Condition monitoring of helical gears using automated selection of features and sensors

The selection of most sensitive sensors and signal processing methods is essential process for the design of condition monitoring and intelligent fault diagnosis and prognostic systems. Normally, sensory data includes high level of noise and irrelevant or redundant information which makes the selection of the most sensitive sensor and signal processing method a difficult task. This paper introduces a new application of the Automated Sensor and Signal Processing Approach (ASPS), for the design of condition monitoring systems for developing an effective monitoring system for gearbox fault diagnosis. The approach is based on using Taguchi's orthogonal arrays, combined with automated selection of sensory characteristic features, to provide economically effective and optimal selection of sensors and signal processing methods with reduced experimental work. Multi-sensory signals such as acoustic emission, vibration, speed and torque are collected from the gearbox test rig under different health and operating conditions. Time and frequency domain signal processing methods are utilised to assess the suggested approach. The experiments investigate a single stage gearbox system with three level of damage in a helical gear to evaluate the proposed approach. Two different classification models are employed using neural networks to evaluate the methodology. The results have shown that the suggested approach can be applied to the design of condition monitoring systems of gearbox monitoring without the need for implementing pattern recognition tools during the design phase; where the pattern recognition can be implemented as part of decision making for diagnostics. The suggested system has a wide range of applications including industrial machinery as well as wind turbines for renewable energy applications.