Unpublished working draft.
Not for distribution.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
Context-aware Anomaly Detector for Monitoring Cyber Atacks
on Automotive CAN Bus
Harsha Kumara Kalutarage
Omar Al-Kadri
h.kalutarage@rgu.ac.uk
o.alkadri@rgu.ac.uk
Cyber Security Group, School of Computing Science and
Digital Media, Robert Gordon University
Aberdeen, United Kingdom
Madeline Cheah
Garikayi Madzudzo
madeline.cheah@horiba-mira.com
garikayi.madzudzo@horiba-mira.com
HORIBA-MIRA Ltd
Nuneaton, United Kingdom
ABSTRACT
Automotive electronics is rapidly expanding. An average vehicle
contains million lines of software codes, running on 100 of elec-
tronic control units (ECUs), in supporting number of safety, driver
assistance and infotainment functions. These ECUs are networked
using a Controller Area Network (CAN). Security of the CAN bus
has not historically been a major concern, however, recent research
demonstrate that CAN has many vulnerabilities to cyber attacks.
This paper presents a contextualised anomaly detector for monitor-
ing cyber attacks on the CAN bus. Proposed algorithm is based on
message sequence modelling, using so called N-grams distributions.
It utilises only benign data (one class) for training and threshold
estimation. Performance of the algorithm was tested against two
diferent attack scenarios, RPM and gear gauge messages spoofng,
using data captured from a real vehicle. Experimental outcomes
demonstrate that proposed algorithm is capable of detecting both
attacks with %100 accuracy, using far smaller time windows (100ms)
which is essential for a practically deployable automotive cyber
security solution.
KEYWORDS
In-Vehicle Networks, CAN bus, Automotive Cyber Security, Context-
aware Anomaly Detection
ACM Reference Format:
Harsha Kumara Kalutarage, Omar Al-Kadri, Madeline Cheah, and Garikayi
Madzudzo. 2018. Context-aware Anomaly Detector for Monitoring Cyber
Attacks on Automotive CAN Bus. In Proceedings of CSCS ’19: ACM COM-
PUTER SCIENCE IN CARS Symposium (CSCS ’19). ACM, New York, NY, USA,
8 pages. https://doi.org/10.1145/1122445.1122456
1 INTRODUCTION
Modern automobiles are increasingly becoming intelligent and
smarter, ofering range of exciting new features such as telemat-
ics, advanced driver assistance and augmented reality displays. An
average vehicle contains a million lines of software codes running
Unpublished working draft. Not for distribution.
Permission to make digital or hard copies of all or part of this work for personal or
classroom use is granted without fee provided that copies are not made or distributed
for proft or commercial advantage and that copies bear this notice and the full citation
on the frst page. Copyrights for components of this work owned by others than ACM
must be honored. Abstracting with credit is permitted. To copy otherwise, or republish,
to post on servers or to redistribute to lists, requires prior specifc permission and/or a
fee. Request permissions from permissions@acm.org.
CSCS ’19, October 08, 2019, Kaiserslautern, Germany
© 2018 Association for Computing Machinery.
ACM ISBN 978-1-4503-9999-9/18/06. . . $15.00
https://doi.org/10.1145/1122445.1122456
on 100 of micro computers (known as ECUs) to facilitate these
services [28]. These ECUs spread over the entire vehicle and largely
connected to one another using bus-based network called CAN, low
latency, low overhead high performance bus standard. Moreover
modern vehicles have number of external communication inter-
faces to communicate with the outside world, for example, with
personal devices, vehicular ad-hoc networks and the Internet. Esti-
mates show that 75% of cars shipped globally by 2020 will be built
with the necessary hardware to connect to the internet [1]. Despite
the fact that security of some of these connections and software
codes may be strengthened by automotive manufactures or original
equipment manufacturers (OEMs), having so many lines of codes
and increased connectivity extends the potential attack surface that
can be exploited by a cyber criminal. Security researchers demon-
strate that their ability to implement attacks to real vehicles [30].
Vehicle hacks are potentially disastrous. Illegitimately accessing
and modifying data in a vehicle is not only a security issue but also
a safety issue. For example, corrupted ECU driving the brakes can
lead to an accident with serious consequences for passengers, peo-
ple and goods in the surrounding environment. Therefore security
of connected and autonomous vehicles is a big concern for auto-
motive manufacturers and OEMs who are now seeking methods to
secure their products against Cyberattacks.
Security research in this area has taken many forms, encom-
passing anything from hardware security to encryption of various
aspects of the vehicle (see Section 2.2). One of the larger areas of
research identifed was the need for the trafc stream of the inter-
nal vehicle to be in some way monitored for potentially malicious
behaviour. This paper focuses on contextualising anomaly detec-
tion on the intra-vehicular network bus (see Section 2.1). Anomaly
detection for security monitoring on the CAN bus has been difcult
due to the fact that many actions or reactions on a vehicle can be
construed as anomalous; for example, an emergency braking event
carried out by the driver, whilst legitimate, is always anomalous in
day-to-day driving scenarios. To mitigate or avoid false positives,
context is required to tell between a legitimate anomaly and one
that could be interpreted as a potentially malicious action.
The contribution of this paper starts by modelling of normal CAN
behaviour, then we propose a novel context-aware anomaly detector
using n-gram distributions. The main features of the proposed
algorithm can be summarised as follows.
(1) The algorithm depends only on benign data (one class) for
the training purpose and threshold estimation. This avoids
the need of large amount of realistic attack data for model
2019-09-02 19:24. Page 1 of 1ś8.