Validation of Friction Estimating System
Erik Andreasson
Automotive Engineering, bachelor's level 2017
Luleå University of Technology
Validation of Friction Estimating System
Erik Andreasson
A thesis submitted in partial fulfillment for the
degree of B.Sc in Automotive Engineering
at
Luleå University of Technology
Supervisor: Assistant Lecturer Johan Casselgren
Abstract
This thesis presents an evaluation of NIRA Dynamic’s Tyre Grip Indicator system. The evaluation involves performance tests, accuracy tests and reliability test. The Tyre Grip Indicator system (TGI), uses the vehicle’s internal sensors to estimate to friction of the road surface. Commonly and almost exclusively, analyzing friction requires some sort of external or additional hardware. The innovative concept of the TGI is that the additional, often bulky, hardware now is obsolete.
The data used for this thesis was collected from tests in cooperation with the Norwegian Public Roads Administration (NPRA). All friction values were collected around the vicinity of Björli Norway during the period of 6th to 9th of February 2017 which in turn entails winter conditions. All evaluation is made in comparison to NPRA’s two standardized systems ROAR and ViaFriction.
Acknowlagement
This thesis was produced at Luleå Technical University as the closure of my studies of the program B.Sc Automotive Engineering. My time at the university has been motivated by my dream of a future within the business of automotive engineering. The knowledge I’ve gained and the skills that I’ve polished would not have been possible without the great teachers of Luleå Technical University.
I wish to humbly express my gratitude towards my supervisor Johan Casselgren. Both his knowledge and his positive attitude has been an extraordinary boost, not only during the writing of my thesis, but all throughout the length of my education. He inspires to chase dreams and achievements and I will continue to do so after my studies are concluded.
I do also want to thank Torgeir Vaa and all his colleagues from NPRA for hosting the tests in Björli. Their invitation and cooperation made this thesis possible. And of course, thanks to NIRA Dynamics for giving me the opportunity to work with their product.
Contents
1 Introduction 1 1.1 Background . . . 1 1.2 Objectives . . . 2 1.3 Thesis Boundaries . . . 3 2 Theory 5 2.1 Friction . . . 5 2.2 Traditional Measurements . . . 5 2.3 Evaluation technique . . . 6 2.3.1 Average (AVE) . . . 6 2.3.2 Standard Deviation (STD) . . . 72.3.3 Root Mean Square Error (RMSE) . . . 7
3 Equipment 9 3.1 NIRA . . . 9 3.1.1 Dongle . . . 9 3.1.2 Harness . . . 9 3.2 NPRA . . . 10 3.2.1 OSCAR . . . 10 3.2.2 ViaFriction . . . 11 3.2.3 ROAR MkIII . . . 11
3.3 Luleå University of Technology . . . 12
4.1 Pre-Bjorli . . . 13 4.2 Bjorli . . . 13 4.2.1 Circuit . . . 13 4.2.2 Road . . . 15 5 Methodology 17 5.1 Matlab . . . 17 5.2 Handling files . . . 17 5.3 Data Management . . . 18 5.4 Analyze . . . 18
5.5 Visualization and Data Representation . . . 18
6 Results 19 6.1 Airstrip . . . 22 6.2 Motocross circuit . . . 23 6.3 Road to complex . . . 24 6.4 Rånå . . . 25 7 Conclusion 27 Bibliography 29
Chapter
1
Introduction
1.1
Background
Most conventional methods to measure road friction depends on either physical och optical sensors. This often implies some issues and inconveniences. First of all, all systems are not calibrated to one and the same standard. Secondly, these sensors demands a significant amount of space. Thirdly, the cumbersome systems often require service and trained personnel to operate it. NIRA Dynamics’ TGI-system aims to eliminate the problem of both price, girth and availability to create a system that can be used for many vehicles simultaneously. By uploading the data from all vehicles to a collective database, NIRA Dynamics are able to plot friction values on a regular road map. Only a handful of vehicles connected to the up-link are able to map large areas with continuous updating. The "cloud", in a larger scale, could help road maintenance to focus on critical areas where the friction values are below what is deemed safe. This means that specific road sections can receive a relevant form of deicing, without having to treat entire area. This can lower the acidification of roadsides and also reduce the cost of road maintenance. Also it gives the driver a chance to be aware of impending situations that may require the driver to take action in advance.
Figure 1.1: NIRA’s online database displaying data collected March 16th 2017
Displayed in the Figure ( 1.1) are data collected by several vehicles located in Stockholm during March 16th 2017. It shows primarily green areas, indicating clear and dry high-friction sections of road. The blue areas indicate water on the surface. The areas of interest are indicated by the colour turquoise (snow) and pink (ice). These are the road sections in need of attention. Both by road users and by road entrepreneurs.
1.2
Objectives
The main objective is to analyze the performance of the TGI-system and to determine if it is a viable option that provides reliable data. The data will be compared to measurements calibrated to the Norwegian standard OSCAR. The standard will provide reliable data ideal for comparison. Also, the data from the TGI-system will be evaluated for consistency. For credible results, the data must be repeatable for different kinds of driving style.
The thesis will resolve the following objectives:
1. Is the data comparable to the set standard? 2. If not, how does it differ?
3. Is the results repeatable during different driving styles on the same surface? 4. If not, what makes it differ and how much does it deviate?
1.3
Thesis Boundaries
All measurements has been performed in winter climate dominated by snow- and ice covered surfaces. Both public and private roads has been used for the tests, but due to the requirement of occasional flamboyant driving, objective #3 has been exclusively performed on a closed track. More tests concerning high-friction surfaces should be performed to accurately evaluate the performance of the system. Even though the system is primarily created to evaluate low-friction surfaces, high-friction tests would be an interesting addition for the matter of validating its performance.
Chapter
2
Theory
2.1
Friction
Friction can be found everywhere. The force of friction resists lateral relative motion between two objects. It converts kinetic energy to heat. Most commonly, friction is viewed as an unnecessary waste of energy and therefor minimizing it is a primary objective. However in the case of friction between the tyre and the road surface, higher friction is desired. The friction force is a product of the normal force exerted by each object on to the other, directed perpendicular to the surfaces and the coefficient of friction.
The coefficient of friction is usually referred to as the Greek letter µ. It is determined by the combination of materials in contact, where the combination results in a dimensionless scalar. Friction can be divided up in to separate entities, static friction and kinetic friction. Static friction is, as a rule of thumb, greater than kinetic friction. Because of this, in the case of friction between tyre and road, static friction is desired to ensure optimal control of the vehicle. [1]
2.2
Traditional Measurements
Traditional friction measurements are dependant of a friction wheel that are subjected to the ground. Upon contact with the ground, the wheel is set in rotation. The most common of frictionwheelsystems are connected to a break which applies a negative momentum. The wheelspeed is monitored while the break is engaged. Upon wheelslip, the wheelspeedsensor records the slip. At that instant the force applied to the break is also recorded. When the applied force is known, along with the systems own parameters e.g wheel diameter and braking disk
Figure 2.1: Friction graph displaying static and dynamic friction
diameter, enables the system to calculate the friction coefficient for that particular instant.
2.3
Evaluation technique
To evaluate the results, techniques such as average, standard deviation and root mean square error are implemented. These relatively simple techniques enables analysis considering how different variables might effect the result.
2.3.1
Average (AVE)
The average is the quickest and most simple way to compare different results. The average is the sum of any given number of measurements, divided by the number of measurements.
Average = 1 n n X i=1 xi (2.1)
By comparing averages, it is possible to make a quick estimate how the two measurements correlate. [2]
2.3.2
Standard Deviation (STD)
Standard deviation, or STD, is a way to quantify the amount of variation of any given set of measurements. A low value STD indicates a low spread of the data points. Standard deviation is most commonly referred to as the Greek letter σ. [3]
σ = v u u t 1 n n X i=1 (xi− a)2 where a = 1 n n X i=1 xi (2.2)
2.3.3
Root Mean Square Error (RMSE)
Root mean square error is yet another tool used to measure difference between values. It is a way to evaluate the accuracy of measurements, to compare forecasting errors of different models for a particular data and not between data sets, as it is scale-dependant. RMSE is the square root of the average of all squared errors, thus RMSE confounds information containing the average error with information containing variation in the errors. The effect of each error on RMSE is proportional to the size of the squared error. Thus larger errors have a disproportionately larger effect on the RMSE. [4]
RM SE =
s Pt=1
n ( ˆyt− yt)2
Chapter
3
Equipment
3.1
NIRA
NIRA Dynamics based in Linköping, provides both the hardware and software to be tested against the norm. The hardware comes in two different forms, as a dongle and also as a harness-kit. Both were installed into the subject vehicle in advance to the main testing period.
3.1.1
Dongle
The dongle NIRA provided for the evaluation measures roughly 5.5x6.5x3 cm with a 3D-printed casing. It uses the OBD-connector to access the vehicle’s internal signals. The signals are processed and results in a value of the calculated friction. The dongle contains hardware to connect to the surrounding cellular network which allows information upload to the cloud. The sampling speed is roughly 1/3 Hz which provides sufficient number of data points to make an estimate of the condition of the road. However, for more accurate comparison to the norm, more data points are needed. Due to hardware design limitations, the dongle is not able to collect and upload data simultaneously.
3.1.2
Harness
The harness is NIRA’s own development hardware which plugs in inside the engine compartment. It also is installed with its own GPS-puck which provides good and accurate location data. The harness has better resolution than the dongle (20Hz). The data is encrypted and written onto a small memory card. All data is sent to NIRA for processing.
3.2
NPRA
NPRA or the Norwegian Public Roads Administration, hosted a couple of days on location in Norway meant for collecting measurements. On site, NPRA had provided two of the standardized systems, i.e ROAR MkIII and ViaFriction, for comparison.
3.2.1
OSCAR
The OSCAR-system was not present during the tests but is still relevant considering the additional systems are calibrated to the OSCAR. Because of its singularity, it provides an unprecedented standard for all other systems. OSCAR uses a traditional wheel that measures slip where both locked- and variable slip-rate are used. The vehicle carry a water tank, which innards are applied onto the road surface in case of dry conditions. This minimize the wear and tear of the measurement wheel. Since it is the only one of it’s kind, it is primarily used for reference measurements and in association with research and development. [5]
Figure 3.1: The OSCAR measurement vehicle
3.2.2
ViaFriction
Via Friction is a relative lightweight measurementsystem that were present during the testing. It utilizes the traditional frictionwheel for the measurements. Because of its low weight, it suffers from different problems e.i limitations for measurements on high µ, and sensitive result on rough surfaces. Therefor, the data collected from ViaFriction is primarily used as an emergency reference for particular cases where other data may be questionable.
Figure 3.2: ViaFriction
3.2.3
ROAR MkIII
ROAR is the other measurementsystems that were present during the test, and is also the paramount standard to which the data of NIRA’s system will be compared to. ROAR, in the same way as OSCAR, uses a wheel with both locked and variable slip-rate to measure the surface friction. It also has a tank to apply water to the road surface if necessary. Because it weight is considerably greater than the ViaFriction, it does not suffer as much from the rough surfaces. NPRA uses five ROAR-systems, one in each region, all of which are calibrated regularly to match OSCAR.
Figure 3.3: ROAR MkV
3.3
Luleå University of Technology
The TGI-system is tailored to and required a specific type of vehicle for it to function. This vehicle was provided by Luleå University of Technology and was a front wheel drive Volvo V70 -16. All tests were preformed with the same vehicle,
Figure 3.4: The Volvo V70 -16 provided by Luleå University of Technology
Chapter
4
Measurements
4.1
Pre-Bjorli
All comparative measurements were made on location in Bjorli, Norway, during week six of 2017. Prior to this hardware were installed into the subject-vehicle. This enabled the opportunity to make reliability tests before the comparison evaluation. The reliability tests were executed during the two weeks prior to the Bjorli-excursion. Roughly 800 km of everyday driving on public roads gave a fair representation of on-raod reliability and also exposed potential issues that could turn problematic during the vital part of the testing.
4.2
Bjorli
During the actual testing in Bjorli, data for around 3400 km was logged. Both using the dongle and the harness. Due to the objective to validate the system’s performance, many different scenarios needed to be explored. The versatile surroundings of Bjorli gave opportunity for a broad spectrum of tests.
4.2.1
Circuit
For testing how aggressive and passive driving affects the result, and also preforming basic control tests on straight roads at different speeds, the Airstripcomplex in Bjorli was utilized. The complex contains both the Airstrip and the Motocross circuit. This provides a suitable environment for controlled tests where public traffic does not affect the surface conditions. Surface conditions varied between hard packed snow, ice and graveled ice.
Airstrip
The airstrip measured roughly 800 m and had two lanes. The first lane was sheet ice with spots of snow, while the second lane had the same surface underneath but gravel ontop.
Figure 4.1: Satellite view of the Airstrip [6]
Motocross circuit
The Bjorli Motocross circuit gave the opportunity to push the test vehicle to the grip limit. The surface was bumpy with hard packed snow. The tight corners and steep inclines made forces in all axis accessible.
Figure 4.2: Satellite view of the Motocross circuit [6]
4.2.2
Road
The roads in and around Bjorli supplied excellent conditions for tests of all different scenarios. Low friction, high friction, steep inclines etc. The main function of the system is to detect when the roadsurface is dangerously slippery in normal driving conditions, hence its very important to make reliabiltytests in such environment. Therefor the bulk of all measurements were made on "normal" winter roads to ensure good performance and good reliability.
Road to Complex
The main road leading up to the complex was covered in ice. The road had a slight camber on it which in combination to the slippery surface meant that at some points, the vehicle’s stabilization systems were engaged while driving in a straight line.
Figure 4.3: Satellite view of the road to the Complex [6]
Rånå
To differ from the main road, the road to Rånå was not completely covered with ice. Because of its westerly outlook, the sun kept the road clear form ice and snow for the majority of the section. This lead to a road combination of dry and wet asphalt, snow and ice.
Figure 4.4: Satellite view of the road to Rånå [6]
Chapter
5
Methodology
The following chapter describes the work-flow of the project. How the data goes from being collected to analyzed and finally visualized. The process started while on site in Norway. The importance of gathering data in an organized manner proved helpful later on in the project. Post-Norway, the data files needed processing to be readable. Once the data files are presentable, analyzing and visualization are possible.
5.1
Matlab
All calculations and comparisons are made using Matlab. The created script utilizes functions for different purposes. The first of which compares the collected datapoints to the reference. Because the reference creates considerably more datapoints, the closest point to the TGI-point needs to be determined for an as accurate result as possible. Their GPS-location is analyzed and the closest point is then selected for comparison.
5.2
Handling files
To properly handle the files, all files were named and categorized. The referencedata from NPRA was labeled with a time stamp and location of the test i.e whether it was the airstrip or any of the other road sections. NPRA’s data files contained alot more information than the relevant GPS-locations and friction value required. Therefor, most of the actual data contained within the file has not been taken into account.
NIRA contained friction for each of the front wheels, speed, GPS-location and a time stamp. The data was contained within four files, representing the four days of testing.
5.3
Data Management
In order to properly compare the values from the TGI to the reference, certain filtration must be performed. Because of the difference in resolution between the TGI and the reference, certain data points must be specified to acquire a fair representation of data from both. Therefor the Matlab-script isolates the closest GPS-position of the ROAR-data to any given TGI-point. To exclude any rouge data points with a non-plausible GPS-location, the limit for the distance between points are set to 5 meters. If the requirement is not fulfilled, the point will be discarded. This is a tool to ensure that the two measurements are as close to each other as possible. Even though both the measurement are made on the same surface with only a couple of seconds interval, exact placement of the measurement wheels remain a small possible error source. The data management results in two equal sized arrays of data, ready for comparison.
5.4
Analyze
Once all the data files has been managed, they are ready for analysis. All files are evaluated individually in case of corrupt data. If the results are unrealistic, further data management may be needed. If not, the data is analyzed and evaluated. The Matlab-script calculates mean values, STD and RMSE values. It also plots the compared friction values for each run.
5.5
Visualization and Data Representation
The Matlab-script used for the analysis also visualize the results by plotting the TGI-values next to the reference. The basic comparison plot contains three subplots. One for the GPS location, the second for the actual friction values for each separate data points and the last one for the distribution of the friction.
Chapter
6
Results
In the following chapter the results are presented by graphs and tables containing explanatory data. Table 6.1 shows maximum, average and minimum values for both TGI and ROAR from the twenty one runs taken into consideration. It also contains the calculated RMSE-value for each run. Studying the RMSE-values, it quickly becomes apparent that high friction values results in a substantial error, while error for low friction values are tiny. The average RMSE value is calculated to 4.59e−2.
Table 6.2 displays different sets of data, where ROAR and ViaFriction are compared. These eighteen data sets are not the same used in the previous comparison due to long intervals between measurements. All compared data sets are within reasonable time frame. The average RMSE value is calculated to 4.62e−2.
Table 6.1: Maximum and minimum values for TGI and ROAR for each run
TGI ROAR
Run Max Ave Min Max Ave Min RMSE Section 1 0.307 0.292 0.259 0.37 0.26 0.11 0.0271 2 0.376 0.299 0.261 0.43 0.33 0.1 0.0296 3 0.294 0.241 0.224 0.37 0.29 0.16 0.0514 4 0.379 0.294 0.233 0.37 0.29 0.17 0.0011 Airstrip 5 0.342 0.284 0.259 0.58 0.27 0.1 0.0118 6 0.432 0.412 0.393 0.65 0.47 0.34 0.0619 7 0.32 0.186 0.124 0.32 0.23 0.19 0.0471 8 0.306 0.273 0.234 0.56 0.29 0.18 0.0124 9 0.363 0.33 0.311 0.36 0.30 0.24 0.0281 Motocross 10 0.357 0.291 0.23 0.36 0.24 0.1 0.0491 Circuit 11 0.324 0.298 0.265 0.35 0.27 0.1 0.0238 12 0.375 0.266 0.174 0.49 0.26 0.1 0.0055 13 0.676 0.662 0.601 0.71 0.48 0.26 0.184 Road to 14 0.318 0.252 0.218 0.35 0.22 0.12 0.0363 Complex 15 0.307 0.276 0.25 0.37 0.21 0.17 0.0693 16 0.282 0.245 0.205 0.34 0.21 0.17 0.0387 17 0.385 0.358 0.311 0.35 0.27 0.17 0.0839 18 0.421 0.265 0.205 0.53 0.24 0.15 0.0274 Road to 19 0.359 0.245 0.193 0.5 0.23 0.17 0.0141 Rånå 20 0.402 0.319 0.255 0.69 0.44 0.29 0.1223 21 0.537 0.524 0.517 0.66 0.49 0.27 0.0389 Brøste
Note: Measurements to Brøste is not included in the results due to only one test run Table 6.2: Maximum and minimum values for TGI and ROAR for each run
ROAR ViaFriction
Run Max Ave Min Max Ave Min RMSE 1 0.47 0.26 0.1 0.462 0.302 0.08 0.0381 2 0.45 0.27 0.1 0.471 0.284 0.08 0.0194 3 0.39 0.29 0.11 0.409 0.333 0.203 0.0414 4 0.42 0.29 0.1 0.419 0.338 0.192 0.0458 5 0.6 0.25 0.1 0.469 0.232 0.123 0.0226 6 0.77 0.48 0.11 0.822 0.433 0.19 0.0432 7 0.43 0.25 0.1 0.457 0.242 0.059 0.0111 8 0.42 0.25 0.1 0.486 0.242 0.055 0.0077 9 0.39 0.29 0.1 0.42 0.317 0.141 0.027 10 0.42 0.28 0.1 0.402 0.323 0.17 0.0403 11 0.62 0.24 0.1 0.751 0.224 0.11 0.0191 12 0.74 0.46 0.1 0.808 0.231 0.104 0.2277 13 0.67 0.28 0.1 0.681 0.293 0.102 0.0087 14 0.69 0.31 0.1 0.62 0.325 0.1 0.0191 15 0.59 0.25 0.11 0.772 0.398 0.28 0.1525 16 0.38 0.29 0.1 0.425 0.322 0.134 0.0297 17 0.39 0.29 0.1 0.495 0.228 0.082 0.0587 18 0.39 0.24 0.1 0.488 0.221 0.071 0.0202 20
Figure 6.2: Correlation between TGI and ROAR, dependent on magnitude of µ
As seen in figure( 6.1), the majority of runs, the TGI has both a lower maximum and a higher minimum compared to ROAR. Due to the slower response for the TGI, this is to be expected. The measured value from the TGI is however still within the realm of the values taken from ROAR.
Window two of figure( 6.2) shows the average value of each run and compare the TGI to the ROAR. The first window displays the correlation between the two averages dependent on the magnitude of the friction. E.i the closer the each dot is to the diagonal line, the more they correlate. By studying the graph in question, the grouping of the dots shows better results for low friction values. Once the friction becomes greater than µ > 0.5, the values start to deviate.
6.1
Airstrip
Testing on the airstrip is presented by five separate runs. Figure ( 6.3) shows one of the five runs. First window plots the GPS coordinates, second window plots the friction value throughout the test run and the third window displays the distribution of friction values. In all graphs, blue and black represent the reference while TGI-values all are in orange.
Figure 6.3: Result illustration of tests performed on the Airstrip
The friction plot shows a well-correlated result where the TGI follows the reference through-out the run. However, there is a delay resulting in that quickly changing friction is overlooked. Therefor the distribution for the TGI is much more even hence it filters out most peek-values. By just visually analyzing the the plots, it would appear that the TGI figures out that the return run is made in the lane not graveled. The change of friction is not noted instantaneously, but is gradually settled on the correct level.
6.2
Motocross circuit
Figure ( 6.4) displays the result from the motocross circuit in the same manner as previously done for the Airstrip. Here the bumpy road conditions comes in to play. As seen in the second window, the blue line representing ROAR, is rather uneven. This is due to bouncing of the friction wheel. However, the measurement is still correlating to each other. This is one of four runs for the Motocross circuit.
6.3
Road to complex
The results from the tests performed on the road to the complex shows a rather clear change in surface conditions. Where the first section is graveled and the second is sheet ice, there is a noticeable difference between the two. ROAR picks up the difference right away while it takes a several meters for the TGI to fully compensate for the difference. Still, it does find the correct value eventually and returns a good result.
Figure 6.5: Result illustration of the tests performed on the Road to the Complex
6.4
Rånå
The road to Rånå proved difficult for the TGI to evaluate because of its rapidly changing surface conditions. The results for the early part of the section are credible. The surface is initially more or less homogeneous hence the good result. Once the subject vehicle reaches the colder part of the section containing ice, water and snow, difference appears. ROAR responds right away but the TGI does not get enough time to adapt. Therefor the latter part of the plot, the results differ more.
Chapter
7
Conclusion
Considering the simplicity of the hardware, the TGI performed rather well. The values are credible even though it misses out on quick changes of the surfaces conditions. While it might struggle with rapidly changing surface conditions, it does not get bothered by uneven roads. Where the ROAR system suffered from inaccurate values due to bouncing of the frictions wheel, the TGI remains stable throughout.
An area where the TGI struggles is when it comes to communicating to the cloud. While running in the vicinity of a city, the telecommunication works fairly well. But once driven rural, communication errors becomes a continuous issue.
During the drive to the testing area in Bjorli, the TGI was recording the entire way. The resulting data extracted from the journey was incomplete to such a degree that roughly half was missing due to communication error. Figure ( 7.1) shows the recorded GPS coordinates from the trip. Where there are unrealistically straight lines, is where the system loses connection.
Finally, to compare the correlation between TGI and ROAR, by also comparing ROAR to the ViaFriction. ROAR and ViaFriction are after all calibrated to each other, so they should give a good representation of an accurate result. Where the RMSE between the ViaFriction and ROAR gives an average of 4.62e−2, roughly 16 percent of the friction value, the TGI performed excellent with an average RMSE of 4.59e−2. So, on average, the TGI performed within the range of two systems calibrated to each other.
Bibliography
[1] A.Ruina, R.Pratap: Introduction to Statics and Dynamics, Oxford University Press, p. 713, 2002, ISBN: 9780534029296
[2] George E.P;Box, Time Series Analysis: Forecasting and Control (revised ed.), Holden-Day, 1976, ISBN: 0816211043.
[3] H.Walker: Studies in the History of the Statistical Method, Baltimore, MD: Williams & Wilkins Co, 1931, ISBN-13: 9780405066283
[4] R.Hyndman, A.Koehler, Another look at measures of forecast accuracy, International Journal of Forecasting. 22 (4), p.679–688, 2006.
[5] Statens Vegvesen, Vegdirektoratet, Publikasjonsekspedisjonen: Lærebok Drift og vedlikehold av veger. Postboks 8142 Dep 0033 OSLO, June 30th 2015, ISSN: 1893-1162.
