submit a 1500 words paper on the topic Analysis of Logged OBDII Data. The rest of the analysis follows the method of Holmén & Niemeier (1998) whereby the logged data is separated into events. Each event was determined to last for one linearised sample interval (1 second).

Hi, need to submit a 1500 words paper on the topic Analysis of Logged OBDII Data. The rest of the analysis follows the method of Holmén & Niemeier (1998) whereby the logged data is separated into events. Each event was determined to last for one linearised sample interval (1 second).

Events are separated firstly into the cruise, positive acceleration, and negative acceleration. A cruising event is defined as occurring when the vehicle is non-stationary and the acceleration is below a set threshold. This was set to 0.5 ms-2 which provided a good balance of cruise/acceleration events in the datasets considered.

Information on gear selection is not directly available via OBDII PIDs. It is possible however consider the ratio of vehicle speed to engine speed. There are however three potential unknowns required in addition to deriving the gear result. These are gear ratios, differential ratio (for rear-wheel drive), and tire radius. Even with a priori knowledge of a given car gearbox/differential, there is no way of deriving tire radius as these may be changed as aftermarket items. A more generic solution to determining gear ratio is thus preferential.

It is possible to use a cluster type analysis in order to derive the current gear. An analysis of the distribution of vehicle to engine speed ratio will yield peaks at each valid gear. There will be some noise associated with actions during clutch depression (e.g. revving, coasting) or wheel spinning caused by traction loss.

In order to establish gear positions a normalized distribution of vehicle to the engine, the speed ratio was calculated (200 bin histogram). Calculating the ratio this way avoids the divide by zero that occurs when the vehicle is stationary. A threshold was then used to effectively mask the random noise which is spread fairly across the distribution. Any value below this threshold is zeroed. Experimentally, a threshold value of 0.05 worked well for this purpose. The resulting maxima were then detected for a significant change in the first-order differential of the data.