Wo! this is like asking do you need to go to London to see if itt is really there, or if a tree falls down when no one is to hear does it make a noise? The whole issue revolves around do you trust your model, if you don't there is your answer.

p

I posed this question because the need for model-based approach is a really a limiting factor of vibration-based rotor eccentricity detection. And, this is a conceptual conditio-sine-qua-non that affect mechanical engineers. What if, instead, one proposes a method for assessment of rotor percentage eccentricity in sunchronous AC generators without complicate mathematical mechanical models of the generator? I mean by using split-phase currents. :)

please refer to https://www.researchgate.net/publication/256274497_Field_Experience_With_the_Split-Phase_Current_Signature_Analysis_%28SPCSA%29_Eccentricity_Assessment_for_a_Stand-Alone_Alternator_in_Time-_Varying_and_Unbalanced_Load_Conditions

Conference Paper Field Experience With the Split-Phase Current Signature Anal...

Practically it is not necessary for damage diagnosis of the rotor, but for better interpretation of the FFT or spectrum of the rotor vibration signal, it is necessary to know which fault can excite which frequency. Therefore by looking at to the related papers in literature by which the model has been used for understanding the exact behavior of the rotor vibration when there is a fault, you can use just practical spectrum of the vibration signal for finding the rotor fault.

Thus by deriving the spectrum of the vibration signal and finding the fundamental frequency and studying the literature and related book you can find the fault of the signal

The question is in my opinion interesting. In my opinion a "model" is always neessary: one must be able to compare the state of a machine to an ideal good state in order to determine if the machine is either healthy or in pre foul state, but the real question is if this "model" must be a mathematical model or for example an Artificial Intelligence (AI) based model. My opinion is that an AI model can be used, but it is not definitely sure that this model is simpler to be defined and to be used.

So an affordable mechanical model of the particular machine of interest, a precise tuning of model parameters, a deep understanding of the machine rotor-dynamics, a dedicated set of sensors (accelerometers, etc.) with signal conditioning hardware, an historical data-base and finally the judgment of an expert is needed for classical rotor eccentricity fault assessment. However, a much simpler way to obtain %of both static and dynamic eccentricity in rotating-field electrical machines may be the signature analysis of internal currents in split-phase windings (i.e., the machine itself acts like a sensor). I have experimented this technique in the paper above, and it appears working well. The only machine data required is...the synchronous reactance!

Model-based approach is not really necessary...however, it is an interesting procedure for the identification of rotor parameters for certain applications. Also, some balancing techniques are model-based and do not require trial weights.

In my opinion, basically yes. You can always use 'off-the-shelf' algorithms available e.g. from monographs (a book by Bently and Hatch is worth recommending). Yet you have to remember that these had been mostly developed on the basis of models. Besides, it depends on what is meant by condition monitoring. If the fault is to be diagnosed quantitatively (i.e. not only detected and identified), a pretty complicated model will most probably be necessary.

No, not at all - an accelerometer (or two) and an FFT will do very nicely.

This is wicked interesting and the responses all seem to point towards the question: are you doing this to alert the user or are you using this to understand your machine.

The really interesting part to me is the ramification of what we often try to do in controls. Maybe if we can learn more if we looked at what is being done today in model reference controls.

Sorry I didn't really add anything but this is causing some old gears to be turning in my head.

Please don't mistake my real intentions. I'm not skilled with vibration monitoring (I'm an electrical engineer, not mechanical), and I really should like to understand what can be done in practice on the workshop-floor with a practical machine to grab some information on the rotor center displacement. This site permits to catch opinion from non-electrical engineers, and this is really interesting for me (but I think also for other el-engs. used to read about electrical machine condition monitoring mainly from other el.-eng.'s papers). So opinions from mechanical engineers are the most welcome.

If you can measure vibration (frequencies and modes, or time histories) in sufficient accuracy, you do not have to have a model. However, the real work environment may not allow such an accurate measurement to be done. In this case, a model-based approach is still needed.

My colleageus and I are developing non-model based approach for identification and control. One idea is to use measured receptances. Please see for example:

http://dx.doi.org/10.1016/j.jsv.2009.12.027

The work is being extended to rotor systems but this new work has not been completed yet.

Rotor dynamics is a well published area of technology and you would do well to start with basic condition monitoring texts such as Mitchell's Introduction to Machinery Analysis and Monitoring. Understanding the rotor modes is important but models are only as good as the data and are heavily dependent on well defined characteristics of the bearings. Models are only approximations. Proximity probes, accelerometers, pressure sensors and thermocouples together with oil analysis are typical in monitoring systems. Read Mitchell and decide where you want to start.

In fact, model-based approach is not necessary for fault diagnostics/condition monitoring. It is however helpful in diagnostics of some unconventional vibration problems. For example, when it is difficult to pinpoint the root cause of vibration, a rotor model will be needed to give more insight into the rotor’s dynamic behavior. Yet, the fidelity of the dynamic model remains a challenging requirement. For instance, we were resolving a VPF vibration problem in a multistage centrifugal pump, which all standard vibration suppression remedies fail to succeed. We had to resort to model-based diagnostic procedure, by which we concluded that the V-cut in the volute needs to be corrected (source is attributed to a design problem). The problem was resolved by successfully in the field, and vibration was reduced to 50% after correction.

ask this question from Hamid Khorrami, PhD student of Concordia on this area.

The conference proceedings of the 9th International Conference on Damage Assessment can be accessed at http://iopscience.iop.org/1742-6596/305/1 free of charge. There are papers on condition monitoring of rotating machines.

Thank you Huajiang Ouyang! Very useful indication.

I am just a beginner like you in the field of Vibration Control. Until now, what i realized is we cannot really prefer the traditional approach of performing the vibration analysis of a physical system. i.e., Deriving a mathematical model to represent the behavior of a physical system and then trying to match the real-time data with that of the model. To a certain extent, it kills the time.

For such kind of scenarios, the best possible approach is to perform an Online System Identification to obtain the Modal Damping and Modal Frequencies. For further details, you can contact me through my E-mail Id: [email protected]

A new high speed boiler feed pump had excessive vibration on its coupling. All at 1X, the engineer tried balancing it with washers under the connecting bolts. The problem was solved by removing the coupling and installing a similar design coupling, but another model, from another plant then under construction.

Much later, a university professor made up a dynamic mathematical model to show how the problem should have been solved and criticised those involved.. Elegant solution- yes- but he was not there at the machine when the project manager, the power plant manager, the commissioning engineer were all standing around asking ":how long will you be?"

An answer like "about 2 weeks while I set up a model" would NOT have been acceptable!

Thank you Ray Beebe for your comment. This is because real-world problems often are not matched by academic studies. Personally, I'm much more interested in application-oriented methods for practical machine condition monitoring. That is, practical and economical-efficient methods (although not the more precise) may be more valuable in real-world situations.

There is the analysis and then there is the synthesis. This is true with all problems. The synthesis has to do with figuring out how to solve the problem by using all of the existing methods and theories. Real world engineering as it were.

The analysis portion has to do with the why and giving us the ability to predict and providing a basis for synthesis. Both are necessary, just flip sides of the same coin.

Depends on what the speed range is. If it is the usual 1000 rpm and up, no need for model based technique. If it's not then you need a full dynamic model to study situations at low speed.

If you plan to use dynamic characteristics as a measure for fault detection, you will be using a model-based approach. You don't need to use a model based approach if you employ techniques such as artificial neural networks, that are not model dependent. I do not recommend, however, a black box ANN approach.