Technical Artist
Thor.eli.02@gmail.com
Tower Destruction
This project is a destruction simulation of a radio tower collapsing after being hit by an explosion. The initial inspiration came from the large-scale level destruction events in Battlefield 4, where major structures collapse during gameplay.
My original goal was to recreate a similar large tower collapse. However, due to time constraints I decided to focus on creating a radio tower that can be destroyed by shooting its support structure, triggering a controlled collapse simulation.
The project focuses on creating believable destruction using physically based simulations while still considering how the result could be implemented in a real time engine.
Base Model
Material IDs
Model Work
The tower was modeled in Blender rather than Houdini, since the structure itself was easier to design using traditional modeling tools.
During the modeling phase I researched how real world radio towers are constructed. Many of these towers are built in modular sections, especially in remote locations, where individual beams are assembled and welded together on site. I reflected this construction style in the model by dividing the tower into smaller structural segments.
This segmentation served two purposes:
- It made the structure more realistic.
- It allowed the pieces to break apart naturally during the destruction simulation.
To showcase different destruction behaviors, I included several types of materials:
- Metal– structural beams of the tower
- Wood– smaller supporting structures
- Electronics– satellite dishes and equipment
Each material was assigned its own material slot so that when imported into Houdini the pieces could easily be grouped and controlled separately when creating constraints.
Fractures
Most of the objects were fractured using a fairly standard material fracture workflow.
The metal structure required a more advanced setup and was fractured in two stages:
- Small fragments– to allow the metal beams to deform and bend.
- Larger chunks– to allow structural pieces to snap apart under stress.
Constraints
The constraint setup was the most complex and time consuming part of the simulation.
For most objects, such as wood and the satellite dish, I used custom glue constraints to hold the pieces together until a certain amount of force was applied.
The metal structure used two different constraint systems:
1. Bend constraints
These held the smaller fracture pieces together, allowing the beams to bend before breaking.
2. Break constraints
These controlled when the larger structural segments would snap apart.
To make the simulation more physically believable, constraint strength was influenced by the materials connected at each point. Using proximity checks, the system evaluated which materials were interacting and adjusted the constraint strength accordingly.
One of the biggest challenges was simulating metal behavior. Unlike materials such as wood, metal typically does not spring back once bent. I had to carefully tune the constraint system so the metal would deform realistically without returning to its original shape.
Main Constraints
Sekandery Constraints
Constraints Overview
Weakpoint and strenght
To improve realism, I introduced variation in the tower's structural strength.
A gradient system was used so that constraints closer to the base of the tower were weaker. Since the base supports the entire structure, this made it more likely to bend and fail first, producing a more dramatic and cinematic collapse.
This also helped guide the direction of the tower’s fall in a more believable way.
Controlling Collaps
To trigger the destruction event, a sphere with velocity was launched toward one of the tower's legs in the simulation.
This acted as the initial impact force, similar to an explosion damaging the tower’s structural support. Once the constraints in the support structure broke, the rest of the tower began collapsing under its own weight.
Vertex Animation Texture
Vat Texture
Alambic File
Implementation
Importing the simulation into Unreal Engine turned out to be more challenging than expected.
My original plan was to export the simulation as a skeletal mesh animation. However, due to the high number of fractured pieces, the number of bones required would have been too large for the engine to handle efficiently.
As an experiment I exported the simulation as an Alembic (.abc)cache. While this method worked, the file size ended up being around 3.7 GB, making it impractical for real time use.
Instead, I converted the simulation into a Vertex Animation Texture (VAT). Because of the high point count of the simulation, the VAT required an 8K texture. Although still relatively large, this approach was significantly more optimized and made the destruction sequence viable for use in a real-time environment.
If I had more Time
With more time, I would expand the system to allow for multiple destruction variations.
The idea was to create several possible collapse paths depending on which structural elements were destroyed. This would allow players to shoot different parts of the tower and trigger different collapse behaviors.
Technically this would not have been very difficult to implement, but due to the size of the simulation it took a long time to bake and cache each version. Because of this, I decided to focus on polishing a single high-quality simulation instead.