Dual-rotor system is an important rotor form in rotating machinery like gas turbine engine. Its complex structure results in rich dynamical behaviors and more probability to failure. The modeling and simulation of the dual-rotor system can help to understand its dynamic characteristics and provide theoretical support for the design, operation and maintenance. In this paper, a dynamic model of a dual-rotor system with multiple rotor faults is established. The dual-rotor vibration model without any fault is built by finite element method, where the two shafts are connected by an inter-shaft bearing and nonlinear models of rolling element bearing and squeeze film damper are considered. The numerical integration method of Newmark-? is used to obtain the steady-state vibration response of the system. Then rotor faults are introduced to the system model, including unbalance, misalignment, looseness and rub-impact. The steady-state responses of single faults in the dual-rotor system are analyzed and typical fault features are obtained. Then coupling characteristics between different rotor faults, and the influences of the squeeze film dampers on the dynamic characteristics and fault features of the system are studied.
This paper presents models for complex systems or subsystems to transform conditioned-based data (CBD) signatures into functional-failure signatures (FFS) that are particularly amenable to processing by state-estimator prediction algorithms. A CBD-based approach overcomes certain problems associated with conventional approaches, such as model- or statistical-based methods, for producing prognostic information. Failure modes generate characteristic CBD signatures that are correlated to changes in value of a parameter, such as capacitance, as degradation progresses. Features are extracted from CBD signatures and transformed into fault-to-failure progression (FFP) signatures: a function of the feature data (FD). Then FFP signatures are transformed into degradationprogression signatures (DPS): a function of the change in the value of a parameter. A DPS, absent noise, has a characteristic linear curve: a straight line progressing from 0 (no degradation) to a value defined as a level of degradation at which the degrading component, and the assembly it is in, no longer functions within specification(s): functional failure occurs.
A DPS is then transformed into a functional-failure signature (FFS) for input to state-estimator prediction algorithms to support Prognosis for Health Monitoring/Management (PHM): (1) an FFS approaches an ideal straight-line transfer curve as noise is ameliorated and/or mitigated; (2) has negative values in the absence of degradation; (3) has positive values below 100 when there is degradation below a defined level of functional failure; and (4) has values at or above 100 when the level of degradation is at or above a level defined as functional failure. Even in the presence of noise and feedback effects, and even when the rate of degradation is nonlinear, an FFS is still a very linear transfer curve. Seven different families of signatures and models are presented to transform CBD-based signature data into FFS data, and when that data is so transformed and used, the estimation accuracy of prediction algorithms is greatly improved.
This paper presents a method to transform raw sensor data into data that is properly conditioned to be analyzed on a real time basis by anomaly detection and state estimator routines. This capability, in turn, provides a solid foundation that can detect and mitigate faults that occur in complex systems. An example of an electromechanical actuator will be analyzed using the algorithms discussed in this paper.
Vibration analysis is an important part of predictive maintenance and is widely used to diagnose a wide range of machine component faults. One of the major limitations of vibration analysis as it stands today is that, in many cases, it still relies on experts to interpret vibration signatures manually. As such, it is at the mercy of human error and biases. To build an automatic machine learning system for vibration analysis a high-quality dataset is needed. While there are many ways to achieve this, one commonly used technique is to crowdsource data labeling. In our paper, we will present the application of crowdsourcing for vibration analysis on common HVAC machinery.
The paper proposes an optimized testing agenda for centrifugal compressor with a compact manufacturing and testing schedule. As API mandates to conduct spin tests ( MRT ) of all compressors of similar deign and geometry , this paper proposes to conduct only one ASME PTC 10 modified Type 1 or Type 2 Full / Part load test (as all test beds may not have required hydrocarbon gas or power for type 1) combining mechanical run test at vendor works among the full lot.
The intention of the paper is to classify the criticality of rotor in terms of rotor stability ratio and then deliberate on the extent / type of tests taking account of OEM test bed capability and schedule of delivery of machines .To supplement the tests , the paper proposes to undertake extensive design audit activities in terms of rotor-dynamics , aerodynamics taking account of case histories of past failures. Assurance of dimensional repeatability in terms of metrology backed by latest methodology of fault identification of multi-layered manufacturing process with PQM ,which can avoid multiple tests of similar machines can avoid multiple third party inspection at various stages.
In No load spin test and PTC10 type 2 tests , site conditions such as influence of piping loads , gas pressure / density , foundation dynamics are not replicated . The above tests do not identify the region of incipient surge, torsional instability region as well . The paper proposes to introduce various instruments and sensors to detect the above instability region with some diagnostic flow charts for the extensive test .With the rotor-dynamic data taken from the proposed test set up, it shall be easy to further enhance the base line data after site performance test .The same can be used for pre-alarm configuration based on zone wise amplitudes in vibration spectrum which can be very useful for reliability engineer engaged in diagnostic and prognostics. For critical machines located in hostile environment, anomaly detection may be carried out with shape identification of plot / spectra.