Forums Anyscript.org  

Go Back   Forums Anyscript.org > Professional Forum > AnyBody Managed Model Repository

Notices

Reply
 
Thread Tools Display Modes
  #1  
Old 01-30-2018, 01:14 PM
dave1111 dave1111 is offline
Junior Member
 
Join Date: Mar 2016
Posts: 16
Default Outputs of TLEM2 and TLEM1

Hi

Since the update to AMS 7.1 and AMMR 2.0, I tried to update my models. However, I found a substantial difference in the outputs and I am not sure why is it happening. Specifically, I am interested in the outputs of the leg, which has been completely revamped with the TLEM2.1 model in AMMR 2.0.

Therefore, to understand the differences due to the updates in AMS and AMMR, I compared the outputs of the Human Standing template from AMS 7.0/AMMR 1.6.6 and AMS 7.1/AMMR 2.0. I have not made any changes to the template except for changing the body parameters for the leg and muscle models.

The comparisons that I have done are:
  • AMS v7.0 AMMR 1.6.6 TLEM 1
  • AMS v7.1 AMMR 2.0 TLEM 1
  • AMS v7.1 AMMR 2.0 TLEM 2

For all the 3 models I also checked the outputs with the Simple Muscle model and the Hill Muscle model. The outputs that I have compared are Muscle activity envelopes, Joint Moments and Joint Reaction Forces, with a focus on the leg model. Please find the results in the attached file.

As you could see, the joint moments and joint reaction forces at the Knee and Hip are quite different in the TLEM1 and TLEM2 models. The change to TLEM2 also affects the forces at L4-L5. I also noticed that this behavior is indifferent to the use of Simple or Hill muscle models.

The questions are:
Why is there such a notable difference between TLEM1 and TLEM2 and which of these is a more accurate/reliable model?
The outputs from which of these 6 combinations should I trust (or rather 3 combinations. I suppose it's reasonable to say that the Hill muscle model is more accurate than the Simple muscle model)?

Regards
-Dave
Attached Files
File Type: pdf StandingHuman_ComparisonOutput_V7.0-7.1.pdf (123.5 KB, 23 views)
Reply With Quote
  #2  
Old 02-09-2018, 03:58 PM
melund3 melund3 is offline
Junior Member
 
Join Date: Jul 2017
Posts: 6
Default

Hi Dave

Sorry. It seems your post was completely overlooked. Let me see if I can explain some of the differences.

First, there shouldn't be any differences by going from AMS 7.0 to 7.1. All the differences you see are likely from the models.

It is not surprising that the TLEM1 and TLEM2 produce different results. They are really two different models. Which one is the best? I would say the TLEM2 model. The TLEM2 model is built on the most consistent dataset, and I think there were a lot of measurements error in the original TLEM1 cadaver study. On the other hand, the TLEM2 implementation is relatively new, so there could be some implementation bugs which we haven't thought about.

You also point to differences between TLEM1.1 in AMMR1.6 and TLEM1.2 in AMMR2.0. Some of those difference are the accumulations of small bug fixes to TLEM1 model. But probably the biggest difference that you observe is due to the default scaling, which changed from AMMR 1.6 to AMMR 2. I think the model had the original cadaver size in AMMR1.6. In AMMR2 all models are scaled to the default size (50 percentile male), unless anything else is specified. I think that accounts for the biggest difference you see between the two TLEM 1 models.

Finally, I wouldn't automatically assume the 3-element (hill) muscle model to be the most accurate one in all situations. Using the hill muscle model has the potential of capturing the variation in muscle strength as a function of muscle length. This relies on the calibration of the tendon length so the muscle fibers achieve their maximum strength in certain postures. But I have never seen really good validations where the muscle fibers are strongest. And if you input is wrong, then a more complicated model is not always better. Hill-type muscle models are complicated. I can recommend you read this review on different approaches for setting their parameters:

http://journals.sagepub.com/doi/10.1...54411916659894

The paper mentions a calibration procedure by Winby et al, which maintains the force-length relationship across the range of motion regardless how the models are scaled. This, two-parameter calibration, was actually implemented in the AMMR2 model. It can be enabled by setting the `BM_CALIBRATION_TYPE` parameter. See https://anyscript.org/ammr-doc/bm_co...libration-type

The question still remains how to set parameters, and what part of the force-length relationship curve the models work on. We still don't know that, and that is why these new calibration procedures are still fairly hidden.

Please write again if we should help you dig into specific differences you find between the models.

Regards
Morten
Reply With Quote
  #3  
Old 02-14-2018, 05:47 PM
dave1111 dave1111 is offline
Junior Member
 
Join Date: Mar 2016
Posts: 16
Default

Hi Morten

Thank you very much for the detailed reply (and don't worry about the delay - it seems I am much worse).

I was quite certain the differences are due to the AMMR versions and the TLEM versions rather than the AMS version. Naturally, one expects things to improve with updates. I just found the differences too much to not raise a question here on the forum. One of the two models has to be quite erroneous (rather, the right word should be simplified) due to the assumptions that have been made in the process. As you say, I would stick with the TLEM 2 model.

You were right about the scaling. Somehow, I missed that detail. Now, I re-ran all the simulations scaling the manikin to the default AnyManUniform file (1.8m, 75kg, and 20.7% fat) using the Length mass fat scaling law. I am not sure if this would be the right approach to eliminate the differences due to scaling (perhaps because the scaling laws have changed?), but I didn't know any better. Anyway, the results are closer than before. I attach the scaled results.

On a side note, I appreciate how the Anthro data is easily found in AMMR 2.0

Your comments on the complexity-accuracy of the Hill muscle model are interesting. Thank you for the reference. I read the article but found it difficult to follow through. Unfortunately, it's neither my background nor the scope of my work to get into so much detail. Also, you read my mind about the calibration type. I had noticed the new 2 Parameter Type calibration in the options but didn't find sufficient details in the tutorials/AMMR manual. I had thought of asking about it but didn't because it's not a topic that I understand in depth in the first place.

However, I would ask you to elaborate a little bit more on what you wrote about the 3 element model not being the most accurate in all the situations. What situations are these and which model would you advise in these - the Simple muscle model completely ignores the muscle strength and length relationship, doesn't it? I am working in the field of ergonomics at the workplace. Specifically, I am using AMS for the biomechanical analysis of the use of exoskeletons. Currently, I am modeling static simulations where I analyze the advantages of the use of exoskeleton in a static posture. In the future, I hope I would be modeling dynamic activities as well. Usually, I scale the manikin to the subject using the Length Mass fat law and the AnyManUniform file. I do not get into more subject-specific details. In these simulations (both static as well as dynamic), which muscle model would you recommend?

Thank you for your support.

Regards
Dave
Attached Files
File Type: pdf Scaled_StandingHuman_ComparisonOutput_V7-7.1.pdf (124.5 KB, 8 views)
Reply With Quote
  #4  
Old 02-19-2018, 09:29 AM
melund3 melund3 is offline
Junior Member
 
Join Date: Jul 2017
Posts: 6
Default

Hi Divyaksh

I will always recommend that you start with the simple muscle models. But you are of course right. You do need to consider the 3-element muscle model when you have very dynamic movements.

Unfortunately, the process of calibrating the muscle models is not an area where you should just close your eyes and press a button. It can potentially give you the wrong strength. Most of the research projects in which the different body parts were developed haven't focused on that part of the model validation.

As I wrote, we have recently begun to look at other ways calibrating the output from the 3-element muscle models. That was the 2 parameter calibration I mentioned. I can't really explain the whole concept in this post. But I would be happy to take a small web-meeting with you, where I could explain the details and how it allows you control the output of the 3-element muscle model more intuitively. It is a work in progress, so I would also like to invite you to join our AMMR beta program. That will give you access to the next AMMR repository before it is released. If you are interested, just send me an email (I think you have it).

/Morten
Reply With Quote
  #5  
Old 02-21-2018, 11:32 AM
dave1111 dave1111 is offline
Junior Member
 
Join Date: Mar 2016
Posts: 16
Default

Hi Morten

Thank you for the reply. I have just sent you an email.

Divyaksh
Reply With Quote
Reply

Thread Tools
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off

Forum Jump


All times are GMT +2. The time now is 03:45 AM.