| dc.description.abstract | In this work, theoretical, hardware and software implementation for an embedded inertial measurement unit that fuses the data signal from 4 redundant sets of low-cost tri-axial accelerometer/gyroscope inertial sensors to create an output that is better than each of the individual sensors in the cluster.
It was shown that the sensor placement is an important consideration for the accelerometer ouput, which is dependent on the displacement distance from the desired measurement point and angular velocity and acceleration, however, by placing the four tri-axial accelerometer in certain geometric configuration a gyroscope free inertial measurement unit can be created. Four feasible sensor configuration were explored analytically and the best configuration among the four were picked. Noise variance for acceleration measurement were reduced to a quarter compared to a single accelerometer with no filters. While the gyroscope free IMU is capable of outputting rotational acceleration and velocity, quality of the signal is the dependent on the cube size. The maximum practical cube size of $4.5\times 4.5\times 4.5cm^3$ made the angular rate and acceleration very noisy due to bad noise to signal ratio. However, the relation between acceleration measurements and angular rate and acceleration were instead used in an Unscented Kalman Filter to further filter out noise on both gyroscope and accelerometer data.
An indirect Kalman Filter were designed to combine GNSS data with the integrated inertial sensor measurement to prevent the error in velocity, orientation and position estimates to grow without bounds, resulting a complete navigation solution. The indirect Kalman filters main advantage is it easy to in-cooperate sensors with very different update rates, such as 100HZ for the accelerometer and 5HZ GNSS unit.
In this work, the hardware for the suggested were designed and manufactured for use in formula student racing car. It consist of a stack of two circuit board with 4 sets of accelerometer/gyroscope sensors in a cubical configuration, a GNSS unit that supports both Glonass and GPS, and an Cortex-M4 processing unit from Atmel. The device communicates over both CAN and an USB interface, and is designed to be powered by 19-28V input on the main power input, or alternatively, 5V from the USB port. Final hardware has been concluded to work properly, as we are able to both read and output sensor data. Stability test of hardware have been conducted, where the device have been left to read data from sensors and output them on to the CAN-bus continuously for 24 hours with no failure detected.
The software is built on FreeRTos, an open source real time kernel that provides several tools that simplifies and increases modularity of code. Low-end drivers are based on the Revolves NTNUs own internal code base, to which the author have contributed. Device drivers for MPU9250 and NVC08-CSM, which are based on the low-end drivers, have been developed. | en |
