Advancing Spatial-Temporal Rock Fracture Prediction with Virtual Camera-Based Data Augmentation

---

Diagram

Diagram predicting rock fractures in unexcavated areas based on VCT and actual rock face data

Capture or Slice Rock Fractures

---

Usage for PredRNN-like Models

If you are using a sequence forecasting model such as PredRNN for rock fracture mapping prediction, you may require a substantial amount of data. Here, we provide a tool to synthesize all the data you need!

1. Set Fracture Set and Virtual Camera Parameters:

   • Configure the fracture set and virtual camera parameters in SRM_linedata_no_image.py. Please read the functions' input parameters carefully!
   • Fractures in SRM sliced by a moving target plane and captured by a virtual camera will be saved in the Fracture directory as .txt files. These files will then be converted to .npy files for model training and testing. The DataFactory is used for data loading. Refer to other scripts for more details.

2. Data Factory:

   • A custom DataLoader for sequence prediction, see DataFactory.

---

Data Description

The data includes the following attributes: fracture surface ID, center point coordinates, normal vector, radius, image fracture line center point coordinates, fracture line direction vector, fracture line length, observation distance, target plane position, distance between the target plane and the previous plane, image fracture line segment start point, image fracture line segment end point, actual fracture start point, and actual fracture end point.

| 0 | 1,2,3 | 4,5,6 | 7 | 8,9 | 10,11 | 12 | 13 | 14 | 15 | 16,17 | 18,19 | 20,21 | 22,23 |

• fracId: 0
• fracCenter(3): 1, 2, 3
• fracNormal(3): 4, 5, 6
• fracRadius: 7
• segCenter(2): 8, 9
• segDirection(2): 10, 11
• segLength: 12
• ObserveDist: 13
• targetZ: 14
• StepSize: 15
• segStartX: 16
• segStartY: 17
• segEndX: 18
• segEndY: 19
• fracStartX: 20
• fracStartY: 21
• fracEndX: 22
• fracEndY: 23

To eliminate the influence of image size, the coordinates are normalized. For details, see SRM_linedata_no_image.py:

$$ u = \left[ \frac{u_{1x}}{ResX}, \frac{u_{1y}}{ResY} \right] $$

$$ segCenter = \left[ \frac{u_{1x} + u_{2x}}{2}, \frac{u_{1y} + u_{2y}}{2} \right] $$

$$ segDirection = normalize \left( \left[ u_{2x} - u_{1x}, u_{2y} - u_{1y} \right] \right) $$

$$ segLength = \sqrt{(u_{1x} - u_{2x})^2 + (u_{1y} - u_{2y})^2} $$

Where ResX and ResY are the image resolutions.

To convert back to image coordinates:

$$ u_x = s_x \times ResX $$

$$ u_y = s_y \times ResY $$

---

Demo Dataset Parameters

DataSet	Type	DipDirection	DipAngle	Fisher Constant	Size	P30	Num
Set 1	Single	60	40	25	Exp 10 2	0.01	16
Set 2	Single	210	70	20	Exp 10 2	0.01	16
Set 3	Single	300	60	15	Exp 10 2	0.01	16
Set 4	Fusion	All above	All above	All above	Exp 10 2	0.01	16
SetUnion123	Union123	-	-	-	-	-	-
Set All	Union1234	-	-	-	-	-	-

---

you can download it from Baidu,code:zyzs.

Sequences GIF

Figure	Description	Animation
Figure 1	FixStep Mapping with color-image
Figure 2	FixStep Mapping with binary-image
Figure 3	NRandStep Mapping with color-image
Figure 4	NRandStep Mapping with binary-image

---

---

Usage

UBG for DFN Model Generation is compiled under Windows with Python 3.10. Python 3.10 is required if you want to generate custom data.

1. Set Fracture Set and Virtual Camera Parameters:

   • Configure the fracture set and virtual camera parameters in SRM_linedata_no_image.py. Please read the functions' input parameters carefully!
   • Fractures in SRM sliced by a moving target plane and captured by a virtual camera will be saved in the Fracture directory as .txt files. These files will then be converted to .npy files for model training and testing. The DataFactory is used for data loading. Refer to other scripts for more details.

2. For Training PredRNN-like Models:

   • See repository or subdiroctory

---

Data Factory

A custom DataLoader for sequence prediction, see DataFactory.

---

Acknowledgements

• UnBlocks-gen
• PredRNN

---

Citation

If you use this tool for your research, we would appreciate it if you cite it as follows:

@Misc{Baolin Chen and Jiawei Xie,
  title ={SRM-Fracture Traces Generator},
  author={Baolin Chen and Jiawei Xie},
  howpublished ={\url{https://github.com/GEO-ATLAS/Rock-Camera.git}},
  year={2023}
}

@article{ASTRFPwithVC,
    title={Advancing Spatial-Temporal Rock Fracture Prediction with Virtual Camera-Based Data Augmentation},
    author={Jiawei Xie, Baolin Chen, Jinsong Huang, Yuting Zhang, Cheng Zeng},
    journal={Tunnelling and Underground Space Technology},
    year={2025}
}