AnsweredAssumed Answered

Mates fail in motion

Question asked by John Sutherland on Jan 25, 2011
Latest reply on Feb 24, 2011 by John Sutherland

Like • Show 0 Likes0
Comment • 10

I have not seen this before, but it may have occurred without being captured so dramatically on video.

Frame 15 shows the mechanism as it should be; frame 16, half a second later, shows all hell breaking loose. 4 coincident mates and 2 hinge mates have let go, and 3 bodies are tumbling through space.

In between the two frames, the integration halted, but the timeline clock kept ticking, the video kept rolling although the image was frozen, and the frames were rendered by PW. No messages were displayed and the logs give me no clue.

This is not supposed to happen.

Does it ring any bells for you?

Attachments

AD61H-0015.png43.0 KB
AD61H-0016.png48.0 KB

No one else has this question

Outcomes

10 Replies

Shrikant Savant Jan 31, 2011 3:39 PM
Mark Correct
Correct Answer
Hi John,

Without looking at the model itself, it is hard to diagnose the problems you are having. But here are some things to check:
- By accident do you have any keys in the animation panel at the time corresponding to frame 16. This can happen if you edited any simulation elements (motors, forces etc.) and changed their properties without moving the timebar to t=0.
- Sometimes if you run the simulation for a long time and if your integration tolerance is not tight enough, integration errors keep on accumulating until the constraints break. What is the time associated with these frames? If it is more than around 5-6 sec, try tightening the integrator tolerance. This will obviously slow your simulation, but that is the standard trade-off - accuracy vs. speed. You can turn off the "Animate during simulation" option to speed up the computation.
- It is also possible that there is some force that is increasing in magnitude in a big way at that time that may be tearing the parts apart. Plotting relevant joint forces/torques or external forces/torques will reveal this.
Hope this helps,
Thanks,
Shrikant
Like • Show 0 Likes0
Actions
- John Sutherland @ Shrikant Savant on Feb 1, 2011 3:02 AM
  Mark Correct
  Correct Answer
  1)     The integration terminates at anytime, independently of keys.
  
  2)     Integration settings are at their limits (SW limits the values I can enter) for accuracy. Do you have any more specific tips?
  
  3)     Virtual models have infinite stiffness and strength; they cannot break even under irresistible forces. The integration terminates at any time, not just when forces are higher.
  Like • Show 0 Likes0
  Actions
  - Shrikant Savant @ John Sutherland on Feb 1, 2011 10:13 AM
    Mark Correct
    Correct Answer
    1) What simulation elements do you have? Which elements have keys and which don't? What do the keys do - suppress, unsuppress, on, off, change values? Have you tried isolating the offending simulation element(s) (mate, motor, force, spring, damper, 3d-contact etc), which when suppressed the simulation works and when unsuppressed it fails?
    2) The highest accuracy setting should be sufficient (1.0e-10). Beyond that you start getting into machine precision range. If the model still fails at the highest accuracy setting then it is not a "poor accuracy" problem. It could be the other way too. The accuracy setting maybe too tight and the integrator is unable to converge within such a tight tolerance. You can try loosening the accuracy (i.e. increase the tolerance - move the slider a few notches to the left). If there are a lot of intermittent contacts, you can switch the integrator to WSTIFF. Also try SI2-GSTIFF which has better velocity and acceleration error control. It all depends upon the problem.
    3) Parts in the model are treated as rigid and do not flex, but I was referring to joints. Do any of the mates have flexibility i.e. are they defined as bushings (Analysis tab of the mate definition dialog)? Do you have the "Replace redundant mates with bushings" option checked? In either of these cases, some rigid joints will be replaced with stiff bushings - which have large but finite stiffness. It is possible that other forces in the model can increase to such a level that they may break the bushings and the parts constrained by that mate will come apart. Large forces also cause large accelerations and hence large velocities and large displacements when integrated over a long period of time. The integrator may not be able to satisfy the joint constraints and integrate large forces at the same time - and will eventually fail.
    
    It all depends upon the model. Without the exact details, it is difficult to troubleshoot.
    Hope this helps,
    -Shrikant
    Like • Show 0 Likes0
    Actions
    - John Sutherland @ Shrikant Savant on Feb 1, 2011 5:30 PM
      Mark Correct
      Correct Answer
      Simulation elements are gravity, contact, contact friction, coincident (ball joint) mates, hinge mates and initial velocity.
      
      There are no bushings and no redundancies.
      
      Keys exist only for changing cameras.
      
      Suppressing any mate causes collapse of the whole mechanism.
      
      Reducing gravity makes life easier, but there are no customers for this machine on the moon.
      
      Suppressing contact friction makes life easier, but then the wheels have no adhesion and do not rotate and there are no customers for downhill only trains.
      
      Reducing initial velocity makes life easier, but there are no customers for slow transport.
      
      Reducing accuracy causes the machine to penetrate its contact supports and plummet to the centre of the earth.
      
      Contact forces cannot exceed weight. Ultimately the train will relieve the load by jumping the tracks.
      
      The measurements of contact force always display noise. I found that I had to compromise between damping force perturbations and minimising noise. Measurements of other displacement do not contain noise.
      
      I suspect the integrator takes one frame at a time and resolves all issues before moving to the next frame. It is hard to imagine that anything accumulates or carries forward. The integration is not constrained to calculating a result in real time; it takes all the time it needs.
      
      Logical explanations do not address the random nature of the failure.
      
      What would you look for when troubleshooting? Would the Rx logs help?
      Like • Show 0 Likes0
      Actions
      - Shrikant Savant @ John Sutherland on Feb 3, 2011 4:08 PM
        Mark Correct
        Correct Answer
        The main objective of the exercise to suppress one simulation element at a time like gravity, mates etc. is not to find customers on the moon, or customers for downhill only trains or customers for slow transport. Rather it is to help you troubleshoot a failing simulation so that you can identify where the problem lies.
        
        Any book on integrators can tell you how integration errors accumulate over time and how it can make the integrator fail.
        
        Nonetheless, based on your description it appears to me that the problem lies in the contact elements between wheels and track. The main purpose of the contact element is to model collision between bodies that happens intermittently. When bodies are always in contact, as is the case in your model where the wheels are always in contact with the track, using 3d contact elements is not the best way to model them. Contact computation engine adds a significant overhead to the solver and because of the inherent "numerically stiff" nature of this kind of a problem, 3d contacts need to be used sparingly. The noisy contact force is an indication of this problem. You could replace the 3d contacts between wheels and track by appropriate rack-pinion mates whose ratio is decided by the wheel radius. To model friction at the contact, you could add external torque on the wheel at the wheel axle in the proper direction or just define friction in the axle joint mate. Similarly, you can define damping forces and torques, if needed. The overall net effect will be very similar.
        
        Hope this helps,
        -Shrikant
        Like • Show 0 Likes0
        Actions
        John Sutherland @ Shrikant Savant on Feb 5, 2011 4:29 AM
        Mark Correct
        Correct Answer
        You have convinced me that integration is failing randomly because of an accumulation of something in Contact.
        
        Yes I can successfuly simulate a different model, but what is the point of simulating something I cannot sell? I am not an academic; I am in business.
        
        Rails provide support and guidance to the wheels. Gear mates provide neither. Trains that employ rack and pinion for adhesion on steep inclines still have wheels and rails. I have been playing with gears and found that gear mates alone do not hold the gear wheels in place. They need additional mates.
        
        Rack & pinion mates do not provide a very similar overall net effect to Contact.
        
        Wheel/rail friction is a very different matter to wheel/axle friction.
        
        I looks like SWP cannot do the job, even though the menus and dialogs invite me to attempt it.
        Like • Show 0 Likes0
        Actions
        Shrikant Savant @ John Sutherland on Feb 7, 2011 6:02 PM
        Mark Correct
        Correct Answer
        Rack and pinion mate helps you model the "rolling without slipping" effect. Mathematically, it just adds one scalar constraint. Rack and pinion sub-system has only one relative degree of freedom. But there are two bodies involved, which together have 12 coordinates. The rack-pinion mate will add one scalar constraint and you need additional mates to constrain the other 10 coordinates so that you have a 1 dof system. These additional mates will hold the rack body and the pinion body in place.
        
        No model can capture the reality exactly. The reality is far too complex - there are temperature effects, there are material damping effects, non-linearity effects, plasticity effects, humidity effects and so on. Approximations are inevitable and must be understood and taken into account appropriately. The key question an engineer has to answer is "Can my model explain the behavior I am interested in?". For this, the model has to be as simple as possible but no simpler.
        
        The contact reaction forces like the normal force and the friction force can be recovered from the joint axle forces and torques fairly easily, knowing the geometric dimensions. Whether that is sufficient or not will depend upon your application and what you are trying to get out of the modeling exercise.
        
        The task of simplifying the reality into a mathematical model such that it correctly simulates the desired behavior is the most important task in any modeling exercise. It requires engineering judgment and it cannot be automated. Switching software products does not obviate the need for doing this.
        
        Good luck,
        -Shrikant
        Like • Show 0 Likes0
        Actions
        John Sutherland @ Shrikant Savant on Feb 21, 2011 10:34 PM
        Mark Correct
        Correct Answer
        I tried two tactics for getting a better handle on what is happening.
        
        1) Zooming in (+ arrow) on the timeline enables displaying fine gradations of motion by dragging the timeline. This is much more instructive than video capture of intermittent frames. It showed that the train separated from the rails in a smooth manner. It certainly fell vertically through space, and probably continued horizontally tangent to the curved track. Any suggestion of tumbling was caused by the camera rotation not being locked, and SW failure to render the solid rails after a certain point in time. Mousing over empty screen caused the rails to display in orange colour. When I could not see them I assumed that the train was a great distance from the rails, but it was actually close by.
        
        2) Exporting the contact force results to Excel and eyeballing the sheet of data revealed that all contact forces collapsed to virtually zero (nano Newtons) in the space of one microsecond. Before the collapse the forces were steady and not sufficient to derail the train; and unlikely to overpower the 3D Contact resistance, especially as the rear bogie "derailed" before it reached the change in track radius that seemingly sent the front bogie off the track.
        
        When I commanded to save the assembly after a "derailment", the What's Wrong dialog advised that the stock/parent of the two fixed rail parts had rebuild errors, and they flowed on to the two rail parts and to measurements of contact force on the rails. Now the stock part is happily residing on my hard drive in Read Only mode, and loaded in memory concurrently with the assembly. The only way it could develop a rebuild error is if SW corrupts its own memory. No other apps are running.
        
        I don't know whether the memory corruption coincides with the collapse of measured contact forces, but it seems likely. Would it show in a log anywhere?
        
        I halved the initial linear velocity of the train with a view to reducing the radial acceleration in the curves. However the train "derailed" very early before reaching a tight curve in the track.
        
        I checked the BIOS error log again and there were no entries.
        
        So does SW randomly corrupt its own memory, or what?
        Attachments
        WhatsWrong.png209.8 KB
        Like • Show 0 Likes0
        Actions
        Shrikant Savant @ John Sutherland on Feb 22, 2011 4:19 PM
        Mark Correct
        Correct Answer
        Do your wheels have flanges on the inside? If you are modeling using solid body contact then these flanges will prevent the cart from flying off. If the carts are flying off the track despite the flanges, then you need to investigate the cause of that. In that case, try reducing the maximum integrator step size.
        
        The rebuild errors are due to problems in satisfying all the mates. If the parts are flying off, it is unlikely that all the mates will be satisfied and rebuild erros are to be expected. You need to fix the problem that causes the parts to fly off first. It has nothing do with memory corruption.
        
        Hope this helps,
        -Shrikant
        Like • Show 0 Likes0
        Actions
        John Sutherland @ Shrikant Savant on Feb 24, 2011 4:29 AM
        Mark Correct
        Correct Answer
        "try reducing the maximum integrator step size."
        
        I halved the initial linear velocity to cut the radial acceleration in the curves, and reduced maximum (and initial) integrator step size by a factor of 100.
        
        From the Windows Performance Monitor I inferred that the Contact force calculation collapsed before the train got out of the straight. I was not willing to wait a week for the timebar to reach the end of the timeline.
        
        I think it is time to abandon this task and move on to something else. It has been interesting but financially unrewarding.
        
        It seems I will have to build and test a physical prototype. I suppose SW can produce drawings of the parts, but I did not need to pay the extra cost of Premium to do that.
        Like • Show 0 Likes0
        Actions

Mates fail in motion

Attachments

Outcomes

Attachments

Related Content

Recommended Content