Load Cell Signal Filtering Toolkit

A comprehensive Python toolkit for filtering and analyzing load cell data with three powerful filtering techniques.

📊 Overview

This toolkit provides ready-to-use scripts for cleaning noisy load cell measurements using:

• Moving Average Filter - Simple, robust smoothing
• Butterworth Filter - Precise frequency-domain filtering
• Savitzky-Golay Filter - Preserves peaks and features

🚀 Quick Start

Requirements

pip install numpy pandas matplotlib scipy

Data Format

Your CSV file should have these columns:

• Timestamp (ms) - Time in milliseconds
• Weight (g) - Load cell readings in grams

Example:

Timestamp (ms),Weight (g)
0,245.3
10,246.1
20,244.8

Usage

1. Place your CSV file in the same directory as the scripts
2. Update the filename in the script:

   df = pd.read_csv('your_data.csv')

3. Run the desired filter:

   python moving_average_filter.py
   python butterworth_filter.py
   python savitzky_golay_filter.py

📈 Filter Comparison

Filter	Best For	Pros	Cons
Moving Average	Quick smoothing, real-time	Simple, fast, no ringing	Blunts sharp features
Butterworth	Frequency-specific noise	Precise cutoff, flat passband	Can overshoot
Savitzky-Golay	Preserving peaks/valleys	Keeps features intact	Computationally intensive

🔧 Tuning Parameters

Moving Average

• WINDOW_SECONDS: Averaging window (5-20s typical)
  • Smaller = less smoothing, preserves details
  • Larger = more smoothing, removes noise

Butterworth

• CUTOFF_HZ: Frequency cutoff
  • 0.001-0.01 Hz: Remove drift only
  • 1-3 Hz: Smooth noise, keep signal shape
• ORDER: Filter steepness (2-8, typically 4)

Savitzky-Golay

• WINDOW_SECONDS: Fitting window (5-20s typical)
• POLYORDER: Polynomial order (2-5, typically 3)
  • Higher = better peak preservation

📁 Output

Each script generates:

• Filtered signal added to DataFrame as Weight_filtered_g
• Analysis plot saved as PNG
• Statistical summary printed to console

💡 Tips

1. Start with visualization: Run visualize_data.py first
2. Compare filters: Try all three on your data
3. Iterate parameters: Adjust based on your noise characteristics
4. Check sampling rate: Scripts auto-detect and warn if issues

🐛 Troubleshooting

"Signal too short" error

• Need at least 50 data points

"Non-uniform sampling" warning

• Your timestamps aren't evenly spaced
• Filters still work but may be less accurate

Filter artifacts at edges

• Normal behavior, especially for Butterworth
• Data edges are padded/extrapolated

📝 License

MIT License - Free to use and modify

🤝 Contributing

Issues and pull requests welcome!

✨ Example Results

After filtering, you'll see:

• Original vs filtered signal plots
• Noise reduction statistics
• Filter frequency response (Butterworth)
• Peak preservation analysis (Savitzky-Golay)