Mechanical Systems and Signal Processing (2000) 00(0), 000}000 doi:10.1006/mssp.2000.1297, available online at http://www.idealibrary.com on TOWARD HELICOPTER GEARBOX DIAGNOSTICS FROM A SMALL NUMBER OF EXAMPLES M. ZACKSENHOUSE, S. BRAUN AND MFELDMAN Faculty of Mechanical Engineering, Technion}Israel Institute of Technology, Haifa, Israel AND M. SIDAHMED University de Technologie de Compiegne, France (Received 30 November 1999; accepted 11 February 2000) The United States Navy funded Westland Helicopters LTD to carry out a series of tests on the CH-46E helicopter transmission for the purpose of generating a database that can be used to evaluate diagnostic tools in general, and neural networks in particular. The database is extensive in the number of accelerometers, load conditions and diversity of faults. However, only 1}4 instances (levels) of each fault were included. Thus, the number of examples that can be extracted to train any classi"cation scheme is limited, and well below the minimum required. We argue that instead the database is best used to develop speci"c signal generation models for the CH-46E helicopter gearbox. Such models can then be used to suggest meaningful features and derive appropriate signal-processing tools. Speci"cally, we demonstrate that the meshing vibrations induced by the large collector gear located on the quill shaft are signi"cant and may interact with the vibrations induced by other elements attached to the same shaft. An appropriate model is developed and the e!ect, termed cross-gear-pair interaction, is studied using di!erent signal-processing tools.  2000 Academic Press 1. INTRODUCTION Helicopter gearboxes are critical systems, where in-#ight malfunctions can lead to cata- strophic results. Many investigations, theoretical and experimental have been conducted in this area for decades. Recently, the main power transmission of a US Navy CH-46E helicopter was tested by Westland Helicopters LTD on their test rig under tightly control- led conditions. Some details of the test rig, controlled test parameters, instrumentation and vibration data were made public domain [1], enabling interested researchers to test various signal-processing techniques. A special workshop was even dedicated to some initial results obtained by independent research groups [2]. The available database includes simultaneously recorded vibration signals from eight accelerometers, and a tachometer signal. A multitude of test conditions was conducted under di!erent loads with arti"cially induced (typical) defects. Some tests involved two di!erent instances of the same type of fault at di!erent levels of severity. Other tests were conducted during the initiation and development of the fault. Seven type of faults were included, one at a time, along with a base test with a fault-free transmission. It is indeed rare that such a huge database is made available to the community engaged in vibration signal processing. MSSP=20001297 Susan KGU BG 0888}3270/00/050789#21 $35.00/0  2000 Academic Press