Development of a Physics-based Model for Machine Condition Monitoring

  • By Ken Keith
  • 4 years ago
  • Shares

Condition monitoring of rotating and reciprocating equipment is an important means of ensuring the long-term health and optimal operation of these types of machinery. Many current condition monitoring systems use what may be described as a “data-driven model” to create a performance model for the machinery. In other words, the real machinery will be run over the widest possible range of performance conditions, while sensors are used to pick up various outputs such as RPM, cylinder pressures, temperatures and vibration levels. This data is then processed to create a map of the expected normal sensor outputs for a given performance condition. This method of condition monitoring has some shortcomings when it comes to providing a predictive model of the machinery being monitored. One disadvantage is that creation of a datadriven model requires a significant amount of time to work directly on the equipment to cover the full operating range. If a part of the machine is modified, such as the replacement of an existing cylinder with a different-sized one, the process would have to be repeated for the new configuration, resulting in further machine down-time. In addition, the data-driven model cannot be used to predict the expected operating condition of the machinery under various faults, since attempting to do so by re-creating faults in the machine and then running it would likely damage the equipment.