Project Details

Our project is focused on developing a deep learning model to detect the start time of moonquakes using seismic data from the Apollo missions. The core task involves analyzing long time series data, specifically the velocity measurements of the lunar surface, to identify when a moonquake occurs by learning from past seismic events. This model combines CNN and LSTM (Long Short-Term Memory) network layers, which are ideal for recognizing long-range temporal dependencies, such as the patterns of seismic waves over time.

How It Works:

The model processes seismic time series data to pinpoint the start of a moonquake. We enhance the model by calculating both the velocity and its derivative (rate of change) to help it focus on spikes in the data. The data from multiple Apollo missions is preprocessed, normalized, and padded to ensure consistent input length and then fed into the LSTM+CNN network, which learns to predict the exact time when the quake starts.

Key Technologies and Tools:

• Languages: Python

• Frameworks: TensorFlow, Keras (for model development)

• Preprocessing: Pandas, NumPy (for data loading and manipulation)

• Data Visualization: Matplotlib (for plotting results and visualize detection)

• Seismic Data: Lunar seismic recordings from the Apollo missions

Benefits:

This project offers a highly efficient and accurate method of detecting moonquakes, which could be critical for future lunar exploration missions and seismic studies of other planetary bodies. The model can process long and noisy time series data, making it applicable to broader seismological use cases.

Next Steps:

Future developments include:

1. Data Preprocessing to Remove Noise:

• Apply advanced filtering (Fourier Transform, wavelet denoising) to reduce noise and standardize data across Apollo missions.

2. More Feature Engineering.

3. Tuning Model’s Parameters:

• Use Grid Search or Random Search for hyperparameter tuning, cross-validation, and regularization (dropout, L2) to optimize performance.

4. Two-Stage Detection Model:

• Implement a two-stage model: first to detect moonquake regions, then to fine-tune start time predictions for higher accuracy.

This project demonstrates a promising intersection of machine learning and planetary science, opening up opportunities for further advancements in space exploration technology.

2024 NASA Space Apps Challenge

Debugging the Universe