| The navigation systems of long-endurance remote aerial vehicles require high precision, reliability, autonomy, and endurance, thus placing stringent demands on the performance of quartz flexure accelerometers. Calibrating the static error model is directly linked to key performance indicators of the accelerometer, such as zero-bias stability, scale factor accuracy, and nonlinearity. This study examines calibration methods for accelerometer error models, addressing the limitations of the least squares method in fitting complex nonlinear relationships and the tendency of BP (Back Propagation) neural networks to become trapped in local optima when identifying the static model of quartz flexure accelerometers. To overcome these issues, a Particle Swarm Optimization BP (PSO-BP) algorithm is employed to integrate the advantages of both approaches. After calibration experiments on a precision centrifuge, a dataset of accelerometer inputs and outputs was obtained. The least squares method, BP neural network, and PSO-BP neural network were then applied to fit the accelerometer output, and the prediction results and mean square errors for each method’s test set were evaluated. Experimental results demonstrate that the PSO-BP algorithm effectively calibrates the static model of the accelerometer, achieving a reduction in mean square error by two orders of magnitude compared to BP network results, a smoother prediction error curve with smaller fluctuation amplitude. |
