Gear condition monitoring metrics are very important for gearbox development and its time-based preventive maintenance. The indicators enable detection of gear anomalies, and help prevent catastrophic failure before the fault progresses. Condition monitoring systems deal with various types of input data, for instance vibration, acoustic emission, temperature, oil debris analysis. Systems based on vibration analysis, acoustic emission and oil debris are the most common types.
The following figure illustrates the workflow of identifying gear condition metrics and their further evaluation.
You can use the Diagnostic Feature Designer app and command-line functionality in the Predictive Maintenance Toolbox™ to do the following:
Import measured or simulated data from individual files, an ensemble file, or an ensemble datastore.
Derive new variables such as time-synchronous averaged (TSA), regular, residual and difference signals.
Generate gear condition metrics from your variables.
Rank your features so that you can determine numerically which ones are likely to best discriminate between nominal and faulty behavior.
Export your most effective features directly to the Classification Learner app for more insight into feature effectiveness and for algorithm training.
From your original data, derive the signals and extract the gear condition metrics in the following way:
Extract the time-synchronous averaged (TSA).
Derive the regular, residual and difference signals.
Compute gear condition monitoring metrics from the set of signals obtained from the previous step.
Gear condition metrics that can identify the precise location of faults, include the following:
Computed from TSA Signal
Root-Mean Square (RMS) — Indicates the general
condition of the gearbox in later stages of degradation.
RMS is sensitive to gearbox load and speed changes.
RMS is usually a good indicator of the overall
condition of gearboxes, but not a good indicator of incipient tooth failure.
It is also useful to detect unbalanced rotating elements.
RMS of a standard normal distribution is 1.
Kurtosis — Fourth order normalized moment of the signal
that indicates major peaks in the amplitude distribution. Kurtosis is a
measure of how outlier-prone a distribution is. The kurtosis of the standard
normal distribution is 3. Distributions that are more outlier-prone have
kurtosis values greater than 3; distributions that are less outlier-prone
have kurtosis values less than 3.
Kurtosis values are
higher for damaged gear trains due to sharp peaks in the amplitude
distribution of the signal.
Crest Factor (CF) — Ratio of signal peak value to
RMS value that indicates early signs of damage,
especially where vibration signals exhibit impulsive traits.
Computed from Difference Signal
FM4 — Describes how peaked or flat the difference
signal amplitude is.
FM4 is normalized by the square of
the variance, and detects faults isolated to only a finite number of teeth
in a gear mesh.
FM4 of a standard normal distribution is
M6A — Describes how peaked or flat the difference
signal amplitude is.
M6A is normalized by the cube of the
variance, and indicates surface damage on the rotating machine components.
M6A of a standard normal distribution is 15.
M8A — An improved version of the
M8A is normalized by the fourth power of the
M8A of a standard normal distribution is
Computed from a Mix of Signals
FM0 — Compares ratio of peak value of TSA signal to
energy of regular signal.
FM0 identifies major anomalies,
such as tooth breakage or heavy wear, in the meshing pattern of a
Energy Ratio (ER) — Ratio between energy of the
difference signal and the energy of the regular meshing component.
Energy Ratio indicates heavy wear, where multiple
teeth on the gear are damaged.
Computed from a Set of Residual Signals
NA4 — An improved version of the
NA4 indicates the onset of damage and
continues to react to the damage as it spreads and increases in
Feature selection techniques help you reduce large data sets by eliminating gear condition metrics that are irrelevant to the analysis you are trying to perform. In the context of condition monitoring, irrelevant features are those that do not separate healthy from faulty operation or help distinguish between different fault states. In other words, feature selection means identifying those gear metrics that are suitable to serve as condition indicators because they change in a detectable, reliable way as a gearbox performance degrades.
To perform a rigorous relative assessment, you can rank your features using specialized statistical methods. Each method scores and ranks features by ability to discriminate between or among data groups, such as between nominal and faulty behavior. The ranking results allow you to eliminate ineffective features and to evaluate the ranking effects of parameter adjustments when computing derived variables or features.
|Diagnostic Feature Designer app||Command-line|
Histograms allow you to perform an initial assessment of feature effectiveness. To perform a more rigorous relative assessment, you can rank your features, using the Rank Features option, by specialized statistical methods.
Use the Export option to export the selected metrics to the Classification Learner app in the Statistics and Machine Learning Toolbox™.
You can choose from the following feature selection functions:
Once you have defined your set of candidate gear condition metrics, you can export them to the Classification Learner app in the Statistics and Machine Learning Toolbox. Classification Learner trains models to classify data, by using automated methods to test different types of models with a feature set. In doing so, Classification Learner determines the best model and the most effective features. For predictive maintenance, the goal of using the Classification Learner is to select and train a model that discriminates between data from healthy and from faulty systems. You can incorporate this model into an algorithm for gear train fault detection and prediction.
Feature Designer |
tsadifference | Classification Learner