A Diversity-Driven Self-Adaptive Genetic Algorithm for High-Precision 24-Position Accelerometer Calibration

To enhance the accuracy of consumer-grade accelerometers in low-cost navigation and attitude measurement applications, this paper proposes a Diversity-Driven Self-Adaptive Genetic Algorithm (D-SAGA) for high-precision calibration of accelerometer error parameters. A comprehensive error model is first established, accounting for bias, scale factor, and non-orthogonal errors. A static 24-position calibration platform is designed using a 3D-printed hexahedral structure to acquire high-quality attitude data.Building on the conventional genetic algorithm (GA), D-SAGA incorporates a population diversity metric and adaptive operator adjustment strategy, thereby improving global search capability and convergence stability. To validate the proposed method, experiments were conducted using two accelerometer modules: the consumer-grade MPU6050 and the higher-performance accelerometer in the CMP10A IMU. Comparative evaluations were performed against Gradient Descent with Line Search, Newton + Gradient Descent + Line Search, Particle Swarm Optimization (PSO), and traditional GA.Results demonstrate that D-SAGA consistently achieves superior calibration accuracy, convergence efficiency, and robustness. For the MPU6050, post-calibration RMSE was reduced to 1.41 mg, representing a 95.81% error reduction, while CMP10A experiments confirmed the generalizability of the approach. Repeatability experiments with 50 independent runs further verified robustness: D-SAGA achieved a mean RMSE of 1.4254 mg with a standard deviation of only 0.0213 mg, compared to 0.1222 mg for GA. Statistical analyses (two-sample t-tests, p < 0.01) confirmed that the improvements are significant.Overall, the proposed D-SAGA method offers an efficient, stable, and scalable calibration solution for consumer-grade inertial measurement units (IMUs), with promising applicability for practical engineering use, though further validation in large-scale field scenarios is warranted.