In an era where AI has taken over, it's essential for engineers to leverage this technology if they wish to stay relevant in our fast-changing world. Join us as we delve into cutting-edge products from Hexagon, Odyssee A-eye, and Odyssee CAE, which puts AI and ML in the hands of any engineer. Drawing from a recent presentation at the South African Ballistics Organisation (SABO) symposium, we explore a real-life use case that demonstrates the power of Reduced Order Modelling (ROM) in evaluating alternative blast energy dissipation designs rapidly. Learn how ROM, unlike traditional DOE methods, allows for complex crash and blast models to be replicated in seconds, revolutionizing the way we approach design space exploration, optimization and robust design.
Regards
Managing Director
SIMTEQ Engineering Users' Conference
Join us at the SIMTEQ Users' Conference and experience the pinnacle of simulation applications! Learn how South African engineers have leveraged Hexagon's MSC Software tools to solve engineering problems more efficiently. Register now and join us at the forefront of simulation innovation!
When - 27 July 2023 Where - CSIR International Convention Centre
The bi-annual South African Ballistics Organisation (SABO) symposium was held at the CSIR International Convention Centre from 9 to 11 May 2023. We at SIMTEQ extend our congratulations to all the presenters for their insightful, illuminating and sometimes thought-provoking papers.
SIMTEQ’s own Paul Naudé presented how artificial intelligence (AI) and machine learning (ML) are used to develop blast protection structures, featuring Odyssee A-eye and Odyssee CAE from Hexagon (available to MSC One customers). Read more about this process in our newsletter's Product Of The Month section.
In his presentation titled Artillery in the Challenges of Future Warfare, Louis Du Plessis of Ballistics cited Immerse from his Essay on "Causes of War": "Conflict is comparable to a bonfire: the wood is unrelenting economic inequality, differences in ethnicities and religions are the gasoline, and the sparks needed to ignite the fire are the reckless actions of self-interested individuals."
In a powerful conclusion and call for peace, he cited George Washington and Desmond Tutu who said: “Don’t you think I have concurred my enemy if I managed to make him a friend?” - GW and “If you want peace, do not talk to your friends – talk to your enemy.” - DT.
AI has taken over and how to design a blast protection system using it.
In a matter of mere months, AI has taken over every conversation on the planet and suddenly everyone with access to the internet knows it can write articles, generate unique images, and even have lengthy conversations. But what does this mean for engineers, particularly those specializing in simulation and analysis? How can AI be practically utilized in the design of new products and systems?
In this month's feature, we shine the spotlight on two products from Hexagon: Odyssee A-eye and Odyssee CAE (Lunar) and we will be using a practical example as shared at the bi-annual South African Ballistics Organisation (SABO) symposium earlier this month to demonstrate a use case.
The fields of Artificial Intelligence (AI) and Machine Learning (ML) are vast and developing rapidly, spawning new definitions every day. In the context of simulation and analysis, however, there is one crucial definition you should know: Reduced Order Modelling (ROM).
In a survey conducted by Rescale, they reported that only 7% of their users only conduct final design analyses. 78% of respondents indicated that they use simulation and analysis on broad design space exploration and evaluating alternative designs.
This, however, becomes a problem in terms of computing time and costs since at the same time models are becoming more detailed and employ multi-physics. Even on clusters, simulation times are a limiting factor to properly evaluating alternative designs. This is where ROM comes into play.
If you ever heard of super-elements (Nastran) or flexible bodies (Adams), you already understand in principle what Reduced Order Modelling is. While super-elements and flexible bodies are mathematically precise equivalent models, ROM is not exact, although the aim is, of course, to have it as close to exact as possible to actual steady-state and/or dynamic responses.
Another important similarity between these technologies is that they can be used in the time domain. However, while super-elements and flexible bodies have their limitations when it comes to non-linear cases, ROM can even be used to replicate crash or blast models.
The analogy here is that a ROM can represent a very complex model, however, with far less computation effort. In fact, numerous case studies exist showing how even crash analyses can be replicated in seconds rather than hours or days.
This opens the door not only to broad design space exploration but to rapid design optimization and robust design.
A mind shift is required.
This is however where the similarities end. Unlike super-elements and flexible bodies which are exclusive to Finite Element Methods, Reduced Order Modelling is technology neutral. It can be used with any data from any software or analysis type, test or sensor data and it can even use CAD and images as input.
When evaluating different concepts, it is often not possible to define the differences with model parameters. However, with Odyssee A-eye, this is no longer the case.
Energy dissipation device development.
In the following example, several energy dissipation devices were to be evaluated using Dytran. The problem was defined as a single blast (explosion) event some distance from (below) an object with this energy-dissipating device in between, to reduce and delay the peak force transferred as much as possible.
Energy dissipation device between blast wave and object.
The premise of this demonstration was to show that these models can be analysed in detail using explicit FEM and although the dynamic event lasts no longer than 2 milliseconds, the detailed analyses take a considerable amount of time and cannot be neglected in a project timeline.
The need, therefore, arises to replace the detailed FE analyses with fast representative calculations with which many more designs can be evaluated in seconds, rather than minutes, hours, or days.
Starting Concepts.
The next step is to provide the created project with learning data to learn from. In this example, the 10 images representing each design were provided as input and the force transfer time histories files (as obtained from the blast analysis using Dytran) as well as the mass of each model were provided as output.
Concept designs to evaluate their blast force transfer characteristics
The first challenge that may appear to arise, is how to provide the designs as input to an AI model. Even with these basic concepts, the designs are too different from one another to be defined with a single parametric model and therefore, to make a connection between the design and results from each design, the only useful (comparable) data available to feed into the AI is the CAD files themselves or images of the designs.
Using images as input data.
Odyssee A-eye provides several options for evaluating images, from filters to batch correct or enhance images, to pixel counters and various image complexity indicators.
With such an analysis, the attributes derived from these images are then used as effective model inputs which can be linked to the results from each design’s analysis.
The magic lies in connecting the dots between the input and output. The Odyssee products simplify the process and make it accessible to any engineer. This frees the engineer to focus on what he already knows and excel in to focus on extracting more insight from existing work and improving on what he already does best.
From experiments to improved designs
Defining and populating the AI project
For the above example, the first step is to define the project in terms of what types of data will be provided for the input and output to learn from.
Create a project in Odyssee A-eye by defining the project input and output data types
Once the data is loaded into the dataset, the chosen image processing will automatically start, and the dataset appended with the extracted data.
Loading learning input and output data into the project
Insights from available data
Before predicting new results based on the provided data, valuable insight can be obtained by simply viewing the database in Odyssee CAE (Lunar).
Apart from visualising the provided input and output data, other valuable information such as regression coefficients can also be viewed at each time step for each input parameter as an indicator of its influence (and weight) on the differences in output between the different dataset values.
Scatter plots are also provided of the input vs output at the chosen point in the time series from which trends (if present) can be identified, providing the user ample tools to digest the available data before any predictions are made.
Viewing the available data and trends in the provided learning dataset
Evaluating different prediction algorithms
In contrast to what most people may think today about the autonomy and power of AI, they are still driven by different algorithms and there isn’t one algorithm that is superior to all others, for all scenarios. Different algorithms will perform differently on different datasets and for the user to learn and understand everything can be daunting and undesirable.
While understanding the pros and cons of each algorithm carries enormous value to understanding what is feasible and what is not, Odyssee CAE (Lunar) can evaluate all its own (and custom) algorithms and their parameter settings in an automated way. Depending on what the most important aspect of the output data is to the user (e.g. the R2 values, RMS, maximums etc.), it can use that as a measure to evaluate the accuracies of each method and provide a suggestion of the best algorithm and its parameter values to use.
Checking prediction accuracy
The most popular method is the Cross Validation method where each provided row of output is predicted in turn using the other provided rows of data and comparing this for all the available algorithms. This not only identifies the best algorithm for the most accurate predictions but also provides a reliable indicator of what the expected accuracy can be for new predictions.
Accuracy check and suggested algorithm settings
Once an acceptable prediction error is achieved, the user can comfortably continue knowing that enough data was provided for accurate predictions.
Predicting results
With the accuracy known, new predictions can be made by simply supplying new images of other design concepts.
New design concept image provided for prediction
Results predicted based on the provided image
Defining a refined design set to evaluate
Once the best concept is identified for the scenario, a refined approach can be taken where the model is parameterised and a new design of experiments (DOE) is conducted on only those changing parameters.
The aim is to then optimize the design further without further detailed analyses.
Defining a new DOE from the best design candidate
With the new design sets, new images are not required as input to the AI since all the variations can be defined with parameter values, in this case, various plate thicknesses throughout the structure.
Once the input range values and distribution patterns are set, Odyssee CAE (Lunar) can provide randomly scattered values for use in the next batch of Dytran runs.
The chosen sets of input parameters are then used to analyse the blast again using Dytran and report the transferred force time history, however, only for a small sample set.
New parametric design set and Dytran results
Dytran transferred force results with regression coefficients and input parameter trends
Using these results, a new ROM could be created and further design improvement be investigated.
The results can also be viewed in Odyssee CAE and the peak force values are plotted against the various input values to show trends and dependencies of the peak force transferred against the various design parameters.
New results predicted instantaneously for newly chosen parameter values
Since new results can be generated instantaneously, Odyssee CAE can be used to optimize, or target required results in a fraction of the time.
Optimization setup.
By repeating the process to find the best prediction algorithm for this new set of data, not only can new design results be obtained instantaneously, but it can also now be used to find an optimum design given certain design goals and constraints.
New optimum design identified and results predicted in seconds
When was the last time that you played with a magnet, and did you know you can make music with them?
Almost everyone has played with magnets and most people have figured out the north-south attract and repel forces, but how many of you have used arrays of magnets to visualise wave propagation through magnetic field interference between magnets?
Magnets are used everywhere around us, from the door seal on your refrigerator to the speaker in your phone or the electric motor in your washing machine. But when was the last time that you played with a magnet? This is basically the whole premise of the video we have for you this month as well as that of the Magnetic Games YouTube channel in general.
In the video we’ve featured here he makes a magnetic musical instrument (which is not a theremin) he then proceeds to show how multiple magnets interact with each other through their magnetic fields. First through a circular array of various magnets and then a square array. What is specifically interesting is to see how the motion of a single magnet is propagated through the system in a wave as each magnet’s magnetic field interacts with the next.
Click here to learn about What’s New in Adams 2023.1.
Hexagon Manufacturing Intelligence is pleased to announce Adams 2023.1. This release provides several new features and improvements including:
Ride height adjustment Improvements support runtime ride height adjustment capability
Car Nonlinear Bushing in AView Provide Adams View users the ability to create/modify bushings with the same physics as the ACar nonlinear bushing (ac_bush)
Emag AT The EMag toolkit supports the integration of Maxwell motor models
Auto Save for AView and Acar Set frequency at which backups are automatically made of the current model
Name: Going the extra mile: Multiphysics and co-simulations with Hexagon
Date and Time: 4 October 2022 10:00-11:00
Name: Get CFD results in real-time and leverage it to systems of system analysis by using Machine Learning Technics
Date and Time: 11 October 2022 10:00-11:00
Name: Using AR to increase immersive knowledge
Date and Time: 18 October 2022 10:00-11:00
MSC Apex Tip
Define one, delete them all
When preparing a model for FE analysis, cleaning small features are fun in MSC Apex and it’s easy to forget that you don’t have to delete all of them, one by one.
Define a feature and delete all similar ones in one step
Use the custom feature tool to clean up repetitive patterns on surfaces.
The process:
Select connected edges on a surface to define the custom feature,
Select the surfaces where the features should be found, and
Delete them!
It works on both internal and external edges of any shape, open or closed:
Custom features can be internal or external, open or closed
