Table of Contents

• Table of Contents
• Waste Bunker Optimization Challenge
• Data
• Problem Statement
  • Problem 1: Bunker Fullness Estimation
    • Objectives
  • Problem 2: Optimal Waste Mix
    • Objective
• Expected Output & Submission
• Infrastructure
  • Connecting to your Server
  • Customizing the Server
  • Vibe Coding

Waste Bunker Optimization Challenge

Waste incineration plays a crucial role in modern energy systems by converting non-recyclable waste into usable energy. However, efficient incineration is not just about burning waste. It requires smart bunker management to ensure stable operations and optimal energy yield.

At Wasteer, our mission is to make waste-to-energy operations more intelligent and data driven. One of the most critical components of this process is the waste bunker, where different waste streams are temporarily stored and mixed before incineration.

For this challenge, you will work with real industrial data to optimize waste bunker management.

Data

You are given the following data which is available here:

1. LiDAR scans of the bunker
   3D surface scans showing the current fill level (visible waste surface) of the bunker. A script for reading and exploring the LiDAR data is available at the file downloads. Note that the filename is indicating the timestamp of the scan YYYYMMDDhhmmss.
2. Truck shipment records
   Timestamped list of trucks shipping waste. Each entry includes the waste type and quantity (in kg) delivered into the bunker.
   • created_at: Creation timestamp (UTC, with microseconds).
   • entry_timestamp: Time when vehicle entered/was first weighed.
   • entry_weight: Gross weight on entry (kg, unless otherwise noted).
   • exit_timestamp: Time when vehicle exited/was second weighed.
   • exit_weight: Tare or final weight on exit (kg, unless otherwise noted).
   • waste_code: Material/waste EWC code (e.g., 200301, 180104).
3. Waste code properties
   Mapping from waste type to calorific value (MJ/kg) and some more info.
4. Images Images which can be used either to train or enhance your model or just for you to get a better understanding.

Note:
The dataset does not explicitly include information about material removed from the bunker through incineration. Participants are encouraged to make and justify reasonable assumptions where necessary. Partially observable systems and incomplete operational data are common in real industrial environments and are part of the challenge 😉.

Problem Statement

There are two related problem statements that reflect real challenges faced by Wasteer today. Solving them would have a real-world impact by making the process more efficient. It is recommended that you start with Problem 1 and apply your learnings to Problem 2.

Problem 1: Bunker Fullness Estimation

The first task focuses on understanding how much waste the bunker can still accept until it is "full".

Objectives

1. Estimate how much waste can still be added to the bunker

   • Calculate how much of each waste type (in kg) can still fit in the bunker until the full threshold is reached.
     • Think of your own definition of what "full" really means. There is an indication on the slides shown at the start of the event. See Slideset at slide 34 and 35.
   • Note that each waste type has different density and compression properties, so the same mass of different waste types will change bunker fullness differently.
     • Use the LiDAR data along with truck shipment records to perform this estimation.
   • Also note that as waste accumulates in the bunker density will be changing over time, if waste is covered with other waste for a certain amount of time the density increases and the volume decreases.
   • Showcase your approach at different points in time by selecting several LiDAR scans and demonstrating how much of each waste type could still be added at those times.

   Bonus points:

   • Investigate whether LiDAR data alone is sufficient for estimating bunker fullness, or whether the type of waste already present in the bunker should also be considered.
     • Different waste types may affect compression differently (for example, heavy items on top of lighter materials can compress more). Explore whether this effect can be observed in the provided data.
   • Note: The starting point of the data is a "black box" regarding the exact waste composition in the bunker.

2. Estimate material density per waste type

   • Determine the density (kg/m^3) of each waste type based on the data, taking into account how it contributes to overall bunker volume and fullness.
   • This may be required to solve the first objective, but density estimation per waste type is also an explicit goal.

Problem 2: Optimal Waste Mix

Building on problem 1, the second task focuses on optimizing the energy potential of the bunker. The bunker operates most efficiently when the average calorific value is around 10 MJ/kg. Some waste types have lower calorific values, while others have higher values.

Your task is to estimate the optimal mix of waste required to maintain a stable average calorific value close to 10 MJ/kg, while considering the remaining bunker capacity. Minimizing fluctuations is important, as variations in calorific value reduce bunker operation efficiency.

Objective

Determine the optimal waste composition:

• Each waste type has an associated calorific value, as shown in the file waste_code_properties.
• Identify which waste types should be added to the bunker, to maintain an average calorific value of 10 MJ/kg.
• Showcase your approach using truck delivery record data:
  • Assume that at the start of the data, the average calorific value inside the bunker is 10 MJ/kg.
  • You may define your own reasonable assumption for the initial bunker mass.
  • Select several truck shipment records from the dataset to showcase your approach. Estimate how individual deliveries change the average calorific value inside the bunker and determine which waste type(s) and quantities would be needed to bring it back toward 10 MJ/kg. Remember to consider your learnings from problem 1 to make sure the bunker does not exceed its limits.

Expected Output & Submission

• Each team will be given a 5-minute presentation slot to convince the jury of the value and effectiveness of their solution.
• Evaluation criteria were provided during the initial presentation and should be considered.
• Please submit your slides and solution before the deadline at 1:00 PM. The submission should be done in your private team chat on discord. This is a strict deadline, so make sure to submit your solution on time.

Infrastructure

You will be assigned a GPU server (NVIDIA L4 24GB) for the event to be able to train models if you would like to. Each team is assigned one server with a preconfigured JupyterLab environment (including torch and tensorflow)

Connecting to your Server

1. You will receive your keypair and IP necessary to connect to the server in your teams private discord team channel.
2. Connect to your GPU server via SSH ssh -i <path_to_private_key> hacker@<ip_address>
   1. Running chmod 400 /path/to/key might by necessary
3. Activate the preconfigured conda environment on the server conda activate ml
4. Execute start-jupyter on the server to start your environment
5. Execute ssh -i <private_key_file> -N -L 8888:localhost:8888 hacker@<ip_address> on your local machine to establish ssh tunneling
6. Connect to Jupyter Lab by opening your browser and go to http://localhost:8888
7. A subset of the dataset is already present at the server. Upload more data of the dataset if needed.
8. Start hacking

Customizing the Server

• Note that your user hacker has passwordless sudo access to the server with your user. You are free to install additional tools and customize the environment as needed 😉
• The preconfigured environemnt is available as ml conda environment. Feel free to create a new one if you like.
  • You can also install new packages via pip install in jupyter lab in jupyterLab terminal. Make sure to conda activte ml before installation.

Vibe Coding

Each team has access to a OpenRouter API key with 20€ budget, which you will receive during the event in your teams private discord teams channel. See vibe-coding/ for more info.