Excessive torsional vibration in rotating equipment trains can result in damage to, or failure of, equipment, leading to emergency shutdowns and costly downtime. It also increases the potential for major safety incidents, such as turbine over-speed events or coupling failures that generate dangerous shrapnel, as shown in FIG. 1.
When designing a torsional system, engineers typically use comprehensive torsional vibration analysis to avoid excessive vibration. API Standard 617, which specifies torsional system requirements, specifies a 10% separation margin between torsional natural frequencies (TNFs) and any excitation frequency. The method of achieving an acceptable torsional separation margin is generally limited to coupling selection and tuning. In some cases, modification of system inertia is acceptable. If an acceptable separation margin is not achievable, then design engineers must use stress analysis to demonstrate torsional system design acceptability.
The validity of the predicted TNF and associated stress analysis is dependent on the accuracy of the model. However, assumptions within the analytical data are always subject to a degree of uncertainty. To address this uncertainty, a case study of 10 torsional systems was conducted, looking at the effect of variation in mass-elastic data, and comparing differences between measured and predicted TNFs.
Modeling and natural frequency prediction. The TNF of a rotating equipment train is predictable through analytical methods. One modeling method is the mass-elastic or inertia-stiffness model, which uses lumped inertias, springs and dampers. Each component in the torsional model has a specific function, accurately modeled by following readily available component modeling guidelines. As torsional damping is usually small, an undamped analysis is generally enough to predict the TNFs of most rotating equipment systems.
Uncertainties in models. Uncertainties are factors that cause a difference between the predicted TNF and the true TNF. If the lumped mass-elastic model accurately represents the equipment train, then the predicted TNFs will match the real system's TNFs.
For the purposes of this study, two types of uncertainties were considered: error and variation. Error uncertainties include an insufficient number of...