Today every new car has an OBDII (On Board Diagnostic II) port that can be used to retrieve vehicle diagnostic data using an ELM327 or STN1110 chip. This microcontroller can be used to determine the currently measured parameters of the vehicle, such as speed, engine and water temperature, battery charge level, and error codes for fault detection. Our research aimed at developing an application and an algorithm for limited HW resources that performs the relevant analysis of the collected data and produces statistics on whether the currently measured value is within the suitable range. Since the algorithm is executed and the data is stored on a mobile phone, it is impossible to store and analyze all measured values. By examining the diferent readings (if they are alarming several times in a row) it would be possible to warn the user that there may be a problem with the vehicle. By monitoring the data, it would be possible to reduce the probability of major faults occurring and provide information about the occurrence of the fault.
On-board diagnostics (OBD) is an automotive term that refers to the self-diagnosis
and reporting capability[
Copyright © 2020 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
The Onboard Diagnostic System (called OBDI) became obligatory in the USA in 1988. The technical specifications were defined by SAE (Society of Automobile Engineers) standards and recommendations. In 1994 OBDI was replaced by OBDII, and from 1996 it was also mandatory for diesel vehicles.
EOBD is the European equivalent of OBDII, which had to be introduced in the member states of the European Union by Directive 98/69/ EC.
A large amount of diagnostic information about the car can be obtained via OBD. To analyze this information, a simple and cheap ELM327 chip was used based on OBD interfaces (Figure 1), which is powered by the Microchip Technology PIC18F2480 Micro Controller. Newer devices use the STN1110 chip, which is fully compatible with ELM. ELM is a preprogrammed microcontroller, and the ELM327 Command Protocol is one of the most popular PC -OBD interfaces.
Many applications process OBD signals and have used Android mobile phones
for that[
Some researchers are also working on automatic error detection. Most of them
look at a single data series. In particular, cases when data that are outside the
expected values are reported to the user as fail[
In contrast, our research aimed at the development of a fully automatic fail detection system. The detection was based on rare data. The good vehicle works with typical values for a long time. During this time the correct working values were recorded, but a failed vehicle usually showed values that were not usual. Our algorithm tries to find automatically the rare data and the parameters indicating the failure.
An Android application was written that communicates with the ELM interface via Bluetooth. On the main screen, you can see the data of the real-time measurement (see Figure 2). Benchmarks can also be performed like Engine Load, Throttle Position, Engine Coolant Temperature, Air Intake Temperature, etc.
Under settings (Figure 3) further details can be set, e.g. the parameters to be measured. You can do this under OBD commands. Under OBD protocol one can select the most frequently used protocols. Supported protocols: • SAE_J1850_PWM • SAE_J1850_VPW • ISO_9141_2 • ISO_14230_4_KWP • ISO_14230_4_KWP_FAST • ISO_15765_4_CAN(_B, _C,_D) • SAE_J1939_CAN
The ratio of successful communication of interactions between interface and Android phone was investigated. As it can be seen in Table 1, that below 800ms too much data are lost. The conclusion is that we can communicate enough data with a density of 1 second.
Data collection on mobile phones is fast enough, but still it is impossible to process this amount of data in the long run on the resources of the phones. However, communication interval 100 ms following section, we recommend a procedure that does not require data to be recorded but uses more recent data to detect errors more accurately.
The data processing is performed with a confidence interval analysis. The
conifdence ellipsoid is calculated for the data with a normal distribution [
To calculate the Confidence Ellipsoid, the sum of the measured data ( ∑︀ 1), a sum of the square of the data (∑︀ 12) and a sum of the product (∑︀ 1 2) must be collected. This information can be calculated with the following methods, where the new data is 1 ( +1) and 2( +1): ︁∑ 1( ) = ∑︁
1( ) + 1( +1)
The eigenvalues and eigenvectors determine the confidence ellipsoids for the normal values with a diferent confidence level. To calculate the confidence interval of the data, the data was transformed into a normal distribution with expected values of zero. The transformation whitening matrix is = √ , where is a diagonal matrix of eigenvalues and the matrix is the one whose columns are the corresponding right eigenvectors of the covariance matrix. This was illustrated for two measured data sets in Figure 4.
This matrix can be calculated from previous information. If the following condition applies to the new pair of values: considered to be not rare. ︁( 1( )
, 2( ))︁ , then the measured data is ⃦ ⃦ ⃦ ⃦ −1 ︂( (︁ 1( ) , 2( ))︁ − ︂( ∑︀ 1 ∑︀ 2 )︂ , ⃦ ⃦ ⃦⃦ > , where is the confidence level.
If the new values are rare, the users can be warned.
After a certain large number of warnings, the application displays errors. In other cases, the new values are inserted into the previously calculated information to improve the statistics of the rare values.
We have completed further measurements with data such as Vehicle Speed, Engine RPM, Engine Coolant Temperature, Fuel Pressure, Engine Load, Throttle Position, and Air/Fuel Ratio.
We have generated some faulty operation, for example, a bad cooling system, an error in the power source or a poor fuel supply.
In these measurements we found that although the data showed mean values, when we looked at 2 data series we could already find pairs of variables where the data were outside the expected range.
In one case the power source for the cooling system was switched of. The measured data of the cooling water and engine compartment temperatures are shown in Figure 5. The cooling water temperature is on the horizontal axis, while the engine compartment temperatures are on the vertical axis. The green dots illustrate the proper operation and the red dots are measured under the cooling failure system. The values for both sets are in the normal range, but in the worst cases they will be shifted to the warmer range. It can be seen that the majority of the values measured for a defective vehicle were outside the 95% confidence interval, while in the good case they are mostly inside.
The other faulty cases had similar results. Unfortunately, it is not trivial which of the two data sets are useful for monitoring. It is a good idea to keep an eye on all of them, especially when other untested bugs are likely to be helpful.
We have developed a system that is capable to detect rare data. If it appears, the users can be alerted. In the future, we would like to expand the application. We also plan to generate more errors in cars and investigate whether higher dimensions lead to better results. The long-term goal is to create a uniform database for comparing vehicles based on these results.
Acknowledgements. This research was supported by the projects “Extending the activities of the HU-MATHS-IN Hungarian Industrial and Innovation Mathematical Service Network” EFOP-3.6.2-16-2017-00015, the János Bolyai Research Scholarship of the Hungarian Academy of Sciences, and the Unkp-19-4-Bolyai+ New National Excellence Program of the Ministry of Human Capacities. Special thanks to Tamás Radványi for the implementation and running the related computer programs, who was supported by the “Integrated program for training new generation of scientists in the fields of computer science”, EFOP-3.6.3-VEKOP-162017-0002.