TL;DR
This paper introduces a cost-effective and efficient method for calibrating triaxial MEMS accelerometers, significantly improving their accuracy by addressing temperature-dependent errors. By optimizing the calibration procedure and applying a novel thermal compensation strategy, they achieved a ~100x improvement in accuracy and drastically reduced temperature drift, making these sensors more reliable for tilt measurement applications.
What Problem Does This Paper Address?
MEMS accelerometers are increasingly used in applications like navigation and posture control due to their small size and low cost. However, their accuracy is significantly hampered by temperature variations, which cause drifts in their output. Traditional calibration methods are often time-consuming and expensive, especially for achieving high accuracy across a wide temperature range. This paper tackles the challenge of developing a practical, low-cost, and highly accurate calibration and thermal compensation method for triaxial MEMS accelerometers, specifically aiming to improve their performance in tilt measurement applications where temperature fluctuations are common.
Methods
The authors propose a two-pronged approach:
- Optimized Calibration Procedure: They leverage the Levenberg-Marquardt (LM) algorithm, a robust optimization technique, to estimate the Sensor Frame Error Model (SFEM) parameters. Crucially, they optimize the number of calibration positions required. Instead of relying on a large, fixed number of positions, they use the Root Mean Square Error (RMSE) to determine an optimal set of positions (24 in their case) that balances accuracy with efficiency. This avoids the need for highly precise, expensive equipment like turntables for every calibration.
- Thermal Compensation: For temperature-dependent errors, they employ a least squares method to fit polynomial models to the calibration parameters (bias, scale factor, non-orthogonality) as a function of temperature. This allows them to create a thermal model that can compensate for drift across a wide temperature range (-40 °C to 60 °C). The process involves collecting data at various temperatures, fitting these polynomial models, and then applying these models to correct the accelerometer's output in real-time.
Datasets
The experiments were conducted using commercially available ADXL355 triaxial accelerometers. The data was collected using a custom-built setup involving a temperature-controlled rotational-tilt platform. The authors collected raw data at 1296 uniformly distributed positions on a sphere to analyze the SFEM and its dependence on the number of calibration positions. They also performed temperature cycling experiments from -40 °C to 60 °C at 10 °C intervals. No new benchmark datasets were introduced; rather, existing hardware was used to generate experimental data for validation.
Key Insights and Ideas
- Optimizing Calibration Positions is Key: The study demonstrates that a significantly reduced number of calibration positions (around 24) can achieve comparable accuracy to much larger sets, drastically cutting down calibration time and cost without sacrificing performance. This is a practical win for manufacturers and users.
- Polynomial Models Excel for Thermal Drift: Fitting simple polynomial models to the temperature dependence of error parameters (bias, scale factor, etc.) proved effective in capturing and compensating for thermal drift across a broad temperature range.
- Synergy of LM and Polynomial Fitting: The combination of the robust Levenberg-Marquardt algorithm for initial calibration and polynomial fitting for thermal compensation creates a powerful, low-cost solution for high-accuracy tilt measurements.
- Significant Error Reduction: The results show a dramatic improvement in accuracy, with errors reduced by approximately 100 times after calibration. Thermal compensation further refines this, reducing temperature drift to very low levels.
Key Results
- Accuracy Improvement: The calibrated results showed an improvement of nearly 100 times compared to the uncalibrated data.
- Reduced Temperature Drift:
- X-axis drift reduced from -13.2 mg to -0.9 mg.
- Y-axis drift reduced from -11.8 mg to -1.1 mg.
- Z-axis drift reduced from -17.9 mg to 1.8 mg.
- Tilt Angle Accuracy: The tilt angle measurements after calibration were significantly more accurate than before calibration, with Sensor Error Optimization Quantity (SEOQ) values showing substantial improvements.
- Efficient Calibration: The optimized 24-position method, combined with the LM algorithm, proved sufficient for accurate SFEM estimation.
Limitations and Future Work
- Stabilization Time: The authors acknowledge that the sensor requires a significant amount of time to stabilize at each temperature point during thermal calibration. This can lead to inaccuracies if not properly accounted for.
- Dynamic Temperature Changes: While the method addresses static temperature variations and slow cycling, the paper suggests that future work could explore handling more dynamic temperature change rates and input accelerations.
- Multiple Thermal Models: For demanding applications, developing multiple thermal models calibrated in various ambient temperatures and potentially a combined thermal model could further enhance performance.
Summary updated