Research On Fuzz Testing Technology Of Vehicle CAN Bus Security

Due to the increasing number of external communication interfaces of the vehicle,the communication safety of the CAN bus network in vehicle is constantly threatened.Fuzz testing technology can be used as a security testing method of CAN bus communication.It can monitor the abnormal state of the network and find the loopholes of the network security by sending random or mutated unexpected message data to the CAN communication network in the vehicle.At present,there is little research on the fuzz testing technology of the vehicle CAN bus.Most test cases are generated randomly,and the application layer protocol format of the CAN bus is not considered,which reduces the pass rate of CAN bus messages and affects the efficiency of fuzz testing.In order to solve the above problems,on the basis of analyzing the CAN bus application layer protocol,this paper proposes a CAN bus protocol field division algorithm based on the message bit flip rate characteristic and conditional random field,and proposes field-based weighting and improved Wasserstein generation adversarial network CAN bus fuzz testing method.Firstly,according to the development trend of vehicle networking and intelligence,the information security threats and attack methods of the connected vehicle are studied;the CAN communication network architecture and the CAN bus application layer protocol of the vehicle are analyzed;the vulnerability of the CAN bus communication is clarified,and the related security protection of the CAN bus is summarized.Secondly,to extract CAN bus application layer protocol field format,based on the analysis of the different field types of the message field of the CAN bus of the vehicle,a field division algorithm based on the characteristics of the message bit flip rate is proposed.According to the characteristics of the message bit flip rate,the field type and boundary are determined.For the commercial vehicle CAN bus application layer protocol based on SAE J1939,a field division algorithm based on conditional random field is proposed.The SAE J1939 protocol is used as a priori knowledge,and the conditional random field model is established by using the bit flip rate feature to divide the fields;in order to analyze the vehicle operating status represented by the signal fields,the method of correlation analysis between OBD diagnostic message data and signal field data is given;the above mentioned algorithm is verified by using the online published message dataset,simulated message dataset,and real vehicle message dataset,respectively.Thirdly,to improve the efficiency of CAN bus fuzz testing,based on the reverse analysis of the CAN bus message protocol,a fuzz testing method based on field weights is proposed.Determine the field weight by calculating the average bit flip rate of each field,and mutate the field according to the weight to generate a fuzzy test case;in order to make the CAN bus fuzz testing universal and avoid reverse analysis of the CAN bus application layer protocol,a CAN bus fuzz testing method based on improved Wasserstein generation adversarial network is proposed.The CAN bus message data set is used to train WGAN-GP to generate a message test case similar to the original message protocol structure;the above fuzz testing method is verified based on a real car.The protocol vulnerability of CAN bus are found successfully.Finally,in order to illustrate the application of fuzz testing technology in the safety test of other CAN bus protection measures,several common CAN bus attack modes are analyzed,and the security analysis of CAN bus intrusion detection algorithm based on machine learning is carried out.Two machine learning algorithms,k-nearest neighbor and Ada Boost,for intrusion detection are studied.Based on MATLAB programming,the above two machine learning based intrusion detection algorithms are implemented and evaluated by fuzz testing technology.