OktoDriveshaft

This repository contains firmware, data, and processing scripts for the Okto (ESP32-based) driveshaft strain measurement project used by McMaster Baja Racing. The project captures torsion and axial strain from an instrumented driveshaft and provides a lightweight Python pipeline to correct timestamps, filter / detrend signals, and save normalized outputs for analysis.

Contents:

• firmware/ — ESP32 code and calibration files
• data/ — raw and processed CSV exports from the device
• scripts/ — Python scripts to correct timestamps, filter signals, and save normalized data

Quick summary:

• The ESP32 firmware records sensor data to an SD card. Two firmware variants are included:
  • millis_code/: samples at ~2 kHz but timestamps are written with a millisecond resolution (so consecutive samples can share the same millisecond value). The data files contain a Timestamp [ms] column.
  • nanos_code/: samples at ~5 kHz and records timestamps in nanoseconds (Timestamp [ns]).
• firmware/calibration/ includes an Excel calibration sheet and example csv used to compute zero-offsets and linear multipliers for converting raw sensor counts to physical units (measured in a pure torsion setup).

Data layout:

• data/0_raw/ — CSV files straight from the ESP32 (filename prefix okto25_). These are unmodified exports.
• data/1_time_corrected/ — CSV files produced by the time-correction step. All timestamp columns are converted to seconds, the first timestamp is zeroed, and known SD-flush/noise discontinuities are smoothed.
• data/2_normalized/ — Final processed CSVs after filtering and EMD detrending. Signals are saved with the same okto25_ prefix and contain a Timestamp [s] column and processed measurement columns (e.g., Torque [Nm]).

Notes:

• The millis_code firmware records timestamps in Timestamp [ms]. Because the sample rate (~2 kHz) is higher than millisecond resolution, there can be duplicate ms values. _correct_time_ms adjusts duplicates and fixes single-sample discontinuities (like SD flush artifacts) by interpolating a local mean and applying an offset to subsequent samples.
• The nanos_code firmware records Timestamp [ns]. _correct_time_ns looks for periodic flush boundaries (1 second) and corrects large jumps near those boundaries.
• The EMD-based detrending (PyEMD) in remove_slow_wave_emd decomposes the signal into intrinsic mode functions (IMFs). The code currently reconstructs the baseline as the difference between the original and the summed IMFs in the specified keep_imfs range — review the keep_imfs tuple to control which IMFs are treated as baseline vs signal.
• Plots are generated when functions are called with debug=True which helps visualize filtering and detrending steps.