test Facility operated by the Defence Science and Technology Organisation (DSTO) of Australia. The test succeeded in initiating and propagating pitting damage in one of the planet gear support bearings. Vibration acceleration signals were recorded periodically for the duration of the test. The time domain vibration signals were converted to angular domain to minimise the effects of speed variations. Auto-Regressive Moving-Average (ARMA) models were fitted to the vibration data and a change detection problem was formulated in terms of the Generalised Likelihood Ratio (GLR) algorithm. Two different forms of the GLR algorithm in window-limited online form were applied. Both methods succeeded in detecting a change in the vibration signals towards the end of the test. A companion paper submitted by the University of New South Wales outlines the corresponding diagnosis and prognosis algorithms applied to the vibration data.

No abstract