Normalisation rgb reh c

f

Answer 1

[Bora Eryilmaz]

There is a set of commands in the Signal Processing Toolbox and the Predictive Maintenance Toolbox for handling time-domain and frequency-domain data associated with a rotation rate so that the data can be "normalized" with respect to a rotation speed.

For time-domain data, see the tsa, tsadifference, and tachorpm commands. This demo can be helpful: https://www.mathworks.com/help/predmaint/ug/isolate-a-shaft-fault-using-diagnostic-feature-designer.html

For frequency-domain data, see the orderspectrum and ordertrack commands. This demo can also be helpful: https://www.mathworks.com/help/signal/ug/order-analysis-of-a-vibration-signal.html

For locating important frequencies to identify faults in the rotating systems, you can also check this example:

https://www.mathworks.com/help/predmaint/ug/analyze-gear-train-data-and-extract-spectral-metrics-using-live-editor-tasks.html

[Bora Eryilmaz]

For healthy data, you can probably use a moving average of the healthy signal values over a fixed window length and compare it to a fault threshold, where the threshold is either a generic value (say, 150% of the normal mean for that pump) or has to be specified per pump (based on pump size, etc.). So, instead of standardizing the amplitudes, you can look into standardizing the change in amplitude and compare that to the threshold, where the threshold could be a function of pump size, power, etc.

Hope this helps.

[Bora Eryilmaz]

For a given pump model, if the fault threshold really depends on the usage history of a particular pump (of that model), you would still need some indicators (also called condition variable or operating point) to differentiate the pumps with respect to their individual use cases.

Could you create some condition variables as differentiators of those use cases, such as:

• Continuously-used vs. intermittently-used pump
• High- vs low-temperature fluid
• Corrosive vs non-corrosive fluid
• Soft vs hard foundation at installation site