Results from Van Testing of Inertial Unit with Innovative Bias Drift Compensation by Kalman Filtering
Rita Fontanella, Rosario Schiano Lo Moriello, Domenico Accardo, Leopoldo Angrisani University of Naples “Federico II”, Italy; Domenico De Simone, Generale Meccatronica Applicata S.p.A., Italy
Location: Big Sur
Alternate Number 3
Inertial sensors are the primary sensors used for air vehicle navigation and control. In particular, the integration of an Inertial Measuring Unit (IMU) with a GPS receiver-chip provides a navigation system that has several advantages over each individual system. GPS alone is incapable of providing continuous and reliable positioning, because of its inherent dependency on external electromagnetic signals. Instead, the IMU has unbounded error growth since errors accumulate at each step and it is necessary to use an external aiding in order to contain these errors. In such integration, the GPS derived positions and velocities update inertial sensors through a Kalman Filter, while the IMU is used for providing the navigation information during GPS signal outages and for fast GPS signal reacquisition. Therefore, it is necessary that inertial sensors have a very good performance in terms of stand-alone navigation accuracy to support longer GPS outages.
Although traditional inertial sensors, such as gimbaled gyroscopes, optic gyroscopes and dynamically tuned gyroscopes, provide high-precision information for navigation and control systems, they can be either expensive or bulky. Recent improvement of advanced micro-fabrication techniques has allowed the development of Micro Electro-Mechanical Systems (MEMS) inertial sensors. These sensors have the advantages of small volume, light-weight, high reliability and low-cost. Consequently, they are well-suited to perform the flight attitude calculation for low-cost and small platforms, such as small Unmanned Aerial Systems (UAS). Recently, the application area of these sensors is expanding also to automotive industry and robotics.
Due to manufacturing limitations, MEMS inertial sensors suffer for some types of errors, such as turn-on to turn-on bias, in-run bias, bias drift, scale factor drift and other environment dependent errors, which are generally small or negligible for higher grade inertial sensors. These errors determine a fast build up over time of the attitude determination error, corrupting the precision of the measurement until the performance does not fulfill the specifications. In particular, the performance of MEMS inertial sensors is greatly affected by temperature variations, which can cause the serious drawbacks of low precision and even errors. This is due to the sensitivity of silicon's material properties and gyro’s packaging and electronics to temperature. Therefore, a significantly large error results from using the output of MEMS inertial sensors without an adequate temperature compensation.
The standard Kalman Filter process consists of two different steps. In the first step, a Field Programmable Gate Array (FPGA) performs the thermal calibration of raw sensor data. In the second step, a Digital Signal Processing (DSP) applies the Kalman Filter to the calibrated sensor data. This paper proposes an innovative method that unifies the steps of calibration and filtering. In the innovative method, the calibration and filtering are simultaneously performed. The Kalman Filter is applied directly on raw sensor data and the thermal calibration process is performed during the filtering step.
The main difference with respect to the standard method is the MEMS gyro bias time model. In the standard algorithm, the MEMS gyro bias is modelled as a zero-mean Gaussian noise. In the innovative method, it is modelled as the composition of two terms: a temperature dependent component and a stochastic component, which is modelled as a zero-mean Gaussian noise. The temperature dependent component can be obtained by using Artificial Neural Networks to estimate the partial derivative of MEMS gyro bias with respect to temperature and the temperature sensor output to evaluate the partial derivative of temperature with respect to time. We have chosen to use Artificial Neural Networks, since MEMS gyro bias can have some local abrupt change of trend with temperature. Artificial Neural Networks are self-adaptive in constructing a mathematical model after several repetitive learning and testing phases. Therefore, they have the advantages of performing non-linear fitting, regardless of the mathematical model of the sensor and various non-linear factors.
The experimental tests have been performed by using the AHRS Axitude AX1-[ ]TM. This device, intended for General Aviation applications, such as advanced flight displays and autopilots, includes:
• Triaxial accelerometer sensor;
• Triaxial gyroscope sensor;
• Triaxial magnetometer sensor;
• Temperature sensor.
The adopted gyroscopes are MEMS ADXRS646TM gyros by Analog Devices Inc.TM, while the adopted accelerometers are the MS8010TM accelerometers by ColybrisTM.
To test our proposed Kalman Filter algorithm, the AHRS has been installed in a car and an extensive test campaign has been executed to acquire significant experimental data. This innovative algorithm has several advantages: it requires a simplified hardware configuration, since thermal calibration and filtering are simultaneously performed in the DSP and it allows to save time, avoid expensive phases of testing (considering test equipment also) and decrease the man-hours intended for the calibration process.