Development of a crash detection device and sensor position estimation

Abstract: The purpose of this thesis was to design a device that could estimate the position and velocity of a vehicle as well as record the data in the event of a crash using off-the-shelf components. To achieve this, the device has to be able to detect crashes. The device utilised measured acceleration and velocity difference during a time window of 100 ms to determine whether a frontal crash has occurred. To estimate the state of the vehicle a Kalman filter was implemented. The amount of false positive errors, i.e. scenarios where the device incorrectly detects a crash when no crash has occurred, was analysed to determine its relation to different parameters.  Depending on the sensor location, recorded accelerations will result in different crash profiles. Therefore, for the recorded crash accelerations to be of use, the sensor-to-vehicle position was estimated with a novel algorithm. The position of the centre of gravity when the vehicle is loaded and stationary is considered to be known while the sensor position is considered to be unknown. The estimation algorithm works by comparing velocities at the centre of gravity and sensor location and using the principle of relative velocities to establish the distance between the two points. The distance estimation is applied during constant turning in roundabouts in both physical tests and in simulated tests. The sensor-to-vehicle position was evaluated by comparing the mean and the standard deviation of the estimated distance during the manoeuvre. The construction of the device was successful in the sense that it is able to estimate the position and velocity during normal driving. If a crash occurred the device was able to detect it and started recording with a higher frequency than prior to the crash. However, at low velocities and at locations where the GPS-signal was obstructed the position estimation was not optimal.  The false positive analysis showed that in normal driving conditions, there will be no false positive errors if the acceleration threshold is set to greater than 1g with a velocity difference threshold corresponding to an average acceleration of 25% of the acceleration threshold. In the case when a harsh stop is performed, the abovementioned thresholds would still falsely detect a crash.  The sensor-to-vehicle position estimation in the physical real world test prove to be too inaccurate for the uses of crash recording. The simulated tests arrived at the same conclusion. However, tests with an additional GPS at the sensor location showed that pure longitudinal displacement could be estimated to a satisfactory degree but not pure lateral. Problems in the methodology and the limitations were analysed and discussed.