Artificial neural network-based MEMS accelerometer array calibration

Abstract Accurate calibration of micro-electromechanical systems (MEMS) accelerometers is crucial for enhancing the performance of low-cost inertial measurement units (IMUs). This paper introduces a novel calibration technique that leverages artificial neural networks (ANNs) combined with data from multiple IMUs to increase the accuracy of the calibration. The proposed method involves a calibrated UR robot, which enables the data acquisition of ground truth data for an effective calibration of IMUs. It enhances the calibration accuracy by utilizing the collective measurements from five IMUs within an accelerometer array. Fourteen sets of measurement data were established in dynamic environments using the robotic arm. The ANN-based approach was trained using ten datasets of dynamic measurements, where the trained model is validated against four unseen test data. The ANN-based calibration performance is evaluated by comparing it to standard methods such as ellipsoid fitting method and arithmetic averaging of the sensor outputs. Results demonstrate that the proposed method achieves superior calibration accuracy, with an improvement of 18.2% over the ellipsoid fitting technique and 23.3% over the averaging method. It also shows that fusing accelerometer measurements with Euler angles calculated from acceleration as input data for the ANN provided the best results for the calibration. The findings suggest that integrating ANN models with data fusion from multiple sensors significantly improves the calibration accuracy of MEMS accelerometers, thereby enhancing their potential for use in precise motion sensing applications.