Fluid Initial Velocity 'Source' incorrect when enabling Inflow #80833

New Issue

Rich Sedman · 2020-09-16T12:19:40+02:00

Rich Sedman commented

2020-09-16 12:19:40 +02:00

System Information
Operating system: Windows-10-10.0.19041-SP0 64 Bits
Graphics card: Intel(R) HD Graphics 5500 Intel 4.4.0 - Build 20.19.15.4835

Blender Version
Broken: version: 2.91.0 Alpha, branch: master, commit date: 2020-09-11 09:14, hash: 1b6dd42803
Worked: N/A

Short description of error
When enabing the 'Use Flow' of an Inflow Liquid mesh, the initial velocity of the fluid is incorrectly calculated.

Exact steps for others to reproduce the error
Use Quick Liquid to create the domain around the default cube. Set the cube to 'Inflow'. Enable Initial Velocity with Source set to 1.0 (the default). Animate the cube moving around the domain. Keyframe Use Flow to start with the inflow disabled. Enable the inflow while the inflow mesh is in motion. The initial frame of fluid has the wrong starting velocity (typically zero for the first frame). If the Use Flow is disabled and re-enabled later in the animation the next frame of fluid seems to have a velocity determined by the difference in position between the current location and that where the inflow was last 'switched off'.

Blend file provided containing an animated inflow fluid inflow switching on initial velocity error.blend
The inflow is switched on at various points - first at 90 degrees into the animation where the fluid is initially stationary (but should be travelling to the left), then at 270 degrees where the fluid is travelling allong the y-axis (but should be travelling parallel to the x-axis as that's the direction of travel of the mesh - it would appear to be following a line between its current position and that where it was last 'switched off') and finally at the 'start point' of the animation - where the fluid should travel parallel to the y-axis - but actually travels diagonally.

**System Information** Operating system: Windows-10-10.0.19041-SP0 64 Bits Graphics card: Intel(R) HD Graphics 5500 Intel 4.4.0 - Build 20.19.15.4835 **Blender Version** Broken: version: 2.91.0 Alpha, branch: master, commit date: 2020-09-11 09:14, hash: `1b6dd42803` Worked: N/A **Short description of error** When enabing the 'Use Flow' of an Inflow Liquid mesh, the initial velocity of the fluid is incorrectly calculated. **Exact steps for others to reproduce the error** Use Quick Liquid to create the domain around the default cube. Set the cube to 'Inflow'. Enable Initial Velocity with Source set to 1.0 (the default). Animate the cube moving around the domain. Keyframe Use Flow to start with the inflow disabled. Enable the inflow while the inflow mesh is in motion. The initial frame of fluid has the wrong starting velocity (typically zero for the first frame). If the Use Flow is disabled and re-enabled later in the animation the next frame of fluid seems to have a velocity determined by the difference in position between the current location and that where the inflow was last 'switched off'. Blend file provided containing an animated inflow [fluid inflow switching on initial velocity error.blend](https://archive.blender.org/developer/F8888651/fluid_inflow_switching_on_initial_velocity_error.blend) The inflow is switched on at various points - first at 90 degrees into the animation where the fluid is initially stationary (but should be travelling to the left), then at 270 degrees where the fluid is travelling allong the y-axis (but should be travelling parallel to the x-axis as that's the direction of travel of the mesh - it would appear to be following a line between its current position and that where it was last 'switched off') and finally at the 'start point' of the animation - where the fluid should travel parallel to the y-axis - but actually travels diagonally. ![fluid initial velocity.gif](https://archive.blender.org/developer/F8888658/fluid_initial_velocity.gif)

Rich Sedman commented

2020-09-16 12:19:40 +02:00

Added subscriber: @BaldingWizard

Richard Antalik commented

2020-09-17 18:29:34 +02:00

Added subscriber: @iss

Richard Antalik commented

2020-09-17 18:29:34 +02:00

Changed status from 'Needs Triage' to: 'Confirmed'

Richard Antalik commented

2020-09-17 18:29:34 +02:00

I can confirm inconsistency here.

I am not sure how object velocity is calculated internally, but it will be likely between 2 discrete steps, so you will never get truly tangential trajectory. How off you will be depends on granularity of steps (this may be from one frame to another). In this case you may be better off with invisible individual flow objects correctly aimed and just trigger them at right time.

I can confirm inconsistency here. I am not sure how object velocity is calculated internally, but it will be likely between 2 discrete steps, so you will never get truly tangential trajectory. How off you will be depends on granularity of steps (this may be from one frame to another). In this case you may be better off with invisible individual flow objects correctly aimed and just trigger them at right time.

Rich Sedman commented

2020-09-22 08:42:40 +02:00

The example (using a rotating inflow) is just to demonstrate the problem - I wasn't actually aiming for realistic tangential flow. The point is that the inflow isn't using the previous frame to determine the velocity but is actually using the last frame that the inflow was switched on. So in the first 'emission' in the above case (in the "12-o'clock" position) the fluid has no velocity - not because the inflow wasn't moving but purely because it's the first frame of emission of fluid (so it seems the calculation doesn't have a 'start' position to determine the offset).

In the second case (at "6-o'clock" position) the emission is moving southwards - not because of the inflows current motion but purely because the last frame it emitted was directly to the "north" of its current position. Similarly for the "3-o'clock" position, it's only emitting diagonally because the last time it emitted it was at the "6-o'clock" position.

If you take the above example and add an additional 1-frame emission of fluid at, say, "9-o'clock", the direction of the next emission ("6-o'clock") changes - purely due to the "last frame of fluid emission" being in a different location. The historic state of emission or not should not affect the velocity when the inflow is next switched on.

It's difficult to explain. I could add another animated GIF to demonstrate the above change in behaviour (adding a 9-o'clock emission) if that would make things clearer.

In summary, pulsing an inflow on/off should not be affected by the location of the last time it was pulsed on/off - and the first pulse should not be always zero velocity (due it it never having been 'on' before).

The example (using a rotating inflow) is just to demonstrate the problem - I wasn't actually aiming for realistic tangential flow. The point is that the inflow isn't using the previous frame to determine the velocity but is actually using the *last frame that the inflow was switched on*. So in the first 'emission' in the above case (in the "12-o'clock" position) the fluid has no velocity - not because the inflow wasn't moving but purely because it's the first frame of emission of fluid (so it seems the calculation doesn't have a 'start' position to determine the offset). In the second case (at "6-o'clock" position) the emission is moving southwards - not because of the inflows current motion but purely because the last frame it emitted was directly to the "north" of its current position. Similarly for the "3-o'clock" position, it's only emitting diagonally because the last time it emitted it was at the "6-o'clock" position. If you take the above example and add an additional 1-frame emission of fluid at, say, "9-o'clock", the direction of the next emission ("6-o'clock") changes - purely due to the "last frame of fluid emission" being in a different location. The historic state of emission or not should not affect the velocity when the inflow is next switched on. It's difficult to explain. I could add another animated GIF to demonstrate the above change in behaviour (adding a 9-o'clock emission) if that would make things clearer. In summary, pulsing an inflow on/off should not be affected by the location of the last time it was pulsed on/off - and the first pulse should not be **always** zero velocity (due it it never having been 'on' before).

Jacques Lucke commented

2020-09-22 12:52:15 +02:00

Added subscribers: @sebbas, @JacquesLucke

Jacques Lucke commented

2020-09-22 12:52:15 +02:00

I can see and reproduce the issue. This is another well-intentioned optimization with unexpected consequences (@sebbas). The issue is that escape_flowsobject tells the simulation to completely skip inflow objects that have their inflow disabled. However, this also skips storing the current vertex positions so that the velocity can be computed next frame.
I can think of two possible solutions:

Store the current vertex positions of inflow objects in every frame, regardless of whether inflow is enabled or not.
Evaluate the vertex positions from the previous frame/subframe in the current frame. This way we would not have to store the previous vertex positions at all. Another advantage of this is that we do not depend on the previous frames information. Disadvantages are that it might be slower (although that might be negligible compared to the actual simulation) and it might not work in some cases that are not supported by `BKE_object_modifier_update_subframe.

A hybrid approach might work as well: In the normal case we store the current vertex positions for the next frame. With that we can also store the frame these positions come from. If they are too old, we can use BKE_object_modifier_update_subframe to compute the vertex positions at a previous frame on the fly.

I can see and reproduce the issue. This is another well-intentioned optimization with unexpected consequences (@sebbas). The issue is that `escape_flowsobject` tells the simulation to completely skip inflow objects that have their inflow disabled. However, this also skips storing the current vertex positions so that the velocity can be computed next frame. I can think of two possible solutions: * Store the current vertex positions of inflow objects in every frame, regardless of whether inflow is enabled or not. * Evaluate the vertex positions from the previous frame/subframe in the current frame. This way we would not have to store the previous vertex positions at all. Another advantage of this is that we do not depend on the previous frames information. Disadvantages are that it might be slower (although that might be negligible compared to the actual simulation) and it might not work in some cases that are not supported by `BKE_object_modifier_update_subframe. A hybrid approach might work as well: In the normal case we store the current vertex positions for the next frame. With that we can also store the frame these positions come from. If they are too old, we can use `BKE_object_modifier_update_subframe` to compute the vertex positions at a previous frame on the fly.

Sebastián Barschkis commented

2020-09-22 17:51:02 +02:00

Yes, that was too much optimization.

I am trying out an approach of idea (1) as getting vertex velocities is not the expensive part. If I remember correctly BKE_object_modifier_update_subframe should just be used if it is really needed. It was invoking quite a chain of calls ..

Also, a similar fix will be needed for escape_effectorobject to prevent issues with Use Effector.

Yes, that was too much optimization. I am trying out an approach of idea (1) as getting vertex velocities is not the expensive part. If I remember correctly `BKE_object_modifier_update_subframe` should just be used if it is really needed. It was invoking quite a chain of calls .. Also, a similar fix will be needed for `escape_effectorobject` to prevent issues with `Use Effector`.

blender-admin commented

2020-09-23 15:54:08 +02:00

This issue was referenced by bda274d1fe

This issue was referenced by bda274d1fef8774383d7e89847726e7dc7b3a9a3

Sebastián Barschkis commented

2020-09-23 15:54:56 +02:00

Changed status from 'Confirmed' to: 'Resolved'

Sebastián Barschkis closed this issue

2020-09-23 15:54:56 +02:00

Sebastián Barschkis self-assigned this 2020-09-23 15:54:56 +02:00

Sign in to join this conversation.

No Label

Download

What's New

Blender Studio

Manual

Developers Blog

Documentation

Benchmark

Blender Conference

Development Fund

One-time Donations

Fluid Initial Velocity 'Source' incorrect when enabling Inflow #80833