Data Analysis Fundamentals: Filtering and Offsetting Data
Watch This Course
$129 USD $64.50 USD
-OR-
Or
8 easy payments of only $8.06
Instant access. Easy checkout. No fees.
Filtering and Offsetting Data
04.30
| 00:00 | - From time to time, you'll end up with data that in its raw form, is too noisy to be used for analysis. |
| 00:07 | This could be caused by using a sample rate for a channel that's too high. |
| 00:11 | Or alternatively it could be a result of vibration from a chassis being transmitted into a G force sensor or a damper position sensor. |
| 00:21 | For example, let's look at the data from an engine speed sensor which has been sampled at a higher rate than what the ECU was transmitting the data at. |
| 00:30 | In this case the logging rate was 100 Hz whereas we would typically sample this channel at 20 Hz. |
| 00:38 | In particular we can see the square pattern generated when multiple samples are taken at what is effectively the same RPM because the channel hasn't been updated by the ECU. |
| 00:50 | Fortunately you'll find that most data analysis packages make it easy for us to filter raw data to remove or reduce the effects of noise. |
| 00:58 | Or in our case, excess data samples. |
| 01:01 | There are a variety of different options when it comes to filters with the most common being a moving average which filters the data using samples either side of the current sample. |
| 01:12 | You will usually have the ability to define how many samples will be used when filtering in order to produce data that's meaningful for analysis. |
| 01:21 | The downside of any filtering is that it does reduce the amount of detail that will be displayed in a channel and this can mask issues that we actually want to analyse. |
| 01:32 | For example excess filtering of a wheel speed input may mask or reduce the obvious signs of a wheel lockup under brakes. |
| 01:39 | For this reason filters should be used sparingly and with the bare minimum amount of filtering possible in order to provide useful data. |
| 01:48 | Along with filtering, we may also need to offset a particular channel. |
| 01:52 | There's a variety of reasons we may need to offset a channel however a good example would be a steering angle sensor that wasn't correctly zeroed before the session. |
| 02:01 | Or perhaps a steering angle sensor that slipped or moved between sessions. |
| 02:07 | This will result in data being offset so that the steering angle won't show zero when the car is travelling in a straight line. |
| 02:14 | While this might not seem to be a big concern, if you're performing some more advanced analysis that requires the steering angle to be accurate, this can affect your results. |
| 02:23 | Fortunately it's not particularly difficult to correct this sort of issue inside the analysis package. |
| 02:30 | All we need to do is select the part of the track where we know the car is in fact travelling in a straight line. |
| 02:36 | Once we've done this we will be able to note what the actual offset is and then we can apply the offset to the channel in order to correct the steering angle channel to show zero when the car's travelling in a straight line. |
| 02:48 | Another aspect which may require an offset is when you have lap data generated using a timing beacon or even a GPS beacon where the beacon location is different between the two sessions. |
| 03:01 | This creates an offset in the data where it appears the same actions by the driver are occurring at different points on track. |
| 03:08 | Understandably this makes any meaningful analysis very tricky. |
| 03:13 | Fortunately this can be corrected easily enough by offsetting the beacon location. |
| 03:18 | To start this process, we need a plot of speed versus distance and overlay a lap from each session. |
| 03:26 | Now we need to choose a location on each lap which is likely to be the same. |
| 03:31 | A good option is to choose the slowest point on a specific corner such as a hairpin. |
| 03:37 | The difference between the two points is equal to the distance we need to offset the beacon. |
| 03:43 | This is an imprecise technique unfortunately since there's no guarantee that the two drivers will drive the exact same line through every corner. |
| 03:51 | So there's some amount of compromise here. |
| 03:54 | Another common option is to use a braking point at the end of a long straight but personally I feel this actually opens us up to more potential error as there can be a significant difference in the braking points between two drivers. |
| 04:08 | However the point where the drivers reach minimum speed in a corner is likely to be more consistent. |
| 04:14 | Probably the best option if the car is equipped with a vertical G force sensor or damper position sensors, then you can often use a prominent bump at a specific point on the racetrack to very accurately offset the beacons. |
