This article demonstrates an example of analyzing data with Simcenter Anovis to define a limit metric for identifying a fault.
In the example shown in the article, production data has been gathered. The 10th order vibration detects a fault that was not previously identified.
This article steps through setting up a limit for the 10th order to sort units. Although 10th order is used as an example in this article, any data type can be used in a similar fashion.
First, open the Anovis Signal Analyzer and use File > Open result file… to select a measurement from the desired dataset. This introduces the available curves, metrics etc. to Anovis.
Then, use the Configuration > Viewer connection... menu to open the Buffer Assignment dialog.
The color of the icons show whether the buffer contains data, and the “Hide empty buffers” filter makes the empty buffers disappear from the list.
Click to highlight the buffer of interest and select the Analysis button. If the existing viewers are no longer of interest, you can choose whether to keep or discard them in the prompt that follows.
The Analysis option creates a new type of viewer for comparing data from multiple .ame files. To populate the viewer with data, right-click in its plot area and select the Class editor… option.
This opens a window for defining which data to be present, as well as for separating the data into groups called “classes.” The class names can be customized using the Configuration > Configure class names… option of the main toolbar.
To select data to display, click the Add button and select the .ame files to add to the class. If you expect in advance for certain files to be from different classes, you can navigate between the class tabs to choose which class the data is added to. If you have a result list (.fil file), you can similarly add the contents using the Load List… button.
Once the data is loaded, you can perform some initial analysis by zooming in and looking for curves that seem to diverge from the pattern of the others. By holding the Alt key and clicking on one of the points of the curve, the line of the Class Editor for the file containing that curve will be highlighted, allowing identification of which curve is which. Once a curve of interest has been identified and selected in the Class Editor, the Move button can be used to move the curve to a different class.
After the curves have been separated into classes, the bottom plot can be used to observe trends such as the “difference” between each set’s own maximum or minimum, or such as the “gap” between the highest value in one class with the lowest of the other or vice versa. These statistics can be selected by right-clicking in the plot and selecting the Viewer settings… dialog.
The same dialog also allows the display to be changed from an overlay of lines to a set of statistical curves, such as showing the mean plus or minus the standard deviation by choosing the “Statistics” toggle button, and the Mean-curve and Std-curve options.
After finding a point of interest that distinguishes between the classes, a metric can be chosen to identify whether a curve exceeds the limit for a class, ideally balancing between rejecting as many true positives as possible while avoiding false positives.
Consider the peak at 10 order shown in this example; the blue Class 0 curves have a narrow spread around 110-111 dBm/s2, while the red Class 1 curves are all above 117 dBm/s2. If we split the difference between the highest blue and lowest red, 114 dBm/s2 might seem like a good separating line between the classes, but we will want to check that it is mathematically enough of a distinction to create a reliable metric from.
To evaluate our choice of limit, we might consider the Processing Capability Index (cPK), in which the difference between the desired limit and the mean is divided between three times the standard deviation. The equation for Processing Capability Index (cPK) is shown in Equation 1.
Equation 1: Formula for Processing Capability Index (cPK). A value of 1.667 or greater based on a standard normal distribution means that 1 part per million would be a false positive.
If a standard normal distribution is assumed, then a cPK of at least 1.667 is expected to have false positives for less than one part per million.
By positioning a cursor at our 10 order location, we can see in the legend that our Class 0 group has a mean of 110.96 and a standard deviation of 0.31. Using the cPK formula, our chosen upper limit of 114 would result in cPK = (114 – 110.96) / (3 * 0.31) = 3.2688, which is much higher than 1.667, indicating that false negatives (Class 0s appearing as Class 1s) will be extremely unlikely. Conversely, considering 114 as the lower limit of Class 1, which has a mean of 117.76 and standard deviation of 0.28, the cPK of (117.76 – 114) / (3 * 0.28) = 4.4762, which is again higher than 1.667 and indicates that false negatives (Class 1s appearing as Class 0) are extremely unlikely. Together, this indicates that a cutoff of 114 dBm/s2 at 10 order would give our dataset a very accurate metric.
Another approach would be to rearrange the cPK formula to find the boundary at which a desired level of accuracy is met; that is, an upper limit ≥ mean + 3 * st. dev. * cPK, while a lower limit ≤ mean – 3 * st. dev. * cPK. If 1ppm is acceptable, then solving for the limit using cPK = 1.667 would give us an upper bound for Class 0 of 110.96 + 3 * 0.31 * 1.667 = 112.5103, and a lower bound for Class 1 of 117.76 – 3 * 0.28 * 1.667 = 116.35972. Since the two limits do not overlap, we have a clear separation between our classes, and picking a value between the two (such as 114) would give us an accuracy that satisfies our false positive/negative criteria of 1ppm each.
To turn this limit into a metric in Simcenter Anovis, we can open the Anovis Signal Processing application, accessing its interface by selecting its icon in the system tray.
First, load the .ams measurement setup that was used to create the data that was analyzed, then use Measurement Setup > Save As to save the setup as a new .ams file – this step is important to avoid overwriting the existing .ame results when reprocessing the data.
Then, use the Parameters menu of the toolbar to select what data the metric will be applied to. For this example, our analysis was of the order curves, so we will select Features > Order Levels.
Then, select an existing item on the list, and double-click it to bring up an editing dialog.
We can give our new metric an identifiable name in the Name / Feature ID field. In this example, we are only interested in one value of one feature, so our Feature Mode is “only feature A.”
Under the Level Configuration A, we can use the Setup… button to select the x axis point we are evaluating. This opens a dialog, in which we can enter 10 order as the location to read.
Then, we set our acceptable range to have a maximum value of our chosen upper limit, which in this case is 114. We typically do not set the minimum to zero; this way, unusual data can be caught in case of an unexpected issue.
Once the settings have been chosen, click “Add” rather than “Change,” to avoid overwriting the entry this was cloned from.
To create .ame result files that include the new metric, we can reprocess the data. Again, our measurement setup should first be saved with a new file name so that we do not overwrite the original results.
Select Measurement > Offline signal replay… to open the Signal File IO window, then click New selection… and choose the .asd measurement files to reprocess.
Click on the Start button to begin the processing. You can set the File List Replay to “single step” if you would like to check the data after each file, or “increment automatically” to process all of the files in succession. Once all of the files have been processed, the bottom message will indicate the completed status.
Now, open the Anovis Signal Analyzer, and load a .ame file from the new results. Open the Configuration > Viewer connection… as before, but this time, select the new metric for analysis.
In the Class Editor, load in the curves from the new result dataset, but this time, load all the curves into Class 0.
If the data is not visible at first, right-click in the plot to access the Viewer Settings and check on the Y axis auto scale and level. This will display our tolerance cutoff, as well as a histogram to the side.
Now, return to the Class Editor and click on the “Sort” button, choosing a class (such as Class 1) as the destination for all curves that did not meet the metric limit.
The sorting will detect which curves did not meet the required upper limit, and will move those curves to the new class. In this example, the “Sort” button will automatically move all of the curves that exceeded 114 dBm/s2 at 10 order to Class 1.
The metric successfully worked on the recorded data and is now ready to be used in the production line.
