Research about Vehicle Faults Prediction System based on SVM
Qiu, Renxiang (2017-09-25)
Research about Vehicle Faults Prediction System based on SVM
Qiu, Renxiang
(25.09.2017)
Tätä artikkelia/julkaisua ei ole tallennettu UTUPubiin. Julkaisun tiedoissa voi kuitenkin olla linkki toisaalle tallennettuun artikkeliin / julkaisuun.
Turun yliopisto
Tiivistelmä
With the development of the Chinese economy, more and more families have their own cars; with the completion of China's infrastructure construction, the car is becoming one of the essential travel tools of many people. According to the report of the Chinese government, China's auto repair industry will grow by 10% per year And the report also shows that the market will increase from 500 billion yuan in 2014 to one trillion yuan in 2020.
Nowadays, the mainstream approach of vehicle fault detection is the combination of instrument detection and artificial diagnosis. But this can’t meet the requirements of the market. So this thesis proposes a fault prediction of vehicle system based on Support Vector Machine. This system can collect the data about the vehicle running status from On-board Diagnostics system and other sensors in the car. Then the system can predict possible failures of vehicles by analyzing a series of data. And the system also can diagnose the type of fault and alert drivers in time. So this system will play an important role in guaranteeing the safety of the cars and promoting the driver's efficiency
In this thesis I design of vehicle fault prediction system from two parts. First of all, the SVM classification algorithm is used to classify the fault types of the vehicle. In this thesis, the car running state is divided into five categories, engine failure, battery failure, cooling system failure, multiple types of failures and running in good condition. The SVM model is trained by using the one part of the data, and then use the remaining part of the data to verify the model. At the end of the experiment, I find that the classifier can basically distinguish the running state of the car that can give the driver an intuitive reminder. The other part of thesis is to predict the engine running state using regression algorithm of the support vector machine. The load characteristics of the engine and the engine temperatures are two main characteristics that the model need to predict. In this thesis, some information like engine intake air temperature, engine cooling system temperature and a series of indicators are extracted from the OBD system to build SVM temperature model. And the model uses those data to predict the engine temperature changes. The same time some info like the engine load, engine speed, car speed and other indicators are extracted from the sensor in the car to build SVM load model. And the SVM model uses those data to predict the load characteristics of the engine. In order to obtain the best prediction model, this thesis used the GA algorithm, grid search algorithm and PSO algorithm to find the optimal parameters. This thesis also compared the effects of various SVM Kernel function. This thesis also takes into account the effect of normalization on the data.
So this system based on the SVM has advantages like wide application range, simple data collection, the detection method is simple and quick, low cost and easy popularization. If the system can combined with some technologies like Cloud Computing and the Internet, it will have a brighter future.
Nowadays, the mainstream approach of vehicle fault detection is the combination of instrument detection and artificial diagnosis. But this can’t meet the requirements of the market. So this thesis proposes a fault prediction of vehicle system based on Support Vector Machine. This system can collect the data about the vehicle running status from On-board Diagnostics system and other sensors in the car. Then the system can predict possible failures of vehicles by analyzing a series of data. And the system also can diagnose the type of fault and alert drivers in time. So this system will play an important role in guaranteeing the safety of the cars and promoting the driver's efficiency
In this thesis I design of vehicle fault prediction system from two parts. First of all, the SVM classification algorithm is used to classify the fault types of the vehicle. In this thesis, the car running state is divided into five categories, engine failure, battery failure, cooling system failure, multiple types of failures and running in good condition. The SVM model is trained by using the one part of the data, and then use the remaining part of the data to verify the model. At the end of the experiment, I find that the classifier can basically distinguish the running state of the car that can give the driver an intuitive reminder. The other part of thesis is to predict the engine running state using regression algorithm of the support vector machine. The load characteristics of the engine and the engine temperatures are two main characteristics that the model need to predict. In this thesis, some information like engine intake air temperature, engine cooling system temperature and a series of indicators are extracted from the OBD system to build SVM temperature model. And the model uses those data to predict the engine temperature changes. The same time some info like the engine load, engine speed, car speed and other indicators are extracted from the sensor in the car to build SVM load model. And the SVM model uses those data to predict the load characteristics of the engine. In order to obtain the best prediction model, this thesis used the GA algorithm, grid search algorithm and PSO algorithm to find the optimal parameters. This thesis also compared the effects of various SVM Kernel function. This thesis also takes into account the effect of normalization on the data.
So this system based on the SVM has advantages like wide application range, simple data collection, the detection method is simple and quick, low cost and easy popularization. If the system can combined with some technologies like Cloud Computing and the Internet, it will have a brighter future.