Intelligent Vibration Signal Processing for Condition Monitoring

Recent advances in pattern analysis techniques together with the advent of miniature vibration sensors and high speed data acquisition technologies provide a unique opportunity to develop and implement in-situ, beneficent, and non-intrusive condition monitoring and quality assessment methods for a broad range of rotating machineries. This invited paper provides an overview of such a framework. It provides a review of classical methods used in vibration signal processing in both time and frequency domain. Subsequently, a collection of recent computational intelligence based methods in this problem domain with case studies using both single and multi-dimensional signals is presented. The datasets used in these case studies have been acquired from a variety of real-life problems 1 Vibration and Condition Monitoring Vibration signals provide useful information that leads to insights on the operating condition of the equipment under test [1, 2]. By inspecting the physical characteristics of the vibration signals, one is able to detect the presence of a fault in an operating machine, to localise the position of a crack in gear, to diagnose the health state of a ball bearing, etc. For decades, researchers are looking at means to diagnose automatically the health state of rotating machines, from the smaller bearings and gears to the larger combustion engines and turbines. With the advent of wireless technologies and miniature transducers, we are now able to monitor machine operating condition in real time and, with the aid of computational intelligence and pattern recognition technique, in an automated fashion. This paper draws from a collection of past and recent works in the area of automatic machine condition monitoring using vibration signals. Typically, vibration signals are acquired through vibration sensors. The three main classes of vibration sensors are displacement sensors, velocity sensors, and accelerometers. Displacement sensors can be non-contact sensors as in the case of optical sensors and they are more sensitive in the lower frequency range, typically less than 1 kHz. Velocity sensors, on the other hand, operate more effectively with flat amplitude response in the 10 Hz to 2 kHz range. Among these sensors, accelerometers have the best amplitude response in the high frequency range up to tens of kHz. Usually, accelerometers are built using capacitive sensing, or more commonly, a piezoelectric mechanism. Accelerometers are usually light weight ranging from 0.4 gram to 50 gram. 1.1 Advantages of vibration signal monitoring Vibration signal processing has some obvious advantages. First, vibration sensors are non-intrusive, and at times non-contact. As such, we can perform diagnostic in a non-destructive manner. Second, vibration signals can be obtained online and in-situ. This is a desired feature for production lines. The trending capability also provides means to predictive maintenance of the machineries. As such, unnecessary downtime for preventive maintenance can be minimized. Third, the vibration sensors are inexpensive and widely available. Modern mobile smart devices are equipped with one tri-axial accelerometer typically. Moreover, the technologies to acquire and convert the analogue outputs from the sensors are affordable nowadays. Last but not least, techniques for diagnosing a wide range