In Simcenter Testlab Neo 2506.0001 and higher, a process called “AnomalyDetection” is delivered with the software. The “AnomalyDetection” process can be used to quickly screen measured data for potential problems or unexpected phenomenon.
Some example anomalies are shown in Figure 1 below:
Figure 1: Some potential data anomalies include spikes (top, red) and drift (bottom, green).
Anomalies that can be screened include:
Intermittent wiring issues that create spikes or dropouts in data.
Transducers that reach their maximum output and flat line.
“Dead” channels that are not measuring anything.
Unusual events that occur during the operation of a product.
And more...
The anomaly check process can be very useful when acquiring a high number of channels for a long duration. Instead of having to manually check each channel, potential anomalies are flagged automatically and summarized in a table.
This article has the following sections, based on the data anomalies that are automatically flagged: 1. Introduction 1.1 Getting Started 1.2 Process Output 1.3 Time Data Acquisition 2. Hardware Checks 3. Spike Detection 4. Flat Line 5. Drift 6. Overrange/Underrange 7. Global Statistics: Dead/Outliers/Unipolar 8. Signal to Noise Ratio 9. Undersampled 10. Spectrum
1. Introduction
The “Anomaly Detection” process is run from the Process Designer (Figure 2) of Simcenter Testlab Neo:
Figure 2: Simcenter Testlab Neo Process Designer with the “AnomalyDetection” process opened.
Each anomaly check has a separate colored box. Some of the anomaly group detections include: Hardware Checks, Flat Line, Drift, Signal to Noise Ratio, and more….
The anomaly checks are explained in further detail in the upcoming sections of this article.
1.1 Getting Started
To run the "AnomalyDetection" process, under "File -> Add-ins" of Simcenter Testlab Neo make sure the following are loaded:
Signature Analysis (36 tokens) - Only needed for Spectrum check
If checking on a subset of the available anomaly checks, not all add-ins are required.
Select the “Open” icon in the Process Ribbon of Simcenter Testlab Neo: Process Designer. Load the “Anomaly Detection” process (available in versions 2506.0001 and higher) as shown in Figure 3.
Figure 3: The “Anomaly Detection” process is found in the Process Ribbon by clicking on the “Open” icon.
Each check (the colored boxes in the Anomaly Detection process) begins with a “Pass” method. The pass method can be used to select specific data channels to be checked for a given method (Figure 4).
Figure 4: The “Pass” method of the drift anomaly set to analyze only acceleration and force data.
Any individual anomaly check can be turned off so it is not performed as shown in Figure 5.
Figure 5: Any individual anomaly can be turned off by clicking on the color bar on the left side of the pass method.
The “AnomalyDetection” process is based on standard methods with parameters that can be adjusted by the user. This is because what might be considered a problem or anomaly in one data acquisition measurement may be actual data in another measurement.
For example, a brief spike might mean that a transducer has come loose when measuring broad band random vibration. However, if measuring a phenomenon where an occasional click sound (that looks like a spike) occurs in an actual product, it may appear to be an anomaly but is an actual event.
To inspect time data for anomalies, put the data in the input basket and press “Run”. Results are summarized in a table as explained in the next section.
1.2 Process Output
The output of the anomaly detection process has cells which contain either “1.11” or “8.88”:
“1.11” no data observed which matches anomaly criteria
“8.88” data observed which matches anomaly criteria
It is best to view the results with a pivot view like “DOF ID versus Function”. Click on the top of the indicators column to see a summary of the results as shown in Figure 6.
Figure 6: Use the “Indicators” column to view the results of the anomaly check. Values of "1.11" indicate no anomaly was found, while "8.88" indicates an anomaly was flagged.
In addition to the summary statistics, additional time traces are also calculated for some anomalies. These time traces have a value between 0 and 1 and mark where the anomaly occurs in time as shown in Figure 7.
Figure 7: Original signal (top, green) with anomalies. Calculated binary trace with values between 0 and 1 that marks where anomalies occurred.
In Simcenter Testlab Neo Time Data Acquisition the anomaly detection can be set up to run immediately after a measurement is finished as shown in Figure 8. The results can be viewed directly in the Measure worksheet.
Figure 8: The “Anomaly Detection” process can be set to run immediately after an acquisition is finished in the menu in lower right corner of the Measure tab of Simcenter Testlab Neo Time Data Acquisition.
If data is collected in Simcenter Testlab Signature Acquisition or imported from another data acquisition system, then the Simcenter Testlab Neo Process Designer can be used to check for anomalies.
“Hardware Checks” is one of the areas that are checked for anomalies. This check requires that the data was acquired using Simcenter SCADAS hardware. The process is shown in Figure 9 below.
Figure 9: Hardware Check in the "AmomalyDetection" process of Simcenter Testlab Neo.
When performing a measurement with Simcenter SCADAS hardware, phenomenon like overloads and other hardware-based issues are logged automatically (Figure 10).
Figure 10: Any hardware overloads are logged automatically and flagged when using Simcenter SCADAS hardware.
More about SCADAS hardware and overloads in these knowledge articles:
The “Spike Filter” method is used to flag potential spikes in the data (Figure 11 below):
Figure 11: Spike Detection in the "AmomalyDetection" process of Simcenter Testlab Neo.
Spikes may be due to dropouts or intermittent connections to the transducer. Some dropout spikes are shown in Figure 12 below:
Figure 12: Top (Red) – Original data with dropouts/spikes. Bottom (Green) - The difference trace with threshold times the standard deviation superimposed.
The potential presence of spikes Is determined in a multiple step process:
Difference between all consecutive data points in the recorded time history is calculated.
A new time trace of the differences is created.
The standard deviation of the difference trace is calculated.
A user defined factor (default is 25) is multiplied by the standard deviation to set as threshold for defining potential spikes.
A duration specified in number of data samples is used to flag spikes which should occur for a brief period of time (default sample length is 11 samples).
Check out these knowledge articles about various transducers:
The signals are also checked if “flat lines” occur in any part of the time signal (Figure 13).
Figure 13: Flat line check in Simcenter Testlab Neo.
By default, the “flat line” checks for any signals that do not change value for 5 seconds (default value can be adjusted by user) or more. This might indicate either an intermittently dead channel or that a transducer has maxed out temporarily (Figure 14).
Figure 14: String pot that maxes out at 63 millimeters would be flagged by the flat line anomaly check.
For example, if a string pot reached its limit of travel during a measurement over the specified time duration, it would be flagged.
Measurements can sometimes develop a drift or offset. In some cases, for example strain gauges measuring plastic deformation, this is not unusual. Other transducers like ICP accelerometers cannot measure an offset and should always be centered around zero.
The drift check of Simcenter Testlab Neo is shown in Figure 15 below.
Figure 15: Drift check in the “Anomaly Detection” process.
To determine if the signal has drift, the mean of the signal is calculated for a short period of time (0.125 seconds by default) at the beginning and end of the time history (Figure 16).
Figure 16: To determine if there is drift in a time recording (light green), the mean value (dark green line) of a small time segment (orange boxes) of at the beginning of the end of the time recording are compared.
If the mean has shifted from the beginning to the end of the measurement, there is drift in the signal. The difference in the mean values must differ by more than 10% (default value, can be adjusted by user) of the measurement range to be flagged as having drift.
Drift can be:
Expected: In some measurements, for example strain where plastic deformation of the test object could occur, drift might be expected.
Unexpected: For other measurements, for example an ICP/IEPE accelerometer, the signal should be centered around zero. An ICP/IEPE transducer is incapable of measuring any offset that may occur in real life. So if the channel is drifting, there must be a problem with the transducer or its grounding.
Any transducer that utilizes AC coupling, rather than DC coupling, should not contain any offset or drift.
These checks only work for data acquired by Simcenter Testlab using a Simcenter SCADAS. There is a field stored in the data called “Limit EU”. “EU” is short for Engineering unit, the measurement quantity (g, m/s2, microstrain, etc) that used for the measurement.
The Statistics check (Figure 17) can be used to determine if data is overranged, underranged, etc. These statistics are based on the entire time history.
Figure 17: Overrange, underange, and improper header anomaly checks based on Limit EU.
When measuring with Simcenter Testlab and SCADAS hardware, the limit EU can be viewed in the channel setup as shown in Figure 18.
Figure 18: “Limit EU” field in the Channels tab of Simcenter Testlab Neo Time Data Acquisition.
The Limit EU is the maximum value in engineering units that can be measured on a given channel on a SCADAS data acquisition system. It is the range (maximum voltage that can be measured) divided by the sensitivity value (the proportional voltage that the transducer outputs for each engineering unit).
The statistical checks performed based on the Limit EU include:
Overrange: Checks if any sample in the data uses more that 90% (default value) of the Limit EU range of the data channel.
Underrange: Checks if all samples in the data use less that 1% (default value) of the Limit EU range of the data channel.
Improper Header: The amplitude of any data sample in the time history exceeds the upper and/or lower EU Limit values.
The global statistics checks are applied based on statistics of the entire time duration (hence the word “Global”) of the measurement (Figure 19).
Figure 19: Global statistics check in Simcenter Testlab Neo.
The global statistical checks can be done on any data, whether it was acquired by a Simcenter SCADAS or not. The global statistics do not involve the “Limit EU” values that are available when data is collected by a Simcenter SCADAS as described in the previous section.
The global statistical calculations are as follows:
Dead: If the maximum of the signal is the same as minimum, this would indicate a dead channel.
Unipolar: Signal is either all positive or all negative values. Some channels, like accelerometers are usually both positive and negative. Other types of signals, like a speed channel, are usually all positive.
Outliers: Compares the RMS to the maximum amplitude of the signal. If the RMS is lower than 5% of the maximum amplitude. If this is the case, it means that most of the signal is very low compared to the extremes data points in the measurement.
This anomaly check will work with any data, not just data acquired by Simcenter SCADAS hardware. For example, this can be applied to RPC files imported from different data acquisition hardware.
7. Signal-to-Noise Ratio
The Signal-to-Noise Ratio (SNR) anomaly check compares the range of the first five seconds (default value) of the signal to the range of the entire signal (Figure 20):
Figure 20: Signal-to-Noise Ratio (SNR) anomaly check in Simcenter Testlab Neo.
If the range of the first five seconds is 20% greater than the range of the entire signal, it is flagged as an anomaly.
This is useful for measurements where initially there is no vibration activity before operating vibration measurements start. This is common practice for durability measurements where an initial “zero” period (and at the end) is intentionally measured as a reference condition (Figure 21).
Figure 21: Top (Green) – A suitable measurement for a signal-to-noise anomaly check that begins and ends at zero. Bottom (Blue) – A measurement with no zero condition at the beginning cannot be used for the “AnomalyDetection” process provided in Simcenter Testlab Neo.
For measurements with no initial vibration activity, the amplitude range of the beginning seconds should be well below the range of the entire measurement. If the measurement levels do not increase after the initial zero period, there is a good chance that the transducer is dead or wiring is broken.
In the case where there is no period of inactivity at the beginning of the measurement, the signal-to-noise ratio measurement check is not useful. If the beginning level of the measurement where there is no vibration activity is lower than the full measurement level, then there is a good chance the transducer or equipment is not functioning.
More information on durability measurements in the knowledge articles:
Another check is on how adequate the sampling rate of the measurement is compared to the frequency content of the data (Figure 22):
Figure 22: Undersampled anomaly check.
Vibration, strain, and sound data are continuous functions. The higher the sampling rate, the less the amplitude difference should be between two consecutive data points as illustrated in Figure 23.
Figure 23: Illustrative figure of under sampling – The same sine wave sampled at a lower rate (left figure) has a higher amplitude difference (Δa1) between consecutive samples than the amplitude difference between consecutive samples (Δa2) at a higher sample rate (right figure).
In the “Undersampled” anomaly check, the amplitude difference between two consecutive samples is compared against 25% of the full range of the channel. If the amplitude is lower than 25% (default value, can be changed) of the full range then the data is flagged as undersampled (value of “8.88” in summary table).
More about sampling rates in these knowledge articles:
The “Spectrum” anomaly check is performed on the frequency spectrum (Figure 24).
Figure 24: Spectrum anomaly check in Simcenter Testlab Neo.
The spectrum anomaly check can be used to determine if a channel is dead and not measuring properly.
If the ratio of the mean value and the maximum value of the frequency spectrum is lower than 20% (default value, can be user adjusted), then the channel is flagged as having an anomaly (Figure 25).
Figure 25: Green – Spectrum of channel with no functioning transducer, Blue – Vibration spectrum with functioning transducer.
The percentage can be adjusted in the “Spectrum Calculate” method in the formula area.
