The commodo way of Detecting Anomalies in Simulation Data

As an expert, checking your model for plausibility is no witchcraft. But it is time consuming to do so on every simulation run. Further, huge and complex models may need expertise from multiple persons or departments and also not every engineer might have insight to all details.

The Expectatoris a software suite for the automated plausibilisation and the detection of errors in numeric data. Normally there is no anomaly or incident same than another one. Therefore we do not believe in an overall one-for-all algorithm for anomaly detection. Instead the Expectator follows the philosophy to define numerous simple and pragmatic criteria, so-called Expectations.

Expectator and commodo offer a variety of methods to check your data for anomalies, but how to start? For this, we offer common solver independent critera for standard finite element modelling issues.

Criteria bases assessment

The base methods are summarized here. In particular criteria may be

Contact assessment
- Contact penetrations can be checked initially or over time. While initial penetrations may be known from the preprocessor and might be acceptable, penetrations over time may come from missing contact definitions. Their influence may come significant but undetected depending on the model complexity.
- Contacts with small overlap may lead to unrobust results due to small parameter variations.
Decreasing mesh quality and distortion over time are of major interest in the vicinity of cracks or at highly loaded areas.
Event detection, like unexpected Intrusion or Buckling initiation
Stress evaluation w.r.t. to a stress-strain-curve, gradient assessment or outlier detection.
Entity Tracking, e.g. the dummy head must not leave a certain trajectory
Distance checks between e.g. High Voltage Contacts
Check if your simulation results are within a already occured pattern or totally new
Find "flying" parts which are not connected to the overall structure by any contact or joint.

Single curves or values are easy and fast to check, but things may get time consuming when multiple data must be combined. Further possible expectations are

you might be aware of some modelling issues when some special material has a specific stress distribution near to a spotweld, or
your spotweld models are not valid in special combined loads, e.g. peeling with a certain amount of torsion, or
the crack prediction is only valid when some mesh quality is present on some special steel, or
you may have a surrogate model from a component. An anomaly might be detected to check deviations from that.

Your model is your expertise, but we can support to formalize your criteria and prove your data in an objective and automatic manner. This procedure does not substitute the expert, but it applies the expert’s knowledge to every dataset.

Integration and Reporting

Expectator can be used in two ways

automatic evaluation as a standard postprocessing step within your SDM.
Interactive checks on your workstation

Result presentation also depends on your needs

Integration of results in the post processor, see e.g. our interfaces for interactive analysis
Standard reports for documentation during a release process (pdf, html, etc.)