powermeter-compare

A Python tool for comparing power data between two cycling power meters using FIT files. Performs automatic time alignment, detects dropout regions, computes power offsets, and generates interactive visualizations.

Features

• Automatic Time Alignment: Uses cross-correlation to align candidate power meter data with reference data
• Dropout Detection: Identifies and excludes regions with missing, zero, or invalid power readings
• Power Offset Analysis: Computes median power offset between meters with statistical metrics
• Interval-Based Analysis: Optional per-interval offset computation using MyWhoosh workout files
• RANSAC Regression: Robust linear modeling of offset as a function of power (detects systematic calibration errors)
• Interactive Visualizations: Plotly-based plots with synchronized zoom and detailed diagnostics

Installation

This project uses uv for dependency management:

# Clone the repository
git clone https://github.com/yourusername/powermeter-compare.git
cd powermeter-compare

# Install dependencies
uv sync

# Run the tool
uv run powermeter-compare --help

Quick Start

Basic Comparison

Compare two power meters using their FIT files:

uv run powermeter-compare \
  --ref-fit data/reference.fit \
  --candidate-fit data/candidate.fit

Interval-Based Analysis

Analyze offsets per workout interval using a MyWhoosh workout JSON:

uv run powermeter-compare \
  --ref-fit data/reference.fit \
  --candidate-fit data/candidate.fit \
  --workout-json data/workout.json \
  --ftp 250

Custom Labels

Override auto-detected power meter names:

uv run powermeter-compare \
  --ref-fit data/reference.fit \
  --candidate-fit data/candidate.fit \
  --label-ref "Quarq DZero" \
  --label-candidate "Favero Assioma"

Analysis Pipeline

1. Load FIT Files: Reads power and timestamp data from both FIT files
2. Time Alignment: Resamples to common grid and estimates time offset via cross-correlation
3. Data Merging: Merges aligned power data with configurable time tolerance
4. Dropout Detection: Identifies invalid/missing data regions
5. Trailing Averages: Computes 5s, 10s, and 30s rolling averages
6. Interval Analysis (optional): Computes per-interval offsets from workout definition
7. RANSAC Modeling (optional): Fits robust linear model of offset vs power
8. Visualization: Generates interactive plots for analysis

Output Plots

1. Power Comparison Subplots

Four synchronized subplots showing instantaneous, 5s, 10s, and 30s averaged power:

• Reference power (blue) vs Candidate power (red)
• Gray background indicates dropout regions
• Colored backgrounds show workout intervals (if provided)
• Each subplot annotated with offset statistics

2. Offset vs Target Power

Scatter plot showing power offset at different target power levels:

• Blue markers with error bars for each interval
• Green dashed line shows RANSAC regression fit
• Identifies systematic calibration errors

3. RANSAC Fit (Detailed)

Enhanced offset vs power plot distinguishing:

• Blue markers: Inlier intervals
• Red X markers: Outlier intervals
• Green dashed line: RANSAC fitted model

4. Cross-Correlation Diagnostics

Shows correlation coefficient vs time lag:

• Identifies optimal time offset
• Visualizes alignment quality

Command-Line Options

--ref-fit PATH              Path to reference FIT file (required)
--candidate-fit PATH        Path to candidate FIT file (required)
--label-ref TEXT            Reference power meter label (auto-detected)
--label-candidate TEXT      Candidate power meter label (auto-detected)
--workout-json PATH         MyWhoosh workout JSON for interval analysis
--ftp FLOAT                 Functional Threshold Power (required with --workout-json)
--rampup-exclusion FLOAT    Seconds to exclude from interval start (default: 5.0)
--min-power-threshold FLOAT Minimum power for analysis in watts (default: 50.0)
--time-tolerance TEXT       Time tolerance for alignment (default: "2s")
--max-time-offset INT       Max time offset search range (default: 60)

Use Cases

Validating a New Power Meter

Compare a new power meter against a trusted reference to verify calibration and accuracy across different power levels.

Detecting Calibration Drift

Use RANSAC analysis to identify if offset varies with power (e.g., 2% error indicating systematic miscalibration).

Interval Training Analysis

Analyze per-interval offsets during structured workouts to understand power meter behavior at different intensities.

Quality Assessment

Identify dropout regions, data gaps, and anomalous intervals that may indicate connectivity or sensor issues.

RANSAC Output Interpretation

The RANSAC linear model fits: offset = slope � power + intercept

• Slope H 0, Intercept ` 0: Constant offset (e.g., +8W across all power levels)
• Slope ` 0, Intercept H 0: Proportional error (e.g., 3% calibration error)
• Slope 0, Intercept 0: Combined offset and scaling error
• Outliers: Intervals with unusual behavior (check for dropouts or anomalies)

Requirements

• Python e 3.11
• FIT files from compatible cycling power meters
• Optional: MyWhoosh workout JSON files for interval analysis

Development

# Run tests
make tests

# Format and lint
make format-lint

# Run with debugging
uv run powermeter-compare --ref-fit data/MyWhoosh.fit --candidate-fit data/Favero.fit --workout-json data/MyWhooshWorkout.json --ftp 320

License

MIT

Contributing

Contributions welcome! Please ensure:

• All tests pass: make tests
• Code is formatted: make format-lint
• New features include tests